12 Á Craig Stark
Functional Role of the Human Hippocampus
Á divided into temporopolar and posterior regions), and
12.1 Overview
This chapter examines data from human amnesic patients,
data from electrophysiological recordings in humans, and
data from functional neuroimaging studies that attempt to
shed light on the question, What does the human hippocampus
do? In broad terms, we have learned a great deal about the
kinds of memory in which the hippocampus is and is not
involved. The hippocampus is part of a system that plays a
critical role in the encoding and retrieval of long-term mem-
ory for facts and events. As a whole, the system is vital for this
“declarative” or “explicit” form of memory but is not involved
in other forms of long term memory, in nonmnemonic
aspects of cognition, or in immediate (or “working”) memory.
Furthermore, its involvement in declarative memory is not
permanent but is time-limited in nature (the bases for each of
these conclusions are laid out later in the chapter).
We therefore know a great deal about the role of this sys-
tem in human memory. Our understanding of the roles
played by the individual components of the system is much
less complete. In the human, the system is often described as
consisting of two major sets of structures. One set of struc-
tures, often defined as the “hippocampal region,” consists of
the CA fields of the hippocampus itself, the dentate gyrus, and
the subiculum. For obvious reasons, most of the studies in
humans do not contain histological analyses. Unfortunately,
without histological analyses, it is quite difficult to isolate
these components from each other (see Sections 12.3.3 and
12.3.5), so most studies that report data from the “hippocam-
pus” include data from the more extended hippocampal
region. The second set of structures consist of the adjacent
cortical structures that lie along the parahippocampal gyrus.
This set consists of the entorhinal, perirhinal (occasionally
parahippocampal cortices.
As is discussed here and in Chapter 13, there have been a
number of attempts to arrive at a description of hippocampal
function. One class of theories proposes a functional distinc-
tion that qualitatively sets the hippocampal region apart from
the adjacent structures in the medial temporal lobe, defining
what it is that the hippocampal region specifically does that
other structures do not and vice versa. Such theories have
linked the function of the hippocampal region to conjunctive
(e.g., Sutherland and Rudy, 1989; O’Reilly and Rudy, 2000),
relational (e.g., Cohen et al., 1997), and/or episodic, recollec-
tive, or associative (e.g., Brown and Aggleton, 2001) forms of
memory. A second theory (the declarative memory hypothe-
sis—see Chapter 13) is more quantitative in nature in sug-
gesting that the hippocampal region “combines and extends”
the processing of the adjacent cortical structures that together
form the medial temporal lobe memory system (Squire and
Zola-Morgan, 1991).
The conclusion on this question is that the data from
human studies do not unambiguously support any of the
existing theories. In particular, the data suggest differentiation
of function within the medial temporal lobe but not along
versions of any of the binary dissociations proposed. The dif-
ferentiation of function is likely to be a complex and graded
one that is most consistent with the second theory—that the
hippocampal region combines and extends the processing of
the adjacent cortical structures. However, this theory is still
underspecified, as we do not have the data at hand to under-
stand the precise nature of the way in which the hippocampus
combines and extends adjacent processing or of the dynamic
interplay among the structures. Thus, although we have cer-
tainly learned a great deal about the human hippocampus and
its role in memory, but it would be a mistake to believe that we
549
550 The Hippocampus Book
have a full and satisfactory answer to the fundamental ques-
tion: What does the human hippocampus do?
Á in a persistent impairment of recent memory when-
12.2 Patient H.M.
It would be almost impossible to begin a discussion of the role
of the hippocampal region in human memory without con-
sidering the patient H.M. In a successful attempt to relieve
otherwise intractable epilepsy, H.M. underwent bilateral
resection of the medial portions of the temporal lobe (Fig.
12–1). Although scattered hints in the literature prior to this
suggested that damage to the hippocampal region might affect
memory, the detailed study of H.M. had an influence on both
the study of memory and on neurosurgical practice that was
unprecedented. Following a brief initial report (Scoville,
1954) describing the profound memory loss that resulted
from the surgery, Scoville and Milner (1957) presented a
detailed neuropsychological assessment of H.M. and nine
Figure 12–1. Top. Extent of H.M.’s lesion is shown by comparing
the coronal magnetic resonance imaging (MRI) scan of H.M. (left)
with that of a matched healthy volunteer (right). The hippocampus
(H), amygdala (A), collateral sulcus (CS), perirhinal cortex (PR),
entorhinal cortex (EC), and medial mammillary nucleus (MMN)
other patients with varying degrees and locations of resection.
This seminal work concluded with the following text.
Bilateral medial temporal lobe resection in man results
ever the removal is carried far enough posteriorly to
damage portions of the anterior hippocampus and
hippocampal gyrus. … In two cases in which bilateral
resection was carried to a distance of 8 cm posterior
to the temporal tips the loss was particularly severe. …
The memory loss in these cases of medial temporal
lobe excision involved both anterograde and some ret-
rograde amnesia, but left early memories and technical
skills intact. There was no deterioration in personality
or general intelligence, and no complex perceptual dis-
turbance such as is seen after a more complete bilateral
temporal lobectomy. It is concluded that the anterior
hippocampus and [para]hippocampal gyrus, either
separately or together, are critically concerned in the
retention of current experience. It is not known
are identified, and all show signs of damage in H.M. Bottom.
Summary of the extent of the lesion. Both hemispheres are dam-
aged, with one shown intact for comparison only. (Source: Corkin et
al., 1997, with permission. © 1997 Society for Neuroscience.)

whether the amygdala plays any part in this mecha-
nism, since the hippocampal complex has not been
removed alone, but always together with uncus and
amygdala. (Scoville and Milner, 1957, p. 21)
Subsequent testing of H.M. was able to reveal not only the
breadth and severity of H.M.’s memory impairment but, crit-
ically, it also showed a vast array of cognitive and mnemonic
function that was unaffected by his large bilateral lesion. The
pattern of data allowed Milner (Milner et al., 1968; Milner,
1972) to make several conclusions that expanded on those
noted above and helped lay the foundation for subsequent
investigation into the amnesic syndrome.
First, Milner noted that damage to the medial portions of
the temporal lobe results in profound inability to acquire
long-term memory for new facts or events (anterograde
amnesia) although memories from early life appear to be
intact. Conversely, there was some clear loss of memory of
information acquired for some time prior to the operation
(retrograde amnesia). Together, these observations indicated
that the medial temporal lobe might not be a permanent
repository for memory but that it plays a time-limited, albeit
critical, role in memory. Second, she noted that there was
no loss in general intellect or perceptual ability, indicating
clear dissociation between memory and other aspects of cog-
Figure 12–2. Patient H.M.’s performance on the Rey-Osterreith
figure-drawing task (top). With the original drawing in front of
him, his direct copy is accurate, demonstrating intact perceptual
abilities and a range of intact cognitive abilities. After an hour delay,
H.M. was unable to produce any drawing and did not remember
having previously seen and copied the figure. (Data provided by
Suzanne Corkin, personal communication, March 6, 2005.) In
contrast, H.M. showed learning that lasted over multiple days in
Functional Role of the Human Hippocampus 551
nition. Third, she noted that short-term memory remained
intact, indicating clear dissociation between immediate, or
“working,” memory and permanent, long-term memory.
Finally, she noted that damage to the medial portions of the
temporal lobe did not abolish all forms of long-term learning
and memory, indicating that the medial temporal lobes were
not required for at least some forms of long-term memory
(Fig. 12–2).
Overall, subsequent data from severely amnesic patients
such as H.M. and data from animal models of amnesia (see
Chapter 13) have supported Milner’s basic conclusions rather
well (although there is still no consensus as to whether each
claim is entirely or only largely true). In a more recent exten-
sion of these conclusions, Squire and Zola-Morgan (1991)
identified what they referred to as a medial temporal lobe
memory system (MTL), consisting of the hippocampal region
(defined as the CA fields of the hippocampus proper, the den-
tate gyrus, and the subiculum) and the adjacent entorhinal,
perirhinal, and parahippocampal cortices (which together
form the parahippocampal gyrus—see Chapter 3). Together,
this system is posited to be critically involved in the acquisi-
tion of new fact (“semantic”) and event (“episodic”) memory.
Notably, this system is not the permanent storage site for this
“declarative” or “explicit” memory, nor is it the locus of other
cognitive functions or other forms of memory. For example, it
the mirror-drawing task (bottom). A typical stimulus is shown on
the left. Participants view the shape in a mirror and attempt to
trace the shape while staying inside the lines (gray line shows the
author’s first attempt). H.M.’s performance improved steadily both
within testing sessions and across 3 days of testing (right), despite
having no conscious memory of having performed the task on pre-
vious days. (Source: Adapted from Milner et al., 1998, with permis-
sion. © 1998 Elsevier.)
552 The Hippocampus Book
is not involved in immediate (or “working”) memory, and it is
not involved in a wide range of “nondeclarative” (or
“implicit”) long-term memory tasks (e.g., delay conditioning,
perceptual repetition priming, habit learning). These ideas
amplify and develop Milner’s suppositions in several ways.
As a component of this interconnected system, the hip-
pocampal region participates in declarative memory in some
way. However, the exact nature or particular memory func-
tions in which it participates cannot be unambiguously deter-
mined from patients such as H.M. or other severely amnesic
patients with extensive damage to the medial temporal lobes
or from animal models with analogous lesions. Thus, knowing
that H.M. is impaired regarding tasks of recognition memory
(e.g., Milner et al., 1968) or that a similar patient (E.P.) is
entirely at chance during tests of recognition memory
(Hamann and Squire, 1997; Stark and Squire, 2000b;
Stefanacci et al., 2000) does not definitively implicate the hip-
pocampal region in recognition memory (or any other indi-
vidual MTL region, as the entire MTL is extensively
damaged). Indeed, many researchers have proposed func-
tional dissociations within the MTL such that extrahippocam-
pal structures (e.g., perirhinal cortex) support recognition
memory (or at least one form of recognition memory) and
that the hippocampal region is not involved in recognition
memory tasks (see Section 12.4.5). Of course, the study of
what is not impaired in patients with more extensive damage
(e.g., H.M. or E.P.) can be informative as well in letting us
determine what forms of memory do not require the hip-
pocampal region (see Section 12.4.1).
To move beyond an understanding of what is and what is
not impaired overall in the amnesia that follows damage to the
medial temporal lobe to an understanding of the specific con-
tributions of the components of the MTL (e.g., the hip-
pocampus), we must shift our focus. We must turn our study
to patients (and animals) with lesions to specific structures of
the medial temporal lobe and to recording or imaging tech-
niques that allow us to measure signals confidently from these
structures. Three sources of such data exist for studying the
human hippocampus. First, anoxia or ischemia can lead to
fairly selective bilateral damage to the hippocampal region.
Although this damage is incomplete and although anoxia is by
no means certain to damage only the hippocampus (see
Section 12.3.3), when it does the study of the memory impair-
ments in such patients can inform our understanding of the
role of the hippocampus itself. Second, advances in diagnostic
techniques for localization of epileptic seizure foci have led to
several cases in which depth electrodes have been placed into
the hippocampal region and recordings made during various
memory tasks. The study of such patients provides a clear
bridge between electrophysiology in animal models (e.g., rats
or nonhuman primates) and studies of the human hippocam-
pus. Finally, recent advances in noninvasive neuroimaging
techniques have led to several methods by which correlates of
neural activity can be localized with sufficient precision to dif-
ferentiate signals from the components of the MTL. In partic-
ular, positron emission tomography (PET) and functional
magnetic resonance imaging (fMRI) have become popular
tools for noninvasive monitoring of neural activity (in the
form of local blood flow or oxygenation) throughout the
brain. When data analysis techniques are applied that respect
and account for the morphological variability of MTL struc-
tures across individuals, fMRI data in particular can provide
highly valuable data for understanding normal human hip-
pocampal function (see Section 12.3.5).
What follows is first an overview of these methods, with
particular attention to their strengths and limitations for the
study of the human hippocampus. This serves as the back-
ground for the rest of the chapter, which focuses on the func-
tion of the human hippocampus. Particular attention is paid
to the role of the hippocampus in declarative memory and
how it might be functionally dissociated from the adjacent
cortical structures of the parahippocampal gyrus.
Á
12.3 Methods for Studying Human
Hippocampal Function
Each of the three sources of data discussed in the following
sections (hippocampal lesions, electrophysiology via depth
electrodes, functional neuroimaging) has its own particular
strengths and limitations in providing data to help us under-
stand the function of the human hippocampus. Moreover,
none of them can produce meaningful insight into under-
standing hippocampal function without an understanding of
the behavioral tasks used to assess memory performance. As
such, the cognitive and behavioral tasks used to assess mem-
ory in humans are discussed first. In addition, none of the
three methods alone is sufficiently free of limitations to be
able to answer the fundamental questions surrounding the
hippocampus. In Chapter 13, general theoretical concerns
from these lines of data are discussed (e.g., the interpretation
of correlational datasets—see also Henson, 2005), but several
specific concerns exist for each. These strengths and weak-
nesses are discussed below (see Sections 12.3.3–12.3.5).
12.3.1 Behavioral Tasks and Terms
Unlike assessing memory in nonverbal animals, assessing
memory performance in humans (or at least adult humans)
can be exceptionally straightforward. If we want to know
whether a participant remembers a previously seen list of
items, we can simply ask him or her to recall as many items as
possible. The simplicity of this free recall task and the fact that
it is an obvious test of human memory but not of animal
memory should not be overlooked. No training on the task is
required, as it is perhaps the archetypal everyday memory
task. Despite the prevalence of free recall, or its cousin cued
recall (in which some portion of the to-be-remembered item
is provided) in everyday life, neither task is easily captured in
animal studies of memory.
A second style of memory task, called a recognition memory
task, is more easily captured in animal models. Like tests of
recall, recognition memory tasks begin with a study phase that
 Functional Role of the Human Hippocampus 553

can either be intentional (participants are explicitly instructed
to learn the item for a later test) or incidental (no instructions
are given to learn the items and no mention is made of a later
test, often using an entirely unrelated task to serve as a dis-
guise). During the test phase, unlike recall tasks, entire items
from the study list (target or old items) and entire items
that were not on the study list (foil or new items) are pre-
sented. In the simplest version of a recognition memory task,
yes/no recognition, single items (either targets or foils) are
presented to the participant with the instructions to indi-
cate whether each item was on the study list. In a second
typical version, two-alternative forced-choice recognition, a tar-
get item and a foil item are presented, and the participant
is instructed to indicate which of the two items was on the
study list.
In making their responses to recognition memory probes,
participants are thought to use two types of memory or
sources of information: recollection and familiarity (for review
see Yonelinas, 2002). Recollection provides information about
a specific prior event or episode and gives one the sense of
being placed back in that moment in time. Certainly, if this is
the reaction one has to an item presented during a recognition
memory task, the item is endorsed as having been on the study
list. In contrast, one can still have a strong sense that the item
is familiar without having a recollective experience. Such a
feeling that the item is familiar, albeit with no specific infor-
mation about the study episode, can still lead to endorsing the
item in a recognition memory test. Thus, two sources of infor-
mation can be used when endorsing an item at time of recog-
nition. A variant on the yes/no recognition memory task,
known as the Remember/Know task (Tulving, 1985), asks par-
ticipants to split their endorsements into recollective
(“Remember”) and familiarity-based (“Know”) responses in
an attempt to isolate the two sources of information.
A key question that arises when discussing recollection and
familiarity is whether these two types of information repre-
sent the operation of two separate neural systems. If they do,
a second question that arises is whether this functional disso-
ciation maps onto a division of labor in the medial temporal
lobe: for example, that the hippocampus is particularly
involved in recollective-based memory and the structures of
the adjacent parahippocampal gyrus are particularly involved
in familiarity-based memory. This possibility is discussed in
Section 12.4.5.
Many variations on these basic tasks exist, each designed to
assess different kinds of memory. For example, for a paired-
associates task two stimuli are presented and the participant is
instructed to learn the association between the two items. For
the test, a single item can be presented and the participant
asked to recall the target member of the pair (cued recall) or
to pick out the target item from a list of choices that include
one or more foil items (forced-choice recognition). Further-
more, pairs of items can be presented at the test that are either
intact or recombined versions of the studied pairs. As such,
each trial presents only familiar components, and perform-
ance must be governed by memory for the associations
between the individual items. Finally, just as the study phase
can be conducted in an incidental way, the test phase can be
an incidental test of memory. For example, participants may
be asked to complete the word-beginning win____ with the
first word that comes to mind (stem-completion task), making
no reference to the prior study episode. Such implicit, or non-
declarative, tests of memory are forms of long-term memory
that do not appear to rely upon the MTL (see Section 12.4.1).
12.3.2 Behavioral Measures
In any of these tasks, performance can be quantified in several
ways. The most straightforward is to calculate an overall per-
cent correct—simply the percentage of correct trials out of the
total number of trials. In a yes/no recognition task with 20 tar-
get items and 20 foil items, 16 correct “yes” responses to the
targets and 16 correct “no” responses to foil items would result
in 80% correct (Table 12–1). Although quite straightforward
and intuitive, percent correct has significant difficulty in accu-
rately capturing the underlying strength of the memory when
participants exhibit a strong bias toward one response. An
uneven distribution of “yes” and “no” responses can arise quite
easily from either natural individual participant biases or task
instruction and result in an underestimation of the true

Table 12–1.
Comparison of Various Measures of Recognition Memory Performance

Memory Hits False %

capability (rate) alarms (rate) Correct Pr Br d’

Normal 16 (0.8) 4 (0.2) 80 0.6 0.5 1.7

Normal, “no” bias 10 (0.5) 1 (0.05) 72 0.45 0.1 1.6
Normal, “yes” bias 19 (0.95) 10 (0.5) 72 0.45 0.9 1.6
Moderate 14 (0.7) 6 (0.3) 70 0.4 0.5 1.0
Chance 10 (0.5) 10 (0.5) 50 0 0.5 0
Chance, “yes” bias 16 (0.8) 16 (0.8) 50 0 0.5 0

Typical behavioral measures—hit rate, false alarm rate, percent correct, “corrected rejection” (Pr), “bias”
(Br), and “discriminability” (d’)—are shown for three levels of underlying memory capability (Normal,
unimpaired memory; Moderate, mildly impaired memory; Chance, no actual memory); and under
conditions of response biases.
554 The Hippocampus Book
strength of the memory. For example, if one were to reward
participants with a small sum of money for each correct “yes”
response but administer a strong shock for each incorrect
“yes” response, participants would be generally less likely to
respond “yes” and do so only when they are quite confident
they are accurate. In our above example, this might reduce the
correct “yes” responses to 10 and increase the correct “no”
responses to 19, with a resulting percent correct of 72%. There
is good reason to believe, however, that the actual memory is
the same in these two cases, and this change in response rates
is the result of a shift in the participant’s response bias.
Several measures are commonly used to measure or
remove the detrimental effect of response biases. Response
biases can be easily seen by dividing the recognition responses
into four categories: (1) hits, defined as correct “yes” responses
to studied target items; (2) misses, defined as incorrect “no”
responses to studied target items; (3) correct rejections, defined
as correct “no” responses to unstudied foil items; and (4) false
alarms, defined as incorrect “yes” responses to unstudied foil
items. Comparing the hit rate and correct rejection rate is one
method for determining whether a bias is present. A second
method uses the “corrected recognition” score, often denoted
Pr. Pr is simply the probability of a correct “yes” response (a
“hit”) minus the probability of an incorrect “yes” response (a
“false alarm”). Pr scores are in a range between 0 and 1, with
0 indicating chance performance. By themselves, Pr scores do
not fully remove the effects of response biases; however, a
complementary measure of bias, Br, accurately estimates any
response bias that exists.
Br  false alarm rate  (1 Pr)
The most common method to correct for any bias that
exists is to calculate d′, the discriminability measure from the
signal detection theory (Green and Swets, 1966). Here, it is
assumed that there are two normal distributions of stimuli (a
distribution of targets and a distribution of foils) that differ
along a single parameter (e.g., some representation of per-
ceived memory strength or familiarity at time of retrieval). It
is further assumed that participants set a threshold or crite-
rion value for this single parameter and respond “yes” if the
stimulus presented for recognition exceeds this threshold and
“no” if the stimulus does not exceed this threshold. Thus,
there are four categories of responses corresponding to the
hits, misses, correct rejections, and false alarms defined above.
With respect to memory tasks, the amount of memory
present is viewed as the distance between the target and the
foil distributions along this axis of familiarity (e.g., Morrell et
al., 2002). That is, the effect of studying the target stimuli is to
shift the distribution away from the unstudied distribution.
The amount it has been shifted corresponds to the amount of
memory the participant has for the studied items. If the dis-
tribution has been shifted a lot, it is easy to discriminate the
target from the foil distributions, whereas if the distribution
has been shifted very little, it is more difficult to discriminate
whether a given item presented at test was drawn from the tar-
get or foil distribution. The measure d′ is the distance between
the two distributions and, as such, does not depend on where
the participant decides to set the criterion threshold. Values of
zero represent no memory, and positive values represent accu-
rate memory (negative values indicate a propensity to say
“yes” to foil items and “no” to target items and, when reliable,
indicate reliably below-chance performance). Different
thresholds result in different biases (different propensities to
respond “yes” or “no”) but merely represent different values of
this single parameter that are used to threshold both distribu-
tions. As such, d′ can index memory in a manner that is inde-
pendent of individual response biases.
Finally, one more means of quantification must be dis-
cussed because simply knowing the overall percent correct or
d′ in a memory task only tells a part of the story. For example,
if we measured recognition memory performance in a task
and obtained an average of 85% correct for a group of healthy
control participants and 75% correct for a patient with dam-
age to the hippocampus, it is still unclear whether this
patient’s performance is impaired. If the control participants’
scores ranged from 80% to 90%, the patient is likely to be truly
impaired, but if the control scores ranged from 60% to 100%,
the participant is likely to be performing at a normal level.
In addition to the ubiquitous t-tests and analyses of vari-
ance (ANOVAs) used throughout behavioral testing, analysis
by z-scores is prevalent in the memory literature and in partic-
ular in the analysis of small numbers of memory-impaired
patients covered in Section 12.4. A z-score is simply the num-
ber of standard deviations away from the mean a given obser-
vation lies. As such, it gives a way to determine how aberrant
a single observation is (e.g., a single amnesic patient’s recog-
nition memory score). Critically, this analysis relies on an
accurate estimate of both the average performance of the con-
trol population and an accurate estimate of the variance in
control performance. A z-score of 0 represents performance
equal to the mean of the control population. A z-score of 1.0
corresponds to a likelihood that the patient’s score was, in fact,
drawn from the normal population (alpha level, two-tailed) of
0.32. A z-score of 1.96 has this probability drop to the tradi-
tional threshold of significance (0.05).
12.3.3 Anoxia and Bilateral
Hippocampal Lesions
Neuropsychology—study of the relation between brain and
behavior—has a long history of studying the cognitive and
behavioral results of brain lesions and using this information
to help understand the function of specific brain regions.
Many common etiologies, and even many etiologies that pro-
duce memory impairments, do not provide data that cleanly
isolate hippocampal function. For example, although a com-
mon and valuable source of neuropsychological data, patients
who have suffered damage due to stroke are not common
sources of restricted bilateral hippocampal damage, as the vas-
culature does not provide a mechanism for selective bilateral
hippocampal damage. For example, although rupture or
blockage of the posterior communicating artery often results

Figure 12–3. Patient R.B.’s performance on the Rey-Osterreith
figure-drawing task demonstrated intact ability to copy the drawing
when it was placed in front of him but clearly impaired ability to
copy the drawing from memory 10 to 20 minutes later (healthy con-
trols typically indicate the figure with only very minor distortions at
in MTL damage (and amnesia), the damage is not restricted to
the hippocampus and is unilateral. Thus, although stroke or
aneurysm can lead to memory impairment, it does not give us
the most powerful means of assessing hippocampal function.
Similarly, although physical infarct can lead to amnesia
(e.g., the case of N.A. who suffered mammillary body damage
following an accident involving a miniature fencing foil), it is
hard to imagine how a physical infarct would selectively lesion
the hippocampus bilaterally given the physical location of the
hippocampus buried within the MTL. Alzheimer’s disease
does result in bilateral damage to the hippocampus, but it also
results in damage to the entorhinal cortex and damage outside
the MTL (particularly as it progresses). The thiamine defi-
ciency associated with chronic alcohol abuse and Korsakoff ’s
disease results in diencephalic damage and amnesia. Many
highly informative and influential studies of amnesia have
been conducted with patients suffering from Korsakoff ’s dis-
ease (e.g., the seminal work of Warrington and Weiskrantz,
1968). However, as with patient N.A., it is clear that the dam-
age suffered in such patients does not allow us to isolate the
function of the hippocampus itself.
In contrast, anoxia (reduction in oxygen) and ischemia
(reduction in blood flow) have the potential to damage the
hippocampal region selectively and bilaterally. Such cases,
even if quite rare, can prove to be highly informative. Several
Functional Role of the Human Hippocampus 555
this delay). Postmortem histological analyses revealed his anoxic
episode resulted in severe cell loss limited to the CA1 fields of the left
and right hippocampus (left shown). Location and extent of damage
are indicated by the asterisk and arrows. (Source: Zola-Morgan et al.,
1986, with permission. © 1986 by the Society for Neuroscience.)
patients have been reported to show quantifiable bilateral
damage to the hippocampal region, little or no damage out-
side the hippocampal region, and behavior consistent with
the amnesic syndrome after anoxic or ischemic episodes
(Cummings et al., 1984; Zola-Morgan et al., 1986; Rempel-
Clower et al., 1996; Spiers et al., 2001; Hopkins et al., 2004). In
one such case, that of R.B. (Zola-Morgan et al., 1986) (Fig.
12–3), postmortem histological analysis confirmed that the
damage was almost entirely limited to the CA1 field of the
hippocampus with only minor damage observed elsewhere
(and located outside the MTL).
The presence of these cases does not indicate that anoxia or
ischemia necessarily result in selective damage to the hip-
pocampus. Although histological analyses such as R.B.’s are
rarely available (and only postmortem), even relatively basic
MRI techniques have been shown to correlate with post-
mortem analyses of hippocampal damage, albeit at a coarser
resolution and potentially with less sensitivity (Rempel-
Clower et al., 1996). Using current quantitative MRI tech-
niques, a study of anoxia resulting from carbon monoxide
poisoning and obstructive sleep apnea (Gale and Hopkins,
2004) reported that only 30% to 36% of the patients were
observed to have hippocampal damage. Cortical atrophy (in
the form of the ventricle/brain ratio) was present in 35% of
the patients with damage resulting from carbon monoxide
556 The Hippocampus Book
poisoning. Furthermore, in a review of 43 patients with dam-
age following anoxia (Caine and Watson, 2000), 32 showed
evidence of cortical damage and 31 showed evidence of dam-
age to the pallidum or striatum. The hippocampal region was
only the third most common site of damage (30 cases).
Damage was also reported commonly in the basal ganglia and
the thalamus. Most strikingly, in only 18% of these anoxic
cases (8/43) was the damage present in and limited to the hip-
pocampal region.
Thus, it should be quite clear that the study of patients with
damage resulting from anoxia or ischemia does not necessar-
ily imply the study of selective hippocampal damage or even
hippocampal damage at all. To use anoxic patients as a route
to studying the consequences of hippocampal damage,
detailed structural neuroimaging assessment of the damage
(whenever possible) and detailed neuropsychological assess-
ment of their behavior is certainly required. Improved tech-
niques for quantifying damage and for addressing the
possibility of “covert” damage not resolved by current tech-
niques are certainly warranted as well. In particular, detailed
comparison of postmortem histological analyses and premor-
bid neuroimaging analyses are required to improve and fur-
ther validate the neuroimaging analyses. However, even if
current neuroimaging techniques do not provide a definitive
assessment of the damage, they are infinitely better than
assessing damage merely by etiology, as the above review has
shown. Unfortunately, although assessments to the best of
current standards have been done for some of studies of
memory, they have certainly not been done for all.
Figure 12–4. Electrophysiological data in the human. Left panel.
Trajectory of a depth electrode with microwires placed in the hip-
pocampus (arrow indicates where wires protrude from electrode).
Both peri-event spike-train histograms (middle) and field potentials
(right) can be collected from such electrodes. Middle panel. A hip-
pocampal neuron recorded from the electrode indicated on the left
demonstrates an increase in activity at retrieval upon presentation of
12.3.4 Depth Electrode Recordings
Implanting electrodes and recording from neurons in the hip-
pocampal region and adjacent cortical structures has been a
common technique for studying rodents and nonhuman pri-
mates (see Chapter 13), but electrophysiology has not been
used extensively to study the human hippocampal region for
obvious reasons. There have been a number of cases, however,
in which depth electrodes have been placed in the human
MTL to help identify the topography of epileptic seizures
when other techniques for neurosurgical planning have not
proved sufficient. In a limited number of these cases, single-
unit recordings from the depth electrodes have been made
during various explicit memory tasks (Heit et al., 1988, 1990;
Fried et al., 1997, 2002; Fernandez et al., 1999b, 2002; Kreiman
et al., 2000; Cameron et al., 2001; Paller and McCarthy, 2002;
Fell et al., 2003) (Fig. 12–4).
Such data are certainly highly important, but they are also
scarce. Unfortunately, the scarcity of the data extends not only
to the small number of studies but to the small number of
recordings per participant in these studies. In electrophysio-
logical studies of the monkey hippocampus, one can expect to
see data collected from 50 to 100 hippocampal neurons in
each monkey (e.g., Wirth et al., 2003). In depth electrode
recordings on humans, one may see data from this same num-
ber of neurons (e.g., Fried et al., 2002), but they are spread
across a large number of participants, yielding single-digit
numbers per participant. This is a limitation of the methodol-
ogy, not of the experimenters, as multiple electrode drops and
the cue item in a paired-associate task. (Source: Left and middle
panels: Cameron et al., 2001, with permission. © 2001 Elsevier. Right
panel: Paller and McCarthy, 2002, with permission. © 2002 Wiley-
Liss, a subsidiary of John Wiley & Sons.) Right panel. Field poten-
tials from electrodes in the posterior portion of the hippocampus
are shown for three patients demonstrating differences in activity
for the sample and match phases of a delayed match-to-sample task.

continual adjustment of the electrode location or even initial
placement of the electrode to isolate as many hippocampal
neurons as possible is not an option. The electrode placement
must be motivated by the neurosurgical planning of which
these data are a by-product.
Furthermore, we must remember that the recordings are
coming from brains that are decidedly not normal. The
recordings are being made from patients with epilepsy severe
enough to warrant neurosurgical intervention that is most
likely going to target the medial temporal lobes. Whether and
how this affects the data cannot be known at present. To
address this concern, data are often analyzed only from sites in
the unoperated hemisphere or otherwise distant from the
epileptic foci.
12.3.5 Neuroimaging
The initial set of neuroimaging studies exploring memory and
the medial temporal lobe using either PET or fMRI presented
a mixed set of results. There were a number of initial PET
studies with positive reports of MTL activity during memory
tasks (e.g., Squire et al., 1992; Grasby et al., 1993; Nyberg et al.,
1995; Schacter et al., 1995). However, at the time there were
also a sizable number of studies that observed memory-
related activity outside the MTL but no memory-related activ-
ity in the MTL itself (e.g., Kapur et al., 1994; Shallice et al.,
1994; Tulving et al., 1994; Buckner et al., 1995). For example,
Shallice et al. (1994) contrasted activity during word-pair
encoding and cued recall with activity during a low-level base-
line control task (e.g., hearing the pair “one thousand, two
thousand” repeatedly). Although the contrasts revealed
numerous regions of brain activity (e.g., prefrontal cortex), no
MTL activity was present. As both contrast activity during
what are clearly mnemonic tasks that are impaired in amnesic
patients with activity during tasks that have no mnemonic
demands and that are not impaired in amnesic patients, the
negative results were unsettling. Many hypotheses for the
unreliability of findings were put forth, such as the idea that
the MTL might always be active (thus reducing the contrast
between states) or that MTL signals may be too weak to
resolve (resulting from either poor imaging signals from the
MTL or inherently weak activity). It was even suggested that
the MTL might not be involved in long-term memory after all.
Developments in neuroimaging techniques and further
studies have now led to many positive results, supporting the
former two hypotheses (and several new ones as well) but not
the latter. In fact, as of this writing, more than 1300 articles
indexed by PubMed during the last 5 years contained the key
words “fMRI” and “hippocampus” or “medial temporal lobe.”
The numbers would certainly be significantly higher if other
neuroimaging techniques were included. The popularity of
neuroimaging attests to the fact that it can provide useful data
to further our understanding of the neural mechanisms that
underlie memory.
However, although useful and compelling data can be
obtained using neuroimaging techniques, these techniques
Functional Role of the Human Hippocampus 557
have constraints and challenges that place limits on what one
can conclude from their results. If the challenges are well met
and experiments designed and interpreted with the con-
straints clearly kept in mind, neuroimaging studies can be
highly informative (Henson, 2005). If they are not, they can
present a confusing or conflicting state of affairs (much as was
present at the time of the early studies).
The first constraint is that, on the whole, neuroimaging
and electrical recording techniques provide correlational data
and cannot provide evidence as to the necessity (or even
actual use of) a structure in a particular function. For exam-
ple, although single-unit recording shows strong hippocampal
activity during delay eyeblink conditioning in the rabbit (e.g.,
Berger et al., 1980) and activity has similarly been observed in
humans using PET (e.g., Blaxton et al., 1996), complete bilat-
eral lesions of the hippocampus do not impair acquisition or
expression of the response (e.g., Mauk and Thompson, 1987;
Clark and Squire, 1998). Therefore, the activity observed in a
region may be incidental to the task at hand or may be the
result of processing in another region that has projections into
the observed area. Whereas the use of parametric designs or
other clever experimental manipulations that attempt to link
imaging data to specific components of behavior (e.g., the
recent attempts to link MTL activity to aspects of implicit
tasks reported by Rose et al., 2002 and Schendan et al., 2003)
can provide stronger evidence for a causal link than the use of
simpler designs, neuroimaging data cannot be resolute in this
regard.
A second problem faced by neuroimaging techniques is the
lack of a baseline. Although particularly problematic for
BOLD (blood oxygenation level-dependent) fMRI (Fig. 12–5),
which has a signal with an arbitrary offset and arbitrary units
of measurement, the lack of a clearly defined level of activity
associated with a region not being involved in a task is
endemic to all neuroimaging techniques. (How many spikes
per second constitute “no activity” in the neuron, and what is
an animal actually doing when this is assessed?) Neuroimaging
techniques are contrastive in nature. Our data take the form of
a higher BOLD fMRI signal, greater regional cerebral blood
flow (rCBF), more spikes per second, a steeper excitatory post-
synaptic potential (EPSP) slope, or a sharper tuning function
during task A than during task B. These numbers can often be
quantified and varied parametrically, but it is frequently diffi-
cult to interpret a result “zero activity.” This contrasts with
behavioral data. For example, in two alternative forced choice
recognition, scores necessarily vary between chance and 100%;
and for free recall, performance necessarily varies between 0%
and 100%. Where there may be floor and ceiling effects to con-
sider, one can, in principle, identify both perfect memory and
absent memory.
As noted, this lack of a standard of comparison may be
especially problematic for BOLD fMRI with its arbitrary units
and its complete lack of an estimate of what measurement
“zero activity” in a region should be. For example, Stark and
Squire (2001b) have shown that when randomly interspersed
3-second periods of rest were used as a baseline to assess “zero
558 The Hippocampus Book
Figure 12–5. Left. A current model of the chain of events that leads
to the blood oxygenation level-dependent (BOLD) effect measured
on functional MRI (fMRI). Underlying neural activity results in
local increases in the cerebral metabolic rate of oxygen extraction
(CMRO2), cerebral blood flow (CBF), and cerebral blood volume
(CBF). Local increases in CBF affect the other two, and all combine
to change the ratio of oxygenated relative to deoxygenated hemo-
globin in a local area that is measured on fMRI. This signal is
called the BOLD effect. Right. Typical BOLD effects. Here, visual
activity was recorded in response to either a single 1-second visual
stimulus (flickering checkerboard) or to two 1-second trials, spaced
5 seconds apart. In both cases, the percent change from a baseline
activity,” viewing novel or familiar pictures failed to elicit any
apparent activity in the hippocampal region. However, when
an active but menial task was used as a baseline (deciding
whether digits were odd or even), robust activity was
observed.
Similarly, Law et al. (2005) collected fMRI images as par-
ticipants learned a concurrent set of arbitrarily paired associ-
ates gradually over multiple trials (Fig. 12–6). Each trial
contained both encoding and cued-recall components as par-
ticipants learned through trial and error which abstract geo-
metric shapes were associated with which response options.
Notably, a number of MTL regions exhibited activity that
increased in conjunction with the strength of the participant’s
memory for a particular region. Equally notable, however, was
of signals during no visual stimulation is plotted. Note how the
BOLD effect is a temporally low-pass filtered response of the
underlying neural activity. One second of neural activity was
recorded as a protracted response that peaked approximately
6 seconds after onset and did not return to baseline until approxi-
mately 12 seconds after offset of the stimulus. Note also, though,
that the response to two trials is a roughly linear summation of the
response to two individual trials. (Sources: Left: Buxton, 2001,
p. 419, with permission of Cambridge University Press. Right: Dale
and Buckner, 1997, with permission. © 2002, Wiley-Liss, a sub-
sidiary of John Wiley & Sons.)
the observation that the choice of baseline task determined
whether activity during the mnemonic trials was above or
below “zero.” When a trivially easy nonmnemonic perceptual
task was used, activity increased with memory strength but
was all “negative.” When a more difficult version of the same
task was used as the baseline, activity again increased with
memory strength but was now all “positive.” Contrasting the
two baseline tasks revealed substantially greater activity in the
MTL for the easy task than for the difficult task, presumably
the result of participants’ minds wandering during the triv-
ially easy trials. Such mind wandering is likely to include inci-
dental encoding and retrieval of information, the hallmark of
MTL function. This result draws into sharp focus the diffi-
culty of not having an estimate of zero activity in a region.

Figure 12–6. Activity in a region of the left hippocampus during a
paired associate task is shown for both strong, highly accurate
memories and for memories that are above chance levels of per-
formance but still only moderately accurate. Two perceptual tasks
were used to estimate “zero” activity, neither of which was overtly
mnemonic. In both, the task was to identify the brightest square.
The task was trivial in the Easy condition (98% correct) and chal-
lenging in the Difficult condition (54% correct). When the Easy
condition was used as an estimate of “zero,” strong memories
yielded small “negative” activity, and moderate strength memories
yielded large “negative” activity. When the Difficult condition
was used as an estimate of zero, the same exact memory task
trials yielded large and small “positive” activities, respectively.
fMRI (and most other imaging techniques) yield purely rela-
tive measures, and no claims can be made about a region’s
absolute level of activity in any task. The only measures avail-
able are relative measures between tasks or conditions.
Furthermore, simply because a task does not require a cogni-
tive component (memory, in this case) does not mean the
component is not actually being used and that the region is
not active during the task.
Finally, if one considers BOLD fMRI, the most popular
neuroimaging technique used in humans, we still have limited
temporal (~ 1 second) and spatial (~ 3 mm3 or ~10,000 neu-
rons) resolution; and we still do not have a complete under-
standing of the relation between neural events and the BOLD
fMRI signal (or the measured by PET). Progress along these
dimensions is being made, with some evidence from experi-
ments that BOLD is more closely linked to synaptic activity
than to spiking activity (Logothetis et al., 2001). One conse-
quence of this is that inhibitory inputs may paradoxically
serve to increase the BOLD effect rather than decrease it.
Strong inhibition in a region results in substantial synaptic
activity (and metabolic activity), even if spiking is reduced.
Direct tests of this hypothesis in the rat have yielded increases
in the CBF, a precursor of the BOLD effect (Caesar et al.,
2003), indicating that inhibition would result in an increase
rather than a decrease in BOLD. Thus, our current under-
standing of how to relate electrophysiological findings to neu-
roimaging findings is certainly incomplete, although it does
appear that BOLD fMRI measures offer a relatively linear
assessment of neural activity (Boynton et al., 1996; Rees et al.,
2000; Logothetis et al., 2001).
Functional Role of the Human Hippocampus 559
These data demonstrate the effect of activity during the baseline
task. Substantial activity in the hippocampus during the Easy
baseline task (presumably resulting from incidental encoding
and retrieval unrelated to this rather boring task) served to deflect
activity during the memory task “below zero.” Note, however,
that the relative differences between the curves are maintained,
irrespective of baseline choice. fMRI has no baseline, and all data
must be interpreted in a relative manner. All we actually know
from these data is that Difficult baseline Strong memories
Moderate memories Easy baseline. This property is often ignored
yet is a clear source of several failures to observe activity. (Source:
Data are from Law et al., 2005, with permission. © Society for
Neuroscience.)
12.3.6 Technical Challenge: Alignment
of MTL Regions Across Participants
Even if the above challenges are met and the constraints
respected, one significant challenge remains if neuroimaging
techniques are to be able to help answer questions concerning
the functional role played by the various structures in the
MTL. For neuroimaging to do so, it must be possible to local-
ize signals of the specific subregions of the MTL. Therefore,
the data must be of sufficient resolution to allow confidence in
localization; and if group analyses are desired, it must be pos-
sible to transform the data from multiple participants in such
a way that cross-participant tests respect anatomical divisions
in the MTL. Techniques such as PET and magnetoencepha-
lography (MEG) have clear strengths. PET can directly quan-
tify CBF and can be used to tag specific chemicals (e.g.,
neurotransmitters) in ways no other technique can; and MEG
has millisecond temporal resolution. The spatial resolution
and localization accuracy of both techniques have improved
over recent years, but they still cannot approach the resolution
possible with fMRI. The resolution in typical fMRI studies is
typically 3 to 4 mm3 but can be pushed into the cubic mil-
limeter range (Hyde et al., 2001; Kirwan et al., in press), mak-
ing fMRI a leading candidate for imaging the small structures
of the MTL.
However, even if the spatial resolution is theoretically suf-
ficient to have some confidence in localization, it is still vital
that any cross-participant analyses respect the structural
boundaries in the MTL. It would be pointless to attempt to
discern what factors affect activity in the perirhinal cortex if
560 The Hippocampus Book
Figure 12–7. Coronal structural MRI sections through the hip-
pocampus of 20 participants, averaged following Talairach align-
ment (a) and region of interest alignment (ROI-AL) (b). White
arrows indicate the location of the collateral sulcus used to identify
the parahippocampal gyrus. Overlay on the left hippocampus indi-
cates the amount of cross-participant overlap of manual segmenta-
tions of each participant’s left hippocampus when brains were
aligned using each technique. The left hippocampus of each partici-
our alignment techniques normalized brains in such a way
that a given voxel in a group analysis was located in the
perirhinal cortex of one participant, the entorhinal cortex of
another, the hippocampal region of a third, and outside, in the
ambient cistern, of a fourth. Unfortunately, human brains are
sufficiently dissimilar from each other that alignment with
popular techniques (e.g., alignment to the atlas of Talairach
and Tournoux, 1988) often leaves us in this situation.
An example of this problem is shown in Figure 12–7a.
Here, 20 structural MRI scans were first individually aligned to
the Talairach atlas before an average of the 20 scans was cre-
ated (shown as a coronal image cropped around the MTL).
The white arrows indicate where the collateral sulcus (a defin-
ing structure for the parahippocampal gyrus) is most likely to
be. As one can see, it is far from clearly defined, indicating that
the averaging across participants blurred this feature (which is
plainly visible in each individual scan) into an undifferentiated
mass. This poor level of cross-participant alignment arises
from both global variability across participants (overall shifts
in the location, orientation, and size of structures) and from
differences in the shape of structures (Preuessner et al., 2002).
In addition to averaging the structural MRI images, Figure
12–7a shows the result of averaging segmentations of the left
hippocampal region. Each participant’s left hippocampal
region was manually segmented, and the same Talairach
transformation was applied to each segmentation prior to
averaging the segmentations across participants. The result of
this averaging is a grayscale overlay that indicates how well the
Talairach transformation was able to align all 20 hippocampi.
In this slice, there are no voxels in which there was overlap
across all 20 manual segmentations of the left hippocampal
region, and there are only three voxels in which 19 of the 20
segmentations overlap. Even if our only goal is to assess activ-
ity in the hippocampal region overall (not in subregions of the
hippocampus), this level of alignment is insufficient as signals
pant was initially manually segmented and then transformed along
with the structural image using both techniques. In this group over-
lay, white is ideal (no voxels in a and 23 voxels in b) and indicates
that all 20 participants’ aligned segmentations identified the voxel as
part of the left hippocampus. Black indicates 1 to 10 aligned seg-
mentations identified the voxel as part of the left hippocampus.
Light, medium, and dark gray indicate that 19, 18, and 16 segmen-
tations overlap, respectively.
from one hippocampal region are combined with another
participant’s entorhinal cortex and a third participant’s ven-
tricle. If one extends this analysis to segmentations of all sub-
regions of the MTL, the picture is worse still. Between any two
participant’s MTLs, only about half of the voxels are typically
identified as belonging to the same structure in both partici-
pants (Stark and Okado, 2003; Kirwan et al., in press).
Imperfect cross-participant alignment results in a blur of
the data that has two detrimental effects. First, the blur
reduces the localization accuracy for any observed activity, as
signals from multiple regions may be combined. Second, the
blur reduces statistical power. If separate regions (or subre-
gions) are behaving differently, the spatial blur smears activity
across regions, introducing noise that prevents a consistent
pattern of activity from being observed. For example, the
main result of a study by Stark and Okado (2003) demon-
strating activity associated with encoding during a retrieval
task was not observed when the data were aligned with tradi-
tional Talairach techniques but was observed when the data
were aligned using more sophisticated techniques.
Recently, there have been three approaches taken to address
this issue: simple anatomical region of interest (ROI) analyses,
cortical unfolding applied to the MTL, and ROI-AL (region of
interest alignment). All three approaches can address the issue
of improving cross-participant alignment and, in so doing,
open up the possibility of using fMRI to help differentiate the
role of individual structures in the MTL. Figure 12–7b shows
the result of aligning the medial temporal lobe structures
from the same 20 participants from Figure 12–7a but using
one of these approaches (the ROI-AL method of Stark and
Okado, 2003). Here, one can clearly resolve the collateral sul-
cus (white arrow) and differentiate it from the hippocampal
region immediately above. The 20 segmentations aligned with
this technique have perfect overlap in 23 voxels; and if one
extends this to near-perfect (19/20) levels, the count rises to 34

voxels (versus 0 and 3 in the case of Talairach alignment).
These three approaches give us hope that fMRI may be able to
isolate hippocampal function from that of adjacent regions
and perhaps even differentiate contributions of hippocampal
subfields.
With the most straightforward technique, several articles
have collapsed activity across all voxels within a set of anatom-
ically defined ROIs (e.g., Stark and Squire, 2000a; Small et al.,
2001; Reber et al., 2002, 2003). Thus, there may be a single
measure that represents activity for each participant’s anterior
left hippocampal region in a given condition. Although this
technique has the potential for perfect alignment across par-
ticipants, it suffers from two drawbacks. First, by combining
all voxels in an anatomically defined region into a single meas-
ure (a single, large, irregularly shaped voxel), any functional
variability in that region is lost. For example, if all data from
the right hippocampal region were treated as if it were one
large voxel, and if two opposing patterns of activity were pres-
ent in different subregions of the right hippocampal region,
this functional variability would be lost. A second, related
drawback is that if only a small subregion in the anatomically
defined ROI is active, noise from other included voxels distort
the observed activity.
A second approach has been to adapt cortical unfolding
techniques to the problem of unfolding the cortical MTL
structures and the spiral structure of the hippocampal region
(Zeineh et al., 2000, 2003). With this approach, anatomically
localized boundaries are defined and used to map the three-
dimensional data onto a common two-dimensional “flat map”
and to align individual participant’s flat maps. Unlike collaps-
ing all voxels in the anatomically defined ROIs, this technique
has the advantage of preserving the topography in each
region. In addition, because of the requirement for very high
resolution of the functional data (1.6  1.6  4 mm) and the
technique’s ability potentially to unfold the spiral structure of
the hippocampal region, this technique holds the promise of
differentiating signal from regions in the hippocampus itself.
However, the unwarping process is not entirely invertible, as
functional voxels that lie within a fold can be associated with
two different regions of the unfolded map.
A third approach has been to use anatomically defined
ROIs to guide alignment directly (Stark and Squire, 2001a;
Stark and Okado, 2003; Miller et al., 2005; Kirwan et al., in
press). This technique (dubbed “ROI-AL” by Stark and Okado,
2003) shares with the unfolding technique the advantage of
preserving topography in regions with the unfolding tech-
nique but does not require very high-resolution functional
voxels. ROI-AL takes a direct approach to aligning regions
across participants. Instead of using structural MRI to align
gray matter, white matter, and cerebrospinal fluid (CSF)
across participants, ROI-AL attempts to align anatomically
defined regions of interest based on rough segmentations of
the regions. Furthermore, instead of attempting to arrive at
the best fitting alignment across the entire brain, ROI-AL
focuses only on alignment of a particular structure (or set of
structures). Thus, all of the transformation parameters used
Functional Role of the Human Hippocampus 561
to align two brains with typical techniques (e.g., rotation,
translation, scaling, shearing) are focused on aligning a single
region (e.g., left hippocampal region). The net result is vastly
improved cross-participant alignment, bringing with it
improved statistical power. Furthermore, by using anatomi-
cally defined regions, ROI-AL (and unfolding) can localize
results from cross-participant analyses to specific anatomi-
cally defined regions of interest (e.g., perirhinal versus
parahippocampal cortices). The exact location of any region
of activity in a group analysis can be compared with the seg-
mentations from each participant (or with a composite
anatomical model based on the individual participants’
anatomically defined regions of interest). Thus, one can proj-
ect backward from the group result to the individual partici-
pants’ anatomy and determine with some precision where a
signal was generated.
Á
12.4 Dissociating Hippocampal Function
In the following sections, five aspects or potential divisions
in long-term memory are discussed with particular atten-
tion paid to isolating the function of the hippocampus. For
each of these aspects, the relevant data from patients with
bilateral damage limited to the hippocampal region, from
depth electrode recordings, and from neuroimaging studies
are discussed.
12.4.1 Explicit Versus Implicit
Perhaps the clearest example of functional dissociation in
long-term memory is that between explicit, or declarative,
memory and implicit, or nondeclarative, memory. Explicit
memory refers to “intentional or conscious recollection of
prior experiences, as assessed in the laboratory by traditional
tests of recall or recognition,” whereas implicit memory refers
to “changes in performance or behavior, produced by prior
experiences, on tests that do not require any intentional or
conscious recollection of those experiences” (Schacter, 1999,
p. 233). These descriptive terms relate to specific task demands
rather than distinctions between memory systems or spe-
cific brain structures. Yet, a wealth of data shown has shown
that this descriptive distinction is strongly correlated with
hippocampal function. The terms declarative and nondeclar-
ative are defined in similar terms when applied to the human.
The declarative/nondeclarative distinction goes a bit further,
however, to embrace the apparent functional dissociation
between multiple memory systems. Declarative memory
“identifies a biologically real category of memory abilities”
(Squire, 1992, p. 232) that require structures in the medial
temporal lobes, whereas nondeclarative memory identifies a
heterogeneous collection of memory systems that bear resem-
blance to each other behaviorally (all are observed with
implicit memory tasks) but appear to rely on many brain
structures.
562 The Hippocampus Book
Despite obvious memory impairments, it may come as a
surprise to some that patients with damage limited to the hip-
pocampal region and those with extensive damage to the
medial temporal lobes demonstrate normal levels of perform-
ance on a wide range of long-term memory tasks. In an early
report (Milner, 1962), H.M. was found to acquire a perceptual
motor skill (learning to trace the outline of a shape when
viewed in a mirror) over several days at normal rates despite
not remembering the prior day’s training (see Fig. 12–2).
Subsequent studies have shown that even severely amnesic
patients exhibit normal rates of delay eyeblink conditioning
(e.g., Weiskrantz and Warrington, 1979; Clark and Squire,
1998). Furthermore, the phenomenon of categorization
appears to be intact. In one task, a random dot pattern
(resembling an imaginary constellation of stars) is created,
and numerous distorted versions of this “prototype” are cre-
ated as well by moving the dots by random amounts (Posner
and Keele, 1968). After studying only highly distorted ver-
sions, even severely amnesic patients such as E.P. incidentally
learn to abstract features of the prototype (or the category) so
they can later correctly classify new patterns as either mem-
bers or nonmembers of the studied category (e.g., Knowlton
and Squire, 1993; Squire and Knowlton, 1995). They do so,
however, without any knowledge that they have ever per-
formed the task and without any ability consciously to recog-
nize any of the previously studied dot patterns (even if the
study phase consisted of seeing the same dot pattern 40
times). Thus, there appears to be a dissociation between the
ability to learn in these implicit tasks gradually and the ability
to remember the study episodes or contents of those episodes
consciously or explicitly.
The term “priming” refers to another class of implicit
memory tasks that are not affected by MTL damage. In per-
ceptual priming tasks, exposure to a word or a picture (often
incidentally, with no instruction to study the item) improves
the ability to perceive the item later, usually taking the form of
increased accuracy or decreased reaction time during
degraded presentation. There is a long history in cognitive
psychology dissociating perceptual priming and recognition
memory in healthy individuals (for review see Schacter, 1994).
This behavioral dissociation is relevant here in that long-term
memory for an item in the form of a repetition priming effect
is normal following hippocampal damage, whereas explicit
memory for the same item is clearly impaired.
Warrington and Weiskrantz (1968, 1974) were the first to
show that despite poor recognition memory after studying
lists of words or pictures, amnesic patients had an intact form
of perceptual memory for these items. When tested with a
fragmented or degraded version of either type of stimulus and
asked to complete this partial cue, studied items were com-
pleted more accurately or at greater levels of degradation than
nonstudied items. This perceptual priming effect was normal
in the amnesic patients. Likewise, if words are presented rap-
idly at a test so correct identification is below ceiling levels,
previously studied words are identified more accurately than
nonstudied words (e.g., Jacoby and Dallas, 1981). This per-
ceptual identification priming is normal even in the case of
E.P., who performs at chance when given a recognition mem-
ory test under similar circumstances (Hamann and Squire,
1997).
In perhaps the most extreme example of this dissociation
between intact priming and impaired recognition (Stark and
Squire, 2000b), E.P. studied a list of words (e.g., “window”)
and after a 10-minute delay was presented with a test of repe-
tition priming (“What is the first word that comes to mind
that begins win____”) and a test of recognition memory
(“Which word did you see on the list 10 minutes ago: window
or winter?”) on each trial. E.P. showed a normal priming effect
in the form of a 26% increased likelihood of generating the
studied word (relative to baseline completion rates of generat-
ing this word). However, his recognition memory perform-
ance was at chance, averaging 48% correct. Therefore, within
seconds of each other, E.P. showed intact implicit memory for
a word (in the form of stem completion priming) and no
detectible explicit memory for that same word. Given E.P.’s
complete hippocampal loss (Stefanacci et al., 2000), it is clear
that the hippocampus cannot be vital for this form of implicit
memory.
We should note that when comparing an amnesic patient’s
level of performance on repetition priming tasks (or indeed
on many implicit memory tasks) relative to healthy controls,
there is always the possibility that the amnesic patient’s per-
formance may be impaired relative to that of the healthy con-
trols because the latter group may make use of covert explicit
memory. Although not the case in the above-mentioned stud-
ies overall, the repetition priming task is susceptible to
“explicit contamination”—the use of explicit memory on an
implicit memory task. For example, if at this point in the
chapter the reader were asked to complete the word stem
win_____ with the first word that comes to mind, you might
follow these instructions faithfully and respond with the first
word that simply pops into your mind. However, you might
also realize that in the preceding paragraph the word window
was used as a sample study item (especially if this were done
in the context of not one implicit probe but an entire list of
them). Therefore, you might respond with “window” not
because it was the first word that freely popped into your
mind, but because you tried to remember what study word
began with the letters win_____. Thus, despite the task
instructions, participants may treat an implicit task as a thinly
disguised explicit task and show enhanced levels of per-
formance. For example, in the above-mentioned study with
E.P. that combined a repetition priming task with a forced-
choice recognition memory task during each trial (Stark
and Squire, 2000b, experiment 4), all but one of the controls
had priming effects ranging from 0.21 to 0.29. The other con-
trol had a priming effect of 0.53, approximately 8 standard
deviations (SD) away from the mean priming effect. Although
we do not know for certain that this particular control
adopted an explicit strategy, it almost perfectly matches
the mean hit rate in the recognition task, leaving this as
the most reasonable hypothesis. Therefore, simply knowing
 Functional Role of the Human Hippocampus 563

Figure 12–8. Visual-paired comparison task. Participants view two copies of the same
image and after a variable delay are shown a new image and the old image. A bias exists to
spend more time looking at the new image. This is an implicit memory task, yet the bias
to look at the novel picture is dependent on the medial temporal lobes. (Source: Data are
from Manns et al., 2000.)

that the task requires no more than implicit memory does
not guarantee that the participants will treat it as an implicit
task when there are multiple routes to solve the prob-
lem. Where implicit tasks or any other tasks are impaired
in persons with amnesia, this possibility must certainly be
considered.
A second form of priming, known as conceptual priming,
is also apparently intact in amnesia. Here, the test presents an
item or a category that is semantically or conceptually related
to the studied item. For example, if the word peach had been
presented at study (again, the study task is often incidental),
participants might be asked to generate exemplars of a cate-
gory (e.g., generate examples of fruits), verify category mem-
bership (e.g., how long it takes to verify that peach is a fruit),
or perform a free-association task (e.g., free-associate to the
word pear). In each of these, amnesic patients with damage
including but not limited to the hippocampus have shown
normal levels of conceptual priming (Graf et al., 1984;
Shimamura and Squire, 1984; Vaidya et al., 1995; Keane et al.,
1997; Levy et al., 2004) while being impaired in explicit tests
for the same type of information.
A task called visual-paired comparison appears to be the
one exception to the rule that implicit memory tasks are non-
declarative in nature and are not impaired by damage to the
hippocampal region. In this task, two copies of an object or a
scene are presented to the participant for several seconds (Fig.
12–8). During this time, the participant is free to examine
either copy of the object at will, and no task instructions are
given. After some delay (in which intervening items may be
presented), a copy of the previously exposed object or scene is
presented along with a novel object or scene; and again the
participant is free to examine either stimulus at will. The task,
is quite implicit in instruction and behavior. Participants have
a tendency to spend a greater amount of time looking at the
novel stimulus than the familiar stimulus. A small bias or
change is made in behavior as a result of experience on a task
that makes no reference to the prior study episode—hall-
marks of implicit memory tasks. In fact, the task is commonly
used to assess memory in infants who would certainly not
understand any explicit instructions even if given (Fagan,
1970). Yet, the task is dependent on the hippocampus in that
no such bias is seen following hippocampal damage in
humans (McKee and Squire, 1993; Manns et al., 2000; Pascalis
et al., 2004), the monkey (Bachevalier et al., 1993; Pascalis and
Bachevalier, 1999; Zola et al., 2000), or the rat (Clark et al.,
2000). Furthermore, the degree of bias shown in the visual
paired comparison task is predictive of subsequent recogni-
tion memory for the items shown, whereas the amount of
repetition priming exhibited is not (Manns et al., 2000). Thus,
even though the task requirements are implicit in nature (no
reference is made to the study episode) and even though the
memory is observed in the form of a change in a behavioral
564 The Hippocampus Book
bias, the task relies on the same underlying mechanisms that
are responsible for declarative memory.
One final task deserves consideration when discussing the
role of the hippocampus in relation to explicit and implicit
memory. Chun and Phelps (1999) embedded an implicit
memory task within a visual search task (locate a rotated T
among numerous rotated L distractors). Although the displays
appeared to be random, a set of 12 displays were repeatedly
intermixed with random displays throughout the experiment.
As one might expect, the reaction time to locate the target
item in the random displays gradually decreased over the
course of the task. This basic practice effect or increased skill
in performing the task was similar in a control group and an
amnesic group (two anoxic patients and two patients with
encephalitic damage that included, but extended beyond, the
hippocampal region), consistent with other implicit or non-
declarative tasks. Furthermore, although approximately half
of the healthy controls reported noticing repetition of several
displays when later asked, not even they could identify which
they were (54% correct). Critically, however, the controls’
reaction times to the repeated items were faster than their
reaction times to the random items in the later epochs of
training. Although they lacked explicit knowledge of the
repeated items, they had implicit knowledge of this repetition,
evidenced by their reaction time. In marked contrast, the
amnesic patients did not show this effect. The amnesic
patients exhibited identical reductions in reaction time for
repeated and random displays.
Thus, a second task exists in which the task demands and
memory are implicit in nature, yet performance requires
structures in the medial temporal lobes. This said, perform-
ance may not require the hippocampus itself. A follow-up
study by Manns and Squire (2001) solidly replicated the basic
findings of Chun and Phelps (1999). Here, however, the
amnesic patient group was large enough to separate into two
groups based on size and extent of the lesion. Five patients had
damage limited to the hippocampal region or only mildly
extending into the parahippocampal gyrus. A separate group
of three were encephalitic patients who had extensive, near-
complete damage to the MTL and mild damage that extended
into the lateral temporal cortex (consistent with this etiology).
Notably, the only group that did not show reduced reaction
times specific to repeated displays was the group with exten-
sive MTL damage. These patients performed similarly to the
mixed etiology group in Chun and Phelps’ (1999) study. In
contrast, the patients with damage more restricted to the hip-
pocampal region performed much like the healthy control
volunteers in both studies, demonstrating improved reaction
times for repeated displays. Therefore, our best understanding
of this task is that implicit memory for the repeated displays is
not derived from the hippocampal region itself but, rather, is
derived from some other temporal lobe structure.
In summary, we can conclude that although the hip-
pocampus plays a role in explicit or declarative memory tasks,
it does not play a role in implicit or nondeclarative memory
tasks (at least when the task is solved in an implicit way). The
visual-paired comparison task is the one apparent exception
to this rule, highlighting the importance of extending or refin-
ing the descriptive distinctions created to understand human
behavior (e.g., implicit/explicit) to more fully encompass
data from other sources. The simple description of the task
requirements does an admirable job of dissociating two
classes of tasks, yet there is clearly something about the visual-
paired comparison task that, despite its “implicit” nature,
makes it dependent on the hippocampal region and the adja-
cent medial temporal lobe cortices.
12.4.2 Encoding Versus Retrieval
Whereas studies of patients with damage to the hippocampal
region have been able to demonstrate a dissociation between
declarative memory tasks that involve the hippocampus and
nondeclarative memory tasks that do not, studies of patients
are not particularly well suited to isolating the role of the hip-
pocampus in encoding (or storage) versus retrieval processes.
The inability to make accurate judgments in a recognition
memory task or to recall items from a study list could be the
result of failure to encode the items initially or failure to
retrieve a well encoded item. The observation that amnesic
patients can retrieve fact or event-type memory that was
learned well prior to the onset of their amnesia (see Section
12.4.3) can be taken as evidence that the hippocampus is not
continually required for declarative memory retrieval; how-
ever, it still leaves open the possibility that when anterograde
and retrograde amnesia are observed the central impairment
is one of retrieval rather than encoding or storage (e.g.,
Warrington and Weiskrantz, 1970).
Thus, in humans, determining whether the hippocam-
pus plays a differential role in encoding or retrieval relies
heavily on electrophysiological and neuroimaging data.
Complementing this, studies of animals offer the opportunity
of employing reversible lesions, blockade of long-term poten-
tiation (LTP), and other manipulations (discussed in Chapters
8 and 13). One of the challenges that faces all such attempts is
the observation that during memory retrieval tasks partici-
pants unwittingly encode the test items and can often accu-
rately remember whether an item was present during the test
(for review see Glover, 1989). fMRI activity associated with
this incidental encoding has been observed in numerous
frontal and parietal regions (Buckner et al., 2001) and bilater-
ally in the hippocampal region, perirhinal cortex, and
parahippocampal cortex (Stark and Okado, 2003).
This challenge aside, hippocampal activity has been
observed relating to both encoding and retrieval success. A
number of fMRI studies (Small et al., 2001; Davachi and
Wagner, 2002; Reber et al., 2002; Strange et al., 2002; Davachi
et al., 2003; Stark and Okado, 2003; Kirwan and Stark, 2004)
have observed hippocampal activity correlated with encoding
success. In this so-called Dm (differences due to memory)
effect (Paller and Wagner, 2002), there is greater activity dur-

ing the successful encoding of items that are later remembered
than during the unsuccessful encoding of items that are later
not remembered. Similar Dm effects have been observed in
several studies employing recording from the hippocampal
region using depth electrodes as well (Fernandez et al., 1999b;
Cameron et al., 2001; Fernandez et al., 2002; Fell et al., 2003).
We should note that these Dm effects have been frequently
observed in the parahippocampal cortex (e.g., Brewer et al.,
1998; Wagner et al., 1998; Fernandez et al., 1999b; Otten et al.,
2001; Davachi and Wagner, 2002; Fernandez et al., 2002;
Strange et al., 2002; Davachi et al., 2003) and the entorhinal or
perirhinal cortices (Fernandez et al., 1999b; Cameron et al.,
2001; Davachi and Wagner, 2002; Fernandez et al., 2002;
Strange et al., 2002; Davachi et al., 2003; Fell et al., 2003;
Kirwan and Stark, 2004) as well.
Similarly, a number of fMRI studies have demonstrated
activity in the hippocampal region related to retrieval success
(e.g., Gabrieli et al., 1997; Eldridge et al., 2000; Stark and
Squire, 2000a, 2000c, 2001a; Stark and Okado, 2003; Kirwan
and Stark, 2004). Here, successful retrieval (e.g., responding
“old” to previously studied targets) usually elicits greater
activity than unsuccessful retrieval (e.g., responding “new” to
unstudied foil items). Similar effects have been observed using
depth electrode recordings (Paller and McCarthy, 2002) and
magnetic source imaging (Papanicolaou et al., 2002).
Although apparently engaged in both successful encoding
and retrieval processes, a dissociation between the two is still
possible; and several studies have observed differences in the
hippocampal region between memory encoding processes
and retrieval processes. For example, Zeineh et al. (2003) used
the above-mentioned unfolding techniques to isolate encod-
ing and retrieval-related activity for face–name pairs. They
reported above-baseline (visual fixation on a cross) activity
during encoding but not retrieval in CA2, CA3, and the den-
tate gyrus, with this activity decreasing across repeated pre-
sentations (evidence of encoding-related activity was seen in
the parahippocampal cortex as well). In contrast, retrieval
(and to some degree encoding) was associated with activity in
the subiculum, also showing a decrease in activity with
repeated presentations. Thus, some differentiation between
encoding and retrieval was observed in components of the
hippocampus.
In addition to differentiating encoding from retrieval
across these subregions (see Chapter 3), several attempts
have been made to differentiate activity along the anterior-
posterior (or longitudinal) axis (Gabrieli et al., 1997; LePage et
al., 1998; Fernandez et al., 1999a). However, it appears as if the
initial support for this hypothesis has not been confirmed in
the further analysis of these and subsequent studies (Schacter
and Wagner, 1999). For example Small et al. (2001) used an
anatomical ROI analysis to examine encoding- and retrieval-
related activity along the longitudinal axis of the hippocampal
region (eight ROIs for each participant along this axis). Here,
they observed activity in different locations along the longitu-
dinal axis for encoding faces and for encoding names. They
Functional Role of the Human Hippocampus 565
also observed a combination of these locations (and several
others) during the encoding of face–name pairs. At retrieval,
when participants were cued with a face and asked to recall the
name, the pattern of activity in the hippocampal region was
quite similar to the pattern of activity observed when encod-
ing the face–name pair. Similarly, using an anatomical ROI
analysis, Reber et al. (2002) observed encoding effects for both
words and pictures throughout the longitudinal axis of the
hippocampal region and the adjacent cortical structures (dif-
ferences were observed between picture and word encoding,
however).
Furthermore, in a study that attempted to reduce the
effects of incidental encoding during retrieval (Stark and
Okado, 2003), encoding and retrieval each resulted in activity
in the hippocampal region and in the perirhinal and parahip-
pocampal cortices. It should be noted, however, that in this
study the exact voxels associated with intentional encoding,
incidental encoding, and retrieval were not always the same.
Likewise, Pihlajamaki et al. (2003) found evidence for greater
activity during retrieval than encoding and greater activity
during encoding than retrieval in areas of the perirhinal and
parahippocampal cortices and the hippocampal region. Thus,
it is apparent that although there may be differentiation
between encoding- and retrieval-related functions on a fine
scale, encoding and retrieval processes appear to be present
throughout the MTL.
12.4.3 Time-limited Role in Declarative Memory
At the beginning of the chapter, we noted Milner’s observa-
tion that patient H.M.’s remote memory appeared to be intact
in the face of both his profoundly impaired ability to learn
new information and a profound loss of information that he
had been exposed to for some amount of time prior to his
operation. Given his intact childhood memories, she con-
cluded that the medial temporal lobes were not the ultimate
storage site for what we now refer to as declarative memory.
Therefore, some form of systems-level “consolidation” occurs
by which memories that initially rely on structures in the
medial temporal lobe become independent of the medial tem-
poral lobe over time. (This is not to be confused with an alter-
native use of the term “consolidation” to refer to the fixation
of a memory over the course of seconds to hours.) This phe-
nomenon, termed “temporally graded retrograde amnesia,”
has been observed frequently in amnesic patients since first
being described by Ribot more than 100 years ago (Ribot,
1887).
This observation has been frequent but not entirely consis-
tent. Although some have described it as temporally graded
(e.g., Squire and Alvarez, 1995), others have described it as
constant (e.g., Warrington and McCarthy, 1988). Some have
found evidence that memory for both facts (semantic mem-
ory) and events (episodic memory) acquired before the onset
of amnesia are similarly impaired (e.g., Verfaellie et al., 1995;
Reed and Squire, 1998), whereas others have found more
566 The Hippocampus Book
selective (and constant) impairment in episodic memory
(e.g., Nadel and Moscovitch, 1997). One potential source of
variability in the data stems from the fact that all studies of
retrograde memory in human amnesic patients are retrospec-
tive and quasi-experimental in nature. When testing a
patient’s memory for knowledge acquired before the experi-
ment (and potentially years or decades before the experi-
ment), we cannot know how well the information had been
learned prior to the onset of amnesia or even if it had been
learned at all. Furthermore, we cannot know whether the
information had been retrieved, and therefore reencoded, at
some time following the initial learning. Such retrieval-
induced encoding could easily affect the presence or absence
of a temporal gradient. Likewise, it is a matter of debate
whether such retrospective tests should employ information
that shows a forgetting gradient in healthy controls or they
should employ information that can be retrieved at a constant
level of performance across the supposed delay.
Additionally, the exact location and extent of the hip-
pocampal lesion is often not known nor whether it extends
beyond the medial temporal lobes. There are a relatively small
number of cases in which damage appears to be limited to
the hippocampal region and retrograde amnesia has been
assessed (Zola-Morgan et al., 1986; Reed and Squire, 1998;
Kapur and Brooks, 1999; Holdstock et al., 2002a). In all but
one patient (Y.R. in Holdstock et al., 2002a), a temporally
graded retrograde amnesia spanning several years was
observed for both episodic and semantic memory. In the case
of Y.R., no retrograde amnesia was detected at all in this study.
However, follow-up testing on Y.R. has revealed evidence for
at least some retrograde amnesia. Its precise nature with
respect to temporal gradients and with respect to selectivity
for any kind of information or task is currently not known,
however (unpublished data, Andrew Mayes, personal commu-
nication, February 22, 2005).
Using neuroimaging to study the time-limited nature of
hippocampal function is also associated with a number of
challenges. One clear challenge is the problem of activity asso-
ciated with incidental encoding. It is quite plausible, for exam-
ple, that if memories have been consolidated and no longer
require structures in the MTL the retrieval of such a memory
(e.g., a childhood memory that has not been thought about
for some time) induces encoding in the MTL. If participants
can tell you what memories they had retrieved while inside
the scanner, activity associated with this incidental encoding
could mask (or even reverse) any retrograde gradient (Stark
and Okado, 2003).
A second challenge that perhaps faces studies of consolida-
tion more strongly than most is the difficulty posed by the
choice of baseline tasks. Several studies have shown that there
is significant activity throughout the brain (including the
MTL) during rest or other baseline tasks that do not actively
engage the participant (Binder et al., 1999; Gusnard et al.,
2001; Gusnard and Raichle, 2001; Newman et al., 2001; Stark
and Squire, 2001b). One explanation for such activity is that
during rest and other low-level baseline tasks participants are
actually engaged in an undetermined and uncontrolled task
(e.g., reflecting upon the experiment). When activity during
episodic recollection (e.g., “Reflect upon your visit to Paris as
a child”) is contrasted with rest, it is quite plausible that this
reflection is carried into the rest periods. Such activity during
rest could reduce the magnitude of an effect such that it might
not be observed or, if the amount of reflection during rest or
the actual “task” performed during rest varied with condition
(e.g., participants might be more likely to reminisce or
attempt to further elaborate a memory following retrieval of a
very old than a recent memory), one could create or even
invert a gradient purely as an artifact (Stark and Squire,
unpublished data).
Given these challenges, it is not entirely surprising that in
neuroimaging studies of consolidation the results have been
mixed. For example, Ryan et al. (2001) asked participants to
generate very remote ( 20 years) and relatively recent ( 4
years) episodic memories while outside the scanner. Inside the
scanner, they were asked to recall these memories for 20 sec-
onds. In bilateral hippocampal regions that showed an overall
difference in activity between recollection and rest, there was
no difference in activity between activity for very remote and
relatively recent memories (an alternate baseline of sentence
completion was also used and did not differ from rest, but as
this task required one sentence completion in 5 seconds it may
be open to the same difficulties as explicit rest). Furthermore,
no significant activity was observed in any of the subjects in a
remote-versus-recent contrast.
Similarly, Maguire et al. (2001) collected data as partici-
pants performed a verification task on both public event
information and autobiographical information from periods
ranging from several weeks to more than 20 years prior to
scanning (e.g., “Yes or no: You were at Tim’s wedding in
London”). Although hippocampal activity during both tasks
was greater than baseline (listening to a set of function
words), and it was greater for autobiographical than public-
event verification, it did not vary parametrically with memory
age. However, the participant’s knowledge of this information
had been assessed several weeks prior to scanning, raising the
possibility that they were retrieving the information from the
recent reencoded episode.
Finally, Stark and Squire (2000a) attempted a prospective
study of consolidation by having participants study pictures
of nameable objects at varying delays prior to scanning. At
test, the names of the objects were presented. In so doing,
much of the incidental encoding-related activity appears to be
confined to the left hemisphere, leaving retrieval-related activ-
ity for the nameable objects relatively uncontaminated in the
right hemisphere (Stark and Squire, 2001a). Activity in
anatomically defined ROIs in the MTL did not differ as a
function of study-test interval. It should be noted, however,
that the delays employed were all relatively short, ranging
from an hour to a week.
In contrast to these negative findings, there have been sev-
eral studies that have shown gradients in MTL activity as a
function of memory age. In the first such study, Haist et al.

(2001) used the “famous faces” task (Marslen-Wilson and
Teuber, 1975) in which participants were presented with pho-
tographs of faces from various decades (of both famous and
nonfamous people) and attempted to recall the names of
each. An analysis identifying regions whose activity varied lin-
early by decade (activity determined by a contrast between
famous faces from each decade and rest) isolated activity in a
region that appeared to be the right entorhinal cortex. Here,
activity was greatest when participants attempted to recall
names from the 1990s and 1980s and lowest when partici-
pants attempted to recall names from the 1940s. No other
region showed this linear trend.
In a second study, Niki and Luo (2002) used a version of
the episodic reflection task in which participants attempted
mentally to revisit places they had visited either recently or
more than 7 years prior to scanning. In the direct comparison
between recent and remote memories, activity in the left
parahippocampal gyrus was observed, with greater activity for
recent than remote memories. Although a number of other
regions throughout the brain also showed this pattern, a sub-
stantial number showed the reverse, with greater activity for
remote than recent memories.
Finally, Maguire and Frith (2003) again examined autobio-
graphical and public event knowledge as a function of
remoteness using a task similar to that in their previous work,
described above (Maguire et al., 2001). Although they again
observed greater hippocampal activity for autobiographical
than public event memory, they did observe a gradient in
right hippocampal activity as a function of remoteness of the
autobiographical memory (no gradient was observed in the
left hippocampal region).
Thus, with respect to the expectation that a time-limited
role of the MTL in declarative memories produces a gradient
in the BOLD fMRI signal, the current state of affairs is mixed.
There are several null results in which no gradient was
observed. There are, however, several positive results as well.
None of the studies has been able to address fully the issue
of incidental encoding (and how it might vary with remote-
ness), and many suffer from the difficulty that it is unclear
whether participants are recalling events from several decades
ago or from their retrieval during prescreening sessions.
Furthermore, all studies have ignored the possibility that over
the course of decades the representation of the memory may
change in such a way that the BOLD fMRI signal is affected
without altering the functional role of a particular region
(e.g., recently consolidated memories may have qualitatively
different representations in the tens of thousands of neurons
in a typical fMRI voxel than very long-standing memories).
One might consider this a clear criticism of the current stud-
ies, and the solution to these problems is not clear; nor is it
clear how this question can be better approached with neu-
roimaging. The study of retrograde memory in humans has
always been exceedingly difficult with most tasks being retro-
spective rather than prospective in nature. When combined
with the above-mentioned difficulties presented by the limita-
tions of current neuroimaging techniques, it seems obvious
Functional Role of the Human Hippocampus 567
that our best evidence for any time-limited role of MTL struc-
tures will be gained from animal models of amnesia.
Controlled prospective studies have been conducted in that
arena that, by and large, reveal temporal gradients following
hippocampal damage (for review see Squire et al., 2001).
12.4.4 Spatial Memory
The notion of a hippocampal “place cell” whose activity codes
for the current location in a spatial environment is covered
extensively in Chapter 11. The hypothesis that the hippocam-
pus is primarily or even uniquely involved in spatial process-
ing and spatial storage is laid out there, drawing extensively on
electrophysiological data from the rat. This hypothesis is con-
sidered again in Chapter 13, where a wide array of data from
rats, monkeys, and humans are considered. Before this more
comprehensive treatment, the data concerning the role of the
human hippocampus in spatial memory and spatial process-
ing are considered. In so doing, the principal question is not
whether damage to the hippocampus impairs performance on
spatial memory tasks or whether place cells are observed in
the human hippocampus. In fact, neuropsychological (e.g.,
Holdstock et al., 2000) and electrophysiological data (Ekstrom
et al., 2003) have demonstrated both, implicating the human
hippocampus in spatial memory tasks.
For example, Ekstrom et al. (2003) recorded data from
depth electrodes implanted in the hippocampal region,
parahippocampal gyrus, amygdala, and several frontal sites as
participants navigated in a virtual town. Their task was to pre-
tend to be a taxi driver, picking up and dropping off passen-
gers. Several codes were observed in the data. In particular,
11% of the cells (31/279) could be confidently classified as
place cells, with their distribution skewed toward being found
in the hippocampal region. Approximately 24% of hippocam-
pal neurons could be confidently classified as place cells,
whereas significantly fewer neurons in the parahippocampal
gyrus, approximately 8% of those sampled, could be confi-
dently classified as place cells. A second code was observed pri-
marily in the parahippocampal gyrus. These cells, termed
“location-independent view cells,” were observed to code for
particular objects or landmarks irrespective of their location.
A total of 14 of the 279 neurons (5%) were classified as such
and were observed largely in the parahippocampal gyrus.
Approximately 15% of neurons in the parahippocampal gyrus
were location-independent view cells (7 total) whereas only 1
of 55 neurons in the hippocampal region coded for this infor-
mation. Although this dissociation was observed, one should
note that the numbers of cells and their proportion to the
total number sampled is quite small. For example, of the 55
hippocampal cells recorded, 43 (78%) showed main effects of
factors other than place or coded for combinations of several
factors.
Thus, the question at hand is not whether the hippocam-
pus plays a role in spatial memory. Rather, the question is
whether the function of the human hippocampus can be tied
strongly to spatial processing or the human hippocampus is
568 The Hippocampus Book
better viewed as playing a mnemonic role, with spatial mem-
ory being only one example of hippocampal function.
One approach to this question that has been explored
extensively in the rodent (see Chapter 11) (Eichenbaum et al.,
1999) is to ask whether damage limited to the hippocampal
region impairs memory on nonspatial tasks. If such impair-
ment exists and if spatial contributions to the task can be
eliminated, one could posit a role for the human hippocam-
pus outside of spatial memory or processing. Although there
are many reports of such impairment (see Spiers et al., 2001
for review), it is almost impossible to rule out the option that
healthy human control participants engage in a spatial strat-
egy on such nonspatial tasks. For example, even on tests of
verbal recognition memory, control participants might imag-
ine the stimuli, imagine relations between stimuli, or use other
mnemonic techniques such as the method of loci to improve
their performance. Were such spatial strategies not available to
patients with hippocampal lesions, their performance might
be impaired for spatial, rather than mnemonic, reasons. As
tests of recognition memory in humans have not aimed to
address the spatial hypothesis directly, they have not con-
trolled for these potential spatial aspects of the tests as care-
fully as in much of the work in the rodent literature.
An alternative approach to addressing the question is to
assess the spatial abilities of patients who have lesions that
include the entirety of the hippocampus bilaterally. If spatial
processing appears normal despite a complete hippocampal
lesion, it would be difficult to conclude that the hippocampus
is required for spatial processing in humans. The case of
patient E.P. has provided compelling evidence that spatial
memory and processing may be intact despite hippocampal
loss (Teng and Squire, 1999). Unlike patient H.M., patient E.P.
has what is most likely complete loss of the hippocampal
region bilaterally (Stefanacci et al., 2000), with only a small
“tag” of tissue (~ 10% of the volume) remaining. With the
complete lack of entorhinal cortex in E.P., it is doubtful (even
if this tissue contains healthy neurons, which it may well not)
that it would be functional in any way. Yet, despite this com-
plete loss of the hippocampal region, E.P. can perform a num-
ber of complex spatial tasks at normal levels. When asked to
navigate in the town in which he grew up, E.P. not only can
mentally navigate along familiar routes (from his childhood
house to other local landmarks), he can mentally navigate
along novel routes (paths between randomly chosen land-
marks) and along novel routes with key paths blocked (e.g.,
imagine that a main road is closed). He also can perform a
dead reckoning task (imagine being at one landmark and
point in the direction of another landmark) at a level indis-
tinguishable from that of healthy control volunteers who grew
up with E.P. and who also left during their young adulthood
(Teng and Squire, 1999). The matched control volunteers had
an easier time mentally navigating in their current environ-
ments, scoring at the ceiling (100% correct). E.P., however,
emphatically did not have an easier time navigating in the area
he had moved into 6 years prior to testing (and 1 year after he
became amnesic). His performance dropped from 83% cor-
rect in his childhood environment to 0% correct in his current
environment on the navigation tasks. Despite living on a hill
that overlooks the Pacific Ocean a mere 2 miles away, E.P. was
not even able to point in its direction. Thus, E.P. can retrieve
spatial information and navigate in a spatial environment
learned long before the onset of his amnesia but apparently
has not learned any spatial information about his environ-
ment after the onset of his amnesia. With complete hip-
pocampal loss, spatial processing therefore appears normal
despite complete inability to acquire new spatial information.
Neuroimaging studies have also begun to provide useful
data on the role of MTL structures in spatial processing and,
consistent with lesion evidence (e.g., Epstein et al., 2001), have
clearly implicated the posterior portions of the parahip-
pocampal gyrus in topographical memory tasks. For example,
in an early study, Aguirre et al. (1996) observed MTL activity
that was confined to the posterior parahippocampal gyrus
(likely parahippocampal cortex) as participants learned to
navigate in a three-dimensional maze viewed from a first-
person perspective (activity relative to a low-level control
task). Using a similar task, Maguire et al. (1998) also reported
parahippocampal gyrus activity associated with exploring and
learning the topography of a virtual three-dimensional envi-
ronment (at least when the environment contained objects
that could serve as landmarks). Whether the activity is specif-
ically related to topographical processing or is representative
of more general mnemonic function cannot be determined
from these data. However, it is of interest that they implicate
the posterior portions of the parahippocampal gyrus and not
the hippocampal region in these topographical learning tasks.
Consistent with these findings, Shelton and Gabrieli (2002)
also observed activity in the posterior parahippocampal gyrus
(with this region of activity extending into the posterior hip-
pocampal region) as participants encoded a virtual three-
dimensional environment when viewed from a first-person,
or “route,” perspective. Of interest, the pattern clearly differed
when participants encoded environments from an aerial, or
“survey,” perspective. Despite similar subsequent memory for
environments learned from the two perspectives, activity in
the posterior parahippocampal gyrus was greater when par-
ticipants encoded the environment from a route perspective
than from a survey perspective. The authors suggest that this
difference may be the result of qualitatively different mnemo-
nic demands placed on participants in the two conditions. For
route encoding, participants must continually bind together a
representation of their present position to their past and
future positions to create a layout of the entire environment.
Although still spatial in nature, encoding the environment
from a survey perspective did not place this same sort of
demand on participants. In a subsequent task that asked par-
ticipants to draw maps of the environment, the sequential
order of presentation was preserved following route learning
and was not preserved following survey learning, further
suggesting a difference in the way the two environments were

encoded that may have led to the observed difference in activ-
ity in the parahippocampal gyrus.
Hartley et al. (2003) contrasted activity associated with
navigating in a large-scale virtual environment either by way-
finding (via route-based spatial knowledge gained by free
exploration) or by route-following (traversing the same route
that had been well learned during study). Unlike the simple
route-following task, which required only recapitulation of a
well learned route, the way-finding task required knowledge
of the global spatial relations in the environment and naviga-
tion along a novel path. Within the MTL, Hartley et al. (2003)
reported greater activity in the posterior parahippocampal
gyrus overall during the way-finding task than during the
route-following task. Perhaps of more interest, with the way-
finding condition, activity in the hippocampal region was cor-
related with accuracy of performance (ceiling effects made
this test impossible with the route-following task). Further-
more, in an individual differences analysis that examined cor-
relations between the participant’s overall performance in
way-finding and neural activity differences between way-find-
ing and route-finding, Hartley et al. (2003) reported a signifi-
cant positive correlation in what is likely the perirhinal cortex
and a significant negative correlation in the head of the right
caudate. A positive correlation in the hippocampal region fell
just short of significance. Thus, participants who were better
at navigation showed greater MTL recruitment during the
way-finding task relative to the route-finding task than partic-
ipants who performed more poorly. Conversely, these same
good navigators showed greater caudate recruitment during
the route-finding task relative to the way-finding task than
participants who performed more poorly.
One logical interpretation of these data (although not the
only one, given the contrastive, or relative, nature of fMRI) is
that the better navigators recruited structures in the MTL
more for the way-finding task and that they recruited the
caudate more for the route-finding task. Furthermore, these
results make the tantalizing suggestion that this differential
recruitment might be causally related to their better perform-
ance. Hartley et al. (2003) interpreted their results as being
supportive of a role of the hippocampus in the use of a cogni-
tive map, but McNamara and Shelton (2003) suggested that
the correlations observed in the hippocampal region are also
consistent with the view that the hippocampus processes
memory in a way that allows for its flexible use in guiding
behavior (e.g., Eichenbaum, 2000). Moreover, in the study of
Hartley et al. (2003), the most reliable positive correlation in
the individual differences analysis was observed in the perirhi-
nal cortex, not the hippocampal region. (One reliable correla-
tion was observed in the left hippocampal region, but this was
using a mnemonically laden contrast.) That the contrast
between way-finding and route-finding in the perirhinal cor-
tex would strongly correlate with navigation skill provides a
novel datapoint to help us understand the differentiation of
function across structures in the MTL.
Thus, the evidence from humans suggests that the hip-
Functional Role of the Human Hippocampus 569
pocampus is involved in spatial memory tasks but that its
function is more generally mnemonic and not limited to spa-
tial memory. In humans, spatial memory is an excellent exam-
ple of complex, declarative memory.
12.4.5 Associations, Recollections,
Episodes, or Sources
A large amount of the research on the human hippocampus
has been aimed at functionally dissociating the role of the hip-
pocampus from the role of adjacent cortical structures. As
noted at the beginning of the chapter, a popular idea draws
on the anatomy to suggest that the hippocampus integrates
information from and combines the processing of the adja-
cent cortical structures that feed into the hippocampus. Two
fundamental hypotheses that share this basic idea and that are
both driven by data from human and nonhuman studies have
been proposed and explored. One hypothesis states that there
is a clear dissociation of function between the hippocampus
and the adjacent structures. For example, the hippocampus
has been described as being involved in memory that is
associational, multi-item, spatial, episodic, and recollective,
whereas the perirhinal cortex (and by extension at times the
parahippocampal cortex) is involved in memory that is auto-
matic, noneffortful, single-item, and familiarity, or recency-
based (in contrast to recollective), with this distinction being
qualitative rather than quantitative (Brown and Aggleton,
2001).
The other hypothesis states that the dissociation of func-
tion is more quantitative than qualitative in nature and that
the hippocampus and the adjacent structures in the parahip-
pocampal gyrus are all broadly involved in declarative mem-
ory. This is not to say that the medial temporal lobe is
equipotential in nature. By virtue of being farther up in the
hierarchical structure of the medial temporal lobe (see
Chapter 3), the hippocampus is proposed to “combine and
extend” the processing carried out by the entorhinal, perirhi-
nal, and parahippocampal cortices (Squire and Zola-Morgan,
1991). By virtue of receiving input from both perirhinal and
parahippocampal cortices (via the entorhinal cortex), the
hippocampus is in a position to be able to integrate informa-
tion across these structures and sources of information. Thus,
“associative,” or “conjunctive,” processing can be attributed to
the hippocampus. However, the structures in the parahip-
pocampal gyrus also receive input from a wide range of
sources, putting them in a position to perform “associative” or
“conjunctive” processing as well. As the input to the hip-
pocampus consists of more refined and further processed
information (its major input arrives from the entorhinal cor-
tex, which receives approximately two-thirds of its input from
the perirhinal and parahippocampal cortices), the hippocam-
pus is in a position to perform different, potentially more
abstract or complex associative or conjunctive processing.
This is not to say, however, that there is binary dissociation of
function between the hippocampus and the adjacent cortical
570 The Hippocampus Book
structures according to associative or conjunctive versus sin-
gle-item, episodic versus semantic, recollection versus famil-
iarity, and so on.
The hypothesis that the hippocampus combines and
extends the processing of the adjacent cortex is underspeci-
fied, as the current theories that take this view do not detail
how this process takes place (nor, often, do the theories that
propose a more binary dissociation). Data from amnesic
patients with damage limited to the hippocampus and data
from neuroimaging studies are presented in the following two
sections. Overall, the data are not consistent with a clean,
binary division of labor between the hippocampus and the
cortical components of the medial temporal lobes. They are
more consistent with either the graded division of labor
implied by the concept of “combine and extend” or with a
division of labor along lines that have yet to be explored sig-
nificantly using these two techniques.
Amnesic Patients
The data available from amnesic patients with bilateral dam-
age thought to be limited to the hippocampal region are far
from allowing consensus in support of either hypothesis. For
example, Yonelinas et al. (2002) reported data from 56
hypoxic patients with presumed (although not confirmed—
see Section 12.3.3) bilateral damage to the hippocampal
region. When compared to a similar-sized group of healthy
controls, the patients were impaired on both recall and recog-
nition memory tasks. When transformed into z-scores (so the
patients’ performance is expressed relative to the mean and
standard deviation of the controls’ performance—see Section
12.3.2), the impairment in recall was larger than the impair-
ment in recognition. Thus, if recall tasks place greater
demands on episodic or recollective processing than recogni-
tion memory tasks, these data point to a differential role for
the hippocampus in this kind of memory.
In the same study, four of the hypoxic patients were tested
using the Remember-Know procedure to assess recollective
and familiarity components of recognition (Tulving, 1985). In
this task, participants indicate at the time of retrieval whether
their memory is best described as recollective, containing
episodic components and clear knowledge of the source of the
memory (a “Remember” response), or best described as a feel-
ing of familiarity (a “Know” response). Whereas patients with
MTL damage known to extend beyond the hippocampal
region were impaired in both recollective and familiarity com-
ponents, the hypoxic patients were significantly impaired only
in the recollective component. Thus, barring any concern
about the lesion locations in the hypoxic patients, these data
appear to support the conclusion that the hippocampus plays
a key role in recall and in recollective processing, and that
these processes may not be able to be supported by the adja-
cent cortex. Conversely, if the lack of familiarity impairment
is true (in the four participants the 23% impairment was
unreliable, but this may be the result of insufficient statistical
power), the hippocampal region may not provide any famil-
iarity signal to be used during recognition. Thus, we would
have clear functional dissociation between the hippocampal
region and the adjacent cortex.
The difficulty is that this pattern is not consistently
observed. For example, Manns et al. (2003) tested a group of
seven patients with bilateral damage limited to the hippocam-
pal region (as determined by MRI) using the same tasks
employed by Yonelinas et al. (2002). Consistent with the
results of Yonelinas et al. (2002), the patients were impaired
on both tests of recognition and recall. However, there was no
evidence of disproportional impairment. When recall per-
formance was measured against recognition performance, the
patients’ recall score was in the 30.5th percentile of their
recognition distribution and the controls’ recall score was in
the 30.6th percentile of their recognition distribution. When
the same z-score analysis was performed, the amnesics’ recog-
nition performance was worse than their recall performance,
indicating greater impairment in recognition than recall.
Furthermore, the amnesics’ performance on the Remember-
Know task showed similar impairments for both Remember
(60% reduction) and Know (60–71% reduction, depending
on the scoring method) responses. Thus, although a null
result, the data of Manns et al. (2003) leave little room to sup-
port a clear functional dissociation between the hippocampal
region and the adjacent MTL structures with respect to recol-
lective processes.
Unfortunately, neither finding is entirely atypical of the
results found in the literature. With regard to a specialized role
for the hippocampus in this form of memory, the results are
decidedly mixed, indicating a critical gap in our theories of
hippocampal function or a critical methodological problem
(such as the often-suggested possibility that our assessment of
the damage in these patients is incomplete—see Section
12.3.3). In the particular case of these two studies, there is a
potential solution to the disparity in the findings in the
methodology. The potential solution serves to highlight how
difficult this research can be and how tenuous the observa-
tions of differential impairments often are.
In a detailed reanalysis of the individual participant data
from both data sets, Wixted and Squire (2004) noted that a
single control participant of the 55 tested in the Yonelinas et
al. (2002) study had a recognition score that was a marked
outlier in the distribution of scores. The effect of this data-
point was sufficient to skew the z-score analysis (see Section
12.3.2) and mask the apparent recognition memory impair-
ment.
Results such as these led to a confused state of affairs in the
literature, as it was often unclear what to make of the conflict-
ing data and what factors have influenced the differing results.
Looking at the effects on recall versus recognition, recollection
versus familiarity, associations versus single-items, or episodic
versus semantic memory, conflicting results abound. For
example, Jon, a developmental-amnesic patient (neonatal
hypoxia) with damage limited to the hippocampal region (as
assessed by MRI), has been shown to exhibit clearly impaired
episodic memory but has done relatively well in school and

obtained a solid vocabulary despite his amnesia (Vargha-
Khadem et al., 1997). Jon has also been shown to demonstrate
relatively normal levels of recognition memory (except on
cross-modal tasks), relatively intact familiarity, and impaired
recollection (Baddeley et al., 2001; Vargha-Khadem et al.,
2001). Although Jon’s amnesia was neonatal, it appears that
this is not the source of the dissociation. While examining
early-onset (perinatal to 3 months) versus late-onset (6–14
years) amnesia in children with MRI-confirmed hippocampal
damage, Vargha-Khadem et al. (2003) noted substantial
impairments on several standardized tests of episodic mem-
ory in both groups. In contrast, semantic memory, as assessed
by measures of vocabulary acquired outside the laboratory,
was in the low-average range in both groups, perhaps indicat-
ing only mild impairment (unfortunately, it is difficult to
know if the amount of vocabulary training in the school or at
home was similar to that of normals in these patients. (See
Vargha-Khadem et al., 2001 for a similar finding.)
Furthermore, Y.R., an adult-onset amnesic patient
(Holdstock et al., 2002a,b; Mayes et al., 2002) showed a simi-
lar impairment pattern. Across a series of 34 recall tests,
Y.R. was substantially impaired relative to a group of healthy
controls, obtaining an average z-score of –3.6. However, across
a series of 43 recognition memory tests, Y.R. was only mildly
impaired, averaging a z-score of –0.5 (Mayes et al., 2002).
(Interestingly, by not showing a bias to view the novel item,
Y.R. exhibits impaired performance in the visual paired com-
parison task at 5- to 10-second delays). Thus, even in a case
of adult-onset amnesia, selective impairment in recall relative
to recognition has been observed following hippocampal
damage.
In contrast with the reports on Y.R. and the developmental
amnesic patients, several studies other than that of Manns et
al. (2003) have shown substantial impairments in item-recog-
nition memory tasks in patients with hippocampal damage
(Hopkins et al., 1995; Reed and Squire, 1997; Stark and Squire,
2000b, 2003; Stark et al., 2002). Thus, in some patients, hip-
pocampal damage leads to substantial recognition memory
impairments that are similar to their recall impairments.
Furthermore, in two of these studies (Stark et al., 2002; Stark
and Squire, 2003), single-item recognition (e.g., “Did you see
this object before”?) was compared with associative recogni-
tion (e.g., “You’ve seen both these objects before, but were they
paired together?”) in patients with damage limited to the hip-
pocampal region. In both, single-item and associative recogni-
tion patients with hippocampal damage were impaired—and
to the same degree (e.g., 15% impairment in single-item
recognition and 13% impairment in associative recognition in
Stark et al., 2002, experiment 1).
Thus, we have a number of studies that report a dissocia-
tion showing selective impairment in recall, recollection, or
associative memory following hippocampal damage and a
number of studies that clearly show no such dissociation.
Currently, there is no clear means of explaining the varying
patterns of results. Whether this is due to differences in unde-
tected extrahippocampal damage, damage in the hippocampal
Functional Role of the Human Hippocampus 571
region (location or extent), the stimuli, the tasks, or the man-
ner in which the patients approach the tasks is currently
unknown. Perhaps most critically, the disparate results could
arise from attempting to find evidence for the wrong dissoci-
ation. If the functional role of the hippocampus is correlated
with episodic, recollective, and recall aspects of declarative
memory, but in fact none of these aspects best describes its
role, we could be left with a pattern of positive and negative
results such as we currently face. For each positive result, some
aspect of the task or stimuli correlated better with the
unknown true functional role of the hippocampus than for
each negative result.
Neuroimaging
A significant number of recent neuroimaging studies have
sought to complement the amnesic patient data on the func-
tional dissociation between the hippocampal region and the
adjacent cortical structures with respect to the associative, rec-
ollective, or source components of declarative memory.
Although several patterns are beginning to emerge, the recent
neuroimaging data also do not support a clean distinction
between the role of the hippocampal region and the role of the
adjacent cortical structures along these lines. Clearly, most
studies to date have observed activity in the hippocampal
region associated with associative, recollective, or source com-
ponents of memory. However, in almost all of these studies,
activity in the (presumed) parahippocampal cortex mirrors
that in the hippocampal region. Furthermore, several have
reported associative, recollective, or source components in
(presumed) perirhinal or entorhinal cortices. In contrast,
activity correlated with single-item memory or familiarity
may not be limited to the parahippocampal gyrus, or even the
perirhinal cortex specifically (see Section 12.4.5). For exam-
ple, Henson and colleagues (2003) reported activity associated
with familiarity in several studies within anterior portions of
the MTL that appear to include entorhinal/perirhinal cortices
and the hippocampal region as well. Thus, the conclusion best
drawn from the following review is that the existing data are
certainly not supportive of a clean functional distinction
between the hippocampal region and the adjacent structures
according to the recollective, associative, or source compo-
nents of declarative memory.
Several studies (e.g., Henson et al., 1999; Eldridge et al.,
2000) have examined the dissociation between recollection
and familiarity using the Remember-Know (Tulving, 1985)
task. In this task, Remember responses are used to index rec-
ollective memory, and Know responses are used to index
recognition based on familiarity. Thus, a contrast between the
two might be used to assess whether a region such as the hip-
pocampus is particularly involved in recollective forms of
memory retrieval. Henson and colleagues (1999) reported lit-
tle differentiation between Remember and Know responses
in the MTL, with no differences observed at the time of the
test and only a small region in the parahippocampal gyrus
showing less activity for subsequent Remember responses
572 The Hippocampus Book
than Know responses at the time of the study (significant dif-
ferences were observed in frontal and parietal regions).
In contrast, Eldridge and colleagues (2000) observed
greater activity for Remember judgments than Know judg-
ments, correct rejections, and misses in the hippocampal
region. These data are consistent with a greater role for the
hippocampal region in Remember responses than in Know
responses. However, there are some limitations to the
Remember-Know task. First, it is difficult to perform in the
fMRI scanner because it requires two steps to avoid becoming
a simple confidence rating (Hicks and Marsh, 1999). If two
steps are used (as was done in Eldridge et al., 2000), two cog-
nitive processes are being engaged (first deciding yes/no
recognition and then classifying the retrieval as Remember or
Know), and it is impossible in fMRI data to separate activity
from two events that always immediately follow each other
(see Section 12.3.5). Additionally, it is difficult to ascribe the
enhanced activity for Remember responses entirely to a func-
tional dissociation favoring the recollective component of
recognition. It is quite plausible that any more detail-rich
retrieval would result in more activity than a detail-poor
retrieval. As such, Remember responses might yield more
activity than Know responses in a region not particularly
involved in the recollective component itself. Finally, even if
one reasonably assumes that some portion of the enhanced
activity for Remember responses over Know responses can be
attributed to recollective or associative aspects of processing,
Eldridge et al. (2000) observed greater activity for Remember
than Know responses not only in the hippocampal region but
also in the parahippocampal gyrus (likely to be in the parahip-
pocampal cortex). Thus, this recollective signal was not lim-
ited to the hippocampal region.
Activity tied to familiarity must be considered as well.
Henson and colleagues (2003) reported activity associated
with familiarity in several studies in anterior portions of the
MTL that appear to include entorhinal/perirhinal cortices and
the hippocampal region as well. Likewise, Stark and Squire
(2000c, 2001a) reported activity in the hippocampal region
during simple recognition memory tasks that can rely purely
on familiarity, with no apparent increase in activity when the
task became more associative or recollective (Stark and
Squire, 2001a). Thus, the data do not support a clean func-
tional dissociation between recollective processing in the hip-
pocampal region and familiarity processing in the structures
of the parahippocampal gyrus.
Several other studies have examined activity during the
formation or retrieval of paired associates that may be used to
determine whether the hippocampal region might be more
involved in associative components of declarative memory
than in nonassociative or single-item components. In one case
(Sperling et al., 2001), associative encoding-related activity
was observed in the hippocampal formation (defined as the
hippocampus proper, subiculum, and entorhinal cortex)
without observing significant activity elsewhere in the
parahippocampal gyrus, supporting the notion that the hip-
pocampus is particularly involved in associative aspects of
memory. However, in a similarly designed study, but one that
restricted its analysis to the hippocampal region (Small et al.,
2001), activity was observed not only for the encoding and
retrieval of face–name associations but also for the encoding
and retrieval of individual faces and names (although hip-
pocampal activity for face–name associations was more wide-
spread than the combination of activity for faces and activity
for names alone). In both, a blocked design was used to con-
trast activity associated with viewing novel face–name pairs
with activity associated with viewing repeated face–name
pairs. Unfortunately, the blocked nature of the design
impeded the ability to isolate individual encoding trials based
on the quality or amount of information subsequently
retrieved. Thus, in this study, it is difficult to know whether
the greater activity associated with viewing novel face–name
pairs was the result of encoding the face, the name, or the
association or whether the activity resulted from a novelty
detection (e.g., Strange et al., 1999).
Other studies contrasting associative and nonassociative
memory have observed more widespread activity. For exam-
ple Henke and colleagues (1997) used PET to measure activity
during the encoding and retrieval of face–house pairs. More
recently, Henke et al. (1999) used fMRI to measure activity
during the encoding of semantic associations between words.
In both studies, greater activity for associative relative to
nonassociative memory was observed in both the hippocam-
pal region and the parahippocampal gyrus. Similarly, in an
fMRI study of recognition memory, Yonelinas et al. (2001)
reported greater activity during associative than nonassocia-
tive recognition in both the hippocampal region and the pos-
terior parahippocampal gyrus. Likewise, Pihlajamaki et al.
(2003) observed their most consistent activity during the
encoding of picture pairs not in the hippocampal region itself
(where there was some evidence) but in the perirhinal cortex;
and Sperling et al. (2003) reported activity associated with
high-confidence encoding of face–name pairs in anterior por-
tions of the MTL that covered the hippocampal region and the
parahippocampal gyrus bilaterally (but whose most active
voxels were in the hippocampal region bilaterally and in the
right entorhinal cortex).
In addition to studying arbitrary associations, a number of
attempts have been made to examine the automatic associa-
tions that are formed between memory for an item itself and
the “source memory” or episodic knowledge of when and
where that item was encountered (e.g., Did you see this item?
If so, which study task was it in?). In an elegant example that
looked at encoding activity that predicted subsequent mem-
ory for an item along with the source relative to encoding
activity that predicted memory for the item alone, Davachi et
al. (2003) reported a functional dissociation. Activity predict-
ing subsequent source memory was somewhat widespread,
with this associative activity observed bilaterally in the hip-
pocampal region and in the left parahippocampal cortex. In
contrast, item-only activity was observed solely in the left
perirhinal cortex. Here, activity at encoding predicted subse-
quent memory in general (using as contrast greater activity at
the time of encoding for items subsequently remembered
relative to items subsequently forgotten), but the activity did

not differ as a function of whether the source aspect was sub-
sequently remembered. Thus, the hippocampal region and the
parahippocampal cortex appeared to play a role in encoding
the source component, whereas the perirhinal cortex appeared
to play a role only in encoding the item component of the
memory (see Ranganath et al., 2003 for a similar finding).
In a related study, Kirwan and Stark (2004) examined
activity during both the encoding and the retrieval of
face–name pairs. Like Davachi et al. (2003), greater activity
during the successful encoding of associations relative to their
unsuccessful encoding (an “associative” pattern) was observed
in the hippocampal region and the parahippocampal cortex
(right unilateral in this case). Also like Davachi et al. (2003), a
portion of perirhinal cortex showed a general subsequent
memory effect that did not vary with the associative compo-
nent (a “single item” pattern). Thus, associative signals were
observed in the hippocampal region and the parahippocam-
pal cortex whereas single-item signals were observed in the
perirhinal cortex.
However, unlike Davachi et al. (2003), reliable signals that
ran counter to this simple dissociation were observed as well.
First, the associative pattern was observed in one portion
of the right parahippocampal cortex during encoding,
whereas the single-item signal was observed in a different por-
tion of the right parahippocampal cortex. Thus, at the time of
encoding, there was no clear differentiation observed between
an associative hippocampal region and parahippocampal cor-
tex and a nonassociative perirhinal cortex (nor, of course, was
there clear differentiation between the hippocampal region
and the parahippocampal gyrus as a whole). Furthermore,
when examining activity during recognition memory testing,
the left perirhinal cortex, right entorhinal cortex, right hip-
pocampal region, and right parahippocampal cortex all
showed an associative pattern of activity (greater activity dur-
ing retrieval of the face(name pair and their association rela-
tive to the retrieval of the face and name without their
association). Thus, at time of retrieval, greater activity in the
associative than in the nonassociative condition was observed
throughout the regions of the MTL, with similar amounts of
associative activity (contrast between associative and nonasso-
ciative retrieval) across MTL structures.
Dobbins et al. (2002) reported a similar dissociation dur-
ing retrieval of source information (what task was performed
at the study?) and recency information (which word appeared
more recently on the study list?). When examining activity in
the source retrieval task, activity in the hippocampal region,
the posterior parahippocampal gyrus (likely the parahip-
pocampal cortex), and activity in the entorhinal/perirhinal
cortex all were associated with greater activity during correct
source retrieval than incorrect source retrieval. In the hip-
pocampal region and the posterior parahippocampal gyrus,
an overall task effect was observed, with both regions showing
greater overall activity during source than recencyjudgments.
In contrast, although activity in the entorhinal/perirhinal cor-
tex indexed accuracy in the source retrieval task (it was
affected by accuracy of source judgments), its activity was
similar for source and recency judgments overall. Therefore,
Functional Role of the Human Hippocampus 573
whereas the hippocampal and posterior parahippocampal
gyrus activity could be easily explained as indexing source
retrieval, the entorhinal/perirhinal activity was a more com-
plex combination of source and recency-related activity.
Thus, the current neuroimaging evidence has clearly impli-
cated both the hippocampal region (Henke et al., 1997, 1999;
Eldridge et al., 2000; Small et al., 2001; Sperling et al., 2001,
2003; Stark and Squire, 2001a; Yonelinas et al., 2001; Dobbins
et al., 2002; Davachi et al., 2003; Ranganath et al., 2003;
Kirwan and Stark, 2004) and the parahippocampal cortex
(Henke et al., 1997, 1999; Eldridge et al., 2000; Yonelinas et al.,
2001; Dobbins et al., 2002; Davachi et al., 2003; Ranganath et
al., 2003; Kirwan and Stark, 2004) in recollective and associa-
tive memory encoding and retrieval. Furthermore, there has
been some support for the perirhinal cortex and the entorhi-
nal cortex in this regard as well (Dobbins et al., 2002;
Pihlajamaki et al., 2003; Sperling et al., 2003; Kirwan and
Stark, 2004) (see also the discussion of Hartley et al., 2003, in
Section 12.4.4). Thus, it would be an oversimplification of the
data to conclude that the fMRI data support a specific or
unique role for the hippocampal region in associative or rec-
ollective aspects of declarative memory, as the same patterns
of activity clearly extend into the adjacent cortical structures
(most often the parahippocampal cortex). Likewise, it would
be an oversimplification to conclude that the fMRI data sup-
port a specific or unique role for any of the adjacent cortical
structures in nonassociative forms of declarative memory.
Although there is evidence for nonassociative or familiarity-
based activity in the entorhinal and perirhinal cortices
(Dobbins et al., 2002; Davachi et al., 2003; Henson et al., 2003;
Ranganath et al., 2003; Kirwan and Stark, 2004), there is evi-
dence for such activity in the parahippocampal cortex
(Kirwan and Stark, 2004) and in the hippocampal region
(Stark and Squire, 2000c; Small et al., 2001; Stark and Squire,
2001a; Henson et al., 2003). In addition, each of these struc-
tures has been shown to respond according to the associative
or recollective nature of the task as well. Accordingly, although
these neuroimaging data cannot be resolute with respect to
function (all are correlational data sets), they certainly do not
support clean dissociation of function in the MTL according
to recollective, associative, or source components of declara-
tive memory.
Á
12.5 Conclusions
We began this chapter by asking the question, What does the
human hippocampus do? The significant amount of research
that has followed Milner’s initial hypotheses has allowed us to
reach a number of clear conclusions. First, along with the adja-
cent cortical structures in the parahippocampal gyrus, the
human hippocampal region is critically involved in memory
for facts and events (explicit or declarative memory). Second,
this involvement is time-limited. Third, the hippocampal
region and the adjacent cortex are not involved in immediate
or working memory process and are not involved in a wide
574 The Hippocampus Book
range of implicit or nondeclarative long-term memory process
(although they may be used in working memory or implicit
tasks that evoke declarative processes). Fourth, the hippocam-
pal region is not involved in nonmnemonic aspects of cogni-
tion including spatial processing (although spatial memory is
a clear example of its time-limited mnemonic role).
However, despite these strides, the differentiation of func-
tion between the hippocampal region and the adjacent corti-
cal structures of the parahippocampal gyrus is not currently
apparent. That the hippocampus is involved in associative,
recollective, or source components of declarative memory is
quite clear. Lesions to the hippocampal region yield deficits on
tasks that assess these forms of memory, and neuroimaging
studies have observed activity correlated with them as well.
However, it is equally clear that the hippocampus can be
involved in single-item, nonassociative, and familiarity-based
components of declarative memory as well. Hippocampal
lesions have resulted in deficits on these tasks, and hippocam-
pal activity has been observed during them. Furthermore, the
adjacent cortical structures (most notably the parahippocam-
pal cortex) may also serve to support associative, recollective,
or source components of declarative memory. Thus, the avail-
able data do not support a clean division of labor along any of
the lines proposed.
Some differentiation exists. Not only does the anatomy sug-
gest functional differentiation, the research to date has often
yielded solid evidence of functional dissociations. However,
pushing our interpretations of these dissociations and the
theories that fall out of them has often led to clear disconnects
between the theoretical predictions and additional data.
There is a difficulty faced when one tries to assign function
to a region based on the impairment observed following dam-
age or based on a set of signals recorded from the region. If a
hippocampal lesion impairs task X, does it mean that the hip-
pocampus does task X? Difficulties with dissociations aside
(which are also discussed in Chapter 13), even if we are certain
that the ability to perform task X critically depends on pro-
cessing that occurs in the hippocampus, it is still a large leap
to the conclusion that the purpose of the hippocampus is to
give us the processing required by task X. Yet, this is the kind
of conclusion the dissociation approach can easily encourage
unless one is extremely cautious about the interpretation of
the data.
A more concrete example might make this problem clear.
Suppose there is a task that requires one-trial learning of asso-
ciative, complex, cross-modal, relational, novel, arbitrary
information. Further suppose that a complete bilateral lesion
to the hippocampus entirely removes the ability to perform
this task at above-chance levels. One might conclude quite
reasonably that the hippocampus allows us to, or is necessary
to, perform one-trial learning of associative, complex, cross-
modal, relational, novel, and arbitrary information. However,
it would not be reasonable to conclude that this is the only
thing the hippocampus does. The hippocampus might well be
involved in one-trial associative, complex, cross-modal, and
relational memories between pieces of information that are
not entirely novel or arbitrary. If it were not, we’d have a hip-
pocampus that is rarely doing anything at all, as few experi-
ences would satisfy the extreme end points of the dimensions
laid out here.
Furthermore, it might well be that if one or more of these
constraints were dropped or weakened, adjacent cortical
regions might be able to perform the task to some degree.
With the bias toward visual information found in perirhinal
cortex, perhaps associative, complex, relational, novel, and
arbitrary information could be learned to at least some degree
if it were not heavily multimodal and if the test were not all-
or-none but sensitive to somewhat more gradual learning. By
still being somewhat multimodal, perirhinal cortex may sup-
port learning of multimodal information but only when other
constraints are weakened.
In this situation (which parallels the data at least in spirit),
what would we claim to be the function of the hippocampus?
Are there tasks that only the hippocampus can perform? Yes,
but, is that all that the hippocampus does? No. Is there a sin-
gle feature of a task that allows us to determine that the hip-
pocampus and only the hippocampus can perform the task?
Not really. There is a combination of features, each of which
points toward strengths of the hippocampus that, when taken
together, isolate it as the only structure capable of performing
the required task. By virtue of the anatomy, the hippocampus
receives the most highly processed, multimodal information
and is well designed to learn new arbitrary information rap-
idly. Structures such as the perirhinal cortex receive less
processed, more modal-specific information and are less well
designed to learn new arbitrary information rapidly.
Therefore, despite finding what might be a task clearly to
isolate the hippocampus, we may very well be right back in the
opposite class of theory. Even with a task that dissociates the
hippocampus from the rest of the brain, the hippocampus
may be combining and extending the processing of the adja-
cent cortex in the medial temporal lobe. It is not that the hip-
pocampus is doing anything all that different from the
adjacent cortex, it is merely that it has access to more com-
plete, more refined information and can learn a bit more rap-
idly. Thus, what initially appears as a clear qualitative
dissociation may, in truth, be far more quantitative in nature.
This is not meant to discourage the quest for dissociations.
By finding such dissociations, we get to know in what tasks the
hippocampus (or any other structure) plays a critical role. In
so doing, and when looking across data sets, we can attempt to
discern just what aspects make a task fully dependent on the
hippocampus and what aspects make a task partially depend-
ent on the hippocampus; we therefore can then define just
what the hippocampus may be doing for us. What this is
meant to argue against is the simple extension from dissocia-
tion to functional interpretation. The shorthand of labeling a
structure such as the hippocampus as “performing task X
memory” or “being responsible for” a particular kind of mem-
ory implies that it does not perform task Y memory, and other
structures cannot be responsible for this kind of memory. It is
not the dissociation in the data that is problematical but its

interpretation as a binary functional dissociation that does
injustice to the data and oversimplifies the operation of a
complex, highly interconnected and dynamic system.
What we are then currently left with is incomplete under-
standing of the medial temporal lobe and the role the hip-
pocampus plays in its function. As several have suggested, the
division of labor is likely to be graded rather than absolute
(e.g., Lavenex and Amaral, 2000; Suzuki and Eichenbaum,
2000; Stark et al., 2002; Norman and O’Reilly, 2003). Although
it may best be thought of currently in terms of combining and
extending the processing of the adjacent cortex, this is still an
unsatisfying definition of the role of the hippocampus. We
still do not know the exact nature of this division of labor,
what exactly can be combined and extended, how it is done,
and how the collection of structures interact to perform this
mnemonic function. We also do not know the role of the
perirhinal, entorhinal, and parahippocampal cortices; and we
do not know how they all interact as a dynamic system. These
issues represent currently active areas of experimentation and
theoretical development, and all can benefit from the recent
methodological advances laid out earlier in the chapter. In
short, although our present understanding of human hip-
pocampal function is far from complete, we have certainly
made great strides. This is an exciting time for human hip-
pocampal research.
In conclusion, one final point needs to be addressed.
Throughout this chapter, the hippocampus and the rest of the
medial temporal lobes have been discussed as providing a crit-
ical role in declarative or explicit long-term memory. The
reader should not take this to mean that the structures of the
medial temporal lobe are the only ones responsible for long-
term memory performance. Their role is essential, as they
appear to act as a repository for this kind of information or at
least for some compressed version of this information (i.e., a
“pointer”) for some period of time. However, encoding and
retrieval tasks engage structures well beyond the medial tem-
poral lobes. For example, in cued recall, the cue must be
assembled in some format suitable for probing memory’s con-
tents, and the results of the retrieval must be interpreted and
evaluated. Quite likely, the results of the initial retrieval
attempt leads to an extended or revised cue that is again used
in a retrieval attempt. Much of this appears to be the purview
of structures outside the medial temporal lobe (of frontal
structures in particular), making memory, and even long-
term declarative memory, the operation of a large system of
structures. The medial temporal lobe and the hippocampus in
particular are but one component, albeit a vital component, of
this larger system.
Á ACKNOWLEDGMENTS
The author thanks Shauna Stark, Brock Kirwan, Yoko Okado, Marci
Flanery, and Richard Morris for their comments on drafts of the
chapter; Susan Corkin for providing previously unpublished data on
H.M.; and the National Science Foundation for support.
Functional Role of the Human Hippocampus 575
Á REFERENCES
Aguirre G, Detre J, Alsop D, D’Esposito MD (1996) The parahip-
pocampus subserves topographical learning in man. Cereb
Cortex 6:823–829.
Bachevalier J, Brickson M, Hagger C (1993) Limbic-dependent
recognition memory in monkeys develops early in infancy.
Learn Mem 4:77–80.
Baddeley A, Vargha-Khadem F, Mishkin M (2001) Preserved recogni-
tion in a case of developmental amnesia: implications for the
acquisition of semantic memory? J Cogn Neurosci 13:357–369.
Berger T, Laham R, Thompson R (1980) Hippocampal unit-behavior
correlations during classical conditioning. Brain Res 193:
229–248.
Binder J, Frost J, Hammeke T, Bellgowan P, Rao S, Cox RW (1999)
Conceptual processing during the conscious resting state: a
functional MRI study. J Cogn Neurosci 11:80–93.
Blaxton T, Zeffiro T, Gabrieli JD, Bookheimer SY, Carrillo M,
Theodore W, Disterhoft J (1996) Functional mapping of human
learning: a positron emission tomography activation study of
eyeblink conditioning. J Neurosci 16:4032–4040.
Boynton G, Engel SA, Glover GH, Heeger D (1996) Linear systems
analysis of functional magnetic resonance imaging in human
V1. J Neurosci 16:4207–42212.
Brewer JB, Zhao Z, Desmond JE, Glover GH, Gabrieli JD (1998)
Making memories: brain activity that predicts how well visual
experience will be remembered. Science 281:1185–1187.
Brown MW, Aggleton JP (2001) Recognition memory: what are the
roles of the perirhinal cortex and hippocampus? Nat Rev
Neurosci 2:51–61.
Buckner RL, Petersen SE, Ojemann JG, Miezin FM, Squire LR,
Raichle ME (1995) Functional anatomical studies of explicit
and implicit memory retrieval tasks. J Neurosci 15:12–29.
Buckner RL, Wheeler ME, Sheridan MA (2001) Encoding processes
during retrieval tasks. J Cogn Neurosci 13:406–415.
Buxton RB (2001) Introduction to functional magnetic resonance
imaging: principles and techniques (translated). Cambridge:
Cambridge University Press.
Caesar K, Gold L, Lauritzen M (2003) Context sensitivity of activity-
dependent increases in cerebral blood flow. Proc Natl Acad Sci
USA 100:4239–4244.
Caine D, Watson J (2000) Neuropsychological and neuropathological
sequelae of cerebral anoxia: a critical review. J Int Neuropsychol
Soc 6:86–99.
Cameron K, Yashar S, Wilson C, Fried I (2001) Human hippocampal
neurons predict how well word pairs will be remembered.
Neuron 30:289–298.
Chun MM, Phelps EA (1999) Memory deficits for implicit contextual
information in amnesic subjects with hippocampal damage. Nat
Neurosci 2:844–847.
Clark RE, Squire LR (1998) Classical conditioning and brain systems:
the role of awareness. Science 280:77–81.
Clark RE, Zola SM, Squire LR (2000) Impaired recognition memory
in rats after damage to the hippocampus. J Neurosci 20:
8853–8860.
Cohen NJ, Poldrack RA, Eichenbaum H (1997) Memory for items
and memory for relations in the procedural/declarative mem-
ory framework. Memory 5:131–178.
Corkin S, Amaral DG, Gonzalez RG, Johnson KA, Hyman BT(1997)
H.M.’s medial temporal lobe lesion: findings from magnetic res-
onance imaging. J Neurosci 17:3964–3979.
576 The Hippocampus Book
Cummings JL, Tomiyasu U, Read S, Benson DF (1984) Amnesia with
hippocampal lesions after cardiopulmonary arrest. Neurology
34:679–681.
Dale AM, Buckner RL (1997) Selective averaging of rapidly presented
individual trials using fMRI. Hum Brain Mapp 5:329–340.
Davachi L, Wagner AD (2002) Hippocampal contributions to
episodic encoding: insights from relational and item-based
learning. J Neurophysiol 88:982–990.
Davachi L, Mitchell J, Wagner AD (2003) Multiple routes to memory:
distinct medial temporal lobe processes build up item and
source memories. Proc Natl Acad Sci USA 100:2157–2162.
Dobbins IG, Rice HJ, Wagner AD, Schacter DL (2002) Memory ori-
entation and success: separable neurocognitive components
underlying episodic recognition. Neuropsychologia 41:318–333.
Eichenbaum H (2000) A cortical-hippocampal system for declarative
memory. Nat Rev Neurosci 1:41–50.
Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999)
The hippocampus, place cells, and memory: is it spatial memory
or a memory space? Neuron 23:209–226.
Ekstrom A, Kahana M, Caplan J, Fields T, Isham E, Newman E, Fried
I (2003) Cellular networks underlying human spatial naviga-
tion. Nature 425:184–187.
Eldridge LL, Knowlton BJ, Furmanski CS, Bookheimer SY, Engel SA
(2000) Remembering episodes: a selective role for the hip-
pocampus during retrieval. Nat Neurosci 3:1149–1152.
Epstein R, DeYoe E, Press D, Rosen A, Kanwisher N (2001)
Neuropsychological evidence for a topographical learning
mechanism in parahippocampal cortex. Cogn Neuropsychol
18:481–508.
Fagan J (1970) Memory in the infant. J Exp Child Psychol 9:217–226.
Fell J, Klaver P, Elfadil H, Schaller C, Elger C, Fernandez G (2003)
Rhinal-hippocampal theta coherence during declarative mem-
ory formation: interaction with gamma synchronization? Eur J
Neurosci 17:1082–1088.
Fernandez G, Brewer JB, Zhao Z, Glover GH, Gabrieli JD (1999a)
Level of sustained entorhinal activity at study correlates with
subsequent cued-recall performance: a functional magnetic res-
onance imaging study with high acquisition rate. Hippocampus
9:35–44.
Fernandez G, Effern A, Grunwald T, Pezer N, Lehnertz K,
Dumpelmann M, Van Roost D, Elger CE (1999b) Real-time
tracking of memory formation in the human rhinal cortex and
hippocampus. Science 285:1582–1585.
Fernandez G, Klaver P, Fell J, Grunwald T, Elger CE (2002) Human
declarative memory formation: segregating rhinal and hip-
pocampal contributions. Hippocampus 12:514–519.
Fried I, MacDonald K, Wilson C (1997) Single neuron activity in
human hippocampus and amygdala during recognition of faces
and objects. Neuron 18:753–765.
Fried I, Cameron K, Yashar S, Fong R, Morrow J (2002) Inhibitory
and excitatory responses of single neurons in the human medial
temporal lobe during recognition of faces and objects. Cereb
Cortex 12:575–584.
Gabrieli JD, Brewer JB, Desmond JE, Glover GH (1997) Separate neu-
ral bases of two fundamental memory processes in the human
medial temporal lobe. Science 276:264–266.
Gale S, Hopkins R (2004) Effects of hypoxia on the brain: neu-
roimaging and neuropsychological findings following carbon
monoxide poisoning and obstructive sleep apnea. J Int Neu-
ropsychol Soc 10:60–71.
Glover JA (1989) The “testing” phenomenon: not gone, but nearly
forgotten. J Educ Psychol 81:392–399.
Graf P, Squire LR, Mandler G (1984) The information that amnesic
patients do not forget. J Exp Psychol Learn Mem Cogn 10:
164–178.
Grasby P, Frith C, Friston K, Bench C, Frackowiak R, Dolan RJ (1993)
Functional mapping of brain areas implicated in auditory-ver-
bal memory function. Brain 116:1–20.
Green D, Swets J (1966) Signal detection theory and psychophysics
(translated). New York: Wiley.
Gusnard D, Raichle ME (2001) Searching for a baseline: functional
imaging and the resting human brain. Nat Rev Neurosci
2:685–694.
Gusnard D, Akbudak E, Shulman G, Raichle ME (2001) Medial pre-
frontal cortex and self-referential mental activity: relation to a
default mode of brain activation. Proc Natl Acad Sci USA
98:4259–4264.
Haist F, Gore J, Mao H (2001) Consolidation of human memory over
decades revealed by functional magnetic resonance imaging.
Nat Neurosci 4:1139–1145.
Hamann SB, Squire LR (1997) Intact perceptual memory in the
absence of conscious memory. Behav Neurosci 111:850–854.
Hartley T, Maguire E, Spiers H, Burgess N (2003) The well-worn
route and the path less traveled: distinct neural bases of route
following and wayfinding in humans. Neuron 2003.
Heit G, Smith M, Halgren E (1988) Neural encoding of individual
words and faces by the human hippocampus and amygdala.
Nature 333:773–775.
Heit G, Smith M, Halgren E (1990) Neuronal activity in the human
medial temporal lobe during recognition memory. Brain
113:1093–1112.
Henke K, Buck A, Weber B, Wieser HG (1997) Human hippocam-
pus establishes associations in memory. Hippocampus 7:
249–256.
Henke K, Weber B, Kneifel S, Wieser HG, Buck A (1999) Human hip-
pocampus associates information in memory. Proc Natl Acad Sci
USA 96:5884–5889.
Henson R (2005) What can functional neuroimaging tell the experi-
mental psychologist? Q J Exp Physiol 58:193–233.
Henson RNA, Rugg MD, Shallice T, Josephs O, Dolan RJ (1999)
Recollection and familiarity in recognition memory: an event-
related functional magnetic resonance imaging study. J Neurosci
19:3962–972.
Henson RNA, Cansino S, Herron J, Robb W, Rugg MD (2003) A
familiarity signal in human anterior medial temporal cortex?
Hippocampus 13:301–304.
Hicks JL, Marsh RL (1999) Remember-know judgments can depend
on how memory is tested. Psychonom Bull Rev 6:117–122.
Holdstock JS, Mayes AR, Cezayirli E, Isaac C, Aggleton JP, Roberts N
(2000) A comparison of egocentric and allocentric spatial mem-
ory in a patient with selective hippocampal damage. Neu-
ropsychologia 38:410–425.
Holdstock JS, Mayes AR, Isaac C, Gong Q, Roberts N (2002a)
Differential involvement of the hippocampus and temporal lobe
cortices in rapid and slow learning of new semantic informa-
tion. Neuropsychologia 40:748–768.
Holdstock JS, Mayes AR, Roberts N, Cezayirli E, Isaac C, O’Reilly RC,
Norman KA (2002b) How recall and recognition are affected by
focal damage to the human hippocampus. Hippocampus 12:
341–351.

Hopkins R, Kesner R, Goldstein M (1995) Item and order recognition
memory in subjects with hypoxic brain injury. Brain Cogn
27:180–201.
Hopkins R, Myers C, Shohamy D, Grossman S, Gluck MA (2004)
Impaired probabilistic category learning in hypoxic subjects
with hippocampal damage. Neuropsychologia 42.
Hyde JS, Biswal B, Jesmanowicz A (2001) High-resolution fMRI
using multislice partial k-space GR-EPI with cubic voxels. Magn
Reson Med 46:114–125.
Jacoby LL, Dallas M (1981) On the relationship between autobio-
graphical memory and perceptual learning. J Exp Psychol Gen
110:306–240.
Kapur N, Brooks D (1999) Temporally-specific retrograde amnesia in
two cases of discrete bilateral hippocampal pathology.
Hippocampus 9:247–254.
Kapur N, Craik FIM, Tulving E, Wilson A, Houle S, Brown G (1994)
Neuroanatomical correlates of encoding in episodic memory:
levels of processing effect. Proc Natl Acad Sci USA 91:2008–2011.
Keane M, Gabrieli JD, Monti L, Fleischman DA, Cantor J, Noland J
(1997) Intact and impaired conceptual memory processes in
amnesia. Neuropsychology 11:59–69.
Kirwan C, Stark CEL (2004) Medial temporal lobe activation during
encoding and retrieval of novel face-name pairs. Hippocampus
14:910–930.
Kirwan CB, Jones C, Miller MI, Stark CEL (in press) High-resolution
f MRI investigation of the medial temporal lobe. Hum Brain
Mapp.
Knowlton BJ, Squire LR (1993) The learning of categories: parallel
brain systems for item memory and category knowledge. Science
262:1747–1749.
Kreiman G, Koch C, Fried I (2000) Category-specific visual responses
of single neurons in the human medial temporal lobe. Nat
Neurosci 3:946–953.
Lavenex P, Amaral DG (2000) Hippocampal-neocortical interaction:
a heirarchy of associativity. Hippocampus 10:420–430.
Law JR, Flanery MA, Wirth S, Yanike M, Smith AC, Frank LM, Suzuki
WA, Brown EN, Stark CEL (2005) fMRI activity during the
gradual acquisition and expression of paired-associate memory.
J Neurosci 25:5720–5729.
LePage M, Habib R, Tulving E (1998) Hippocampal PET activation
of memory encoding and retrieval: the HIPER model.
Hippocampus 8:313–322.
Levy D, Stark CEL, Squire LR (2004) Intact conceptual priming in the
absence of declarative memory. Psychol Sci 15:680–686.
Logothetis N, Pauls J, Augath M, Trinath T, Oeltermann A (2001)
Neurophysiological investigation of the basis of the fMRI signal.
Nature 412:150–157.
Maguire E, Frith CD (2003) Lateral asymmetry in the hippocampal
response to the remoteness of autobiographical memories. J
Neurosci 23:5302–5307.
Maguire E, Frith CD, Burgess N, Donnett J, O’Keefe J (1998)
Knowing where things are: parahippocampal involvement in
encoding objection locations in a virtual large-scale space. J
Cogn Neurosci 10:61–76.
Maguire E, Henson RNA, Mummery C, Frith CD (2001) Activity
in prefrontal cortex, not hippocampus, varies parametrically
with the increasing remoteness of memories. Neuroreport
12:441–444.
Manns JR, Squire LR (2001) Perceptual learning, awareness, and the
hippocampus. Hippocampus 11:776–782.
Functional Role of the Human Hippocampus 577
Manns J, Stark C, Squire L (2000) The visual paired-comparison task
as a measure of declarative memory. Proc Natl Acad Sci USA
97:12375–12379.
Manns JR, Hopkins R, Reed JM, Kitchener EG, Squire LR (2003)
Recognition memory and the human hippocampus. Neuron
37:171–180.
Marslen-Wilson W, Teuber H (1975) Memory for remote events in
anterograde amnesia: Recognition of public figures from news
photographs. Neuropsychologia 13:353–364.
Mauk M, Thompson R (1987) Retention of classically conditioned
eyelid responses following acute decerebration. Brain Res
403:89–95.
Mayes AR, Holdstock JS, Isaac C, Hunkin N, Roberts N (2002)
Relative sparing of item recognition memory in a patient with
adult-onset damage limited to the hippocampus. Hippocampus
12:325–340.
McKee R, Squire LR (1993) On the development of declarative mem-
ory. J Exp Psychol Learn Mem Cogn 19:397–404.
McNamara T, Shelton A (2003) Cognititve maps and the hippocam-
pus. Trends Cogn Sci 7:333–335.
Miller MB, Beg M, Ceritogulu C, Stark CEL (2005) Improving statis-
tical power and fMRI data localization using large deformation
metric mapping. Proc Natl Acad Sci USA 102:9685–9690.
Milner B (1962) Les troubles de la mémoire accompagnant des lésions
hippocampiques bilatérales. In: Physiologie de l’hippocampe, pp
257–272. Paris: Centre National de la Recherche Scientifique.
Milner B (1972) Disorders of learning and memory after temporal-
lobe lesions in man. Clin Neurosurg 19:421–446.
Milner B, Corkin S, Teuber H (1968) Further analysis of the hip-
pocampal amnesic syndrome: 14-year follow-up study of H.M.
Neuropsychologia 6:215–234.
Milner B, Squire LR, Kandel ER (1998) The cognitive neuroscience of
memory. Neuron 20:445–468.
Morrell H, Gitman S, Wixted J (2002) On the nature of the decision
axis in signal-detection-based models of recognition memory. J
Exp Psychol Learn Mem Cogn 28:1095–1110.
Nadel L, Moscovitch M (1997) Memory consolidation, retrograde
amnesia, and the hippocampal complex. Curr Opin Neurobiol
7:217–227.
Newman S, Tweig D, Carpenter P (2001) Baseline conditions
and subtractive logic in neuroimaging. Hum Brain Mapp 14:
228–235.
Niki K, Luo J (2002) An fMRI study on the time-limited role of the
medial temporal lobe in long-term topographical autiobio-
graphic memory. J Cogn Neurosci 14:500–507.
Norman KA, O’Reilly RC (2003) Modeling hippocampal and neo-
cortical contributions to recognition memory: a complemen-
tary learning systems approach. Psychol Rev 110:611–646.
Nyberg L, Tulving E, Habib R, Nilsson L, Kapur S, Houle S, Cabeza R,
McIntosh A (1995) Functional brain maps of retrieval mode
and recovery of episodic information. Neuroreport 7:249–252.
O’Reilly RC, Rudy JW (2000) Computational principles of learning
in the neocortex and hippocampus. Hippocampus 10:389–397.
Otten LJ, Henson RNA, Rugg MD (2001) Depth of processing effects
on neural correlates of memory encoding: relationship between
findings across- and within-task comparisons. Brain 124:
399–412.
Paller KA, McCarthy G (2002) Field potentials in the human hip-
pocampus during the encoding and recognition of visual stim-
uli. Hippocampus 12:415–420.
578 The Hippocampus Book
Paller KA, Wagner AD (2002) Observing the transformation of expe-
rience into memory. Trends Cogn Sci 6:93–102.
Papanicolaou AC, Panagiotis GS, Castillo EM, Breier JI, Katz JS,
Wright AA (2002) The hippocampus and memory of verbal and
pictorial material. Learn Mem 9:99–104.
Pascalis O, Bachevalier J (1999) Neonatal aspiration lesions of the
hippocampal-formation impair visual recognition memory
when assessed by the paired-comparison task but not by delayed
nonmatching-to-sample task. Hippocampus 9:609–616.
Pascalis O, Hunkin N, Holdstock JS, Isaac C, Mayes AR (2004) Visual
paired comparison performance is impaired in a patient with
selective hippocampal lesions and relatively intact item recogni-
tion. Neuropsychologia 24:1293–1300.
Pihlajamaki M, Tanila H, Hanninen T, Kononen M, Mikkonen M,
Jalkanen V, Partanen K, Aronen H, Soininen H (2003) Encoding
of novel picture pairs activates the perirhinal cortex: an fMRI
study. Hippocampus 13:67–80.
Posner MI, Keele SW (1968) On the genesis of abstract ideas. J Exp
Psychol 77:353–363.
Preuessner J, Kohler S, Crane J, Pruessner M, Lord C, Byrne A, Kabani
N, Collins DL, Evans AC (2002) Volumetry of temporopolar,
perirhinal, entorhinal, and parahippocampal cortex from high-
resolution MR images: considering the variability of the collat-
eral sulcus. Cereb Cortex 12:1342–1353.
Ranganath C, Yonelinas AP, Cohen MX, Dy CJ, Tom SM, D’Esposito
MD (2003) Dissociable correlates of recollection and famil-
iarity within the medial temporal lobes. Neuropsychologia
42:2–13.
Reber PJ, Wong EC, Buxton RB (2002) Encoding activity in the
medial temporal lobe examined with anatomically constrainted
fMRI analysis. Hippocampus 12:363–376.
Reber PJ, Gitelman D, Parrish T, Mesulam M (2003) Dissociating
explicit and implicit category knowledge with fMRI. J Cogn
Neurosci 15:574–583.
Reed JM, Squire LR (1997) Impaired recognition memory in patients
with lesions limited to the hippocampal formation. Behav
Neurosci 111:667–675.
Reed JM, Squire LR (1998) Retrograde amnesia for facts and events:
findings from four new cases. J Neurosci 18:3943–3954.
Rees G, Friston K, Koch C (2000) A direct quantitative relationship
between the functional properties of human and macaque V5.
Nat Neurosci 3:716–723.
Rempel-Clower NL, Zola SM, Squire LR, Amaral DG (1996) Three
cases of enduring memory impairment after bilateral damage
limited to the hippocampal formation. J Neurosci 16:5233–5255.
Ribot T (1887) Diseases of memory: an essay in the positive psychology
(Smith WH, translator). New York: Appleton.
Rose M, Haider H, Weiller C, Buchel C (2002) The role of medial
temporal lobe structures in implicit learning: an event-related
fMRI study. Neuron 36:1221–1231.
Ryan L, Nadel L, Keil K, Putnam K, Schnyer D, Trouard T, Moscovitch
M (2001) Hippocampal complex and retrieval of recent and
very remote autobiographical memories: evidence from func-
tional magnetic resonance imaging in neurologically intact
people. Hippocampus 11:707–714.
Schacter DL (1994) Priming and multiple memory systems: percep-
tual mechanisms of implicit memory. In: Memory systems 1994
(Schacter DL, Tulving E, eds), pp 233–268. Cambridge, MA:
MIT Press.
Schacter DL (1999) The seven sins of memory. Am Psychol
54:182–203.
Schacter DL, Wagner AD (1999) Medial temporal lobe activations in
fMRI and PET studies of episodic encoding and retrieval.
Hippocampus 9:7–24.
Schacter DL, Reiman E, Uecker A, Polster M, Yun LS, Cooper L
(1995) Brain regions associated with retrieval of structurally
coherent visual information. Nature 376:587–590.
Schendan H, Searl M, Melrose R, Stern CE (2003) An fMRI study off
the role of the medial temporal lobe in implicit and explicit
sequence learning. Neuron 37:1013–1025.
Scoville WB (1954) The limbic lobe in man. J Neurosurg 11:64–66.
Scoville WB, Milner B (1957) Loss of recent memory after bilateral
hippocampal lesions. J Neurol Neurosurg Psychiatry 20:11–21.
Shallice T, Fletcher PC, Frith CD, Grasby P, Frackowiak R, Dolan RJ
(1994) Brain regions associated with acquisition and retrieval of
verbal episodic memory. Nature 368:633–635.
Shelton A, Gabrieli JD (2002) Neural correlates of encoding space
from route and survey perspectives. J Neurosci 22:2711–2717.
Shimamura AP, Squire LR (1984) Paired-associate learning and
priming effects in amnesia: a neuropsychological study. J Exp
Psychol Gen 113:556–570.
Small S, Arun A, Perera G, De La Paz R, Mayeaux R, Stern Y (2001)
Circuit mechanisms underlying memory encoding and retrieval
in the long axis of the hippocampal formation. Nat Neurosci
4:442–449.
Sperling RA, Bates JF, Cocchiarella AJ, Schacter DL, Rosen BR, Albert
MS (2001) Encoding novel face-name associations: a functional
MRI study. Hum Brain Mapp 14:129–139.
Sperling R, Chua E, Cocchiarella A, Rand-Giovannetti E, Poldrack
RA, Schacter DL, Albert M (2003) Putting names to faces: suc-
cessful encoding of associative memories activates the anterior
hippocampal formation. Neuroimage 20:1400–1410.
Spiers H, Maguire E, Burgess N (2001) Hippocampal amensia.
Neurocase 7:357–382.
Squire LR (1992) Declarative and nondeclarative memory: multiple
brain systems supporting learning and memory. J Cogn Neurosci
4:232–243.
Squire LR, Alvarez P (1995) Retrograde amnesia and memory con-
solidation: a neurobiological perspective. Curr Opin Neurobiol
5:169–177.
Squire LR, Knowlton BJ (1995) Learning about categories in the
absence of memory. Proc Natl Acad Sci USA 92:12470–12474.
Squire LR, Zola-Morgan S (1991) The medial temporal lobe memory
system. Science 253:1380–1386.
Squire LR, Ojemann JG, Miezin FM, Petersen SE, Videen TO, Raichle
ME (1992) Activation of the hippocampus in normal humans: a
functional anatomical study of memory. Proc Natl Acad Sci USA
89:1837–1841.
Squire LR, Clark RE, Knowlton BJ (2001) Retrograde amnesia.
Hippocampus 11:50–55.
Stark CEL, Okado Y (2003) Making memories without trying: medial
temporal lobe activity associated with incidental memory for-
mation during recognition. J Neurosci 23:6748–6753.
Stark CEL, Squire LR (2000a) fMRI activity in the medial temporal
lobe during recognition memory as a function of study-test
interval. Hippocampus 10:329–337.
Stark CEL, Squire LR (2000b) Recognition memory and familiarity
judgments in severe amnesia: no evidence for a contribution of
repetition priming. Behav Neurosci 114:459–467.
Stark CEL, Squire LR (2000c) Functional magnetic resonance imag-
ing (fMRI) activity in the hippocampal region during recogni-
tion memory. J Neurosci 20:7776–7781.

Stark CEL, Squire LR (2001a) Simple and associative recognition
memory in the hippocampal region. Learn Mem 8:190–197.
Stark CEL, Squire LR (2001b) When zero is not zero: the problem of
ambiguous baseline conditions in fMRI. Proc Natl Acad Sci USA
98:12760–12766.
Stark CEL, Squire LR (2003) Hippocampal damage equally impairs
memory for single-items and memory for conjunctions.
Hippocampus 13:281–292.
Stark CEL, Bayley PJ, Squire LR (2002) Recognition memory for sin-
gle items and for associations is similarly impaired following
damage limited to the hippocampal region. Learn Mem
9:238–242.
Stefanacci L, Buffalo EA, Schmolck H, Squire LR (2000) Profound
amnesia after damage to the medial temporal lobe: a neu-
roanatomical and neuropsychological profile of patient E.P. J
Neurosci 20:7024–7036.
Strange BA, Fletcher PC, Henson RNA, Friston KJ, Dolan RJ (1999)
Segregating the functions of human hippocampus. Proc Natl
Acad Sci USA 96:4034–4039.
Strange BA, Otten LJ, Josephs O, Rugg MD, Dolan RJ (2002)
Dissociable human perirhinal, hippocampal, and parahip-
pocampal roles during verbal encoding. J Neurosci 22:523–528.
Sutherland RJ, Rudy JW (1989) Configural association theory: the
role of the hippocampal formation in learning, memory, and
amnesia. Psychobiology 17:129–144.
Suzuki WA, Eichenbaum H (2000) The neurophysiology of memory.
Ann N Y Acad Sci 911:175–191.
Talairach J, Tournoux P (1988) A co-planar stereotaxic atlas of the
human brain (translated). New York: Thieme Medical.
Teng E, Squire LR (1999) Memory for places learned long ago is
intact after hippocampal damage. Nature 400:675–677.
Tulving E (1985) Memory and consciousness. Can J Psychol 26:1–12.
Tulving E, Kapur N, Markowitsch H, Craik FIM, Habib R, Houle S
(1994) Neuroanatomical correlates of retrieval in episoidc
memory: auditory sentence recognition. Proc Natl Acad Sci USA
91:2012–2015.
Vaidya C, Gabrieli JD, Keane M, Monti L (1995) Perceptual and con-
ceptual memory processes in global amnesia. Neuropsychology
9:580–591.
Vargha-Khadem F, Gadian DG, Watkins KE, Connelly A, Van
Praesschen W, Mishkin M (1997) Differential effects of early
hippocampal pathology on episodic and semantic memory.
Science 277:376–380.
Vargha-Khadem F, Gadian DG, Mishkin M (2001) Dissociaitons in
cognitive memory: the syndrome of developmental amnesia.
Philos Trans R Soc Lond B Biol Sci 356:1435–1440.
Vargha-Khadem F, Salamond C, Watkins KE, Friston K, Gadian DG,
Mishkin M (2003) Developmental amnesia: effect of age at
injury. Proc Natl Acad Sci USA 100:10055–10060.
Functional Role of the Human Hippocampus 579
Verfaellie M, Reiss A, Roth H (1995) Knowledge of new English
vocabulary in amnesia: an examination of premorbidly
acquired semantic memory. J Int Neuropsychol Soc 1:443–
452.
Wagner AD, Schacter DL, Rotte M, Koutstaal W, Maril A, Dale AM,
Rosen BR, Buckner RL (1998) Building memories: remember-
ing and forgetting of verbal experiences as predicted by brain
activity. Science 281:1188–1191.
Warrington E, McCarthy G (1988) The fractionation of retrograde
amnesia. Brain Cogn 7:184–200.
Warrington E, Weiskrantz L (1968) New method of testing long-term
retention with special reference to amnesic patients. Nature
217:972–974.
Warrington E, Weiskrantz L (1970) The amnesic syndrome: consoli-
dation or retrieval? Nature 228:628–630.
Warrington E, Weiskrantz L (1974) The effect of prior learning on
subsequent retention in amensic patients. Neuropsychologia
12419–428.
Weiskrantz L, Warrington E (1979) Conditioning in amnesic
patients. Neuropsychologia 17:187–194.
Wirth S, Yanike M, Frank L, Smith A, Brown E, Suzuki WA (2003)
Single neurons in the monkey hippocampus and learning of
new associations. Science 300:1578–1581.
Wixted J, Squire LR (2004) Recall and recognition are equally
impaired in patients with selective hippocampal damage. Cogn
Affect Behav Neurosci 4:58–66.
Yonelinas AP (2002) The nature of recollection and familiarity: a
review of 30 years of research. J Mem Lang 46:441–517.
Yonelinas A, Hopfinger J, Buonocore M, Kroll N, Baynes K (2001)
Hippocampal, parahippocampal and occipital-temporal contri-
butions to associative and item recognition memory: an fMRI
study. Neuroreport 12:359–363.
Yonelinas AP, Kroll NEA, Quamme J, Lazzara MM, Suave M,
Widaman K, Knight RT (2002) Effects of extensive temporal
lobe damage or mild hypoxia on recollection and familiarity.
Nat Neurosci 5:1236–1241.
Zeineh MM, Engel SA, Bookheimer SY (2000) Application of cortical
unfolding techniques to functional MRI of the human hip-
pocampal region. Neuroimage 11:668–683.
Zeineh MM, Engel SA, Thompson PM, Bookheimer SY (2003)
Dynamics of the hippocampus during encoding and retrieval of
face-name pairs. Science 299:577–580.
Zola SM, Squire LR, Teng E, Stefanacci L, Buffalo EA, Clark RE (2000)
Impaired recognition memory in monkeys after damage limited
to the hippocampal region. J Neurosci 20:451–463.
Zola-Morgan S, Squire LR, Amaral DG (1986) Human amnesia and
the medial temporal lobe region: enduring memory impair-
ment following a bilateral lesion limited to field CA1 of the hip-
pocampus. J Neurosci 6:2950–2967.
This page intentionally left blank
13 Á Richard Morris
Theories of Hippocampal Function
Á stability over time, and the subsequent retrieval and reactiva-
13.1 Overview
What does the hippocampal formation do? Does it have one
function or many? Can we understand its functions in isola-
tion from that of other areas of the brain? The theories we
review in this chapter all focus on the idea that it is a special
kind of “memory machine.” The link between the hippocam-
pus and memory is an old idea, having its roots in clinical and
experimental observations on patients who sustained brain
damage that included the hippocampus but all too often
extended into other areas of the medial temporal lobe or mid-
brain (see Chapter 2). Careful study of the small number of
patients with more selective damage (see Chapter 12) and that
of animals given experimental lesions or other interventions
(see later in this chapter) has provided further grounds for
believing that the integrity of the hippocampal formation is
causally involved in implementing some but not all types of
memory. Given that multiple “types” of memory have now
been proposed—such as declarative, spatial, and episodic
memory—the focus of this chapter is on identifying which of
these or other putative types of memory are mediated by the
hippocampal formation. Success in this task requires that we
recognize that the hippocampus operates in conjunction with
other networks and brain structures, including neuromodula-
tory afferent systems emanating from the midbrain and
diverse regions of the neocortex. Any claim that the integrity
of the hippocampus is required for one or another type of
memory does not, therefore, mean that other brain areas are
not involved as well. Discussion of types of memory and their
anatomical substrate carries us forward to consider the logi-
cally distinct issue of the various “phases” of memory process-
ing in which the hippocampus may be engaged—the
encoding of information into memory, the storage of memory
traces as biochemical and structural changes in the nervous
system, the consolidation processes responsible for their
tion of hippocampus-dependent memories during recall.
Newer research techniques, such as functional magnetic reso-
nance imaging (fMRI) in humans and invasive techniques
with animals such as ensemble single-cell recording and selec-
tive neuropharmacological perturbations, all point to a role
for specific aspects of hippocampal neural activity in succes-
sive phases of memory processing.
Two theories have dominated research on hippocampal
function over the past quarter century. The first, discussed in
Section 13.3, is that it is involved in the formation of memories
for everyday facts and events that can be consciously recalled—
collectively called declarative memory (Squire, 1992; Squire et
al., 2004). Complementing the human studies of the preceding
chapter, a particular focus has been the development of a pri-
mate model of amnesia built largely on studies of recognition
memory. Other behavioral tasks have also been examined, but
they have been secondary to testing a key issue: whether the
hippocampus is involved in recognizing stimuli that have been
presented before. The other major theory (see Section 13.4),
emerging from observations first made during the recording of
single-cell activity in freely moving rodents (see Chapter 11), is
the idea that it is involved in spatial memory and, more specif-
ically, the formation of cognitive maps and their use in naviga-
tion through space (O’Keefe and Nadel, 1978; Burgess et al.,
2002). This theory makes testable predictions about the cell
activity that occurs during spatial navigation and the impact of
hippocampal dysfunction on spatial memory. Although enor-
mously influential, and often at loggerheads with each other, it
will become clear that neither of these two major theories
offers a fully satisfactory account of the available data.
Section 13.5 discusses a range of alternative theories, par-
ticularly those built around how memory systems handle
ambiguity, associative-relations, and context. In 1984,
Schmajuk identified more than 20 psychological theories of
hippocampal function (Schmajuk, 1984). There have been yet
581
582 The Hippocampus Book
others since. Here we focus on certain particularly influential
ideas of the past 25 years, namely that the hippocampus
implements a learning process involving the formation of
stimulus configurations (Sutherland and Rudy, 1989a; O’Reilly
and Rudy, 2001), that it is involved in the relational processing
of stimuli in a manner that enables flexible retrieval (Cohen
and Eichenbaum, 1993; Eichenbaum and Cohen, 2001), or
that it provides the substrate for encoding and retrieval of
information in relation to the spatiotemporal context in which
that information occurs. Contexts, relations, and configura-
tions may each help disambiguate conflicting associative rela-
tions in which a specific stimulus occurs (Hirsh, 1974; Good
and Honey, 1991). Although perhaps less prominent than the
two major theories, these ideas have been important in
expanding our concepts of what memory is all about and its
mediation by diverse structures in the brain.
The present focus on memory is, arguably, too exclusive.
The hippocampus has also been implicated in a range of other
brain functions, including behavioural inhibition and anxiety
(Kimble, 1968; Gray, 1982; Davidson and Jarrard, 2004), sen-
sorimotor function (Vanderwolf and Cain, 1994), and acting
as a comparator to detect novelty (Gray, 2000). The presence
in the hippocampus of a high density of receptors for adreno-
cortical hormones such as cortisol and corticosterone has also
implicated the hippocampus in the cognitive regulation of
stress and the hypothalamic-pituitary axis (Sapolsky, 1985;
McEwen and Sapolsky, 1995; de Kloet et al., 1999). For clarity,
these issues are considered separately in Chapter 15.
Certain concepts relevant to identifying hippocampal
function in relation to memory are excluded from detailed
discussion in this chapter because it seemed most appropriate
to outline them elsewhere. They include the widely discussed
idea that activity-dependent synaptic plasticity, such as long-
term potentiation (LTP) and long-term depression (LTD),
might play a role in learning and memory (Martin et al., 2000;
Bliss et al., 2004). This important topic is discussed at the end
of Chapter 10. Hebb’s concepts of the “cell assembly” and
“phase sequence” (Hebb, 1949) and Marr’s proposal that dis-
tributed associative memories could be implemented by hip-
pocampal local circuitry (Marr, 1971) were both touched on
in our historical overview (see Chapter 2) and are developed
more fully in Chapter 14. Neural network modeling is a field
in its own right that has matured to a level of conceptual and
mathematical precision that is beyond the scope of this chap-
ter (Willshaw and Buckingham, 1990; O’Reilly, 1998; Rolls
and Treves, 1998; Lisman and Zhabotinsky, 2001; O’Reilly and
Norman, 2002).
The chapter concludes by zeroing in on the idea that neu-
ral activity in the hippocampal formation contributes to
episodic memory (Tulving, 1983; Aggleton and Brown, 1999;
Redish, 1999). This contribution has been variously charac-
terized as remembering the “scenes” in which events take place
(Gaffan, 1994a), or as processing the “automatic” components
of recording and recalling experience that are downstream of
neocortical attentional systems (Morris and Frey, 1997). In
animals, where the phenomenological aspects of memory are
so difficult to capture (Hampton and Schwartz, 2004), this
is now often referred to as “episodic-like” memory (Clayton
and Dickinson, 1998). Although also applied elsewhere, it is
in this section that we see the greatest implementation of
new research technologies, such as novel behavioral tasks,
reversible pharmacological manipulations, and region-specific
gene targeting. This new way of thinking about hippocampal
function also draws more explicitly on the anatomical, physi-
ological, and cell biological ideas discussed in earlier chapters
of this book.
In each section, the main concepts of each theory are out-
lined followed by an examination of exemplar experimental
work that supports or conflicts with it and then a general cri-
tique. Given the scale of the literature, no attempt has been
made to be definitively inclusive, and the choice of studies
cited is inevitably subjective. When possible, an attempt is
made to refer to ideas about hippocampal function that have
emerged from the principles discussed in earlier chapters.
Understanding the hippocampal memory machine is an ambi-
tious exercise that requires integrating “top-down” psycholog-
ical concepts about the distinct types and processing phases of
memory with “bottom-up” physiological and cell biological
ideas about the local networks, cells, and signal transduction
pathways that constitute this beautiful brain structure.
Á
13.2 Cognitive and Behavioral
Neuroscience, Interventional
Techniques, and the Hippocampus
13.2.1 Value of Interventional
Studies to Identify Function
If you want to learn how a machine works, it is a good idea to
watch it working. An appropriate next step is to look at selec-
tive interventions that disrupt one part of the machine or
another and study the various changes in the machine’s oper-
ation that then occur. In the case of a modular organ such as
the brain, it is essential to complement studies that involve its
cells being cut up, put under microscopes, poked with elec-
trodes, or bathed in drugs (see Chapters 3–11) with studies of
selective interventions that may affect its behavior. This is part
of what cognitive and behavioral neuroscientists do to try to
understand the functions of the distinct but interacting
regions of the brain. Lesion studies in animals, on which we
largely focus in this chapter, have been the classic “interven-
tionist” technique for testing theories of function. They have
their methodological and interpretative problems, as do phar-
macological and genetic techniques. Intervention is, however,
the only way to secure definitive information about whether
the integrity of a particular anatomical region of the brain or
specific physiological and cell biological processes in it is nec-
essary for a particular function.
In humans, damage to the brain may arise because of a
tumor, a stroke, or a viral infection, or it may be incurred

through neurosurgery undertaken to prevent further brain
damage (e.g., the surgical management of epilepsy). A few
striking and analytically valuable exceptions aside, lesions that
arise from these causes are rarely restricted to anatomically
circumscribed areas of the brain; nor when they are is the
damage incurred ever complete (see Chapter 12). In animal
studies, in contrast, specific brain regions can be subject to
selective lesions to evaluate whether one or more functions
can still be carried out normally. Postmortem histological
techniques complement the behavioral observations, provid-
ing a vital, relatively immediate check on an experimentalist’s
intentions with respect to the site and size of the lesion. It is
also now possible to use histologically calibrated structural
MRI techniques in vivo to establish the accuracy and extent of
damage before extensive behavioral testing is undertaken
(Malkova et al., 2001).
The lesion method is, of course, used widely in biology, the
modern variants of it being pharmacological techniques
directed at specific receptors or targeted molecular engineer-
ing to investigate gene function (“gene knockouts”). The com-
mon analytical principle underlying such approaches is that
the study of dysfunction offers insights into normal function.
Creating dysfunction does not necessarily require invasive
interventions—the study of visual illusions being a case in
point—but the neurological approach, as twentieth century
neurologist Sir Henry Head put it so well, is one in which “it
should not be the injury that captures our attention but how,
through injury or disease, normal function is laid bare.”
There are, however, problems strewn along the path of lay-
ing things bare. In some cases it is the pitfalls of the approach
that lead an investigator astray. First, a brain reacts to perma-
nent damage with repair mechanisms—cerebrovascular, neu-
ral, glial—that collectively result in dynamic changes from the
moment the damage occurs. It may be some while before
brain function restabilizes after surgically induced diaschisis
(Stein et al., 1983; Finger et al., 2004). Second, even when it
does restabilize, we should not assume that adjacent undam-
aged brain areas necessarily function normally, as the neural
activity carried on input pathways may have been affected by
the nearby damage. Lesions can also induce structural
changes, such as axonal growth and synaptogenesis (see
Chapter 9). This “sprouting” can be compensatory and thus
homeostatic, or it may give rise to abnormal circuitry not
usually seen in the normal brain (Lynch et al., 1973; Steward
et al., 1974). Third, even if postlesion reorganization achieves
relatively little in the way of compensation for lost tissue,
the person or animal may still alter his behavior in a manner
that enables him to compensate for lost function through
the use of undamaged systems. Behavioral compensation can
help mask a neurological deficit. With respect to damage
to the medial temporal lobe and the ensuing amnesia, for
example, the use of memory aids such as electronic devices to
remind people to do things is one example of such “compen-
satory” behavior (Wilson, 1999). Desirable as these aids
are therapeutically, the possibility that behavioral compensa-
tion occurs naturally over the course of time adds difficulties
Theories of Hippocampal Function 583
to the neuroscientist’s efforts to interpret lesion effects in
relation to the functional theories they are intended to test.
Notwithstanding these obstacles along the path, interven-
tional techniques can be used successfully, provided a number
of conceptual issues are also considered.
13.2.2 Lesions, Functional
Hypotheses, and Behavioral Tasks
Dissociations, Double
Dissociations, and Hypotheses
Experimental lesions are widely used to establish functional
dissociations between tasks that are thought to depend on dis-
tinct forms of information processing. Thus, if behavior A is
impaired by a lesion made in area X and, simultaneously,
behavior B is unaffected by lesion X, there are grounds for sus-
pecting that behaviors A and B may, at least in part, depend on
different brain circuits. The confidence in such assertions
grows when a so-called double dissociation is obtained—
when a different lesion Y impairs behavior B but not behavior
A (Teuber, 1955). However, discovering a list of behaviors that
are affected or unaffected by a particular set of lesions (or
other intervention) is only the first step toward such a theory.
An essential parallel step is the development of hypotheses
about information processing in the brain (Shallice, 1988).
Numerous hypotheses have been developed by cognitive
neuroscientists to account for the cerebral organization
of perceptual processes, attention, language processing, exec-
utive function. and memory (Shallice, 1988; McCarthy and
Warrington, 1990; Gazzaniga, 2002). With respect to the hip-
pocampal formation, numerous hypotheses have been devel-
oped and are discussed in this chapter. As we saw in the
overview, these theories differ in their claims about the char-
acter and timing of the information processing that is taking
place in the hippocampal formation during learning and
memory.
Learning, Performance, and Hippocampus-dependent
Learning and Memory Tasks
One important theme in studies of learning and memory is
the distinction between learning and performance. In a behav-
ing animal, performance is what we can see and measure,
whereas learning reflects a variety of covert processes that we
presume are taking place in the brain. It is these covert
processes that are responsible for the acquisition of knowl-
edge. Performance is the expression of this knowledge. Of
course, performance may also itself be learned, as occurs with
sensorimotor skills, but it is often a reflection of learning
rather than being a participant in such a process. Obstructing
performance in some way (e.g., by putting a barrier in a maze
that an animal has been trained to run through) may prevent
the usual expression of learned performance, but it does not
affect what the animal “knows.” The animal generally tries to
find an alternative way in which to express this knowledge,
584 The Hippocampus Book
such as its memory of where some desired food is located
(Tolman, 1948).
The concept of learning itself must also be fractionated as
the brain has evolved to have a number of qualitatively dis-
tinct forms of learning. Most theories of hippocampal func-
tion are statements about the type (or types) of learning and
memory in which this group of structures is engaged. There is
passionate debate about how to characterize them. Over the
years, a range of learning tasks have been developed that are
now widely used in studies with rodents, primates, and birds
to try to identify which of them involve hippocampus-
dependent processes. Widely used tasks for rats and mice
include alternation protocols in T mazes and cross mazes, for-
aging for food in radial mazes, and other spatial tasks such as
finding the way home in an open arena or swimming to safety
in a watermaze. Recognition memory can be studied by pre-
senting novel and familiar objects or by taking advantage
of rats’ superb olfactory sense and then “asking” which object
or smell is familiar. One widely used protocol is the condi-
tioning of fear to the context where testing takes place, and
another examines the social transmission of food preferences.
Certain protocols of some of these tasks seem to be “hip-
pocampus-dependent” (i.e., affected by hippocampal lesions),
whereas others are not. The community that conducts
research on temporal lobe function in primates has its family
of tried and trusted tasks for investigating memory ranging
from manual protocols with three-dimensional objects to
automated touch-screen technology, some of which has now
seen application as neuropsychological tests for humans
(Robbins et al., 1994). Birds are also used extensively, with due
consideration to their natural behavioral ecology. Innovation
has gone into the design and development of rigorous labora-
tory-based tasks such as food caching and recovery (Suzuki
and Clayton, 2000). In using this range of tasks, there is ongo-
ing debate about the relative value of standardization on a
subset of tasks that are then used in a common way within
and across laboratories and of innovation with respect to
procedure in the search for better ways of characterizing
hippocampus-dependent learning and memory.
One complication is that there is rarely, if ever, a simple
one-to-one mapping between a specific task and a single puta-
tive type of learning. All of them rely on effective sensory/per-
ceptual processing in “upstream” brain structures and on
effective motor processing “downstream.” Researchers have
then to be careful that any alteration in task performance
caused by interventions is really due to an interruption of a
neural process mediated by the hippocampus or to some sec-
ondary effect on upstream or downstream processes. The con-
ceptual boundary between input, cognition, and output is
hazy, particularly as it becomes ever more apparent that “bot-
tom-up” processes can be modulated by “top-down” mecha-
nisms. Indeed some researchers of a behaviorist persuasion
even question whether the conceptual boundary between
learning and performance exists, in part because an animal
may appear to be unable to learn when, in practice, a treat-
ment has been given that has affected only its ability to sense
afferent stimuli or to move around effectively (Vanderwolf
and Cain, 1994).
What, How, and When: Different
Aspects of Hippocampus-dependent Memory
The overall aim of the functional enterprise is to find out what
the hippocampus does and how it does it. Identifying which
tasks are hippocampus-dependent and which are not is an
important part of this endeavor, particularly when they shed
analytical light on the relative merits of different theories.
There are, however, a number of ways in which learning
and memory tasks can be theoretically ambiguous. The first
has to do with what an animal learns. In a simple T maze, a rat
might be taught to run from a start box in the stem and then
turn left to find food at the end of the arm—but has it learned
the motor response of turning left? Or that food is to be found
near landmarks positioned on the left? One way to find out is
to rotate the maze through 180 and now start the animal
from the stem of the ⬜ maze. If it has learned the motor
response of turning left, it should move toward what had been
the right-hand side of the maze during training. Conversely, if
it has learned to approach landmarks on the left-hand side of
the room, it should now turn right at the choice point. This
manipulation nicely dissociates control and hippocampus-
lesioned rats early in training; but, interestingly, if training
continues too long, normal rats develop a “turning habit” that
is inflexible to manipulations of the starting location (Packard
and McGaugh, 1996). Another example of ambiguity about
what has been learned concerns recognition memory.
Recognition can be defined as the memory of past occurrence
(Brown and Xiang, 1998), but there are several ways it can be
achieved. For example, recognition of a past stimulus by a
monkey may be mediated by a feeling of familiarity evoked by
the stimulus itself. Alternatively, the stimulus may evoke a
covert mental recollection of past experience. The monkey
cannot tell us about either of these possibilities in the way that
a person would through language, but its sense of familiarity
or its sense of recollection is expressed as appropriate choice
behavior in a memory test (performance). A current challenge
is trying to find a way of distinguishing these two possibilities
of what the animal has learned.
The second kind of ambiguity arises if two or more theories
offer different accounts of how an animal has learned a partic-
ular task. In this case, there is agreement about what the
mouse, monkey, or human has learned but not how it has been
achieved. Allocentric spatial learning is a relevant case in point,
as there is widespread agreement that animals can learn where
things are and that the hippocampus is involved but disagree-
ment about how it is done. Questions include whether there is
anything special about spatial learning that might require spe-
cialized neural circuitry, or is it just an instance of a more gen-
eral form of declarative memory? Are different learning rules
involved for spatial learning than other types of learning? Do
different brain areas have to interact for spatial learning than
for other forms of declarative or relational learning?

A third ambiguity arises from the distinction between types
of learning and memory processes. The classic concept of
a hippocampus-dependent learning task is one that is
impaired by permanent hippocampal lesions. An unambigu-
ous hippocampus-dependent learning task is one that is
affected by such lesions but unaffected by others. However,
this categorization does not capture the subtlety that neural
activity in the hippocampus may be required at one stage of
learning but not at another. It might, for example, be required
at encoding but not at retrieval or during the initial stages of
memory storage but not during consolidation. The logically
orthogonal concept of a memory “process” refers to impor-
tant distinctions that must be made between the distinct
memory processes associated with encoding, storage, consoli-
dation, and retrieval. It is orthogonal, as each of these
processes applies to many types of learning and memory.
They are distinguished experimentally through the various
hemodynamic, physiological, or cell biological correlates that
are seen in association with these various phases, by the
impact of lesions made at different stages of training, and by
the effects of reversible interventions (such as drugs) used to
inactivate brain areas temporarily or to disrupt specific neural
mechanisms.
Thus, the overarching task of dissociating theories of
regional brain function involves zeroing in on what, how,
and when. Many subtle variations in behavioral protocol
are designed to explore the hypotheses. These variations
of protocol are guided by predictions, better still by the coun-
terintuitive predictions, of functional hypotheses about
structure–function relations (Crabbe and Morris, 2004). The
declarative memory, cognitive mapping, and predictive ambi-
guity theories outlined below often make common predic-
tions about standardized versions of specific tasks but distinct
predictions about their variants.
Cerebral Localization or Dissociation
Between Mental Processes?
A frequent non sequitur in the interpretation of an interven-
tional study is the assertion, following a positive experimental
finding, that a function can somehow be “localized” to the tar-
geted brain structure or pathway. A classic and amusing cri-
tique of this “localizational” way of thinking was presented by
Gregory (1961), who pointed out that if removing a part of an
old valve radio causes the radio to howl it need not mean that
the “function” of this bit is to suppress howling. This overly
“symptomatic” way of thinking pervaded the hippocampal
field early on and indeed much of the brain sciences since the
early days of experimental neurology. One early claim about
hippocampal function was that it is involved in “behavioral
inhibition” (Kimble, 1968). This claim rested, in part, on the
observation that hippocampus-lesioned animals are very
inflexible in their behavior; they often persist much longer in
carrying out well learned habits in inappropriate situations
than do control subjects. Analysis of behavior in runways,
mazes, and operant tasks has revealed, however, a number of
Theories of Hippocampal Function 585
reasons why behavioral inflexibility could arise from other
than excision of neural machinery that had (somehow)
evolved to inhibit inflexibility. For example, if the lesioned
animal has a deficit in spatial memory and so cannot learn or
remember that one end of a runway (A) is in a different
“place” than the other end of a runway (B), it cannot acquire
knowledge about their relative location. It learns only to run.
Changing a spatial performance strategy is much easier for
normal animals than the “slamming on the brakes” to con-
strain a habitual running response (Nadel et al., 1975). The
debate about the relative merits of these specific views has
largely subsided, although the inhibitory view still has its fol-
lowers (Chan et al., 2001; Davidson and Jarrard, 2004), but the
logical point remains that the overt phenotype of response
stereotypy need not imply the existence of an inhibition mod-
ule localized in the hippocampus or, indeed, anywhere else in
the brain.
Gregory’s (1961) critique should, however, not be over-
stated. If one were to lesion a car’s exhaust system, it would be
right to infer that its function is to make the engine quieter.
Damaging the earphones of an iPOD limits its ability make a
sound. By the same token, in a modular structure such as the
brain, inferences about localization may often be correct. If
the overarching aim of the enterprise is to develop a theory of
what different parts of the brain do and how they do it, we
must also consider how brain areas interact to realize the
seamless control of all aspects of behavior. It is an article of
faith in the field that different brain areas or neuromodulatory
transmitter systems must have different and therefore, at least
in principle, dissociable functions.
Equally, however, it is also understood that no brain area is
an island; different brain areas work together in networks. In
recent years, there has been growing interest in the issue of
how memory systems interact—when they compete, when
they complement each other, when the interruption of one
enables another to learn faster or slower, and so on (White and
McDonald, 2002; McDonald et al., 2004; Voermans et al.,
2004). This change of emphasis is refreshing, as it counterbal-
ances the near-exclusive stress on “dissociations” of earlier
years. A desire to understand how these systems interact is tak-
ing its place as the brain sciences mature. In any event,
whether the immediate aim is functional dissociation or the
understanding of competition and/or complementarity, data
secured using lesions and other interventions are a powerful
way of testing theories.
13.2.3 Contemporary Lesion Techniques:
Pharmacological and Genetic Interventions
Neurotoxic Lesions
The technique by which a lesion is made can be important
because certain techniques, such as the use of excitotoxins,
first introduced to study hippocampal function by Jarrard
(1989), have the effect of damaging cells while leaving fibers of
passage intact (Fig. 13–1). Thus, if a brain function critically
586 The Hippocampus Book
A. Topological arrangement of HPC Formation
C. Quantitative assessment of partial lesion volume
sham partial
lesion hippocampal hippocampal
50 lesion
% of hippocampus spared
40
30
20
10
0
al
al
ra
ra
pt
pt
po
po
se
se
m
te
te
Figure 13–1. Excitotoxic and knife-cut lesions of the hippocampal
formation in rats. A. Topological arrangement of the structures
comprising the hippocampal formation (from Chapter 3, Figure
3–1). B. Photomicrographs of horizontal sections from a represen-
tative control rat (B1) and rats with neurotoxin lesions of hip-
pocampus (2), subiculum (3), and presubiculum/parasubiculum
(4). EC, entorhinal cortex; HIP, hippocampus; PaS, parasubiculum;
PrS, presubiculum; Sub, subiculum. (Source: Courtesy of Len
Jarrard.) C. Use of three-dimensional volumetric techniques to
depends on communication from area X to area Z, with fibers
passing through area Y, old-fashioned lesions might lead to
the false conclusion that area Y is essential for this function
whereas the modern approach would not. By the same token,
a positive outcome of a lesion study using an excitotoxin gives
comfort to the notion that it is the targeted neurons that are
functionally significant for the task under examination.
Conversely, lesions that cut fibers while leaving cells intact can
also be valuable, especially when the axons of a population of
neurons have axon collaterals. Neurotoxic lesions can be
highly localized (e.g., to distinct components of the hip-
pocampal formation such as the dentate gyrus, subiculum, or
entorhinal cortex). They may even extend to killing specific
cell types in the region to which the toxin is delivered, such as
the use of immunoglobulin G (IgG) saporin to ablate cholin-
B. Selective neurotoxic and knife-cut
lesions in rats
1 2
3 4
complete
lesion
5
al
l
ra
pt
po
se
m
te
establish the quantitative extent of the cell loss associated with par-
tial lesions placed at different points along the septotemporal axis.
(Source: Courtesy of Livia de Hoz.) D. Use of a knife cut through
the longitudinal-association pathway in area CA3. This cuts fibers
(not shown) and damages a small number of cells in CA3 at one
particular point along the axis. Animals to which the same amount
of neurotoxic damage was deliberately applied without damaging
the longitudinal fibers are used as controls in such a study. (Source:
Courtesy of Hill-Aina Steffenach.)
ergic cells afferent to the hippocampus while leaving nearby
-aminobutyric acid (GABA)ergic neurons intact (Gallagher
and Rapp, 1997). It has been claimed that lesions restricted to
area CA1 but not CA3, or vice versa, can be achieved with pre-
cise stereotaxic techniques (Lee and Kesner, 2004). This preci-
sion can certainly be achieved along a small length of the
longitudinal axis of the hippocampus, but it remains unclear
how complete and yet still selective such lesions can be.
Lesions that cut fibers while leaving cells intact can also be
valuable, especially when the axons of a population of neu-
rons have axon collaterals, but it is extremely difficult to do
this in a manner that interferes with one and only one fiber
pathway.
Techniques for making permanent lesions are accompa-
nied, when testing is complete, by histological analysis of the

damage caused. Was the target brain area damaged as
expected? Was the lesion complete? Did it encroach on other
brain areas? Did it damage cells and fibers or only cells? If the
latter, was confirmation obtained through tract tracing tech-
niques that the apparently intact fibers are capable of axonal
transport? Doing this is part of the point of using animals
rather than just studying people. Nissl stains, silver stains that
reveal degenerating fibers, and other histochemical techniques
can all be used to good effect. More ambitiously, one might
enquire whether an apparently intact fiber tract at or near to a
lesion site is electrophysiologically functional. Likewise, the
check might be on whether the lesion has selectively affected
one but not another neuromodulatory system by using in vivo
neurochemistry to monitor changes of particular neurotrans-
mitters. The use of all of these techniques for postexperiment
histology is the “gold standard” to which most laboratories
aspire, but a difficulty shared by all is that it is not always prac-
tical to dot every “i” and cross every “t” in every experiment.
Much has to be taken on trust so research can proceed at a rea-
sonable pace—but this trust is sometimes misplaced, which
sows seeds of confusion. As will become clear, the size of a
lesion can be a major factor to consider when interpreting
experiments on hippocampal function in relation to recogni-
tion memory or spatial memory.
Permanent lesions are often used to realize network “dis-
connections” (Aggleton and Pearce, 2001). Most lesion studies
in animals involve bilateral damage to a brain structure, but,
in an ingenious twist, a unilateral lesion of one structure can
be made in conjunction with contralateral damage to another.
In right-handed people language is largely mediated by struc-
tures in the left hemisphere such that a stroke in the left brain
can interfere with speech, whereas one on the right does
so rarely (McCarthy and Warrington, 1990). Conversely,
right-hemisphere damage to the parietal cortex is associated
with the fascinating syndrome of spatial neglect (Behrmann et
al., 2004; Farah, 2004). Laterality is barely present in animals
(one exception being song birds), such that unilateral lesions
in animals are often behaviorally benign. Thinking is shifting
from a focus on localization toward a “crossed-lesion”
approach to reveal the importance of functional connectivity.
For example, a unilateral lesion might be made to the entorhi-
nal cortex (on the left) and a second unilateral lesion to the
fimbria-fornix (on the right) with, perhaps, section of the hip-
pocampal commissures to prevent cross-talk of information
within the hippocampus via residual interhemispheric con-
nectivity. Such a lesion might produce a syndrome similar to
that of an entire hippocampal lesion (even though the hip-
pocampus and dentate gyrus are actually intact, and neither
brain area that is lesioned has a bilateral lesion) because it has
successfully disconnected cortical and subcortical structures
that normally interact via the hippocampus. The “disconnec-
tion” approach is increasingly popular (Warburton et al.,
2001; Gaffan, 2002; Miyashita, 2004). Disconnection of the
hippocampus need not always be achieved by damaging affer-
ent or efferent structures in opposite hemispheres. It can also
be done by sectioning major interconnecting pathways, such
Theories of Hippocampal Function 587
as the fimbria-fornix, or by selective knife cuts of intrinsic cir-
cuitry such as the longitudinal-associational pathway of area
CA3 (Steffenach et al., 2002).
Pharmacological and Genetic Manipulations
Drugs and genetic manipulations are alternative ways of inter-
vening in brain function. The turn of the twenty-first century
is something of a methodological “tipping point” (Gladwell,
2002) as the major behavioral neuroscience laboratories move
on from deploying classic lesions on their own to using them
in conjunction with single-unit recording and tract-tracing
techniques, together with reversible pharmacological and cell
biological manipulations. With respect to the opportunities
afforded by targeted molecular engineering Wulff and Wisden
quoted Francis Crick’s view that “the lesion approach needs
new methods, especially as the usual ways of ablating the parts
of the brain are so crude. For example, it would be useful to
inactivate, preferably reversibly, a single type of neuron in a
single area of the brain” (Wulff and Wisden, 2005, p. 44). As it
happens, this would not be very useful for behavioral studies,
as a certain minimum amount of tissue must be affected, but
the point is taken nonetheless.
The use of drugs in conjunction with behavioral studies
has a long scientific history (McGaugh, 2000). One aspect
of this approach has been the analytical use of intracerebral
infusions of microiter and nanoliter quantities of active
compounds targeted to specific regions, often (although
not always) given with posttraining to avoid the impact of
known side effects. In relation to hippocampal function, typi-
cal drugs used include excitatory amino acid antagonists
acting at -amino-3-hydroxy-5-methyl-4-isoxazolepropi-
onate (AMPA), N-methyl-D-aspartate (NMDA), and
metabotropic glutamate (mGluR) receptors, and ligands at
inhibitory synapses such as GABAergic agonists (as described
in Chapter 6). These compounds interfere with (or potenti-
ate) normal excitatory or inhibitory neurotransmission for as
long as the drug remains present and active at its site of
administration. Drugs interacting with the synthesis, trans-
port, or degradation of neuromodulatory transmitters—
acetylcholine (ACh), norepinephrine (NE), dopamine (DA),
and serotonin (5-HT), for example—or antagonists of their
target receptors are also analytically powerful. There is
growing interest in drugs acting intracellularly on signal-
transduction pathways downstream of postsynaptic receptors,
such as CAMKII and MAPkinase (see Chapters 6, 7, and 10).
Their use requires compounds capable of penetrating cell
membranes to reach their site of action rather than acting
extracellularly at membrane-bound receptors (Sweatt, 2003).
The exploitation of drug actions in the hippocampal for-
mation as a tool to study learning and memory function is a
relatively new departure. One aspect capitalizes on the devel-
opments in our understanding of synaptic function discussed
earlier in the book. An AMPA receptor antagonist, such as
CNQX, shuts down normal fast synaptic transmission for a
period of time (see Chapter 6). In principle, its effects should
588 The Hippocampus Book
be similar to that of a neurotoxic lesion that spares fibers of
passage—with the analytical advantage of reversibility. In
contrast, an NMDA antagonist such as D-AP5 has the distinct
effect of blocking NMDA receptor-dependent mechanisms,
such as activity-dependent synaptic plasticity (LTP and LTD),
while leaving fast synaptic transmission intact. This should
and does have very different effects on cognition than any per-
manent lesion could have, allowing previously acquired spa-
tial memory to be displayed in performance but preventing
new learning (Morris et al., 1986b). Interestingly, the effects of
an NMDA antagonist such as D-AP5 are “regionally depend-
ent” because NMDA receptors mediate different physiological
functions in different brain areas. In the spinal cord, NMDA
receptors help mediate suprasegmental connectivity, partly by
enabling the rhythmical activation of Ca2-dependent K
currents (Dale and Roberts, 1985; Grillner et al., 1998),
whereas in the hippocampus they serve as a trigger for certain
forms of neuronal plasticity (see Chapter 10).
What virtually all drugs have in common and what makes
them so useful compared to lesions is that their actions are, at
least in principle, reversible. Reversibility has two main rami-
fications. First, the intended dysfunction lasts only a short
time, and thus the chances of brain damage or compensatory
changes are reduced. Second, reversibility makes it possible to
dissect different phases of memory—encoding, storage, con-
solidation, retrieval—in a more exacting way. However, no less
than lesions, the impact of drugs must be studied with care.
Drugs diffuse, raising questions about variation in concentra-
tion across a target brain region. This variation may itself vary
with time, particularly with acute injections. Osmotic
minipumps enable drugs to be infused at a steady rate over
periods ranging from 1 day to 1 month. A steady-state con-
centration is then achieved when the rates of infusion and dif-
fusion of a drug match precisely—a great advance on acute
infusions. Last and by no means least, it is essential to establish
precisely what and where a drug has had its effect. It is aston-
ishing how rarely this is done independently of the effects on
behavior that the drug may have. The use of autoradiography
to establish the local spread of action of a radiolabeled version
of a drug is valuable (Steele and Morris, 1999; Attwell et al.,
2001), but this is a surrogate for an independent marker of its
functional effect. If an AMPA antagonist is used, physiology
should be undertaken to establish the extent to which synap-
tic transmission has been compromised. Monitoring glucose
metabolism is one way of looking at this and direct physiolog-
ical recording another (Riedel et al., 1999).
The last 15 years have witnessed considerable excitement
regarding the introduction of transgenic and knockout mice
as in vivo methods of genetic manipulation. Their application
to study the systems-level and cellular mechanisms of learning
and memory has been particularly imaginative (Tonegawa et
al., 1995). Beginning with pioneering studies that targeted
molecules such as CAMKII and fyn tyrosine kinase, which
were implicated in LTP (Grant et al., 1992; Silva et al., 1992),
an “industrial production line” of transgenic and mutant mice
soon became available in virtually all fields of neuroscience
(Morris and Kennedy, 1992). Knockouts are now (almost)
routine, with many a PhD student in well-founded molecular-
genetic laboratories saying that they “would like to make a
mouse” during the course of their studies. The development
of these techniques has been accompanied by databases and
repositories of mouse lines, such as those at the Jackson
Laboratory in the United States (http://www.jax.org/lab), the
development of testing facilities such as those at the Mary
Lyon Centre in the United Kingdom (http://www.mlc.har.
mrc.ac.uk/aboutUs.html), and meetings at which ideas and
recommendations about breeding different lines of mice are
debated in an open, constructive manner (Conference, 1997).
Neuroinformatics expertise is having a considerable impact
on the field (http://www.neuroinf.de/Members/stefan/mbl).
This molecular-genetic approach has now advanced from
homologous recombination to regionally specific manipula-
tions, using the cre-LOX procedure, enabling studies to focus
on the role of genes in specific brain areas such as the hip-
pocampus (Tsien et al., 1996). The introduction of inducible
constructs, such as tetracycline, has taken the mouse knockout
enterprise into a third phase of sophistication, potentially
allowing the inactivation of specific genes at specific times in
specific cells of specific regions of the hippocampal formation
or other areas (Mayford et al., 1996). For all this enviable
specificity, however, there remain formidable technical prob-
lems to be solved—not least the discovery of region-specific
promoters and the apparent “leakiness” of inducible manipu-
lations. Another problem is that most studies have to be con-
ducted on mice. Many molecular behaviorists look on mice
as “miniature rats,” but this is clearly absurd. Rats are different.
A rarely tested assumption in the field is that hippocam-
pus-dependent tasks for rats may be used similarly in mice.
Other technical issues have to do with running appropriate
controls for knockout mice made using homologous recom-
bination (Gerlai, 1996), pleiotropic developmental abnor-
malities associated with deletion of a gene throughout
embryogenesis and postnatal life (Tonegawa et al., 1995), and
issues associated with the genetic or pharmacological rescue
of a normal phenotype by transgenic expression of a gene in a
mouse with a knockout background (Ohno, 2001). Others are
longstanding conceptual problems such as the need, so rarely
addressed, for the double dissociations that are common in
conventional lesion studies. Fortunately, these problems are
being recognized, and the solutions are forthcoming. The
technological innovations are remarkable; they are continu-
ing, and the molecular-genetic approach has a great deal to
offer.
13.2.4 Biological Continuity
of Hippocampal Function
Virtues of a Comparative Approach
All mammals possess a brain with the same basic pattern
of organization: spinal cord, hindbrain, midbrain, forebrain,
and cerebral hemispheres (Swanson, 2000). The simplest view

Figure 13–2. Genomic sequence similarity in
mammals. Publication of the genomic sequence of
the rat brain, following those of the mouse and of
humans, provided an opportunity to assess the
degree of sequence “similarity” across species. The
outcome places rodents on a line less distant from
primates and humans than that of many other
mammalian species. (Source: Consortium, 2004.)
of this common pattern is that although there has been
substantial encephalization of function in primates and
humans relative to rodents the core functions of various
areas of the central nervous system (CNS)—cerebellum,
midbrain, neocortex—are most likely conserved across all
mammalian species. Even where anatomical similarities
may seem cryptic, there are striking genetic similarities
that justify the use of primates and other mammals (Fig.
13–2), and there may even have been convergent evolution of
certain aspects of anatomy and intelligence in avian species
(Jarvis et al., 2005), as in the case of corvids (Emery and
Clayton, 2004).
If this comparative assumption also applies to the hip-
pocampus, which it surely should, the key advantage of using
animals is that invasive anatomical, physiological, and behav-
ioral studies can be carried out. Although not everyone shares
the view that such work is ethically acceptable—a concern
exacerbated by growing evidence of the complex emotional,
cognitive, and cultural life of higher primates—most neuro-
scientists adopt the utilitarian perspective that such work is
justified on its scientific merits and by its potential to relieve
both human and animal suffering. The macaque monkey is a
particularly suitable subject because Old World monkeys are
closely related to humans, although the last common ancestor
was about 30 million years ago (Kay et al., 1997). Its brain is
remarkably similar to that of humans (see Chapter 3), and
functional imaging studies are revealing striking similarities
in, for example, the dynamic organization and responsiveness
of the cerebral cortex (Van Essen and Drury, 1997). Rodents
may seem less well justified on this account, although the
topological organization of their brain is also quite similar
(Whishaw, 2004b). The comparative similarity argument may
Theories of Hippocampal Function 589
The genetic similarity of mammals
Platypus
Opossum
Elephant
Anteater
Armadillo
Hedgehog
Cow
Dog
Rabbit
Rat
Mouse
Macaque
Chimpanzee
Human
take flight with birds, but unlike rodents they are visual ani-
mals and have a number of other advantages than make them
useful for certain studies.
When using animals, it is essential to be respectful of their
biological needs, to use the minimum number necessary to
realize the required level of statistical power, and to think care-
fully about how to motivate and reward them in an ethically
acceptable manner. Reduction and refinement are laudable
goals in animal experimentation. In general, we cannot ask
animals to do something through language, but mild food
deprivation creates the opportunity to “reinforce” an animal
with rewards for performing a task. This is more charitable
than it sounds because, as it happens, food deprivation actu-
ally lengthens the life of laboratory animals (Nelson, 1988).
The use of mild electric shock to motivate animals in studies
of learning and memory may seem unfortunate to some, but
the neurobiological study of fear and stress is important; and
there are circumstances in which the rapid learning achieved
with shock cannot easily be achieved in other ways. Electric
shock has the advantage of being variable and can be motivat-
ing even when quite mild, but it is artificial. It is probably no
more stressful than the first day of swimming in a watermaze,
although the stress of the latter habituates quickly, and escap-
ing from water may be a more biologically meaningful form of
motivation.
Limitations in the Use of Laboratory
Animals to Investigate Brain Function
Animal work does have its disadvantages. In comparison with
human studies, animal studies include problems associated
with what the animals may be said to “know,” with language
590 The Hippocampus Book
and communication, and the artificiality of the physical and
social environments in which animals are tested.
Can we be confident that animals possess “knowledge” in
the same way humans do? As certain kinds of human knowl-
edge come to us via language (such as semantic, or “fact,”
memory), some observers have questioned whether such
forms of knowledge exist in animals (Macphail, 1998). An
American knows that a “slam dunk” is a shot in basketball,
and he can explain it to a puzzled foreigner; but how do
we teach “knowledge” to animals? Do they have “semantic-
like” knowledge or merely conditioned responses? Should we
assume that a monkey “knows” that a particular object
“means” reward simply because it has been consistently paired
with reward, or that one stimulus leads to another because
they are associated? The presence of pair-asssociate neurons in
the temporal regions of the neocortex of monkeys is sugges-
tive (Miyashita, 2004); and at a strictly behavioral level some-
thing akin to word learning has been demonstrated in dogs
(Kaminski et al., 2004; Markman and Abelev, 2004). The pri-
mary reason for believing that animals “know” things is
because they behave in relation to stimuli—food, water, shel-
ter, danger—as if they know what these stimuli are. A monkey
that has learned to use a rake to collect food that is beyond
arm’s length probably does “know” what a rake is and what it
can be used for. Receptive fields in appropriate somatosensory
areas of the cortex expand in association with learning
(Maravita and Iriki, 2004).
The problem is that, although the possession of knowledge
does not logically require language, communication about it
to others usually does. For example, episodic and autobio-
graphical memories are about specific events (“what”) that
happen in particular places (“where”) at particular times
(“when”) (see Section 13.6). The memory that one person has
of “what, where, and when” may be different from that of
another person who experiences the same event; in fact, much
discourse is concerned with debating such differences. It fol-
lows that it is difficult to investigate episodic memory in ani-
mals. The medium of communication is different and unlikely
to be one that satisfactorily reflects that animal’s personal
experience. Indeed, Tulving took the view that “as far as we
know, members of no other species possess quite the same
ability to experience again now, in a different situation and
perhaps a different form, happenings from the past, and know
that the experience refers to an event that occurred in another
time and in another place” (Tulving, 1983, p. 1). Regardless of
whether this skepticism is justified, we should recognize that
distinguishing both “episodic-like” and “semantic-like” mem-
ory from mere changes in learned behavior is far from
straightforward. Several studies have suggested that animals
probably can remember events as events, and there is no a pri-
ori reason why language is necessary for them to be able to do
so “at another time and in another place.”
The “training” (of animals) and the “asking” (of humans)
may not always yield the same answer in experiments on
learning and memory. The primary ways of teaching animals
to perform particular behavioral tasks are, as we have seen, by
engaging their curiosity and the use of reward and punish-
ment. Such techniques are more painstaking than merely ask-
ing a person to remember something; and they run the risk of
missing important subtleties. The answers you get from
humans who are doing memory tasks depend critically on the
way the questions are put and thus the subject’s awareness of
the problem at hand (Graf et al., 1984; Squire, 2004). This
issue was discussed in detail in Chapter 12: recognizing the
distinction between “explicit” and “implicit” memory process-
ing. Although this distinction may also apply to animals—and
the declarative memory theory argues that it does—it is diffi-
cult to get a handle on it analytically. It seems unlikely that we
shall ever know whether animals are “conscious,” although
here again experimenters have come forward with ingenious
ideas to try tackle the problem.
Finally, behavioral tasks in the laboratory are clearly artifi-
cial, but then so are cyclotrons and supercolliders and, for that
matter, test tubes. Science often has to be artificial to proceed.
Biomedical laboratory work generally requires that animals
are kept in somewhat artificial conditions, trained when they
are experimentally “naïve” rather than after they have accu-
mulated knowledge, skill, or experience, and often trained on
protocols that seem to bear little relation to “normal” behav-
ior in the wild. Many primatologists and evolutionary psy-
chologists see this as a major problem (Tooby and Cosmides,
2000), particularly with respect to the evolution of social
behavior. In contrast, others look upon ecological factors (and
species differences) as somewhat tangential to the primary
task of searching for general principles of brain function
(Macphail, 1982; Bolhuis and Macphail, 2001). This disagree-
ment has provoked much debate (Healy and Braithwaite,
2000; Hampton et al., 2002). What is not always appreciated is
the extent to which certain laboratory tasks are either explic-
itly designed to be analogous to things that animals do in the
wild (e.g., recognizing a previously seen stimulus) or unex-
pectedly turn out to be analogous to naturalistic behavior. An
example is the ability of rhesus macaques to learn arbitrary
“sensory-motor mappings.” These are tasks that require
a monkey to perform action A to stimulus A, action B to stim-
ulus B, and so on. In one laboratory investigation of this abil-
ity, different visual cues were presented to rhesus monkeys,
with the monkeys then having to make different movements
of a computer joystick (Brasted et al., 2002). Such sensorimo-
tor mappings are learned quite slowly; and on the face of it,
this may seem hardly surprising as monkey cognition did
not evolve to play computer games. However, as pointed out
by the authors, the different actions of vervet monkeys to dif-
ferent alarm calls made by troop members (Seyfarth et al.,
1980) are also examples of arbitrary sensory-motor map-
pings. Monkeys climb trees in response to the “barking call”
uttered when a leopard is nearby and look skyward followed
by hiding in bushes to the “cough call” uttered when an eagle
is spotted. Brasted et al. (2002) summarized evidence indicat-
ing that these different actions to distinct stimuli seem to
be learned, just as the computer joystick movements have
to be. An apparently abstract laboratory task may be

somewhat closer to primate “cognitive ecology” than might be
appreciated.
To conclude: interventional studies in animals are a power-
ful way of investigating brain function. When using them in
conjunction with behavioral tasks, we must accept that the
way in which hippocampal function is studied in animals has
to be different from the approaches described in Chapter 12
because animals cannot talk, they cannot tell us of what they
are aware (if anything), and they are unlikely to have a sense of
their past or their future or to have, to quote Tulving (1983)
again, “quite the same ability to experience the past” that
humans do. It does not follow, however, that these differences
of approach mean that the hippocampal formation in animals
is carrying out a different memory function than exists in
humans.
The enlightened possibility, reflecting the commonalities of
anatomy, physiology, and cell biology, is that it carries out
essentially the same information-processing algorithms but
that they might seem to be different because the human hip-
pocampus and the animal hippocampus are operating on dif-
ferent inputs and so provide different outputs to downstream
targets. A metaphor may help to explain the gist of this idea.
Imagine being able to transplant the hippocampus of a mon-
key into the brain of a human. Suppose the neurosurgeon is
the wizard we know all neurosurgeons to be, and that she suc-
cessfully connects up all the nerves and blood vessels to their
appropriate targets. Our perspective is that the result of such
an operation is that all should be well—our monkey hip-
pocampus should work normally inside the human brain.
However, because it would now be receiving different kinds of
information from its cortical and subcortical inputs, the end
result would be appropriate to the human, not the monkey.
Laterality of function in the human brain is a less fanciful
illustration of the same argument because it seems unlikely
that the left and right hippocampi of humans are anatomi-
cally, physiologically, or biochemically very different; yet the
outcomes of their processing are clearly distinct. With this
“optimistic” perspective of the enterprise at hand, we now pro-
ceed to discuss the major theories of hippocampal function.
Á memory is selective. It mediates the memory of facts and
13.3 Declarative Memory Theory
13.3.1 Outline of the Theory
In 1980, Cohen and Squire suggested that the distinction be-
tween “declarative” and “procedural” information processing
could be relevant to the pattern of memory deficits seen with
amnesia. Drawing upon Gilbert Ryle’s distinction between
“knowing that” and “knowing how” (Ryle, 1949), they pro-
posed that amnesic patients are impaired in forming conscious
memories of facts and events (“remembering that”) but have
relatively normal learning motor and cognitive skills (“remem-
bering how”). Following Milner’s observations of successful
motor learning in patient H.M. (see Chapters 2 and 12), this
Theories of Hippocampal Function 591
distinction was prompted by their finding that amnesic
patients could acquire and retain the cognitive skill of mirror-
reading novel words (in which words and all their letters were
printed backward) despite showing no conscious memory of
the training experience itself. Control subjects learned at an
identical rate and, of course, could also recall their training on
the task (Cohen and Squire, 1980). Squire went on to develop
and articulate the “declarative memory” theory of amnesia,
implicating the hippocampal formation and other structures
of the medial-temporal lobe (Squire and Cohen, 1984; Squire
and Zola-Morgan, 1991; Squire, 1992; Squire et al., 2004). Four
key propositions of this theory (Box 13–1) are as follows.
The first proposition—that the hippocampus is primarily
involved in memory—is now relatively noncontroversial. It is
strongly supported by work on amnesic patients and func-
tional imaging studies (see Chapter 12). Observations on peo-
ple have been extensively confirmed and elaborated in studies
of the effects of experimental lesions in animals. In keeping
with the overriding theme of this chapter, we accept this first
proposition of the theory without further comment.
The second proposition defines “declarative memory” as
representing both facts and events (Fig. 13–3). What fact
(semantic memory) and events (episodic memory) have in
common is that both are propositional and can be brought to
conscious awareness. Information about facts and events can
be “declared” (“Paris is the capital of France” or “I ate an
almond croissant for breakfast”). and such declarations can be
either true or false. By virtue of this shared property of fact
and event memory, such memories can be combined inferen-
tially with other information to yield new propositions (a
point developed further in the “relational-processing” version
of this theory—see Section 13.5). This inferential property is
Box 13–1
Declarative Memory Theory
1. Memory: The primary function of the hippocampal for-
mation is in memory.
2. Selectivity: The role of the hippocampal formation in
events, called “declarative memory.” This is the type of
memory that, in humans, can be consciously recalled.
3. Memory systems: The hippocampal formation is one of a
number of structures that comprise a “medial temporal
lobe memory system.” Although the components of this
system may have distinct subfunctions, it operates collec-
tively to mediate the formation and initial storage of
declarative memories.
4. Time-limited: The role of the hippocampus in memory is
time-limited. Memory is gradually reorganized as time
passes after learning. The hippocampus contributes to a
time-dependent systems-level consolidation process such
that, once completed, long-term memory traces are stored
in the cortex and neural activity in the hippocampus is no
longer required for or involved in recall.
592 The Hippocampus Book
Declarative memory theory: a taxonomy of mammalian memory systems
Declarative Nondeclarative
memory memory
Episodic Semantic Procedural Classical Non-associative
Priming
memory memory memory conditioning learning
Striatum;
Hippocampus- Amygdala;
motor cortex; Neocortex
medial temporal lobe cerebellum
cerebellum
an important point of similarity between information about
facts and events, but controversy surrounds the theory’s asser-
tion that there is a common neural substrate for the formation
of both episodic and semantic types of memory.
A key psychological difference is that unique events occur
only once and are specific to particular contexts and moments
in time, whereas factual knowledge is often gradually accu-
mulated and (generally) nonspecific with respect to the con-
text of learning. Certain other neuropsychological theories of
human memory argue that the nervous system encodes,
stores, and retrieves episodic information in a way that is
qualitatively different from that for semantic information
(Shallice, 1988; McCarthy and Warrington, 1990; Schacter and
Tulving, 1994; Aggleton and Brown, 1999). Such theories gen-
erally embrace or, at least, start off from the concept of
episodic memory as a distinct entity (Tulving, 1983) and
question whether it is helpful to subsume it with semantic
memory into the unitary category of declarative memory. In
contrast, declarative memory theorists see the distinction
between facts and events as descriptive rather than funda-
mental, both being initially encoded by structures in the
medial temporal lobe (MTL) that collectively comprise what
they believe to be a unitary “memory system.” Anatomically,
the MTL is positioned downstream of structures in the ventral
“what” pathway of the visual system in primates (Ungerleider
and Mishkin, 1982) and so is well placed to receive attended,
perceptually processed information. Damage to this down-
stream system is therefore held as likely to cause a propor-
tionate impairment in both fact and event processing, at least
in the anterograde domain..
Whether viewed as a merely descriptive or fundamental
distinction, these two forms of memory are not thought (by
anyone) to operate in isolation. There is a dialectic here, such
that our ability to form new event memories depends on both
our general and our personal knowledge of the world (seman-
tic knowledge) and on accumulating information from our
experience of events in the world (episodic memory). As
Figure 13–3. Taxonomy of mammalian memory sys-
tems. The declarative memory theory recognizes many
types of “memory system” and their mediation by dis-
tinct neural brain areas. The scheme was first intro-
duced by Squire (1987) and has been updated many
times since (e.g., Squire, 2004). The relative independ-
Reflex
ence and interdependence of these separate systems
pathway
has not been clearly delineated.
Tulving has long argued, the input to episodic memory must
be partly via semantic memory; equally, we add to that knowl-
edge structure through our experience and memory of events,
so the development of new semantic memories generally
depends on episodic memory. The two forms of memory are
“interdependent.” However, they can also act in parallel and
independently with, for example, direct input of information
from perceptual-representational systems to both semantic
and episodic memory (Graham et al., 2000). Investigations
of developmental amnesia—focusing on a set of young people
referred to a neurological clinic who had grown up being
very forgetful about everyday life but able to do reasonably
well in school—raised the alternative possibility that the
hippocampus plays an exclusive role in the formation of
episodic memory (Vargha-Khadem et al., 1997). However, rig-
orous assessment of this idea is hampered by uncertainties
about the extent of hippocampal damage in these cases and
theoretical uncertainties about the relative degree of “inde-
pendence” and “interdependence” of episodic and semantic
memory. We discuss these cases in the critique below.
Notwithstanding this qualification, the second proposition
of the declarative memory theory implies that a central focus
of the theory has to do with the information processing
and neural representation of what we ordinarily think of
as “memory”—the things we bring to mind as conscious rec-
ollections. For some, the term “explicit” memory is preferred
to describe certain types of memory task in which the subject
is consciously aware of the earlier study episode—in contrast
to “implicit” memory tasks, which lack the requirement
for recollective awareness (Graf and Schacter, 1985). The
differences between the declarative/nondeclarative and
explicit/implicit distinctions are important at some levels
of analysis but are arguably little more than terminological
at another. In practice, the declarative memory theory treats
“explicit” and “declarative” memory as interchangeable
terms (see Chapter 12). In the same vein, no strong philo-
sophical position about consciousness is implied by the
 Theories of Hippocampal Function 593

theory’s supposition that declarative memories entail “aware- • Perceptual and cognitive skills (e.g., the mirror-reading
ness” or “consciousness.” The theory is not tied to any partic- task of Cohen and Squire, 1980); priming phenomena
ular theory of this problematic but fascinating concept (e.g., word-stem completion) (Graf et al., 1984), percep-
(Zeman, 2004); it quite reasonably takes a “folk psychology” tual priming (Hamann and Squire, 1997), and concep-
view of the matter. tual priming (Levy et al., 2004) (see discussion in
Another issue about the second proposition of the theory Chapter 12)
is its skepticism that a categorical subdivision of memory • Simple forms of nonassociative learning such as habitu-
retrieval with respect to familiarity and recollection (Yoneli- ation (waning of responsiveness following repetition)
nas, 2001) can be mapped onto distinct MTL brain structures. and sensitization (augmented responsiveness) (e.g.,
Both forms of remembering involve conscious awareness, the Kandel et al., 2000)
defining attribute of declarative memory. Familiarity appears
The collective attributes of this heterogeneous group of
to be a phenomenological attribute that is acquired by a novel
skills is that they are not propositional (and so can be neither
stimulus, within several sensory systems, after its first presen-
true nor false), are generally learned gradually (although are
tation. A stimulus that has been presented before may then be
sometimes rapidly acquired), do not require conscious aware-
judged “familiar,” but this can happen without the subject
ness of the knowledge implicit in their execution, and are
having any recollection of when or where that stimulus was
simply things that people or animals learn to do. For some
seen, heard, or smelled previously. Recollection is more com-
(Macphail, 1998), they are the sum-total of what animals can
plicated, as it also involves source memory pertaining to
learn to do—such as the proverbial rat that presses the lever in
where or when the stimulus was previously presented.
an operant chamber because doing so in the past he has been
Subjects may even engage in “mental time travel”—the act
rewarded but not because it has any expectation of securing
of moving in their mind’s eye back to the time or place of
reward. Although not what we would ordinarily regard in
prior occurrence while simultaneously remaining in the tem-
everyday discourse as “remembering,” nondeclarative learning
poral present and responsive to their perceived surroundings
is important in constituting “the dispositions, habits, atti-
(such as in conversations about the past with someone while
tudes, and preferences that are inaccessible to conscious recol-
simultaneously driving a car). Certain theorists look upon
lection, yet are shaped by past events, influence our behavior
these two forms of memory retrieval as independent
and our mental life, and are a fundamental part of who we
(Baddeley et al., 2002), with only recollection or “relational”
are” (Kandel et al., 2000, p. 1119).
memories critically dependent on processing in the hip-
The third proposition of the theory concerns the existence
pocampal formation and diencephalon (Cohen and
of distinct brain systems for memory. Systems are distinguished
Eichenbaum, 1993; Aggleton and Brown, 1999; Eichenbaum
with respect to structure as well as function. Different forms
and Cohen, 2001). Yonelinas (2001) has outlined quantitative
of memory are held to depend on anatomically distinct,
techniques derived from signal detection theory to identify
though partially overlapping, brain regions. Different brain
circumstances in which people may be using recollection
systems for memory probably evolved to mediate functionally
rather than familiarity, arguing that recollection is a “thresh-
incompatible purposes (Sherry and Schacter, 1987a),
old” process, whereas familiarity can be better understood in
although the lack of a “fossil record” for memory makes it
signal-detection terms; they then applied this to the analysis of
extremely difficult to draw firm conclusions about their evo-
selected amnesic patients (Yonelinas et al., 2002). However,
lution. However, that different memory systems might be
this approach has been the subject of fierce criticism by
mediated by anatomically different brain substrates nicely
Manns et al. (2003) and Wixted and Squire (2004). As some of
reflects Francois Jacob’s idea that evolution is a tinkerer,”
these ideas constitute major changes to the declarative mem-
adding qualitatively distinct circuits rather than bringing
ory theory as it was originally conceived, they are presented in
about wholescale changes to the brain.
separate sections (see Sections 13.5 and 13.6). For the present,
The putative “medial temporal lobe memory system”
we accept “declarative memory” as a conceptual category on
(Squire and Zola-Morgan, 1991) is held to be one of these
its own merits.
brain systems for memory. It comprises the hippocampal for-
How are other types of long-term memory considered in
mation (as defined in Chapter 3) and both the perirhinal and
the theory? Collectively called “nondeclarative memory,” they
parahippocampal cortices of the medial temporal lobe (Fig.
are thought to reflect acquired information, habits, and
13–4). Damage to these structures in humans and nonhuman
learned dispositions that are expressed in behavior but cannot
primates causes disturbances of declarative memory without
be “declared.” Initially called “procedural” (Squire and Cohen,
necessarily affecting other types of learning or memory.
1984), a term first used in the field of artificial intelligence by
Damage to homologous structures in rats causes apparently
Winograd (1975), these types of learning are held to include
similar dissociations. In an evocative phrase, Squire wrote of
the following.
the distinction between declarative and nondeclarative mem-
• Motor skills and learned dispositions (e.g., the stimu- ory as being “honored by the nervous system.” He suggested
lus–stimulus associations and stimulus–response habits that, in addition to the medial temporal lobe mediating declar-
acquired during simple conditioning tasks) (Corkin, ative memory, the striatum is important for stimulus-response
1984; Cavaco et al., 2004) habits, the neocortex is the substrate for simple perceptual
594 The Hippocampus Book
The medial-temporal lobe memory system
S
Hippocampal
regions CA1
CA3
DG
Other direct Entorhinal
projections cortex
Perirhinal Parahippocampal
cortex cortex
Unimodal and polymodal association areas
(Frontal, temporal, and parietal lobe)
Figure 13–4. Medial temporal lobe memory system. Major compo-
nents of Squire and Zola-Morgan’s (1991) “medial-temporal lobe
memory system” (structures with gray shading), which is believed to
be the neural substrate for the formation of declarative memories
(as updated in 2004). Information enters the system from neocorti-
cal association areas. It is projected via the adjacent perirhinal and
parahippocampal cortices to the entorhinal cortex, which has the
pivotal role of being both the point of entry of information into the
hippocampal formation and a point of exit. Information passes
within the hippocampal formation via a cascade of unidirectional
projections before signals relevant to the encoding of new memo-
ries, or their consolidation, are fed back to the neocortex. CA1,
CA3, and SUB represent areas CA1 and CA3 and the subiculum, as
defined in Chapter 3; DG, dentate gyrus.
representations and priming (see Chapter 12), the amygdala
for emotional and social learning, and the cerebellum for the
learning of skeletal components of classic conditioning. This
anatomically based taxonomy is seen as a crucial advance in
understanding by those neuroscientists who hold a relatively
“localizational” view of brain systems.
However, as outlined in Section 13.2, current ideas about
information processing in the CNS refer to networks of inter-
acting brain areas whose normal operation can be disrupted
by disconnections rather than lesions of specific brain areas
(McCarthy and Warrington, 1990; Aggleton and Brown, 1999;
Gaffan, 2002; Murray and Wise, 2004). Human brain imaging
work has also moved rapidly from a “localizationist” perspec-
tive to one that stresses the importance of “effective connec-
tivity” between brain areas (Friston, 1995, 2004) and the
dynamic interaction of medial temporal and frontal regions
during intentional aspects of memory retrieval (Miyashita,
2004). Accordingly, the “one-to-one” mapping of memory
type to brain area of the original statements of declarative
memory theory is now somewhat dated.
Another complication associated with the mapping of
types of memory onto brain areas, explicit in Figure 13–3, is
that even the proponents of declarative memory theory recog-
nize that brain areas outside the hippocampal formation, and
even outside the medial temporal lobe itself, are also involved
in declarative memory. For example, Shimamura and Squire
(1987) claimed that the frontal lobes contribute memory of
the context in which information is acquired (i.e., “source
memory”) on the basis of what they observed in amnesic
patients with additional frontal damage. As recalling the con-
text where an event happened is a fact that one can declare and
of which one is conscious, it must (by definition) be a type of
declarative memory. Similarly, Squire has written of a “lim-
bic/diencephalic” brain system that incorporates structures
such as the “anterior thalamic nucleus, the mediodorsal
nucleus and connections to and from the medial thalamus
that lie within the internal medullary lamina” (Squire et al.,
1993, p. 462). Squire and Zola asserted that declarative mem-
ory is dependent “on the integrity of medial temporal lobe
and midline diencephalic structures” (Squire and Zola, 1998,
p. 205). This anatomical liberalism wisely recognizes that
medial temporal structures do not operate in isolation from
the rest of the brain (hence the reference to direct projections
in Fig. 13–4), but it leaves skeptics wondering where the
anatomical boundaries of the medial temporal lobe memory
system begin and end. In any event, the key point of this third
proposition is clear: The declarative memory theory places the
hippocampal formation at the apex of declarative memory
processing supported by a network of other brain structures.
The fourth major postulate of the theory concerns memory
consolidation. This is a particularly important aspect of the
theory and a focus of much current research. The basic idea
is that the hippocampal formation has a time-limited role
in memory for an individual fact or event. Some interaction
is held to take place between it and presumed long-term stor-
age sites in the neocortex. Gradually, through some time-
dependent “consolidation” process in which the hippocampus
and neocortex interact, initially labile memory traces in the
cortex become stabilized. This renders them resistant to later
brain damage to the medial temporal lobe itself, although not,
of course, to brain areas that are the eventual storage sites of
lasting memory traces (Fig. 13–5). Implicit in this fourth
proposition of the theory is the supposition that memory for
remote events depends on the strength of memory traces (i.e.,
information encoded within ensembles of spatially dispersed
neocortical neurons), where trace strength can be roughly
thought of in reductionist terms as alterations in membrane
excitability, the strength of synaptic connections, or other
activity-dependent properties of cellular physiology.
Memories with high trace strength can be recalled easily; those
with lower trace strength cannot. Consolidation is thought to
be a process that gradually, sometimes through the interleav-
ing of new traces with existing traces, brings trace strength
from low to high levels.
A key issue when thinking about consolidation is exactly
what is temporarily stored in the hippocampal formation
itself. There are several possibilities. One view is that detailed
sensory/perceptual information is stored there for an inter-
mediate period of time and then literally shuttled to the neo-
cortex. At the start of the consolidation period, the memories
would be “in” the hippocampus; by the end, they would be

Systems-level memory consolidation
encoding and cellular consolidation processes during systems-level consolidation
completed, soon after learning
passage of time
Initial memory traces
New memory traces forming during
systems-level consolidation
Consolidated memory traces
Figure 13–5. Systems-level memory consolidation. Pathways
between neocortical areas representing recent events or recently
acquired facts are held to be weak initially, even after the protein
synthesis-dependent cellular consolidation that sometimes follows
immediately after initial encoding (left). Memory soon after learn-
ing relies on rapidly formed connections between these neocortical
“in” the cortex. No one has much confidence in this possibil-
ity—as if the brain were some kind of railway shunting yard;
and it was never considered seriously by Squire. Another idea,
albeit a somewhat metaphorical one, is that hippocampal
storage consists of “pointers” or “indices” (Teyler and
DiScenna, 1986). These representations do not contain
detailed information; they are more like cartoons and are
thought to do two things. First, they help activate—by point-
ing at them—the relevant but dispersed neocortical neurons
at which traces representing detailed sensory information are
located (presumably through alterations in synaptic strength
between interconnecting neurons—see Chapter 10). This is a
lovely metaphor—neurons pointing at each other—but
exactly how such an addressing mechanism would work is
unclear. How does a neuron in the hippocampus “point at”
one or more neurons in the cortex or, even more difficult, a
subset of their synaptic interconnections? The answer, if this
metaphor is on the right lines, is likely to involve diffuse acti-
vation coupled to local signals, in much the same way that in
the dark a flashlight (diffuse) lights up a prowling cat’s eyes
(local). It is the conjunction of light and the cat that enables
spatial specificity, it not being a property of either feature
alone. More generally, “specificity” is one of the big issues of
modern neuroscience at every level of analysis applying as
much to intracellular synaptic stabilization mechanisms
(Goelet et al., 1986; Frey and Morris, 1997) as to the systems-
level memory consolidation processes that involve several
brain structures (Dudai and Morris, 2001).
Given that areas in the neocortex are the likely sites of per-
ceptual processing and the eventual sites of storage, we might
reasonably wonder why the hippocampus is ever needed. Why
bother with such apparent duplication? Moreover, even
Theories of Hippocampal Function 595
systems-level consolidation complete
Neocortex
Hippocampus
areas and the hippocampal formation. Over time, a consolidation
signal emanating in the hippocampus is thought to strengthen neo-
cortical connections (middle) to the point where hippocampal
activity is no longer necessary (right). The time this consolidation
process takes may be quite long—weeks, months, or even longer
(Squire and Alvarez, 1995).
though systems-level memory consolidation is a process
revealed through experiments in which the hippocampus is
rendered dysfunctional, it is unlikely that the evolutionary
pressure to develop a consolidation mechanism was to avoid
the deleterious effects on memory of brain damage. So what is
the function of consolidation and why the need for a brain
area with a time-limited role in memory? One answer, from
McNaughton et al. (2003), is that the numerous neocortical
interconnections that exist for creating an associational
framework of acquired knowledge are weak, slowly formed,
and/or liable to rapid decay. It is essential to protect against
“everything-becoming-connected-to-everything” during the
long process of learning throughout development and adult
life. More generally, consolidation is held to be a gradual,
selective process precisely because it has to be—ensuring that
only the relevant connections are made across ensembles in
and between cortical networks to represent information about
facts and events accurately (O’Reilly and Norman, 2002). We
revisit the issue of whether consolidation always has to be
gradual in Section 13.6.
The second thing that hippocampus indices and pointers
are supposed to do is guide the consolidation process over
time (Squire and Alvarez, 1995). It may take place over hours,
days, weeks, months, or even years—the virtue of such slow
processing being that it helps avoid catastrophic interference
in distributed associative memory traces stored in the cortex
(McClelland et al., 1995). This is the “teenager’s bedroom”
type of interference that occurs when one set of new memo-
ries (new junk) overlays earlier formed traces (old junk) in
such a manner as to interfere with their subsequent retrieval
(“Has anyone seen my iPOD?”). Cautious interleaving, guided
by hippocampal pointers, prevents this from happening. As
596 The Hippocampus Book
this consolidation process eventually comes to an end for a
particular set of information, the pointers have done their job.
There is therefore no need for extensive long-term memory
retention in hippocampus. Like forgotten movie stars, hip-
pocampal pointers just fade away.
Declarative memory theory therefore asserts that after con-
solidation is complete damage to the hippocampus should
have no effect on memory. However, as consolidation takes
time, the theory makes the important prediction that tempo-
ral gradients of retrograde amnesia should be obtained in
both humans and animals after damage to the medial tempo-
ral lobe. This prediction is attractively counterintuitive: It
states that if the hippocampus is damaged at time t an event
experienced several weeks before time t would be remembered
better than an event experienced only a few days earlier
because the former, despite being older, would have enjoyed a
longer period of consolidation. Testing this prediction has
become a cottage industry in its own right within the field.
This completes our presentation of the major features of
the theory. In the discussion that follows, we first present rel-
evant interventional studies on animals (primarily primates
but with some mention of other species) and then a critique.
As noted earlier, the relevant human studies have been dis-
cussed in Chapter 12, and an important extension of the
declarative memory theory called the “relational processing”
theory (Cohen and Eichenbaum, 1993; Eichenbaum and
Cohen, 2001) is considered separately (see Section 13.5). It has
inspired a body of experimentation somewhat separate from
Squire’s version of declarative memory theory. Similarly, work
on the consolidation of spatial memory is reserved for discus-
sion in Section 13.4.
13.3.2 Development of a Primate
Model of Amnesia
In the historical survey of Chapter 2, we noted that the initial
efforts to model human amnesia in animals were largely
unsuccessful. Monkeys given lesions of the medial temporal
lobe showed deficits in certain tasks. but overall they were
capable of learning tasks that one might not expect an
“amnesic” monkey to learn. Similarly, rats given hippocampal
lesions showed diminished exploration and striking inflexibil-
ity in their learned behavior but were quite capable of learn-
ing many quite complex tasks. The apparent discrepancy
between the human and animal data did not go unnoticed,
and various attempts were made to explain it (Iversen, 1976;
Weiskrantz, 1982).
Development of New Tasks for Monkeys
to Parallel the Types of Memory Lost in Amnesia
An important insight by Gaffan was that the tasks then used
in animal studies were quite unlike those on which amnesics
were seen to fail (Gaffan, 1974). He suggested that new tasks
be developed to capture what he then saw as an important dis-
tinction between “recognition-memory” and “associative-
memory.” Recognition memory refers, at least operationally,
to the ability to discriminate between stimulus items that have
been seen recently and others that have not. A judgment of
prior occurrence (Brown, 1996), it is generally impaired in
global amnesia, as shown by a standard test such as
Warrington’s Recognition Memory Test; however, it is not
always severely impaired in patients with more restricted hip-
pocampal damage (Aggleton and Shaw, 1996). Associative
memory, on the other hand, refers to the ability to learn that
two stimulus items go together, such as a stimulus and a
reward or a response and a reward, without regard to their
familiarity. This is a “habit” type of learning that Gaffan
argued was preserved in the presence of amnesia.
To model recognition memory, he collected a set of 300
junk objects and presented 60 of them to the monkeys each
day. Each trial consisted of a “sample” phase, during which the
experimenter presented one object on its own, and a “choice”
phase in which two objects were presented: the one that had
just been shown as a sample together with another, novel
object. These objects were presented on trays inside a
Wisconsin General Testing Apparatus (WGTA) (Fig. 13–6A).
To motivate the monkeys to perform the task, a “sugar puff ”
reward could be retrieved when they displaced the object pre-
sented during the sample phase and again when they chose
that same object during the choice phase. The only way the
monkeys could perform correctly was if they: (1) learned and
then applied the rule that a reward was available for displac-
ing the familiar object rather than the novel one; and (2)
remembered which object was familiar during the choice
phase of each trial after the memory interval following sample
presentation. This technique of probing recognition memory
was therefore called “delayed matching to sample with trial-
unique cues,” where “matching-to-sample” refers to the rule
for solution, “trial-unique” to the fact that an absolute judg-
ment of familiarity of the objects is required, and “delayed”
because of the memory interval between the sample and
choice phases of each trial. It is usually abbreviated as DMTS.
After learning using the short memory delay of 5 to 8 seconds
(at which the monkeys performed extremely well), two per-
formance challenges were added. One was to increase the
memory delay up to several minutes; the other was to present
a “list” of several sample objects, one after the other, and then
do a seriesof choice trials. Delays and lists are widely used
in modern variants of this task. Finally, to model associa-
tive memory, Gaffan used a small subset of the same junk
objects but a different protocol. The monkeys had to learn
which objects were consistently rewarded and which consis-
tently unrewarded, presenting the objects time and again
until they were all completely familiar. The animals indi-
cated their knowledge of the object-reward relations correctly
by reaching for rewarded objects when they were presented
and toward a small brass disk placed alongside when the
nonrewarded objects were presented. Correct choices were
rewarded; incorrect choices were not (Box 13–2).
The monkeys given fornix lesions showed delay-dependent
impairment in the recognition task (Fig. 13–6B) and list
 Theories of Hippocampal Function 597

A Recognition memory in a WGTA B Delayed matching to sample (DMTS)

100

90

% correct
80 NC
F
70

60

50
10 70 130 delays (s)

C Delayed non-matching to sample (DNMTS)

100 100
C C
90 90
NC NC

% correct
% correct

80 80 A
A
70 H 70 H
A+H A+H
60 60

50 50
10s 30s 60s 120s delays (s) 1 3 5 10 lists of objects

Figure 13–6. Recognition memory in the monkey. A. Rhesus maca-
que in a Wisconsin General Testing Apparatus (WGTA) reaching
for and displacing objects that hide a food reward. (Source: Courtesy
Christopher Coe, Harlow Primate Laboratory.) B. Fornix lesions
cause delay-dependent impairment of delayed matching to sample
(DMS). (Source: After Gaffan, 1974). C. Conjoint aspiration lesions
of the hippocampus and amygdala lesions, including neocortical
structures of the medial temporal lobe, cause delay-dependent
(left) and list-dependent (right) impairment of delayed nonmatch-
ing to sample (DNMS) after the animals reach the initial criterion
of 90% correct with single-sample objects. (Source: After Mishkin,
1978.)

length-dependent impairment, but their associative learning memory or (2) intellectual function (they had learned the rule
was unimpaired. As correct performance at short delays preoperatively and could still execute it postoperatively), but
depends on successful execution of the procedural rule (3) they prevent the formation of memories lasting much
learned preoperatively (matching-to-sample), the deficit longer than about 2 minutes. This conjunction of three
caused by the fornix lesion at longer delays is unlikely to have important characteristics of human amnesia in a single exper-
been due to the monkeys not remembering the rule. Rather, it iment represented the first apparently successful attempt to
suggested that fornix lesions have no effect on (1) short-term model key features of the syndrome in animals. Ironically,
although explicitly introduced on the grounds that it would
cause “the least extrahippocampal damage” (Gaffan, 1974, p.
Box 13–2
1100), fornix lesions are associated with only modest amnesia
Procedures for Testing Learning
and Memory in Primates
in humans (see Chapter 12). This represented something of a
puzzle but one that was soon solved.
Apparatus: Wisconsin general testing apparatus
Computerized touch screens Emergence of Delayed Nonmatching to Sample as the
Procedures: Pattern discrimination Benchmark Test of Recognition Memory in Primates
Spatial discrimination and reversal
Object discrimination and reversal DMTS tasks had been used previously with animals, but
Concurrent object discrimination Gaffan’s innovation of trial-unique cues during each week of
Delayed response
testing appeared to ensure that the familiarity discrimination
Delayed matching-to-sample (DMTS)
demanded of the monkeys was effectively an absolute judg-
Delayed nonmatching-to-sample (DNMTS)
Visual paired comparison
ment (“Have I ever seen this object before?”) rather than, as
Object-in-place memory happens with pairs of cues that are repeatedly given across tri-
Scene memory als, a judgment of relative recency (“Which of these objects
have I seen most recently”?). Relative recency may still be a
598 The Hippocampus Book
factor, even with 300 junk objects, as testing requires that they
be reused within a week or so (Charles et al., 2004a). In any
event, this subtle procedural change successfully engaged a
brain system that operated in the domain of long-term, rather
than short-term, memory. Additionally, the use of lists of
items appeared to make the task analogous to the list-learning
tasks so widely used in human studies. It is, perhaps, a slightly
unrealistic model of human list learning because the stimulus
materials usually presented to people (e.g., words or word-
pairs) are items with which people are already familiar. The
human test is not whether the word “cat” has ever been seen
before, but if this word appeared in the list presented by the
experimenter several minutes earlier. Thus, in the human
experiment, word lists are context-specific, a distinction not
thought to matter at the outset but, as we shall see later,
arguably critical to the anatomical mediation of different
types of memory. For the monkeys, whether a judgment of
absolute familiarity or relative recency is required may not
matter provided a sufficiently large number of objects are
used. However, it most certainly does matter if only two
objects are used, as Owen and Butler (1981) were to show,
because fornix lesions have no effect on remembering which
of two objects has been seen most recently. This is not sur-
prising because normal monkeys can do this well only over a
few seconds, and they use short-term memory to do so; the
task is impaired by a combined orbitofrontal and temporal
stem lesion (Cirillo et al., 1989). Thus, fornix lesions do not
interfere with short-term memory but may affect long-term
memory.
Another, more important difficulty with DMTS relates to
the matching rule. With such a rule, the monkey is rewarded
during the sample trial for displacing a novel object but
rewarded on the choice trial for displacing what is then the
familiar object. This potential source of confusion can be
avoided with several variants of the delayed-matching proto-
col, but the simplest and most influential change in procedure
was the shift to tasks employing a “delayed nonmatching-to-
sample” rule. With nonmatching, the monkey is rewarded for
reaching for a novel object each time he reaches out: on sam-
ple trials for the single sample object and on choice trials
when presented with a pair of objects. Mishkin and Delacour
(1975) developed such a task and found that normal monkeys
could learn nonmatching more easily than matching, proba-
bly because of their natural inquisitiveness about novelty.
Being easier and being later shown to be sensitive to large
lesions of the medial temporal lobe, delayed nonmatching to
sample became the task of choice for many years.
In an important study, Mishkin (1978) exploited both
characteristics of delayed nonmatching to sample to investi-
gate the effects of hippocampal, amygdala, and combined hip-
pocampal/amygdala lesions on recognition memory. He
trained monkeys on the delayed nonmatching to sample task
until they were performing at 90% correct or better. They were
then divided into four groups, with most undergoing bilateral
surgery to the medial temporal lobe; three monkeys were left
as normal controls. Of the nine operated monkeys, three were
subjected to hippocampal lesioning, three to amygdala lesion-
ing, and three to lesioning of both structures. Postoperatively,
the animals that had lesions of the hippocampus (but includ-
ing the caudal perirhinal cortex and parts of the parahip-
pocampal cortex because an aspiration technique was used)
and those with lesions of the amygdala (including parts of the
rostral perirhinal cortex) all relearned the task quickly. Those
with the combined lesion—a lesion analogous to the medial
temporal lobectomy in patient H.M.—were severely impaired.
Mishkin (1978) also found that once criterion levels of per-
formance had been reached by all groups with a short mem-
ory delay the combined lesion group showed pronounced
list-length and delay-dependent impairments in recognition
memory (Fig. 13–6C). The other groups performed well—
indistinguishably from the nonoperated controls. Thus, the
group that had been subjected to lesioning similar to that
applied to H.M. showed strikingly similar memory impair-
ment. This was a great step forward, and the paper has
been justly celebrated as a classic paper of twentieth-century
neuroscience (Aggleton, 1999).
Both lesions involved removing cortical tissue adjacent to
the target structures of the amygdala and hippocampus. At the
time, this was not thought to be of particular significance by
most observers (Horel, 1978), and the results were long
described by Mishkin’s group and by almost everyone else in
the field as lesions of the “hippocampus” and of the “amyg-
dala,” respectively. The pattern in the findings suggested that
these two structures played a parallel role in memory (which
might yet be dissociated by other tasks), the “dual circuit” idea
emerging as part of a theory of memory proposed a few years
later (Mishkin, 1982).
Delayed nonmatching-to-sample (DNMTS) rapidly
emerged as, in Zola’s phrase, the “benchmark test” of recogni-
tion memory. It was used in numerous studies, notably by the
National Institutes of Health (NIH) group led by Mishkin and
colleagues Bachevalier, Murray, Saunders, and others and sep-
arately by the San Diego group of Squire, Zola-Morgan
(Zola), Amaral, Suzuki, and others. A distinctive and laudable
feature of their work over the 20-year period from 1978 to
1998 was partial standardization of the manual testing proto-
cols such that comparisons could be made across studies,
at least within each laboratory. With the sole but impor-
tant exception that the NIH experiments typically involved
extensive pretraining prior to the lesion whereas the San
Diego series always involved surgery prior to training (Zola-
Morgan et al., 1982), this standardization was extremely valu-
able. In the case of the San Diego experiments, it made
possible a comparison of lesion groups with first one and later
a second “standardized” group of unoperated control animals.
Research using primates is not undertaken lightly because of
the ethical, conservation, and financial reasons referred to ear-
lier (see Section 13.2). Thus, establishing a set of repeatable
protocols for a test that models several aspects of amnesia was
an important step forward. If there was a weakness of this
standardization, it was that different experimenters were
inevitably involved in testing the two control groups and the

various lesion groups over the years, a practice that would not
ordinarily be accepted in research on rodents. The experi-
menter was typically not “blinded” with respect to the lesion
created in the animal, whereas this also is common practice in
the best laboratories working with rodents. However, the
robustness of the task and the tight variability of the forget-
ting functions with memory delays make it unlikely that these
weaknesses seriously limit its validity, even if it slightly injures
their elegance.
The outcome of studies conducted throughout the 1980s
and early 1990s indicate that after the combined lesion
(lesions that include the underlying cortical structures of the
posterior entorhinal cortex, perirhinal cortex, and parahip-
pocampal gyrus), a DNMTS deficit is observed whose main
characteristics can be summarized as follows (Box 13–3):
Working through these numbered points in turn, the fol-
lowing additional comments and qualifications should be
noted. First, despite the use of different monkey species and
different surgeons, experimenters, and WGTA designs, bilat-
eral damage to the medial temporal lobe similar to that pro-
duced in the original study of Mishkin (1978) always causes
a memory deficit in DNMTS (Squire et al., 2004). It is, how-
ever, important to stress that this does not mean that the
hippocampus alone is involved in recognition memory, as
the lesions that were used initially typically encompassed
extrahippocampal structures. Second, performance is good at
short delays but appears to decline monotonically as the time
between sample and choice is lengthened (Mishkin, 1978;
Zola-Morgan and Squire, 1985; Overman et al., 1990; Alvarez
et al., 1994a; but see Ringo, 1991, 1993). Securing this claim
involved the development of computer-automated systems for
presenting stimulus material to get around the difficulty that
there was always a small delay in the WGTA when the screen
separating the monkey and the experimenter was raised and
lowered. Notwithstanding this technological improvement in
testing procedure, there was an intense if somewhat technical
debate about delay-dependence, an issue we shall come to
shortly. Third, a deficit occurs in the visual and tactile modal-
ities, and by inference it is probably multimodal. For example,
monkeys tested in the dark so they can only feel the sample
and choice objects also show a memory deficit following
medial temporal lobe lesions (Murray and Mishkin, 1983;
Box 13–3
Delayed Nonmatching to Sample Reveals
an Important Role for the Medial Temporal
Lobe in Recognition Memory
1. The deficit in DNMTS after large medial temporal lobe
lesions is robust and repeatable.
2. The deficit is apparently delay-dependent.
3. The deficit may be independent of modality.
4. The deficit is exacerbated by distracting stimuli presented
during the delay interval.
5. The deficit is enduring.
Theories of Hippocampal Function 599
Suzuki et al., 1993). However, the inference about other
modalities is insecure. The available data on auditory DNMTS
indicates no impact of perirhinal lesions (Fritz et al., 2005), a
finding also observed in dogs (Kowalska et al., 2001). The data
of Fritz et al. (2005) led to the intriguing speculation that lan-
guage may be unique to humans not only because it depends
on speech but because it requires long-term auditory mem-
ory, although, as they pointed out, this appears to be contra-
dicted by field studies showing their ability to recognize alarm
calls that signal predators (Seyfarth et al., 1980). To our
knowledge, the impact of medial temporal lesions on olfac-
tory recognition has not been reported in primates (T. Otto
and M. Munoz, personal communication), partly owing to the
difficulty of making olfactory cues salient for monkeys (M.
Baxter, personal communication). Fourth, the effects of dis-
traction in DNMTS are analogous to what happens with
amnesic patients. They can retain small amounts of informa-
tion for extended periods when not distracted but forget more
rapidly when their attention is diverted (Zola-Morgan and
Squire, 1985)—although interaction between distraction and
the effects of a lesion is not always present (Zola-Morgan et
al., 1989). Fifth, there is no recovery of function; a deficit is
still seen when memory testing is conducted up to 4 years after
surgery as well as when it is carried out shortly after surgery
(Zola-Morgan et al., 1986). These five characteristics of
DNMTS performance after large medial temporal lobe lesions
map neatly onto several features of the amnesic syndrome. It
is not without reason that object recognition memory, as
measured in DNMTS, was championed as an “animal model
of amnesia” (Squire and Zola-Morgan, 1991) and even
declared a “millennial achievement” in neuroscience (Kandel
et al., 2000).
Other Tasks Used in a Comprehensive Test Battery
to Explore Declarative Memory in Primates
Although DNMTS became the most widely used task, it was
complemented by other tasks designed to probe different
types of declarative memory. During the early years, Squire
and Zola-Morgan often used a standardized “test battery”
consisting of several of the tasks shown in Box 13–2. Large
medial temporal lobe lesions cause a learning impairment in
the concurrent object discrimination task involving the inter-
leaved presentation of eight pairs of objects of which one
member of each pair is consistently rewarded, and the delayed
retention of object discriminations (Zola-Morgan and Squire,
1985). Deficits in the delayed response task are sometimes seen,
but they are capricious. This is because lesioned monkeys have
no deficit in short-term memory and even normal monkeys
can tolerate only short memory delays before they become
confused about whether they are remembering the present or
a previous trial. This is never a problem for DNMTS because
of the use of trial-unique cues.
Each of these and other tasks in the battery is held to be
“declarative” in the sense that, with only slight suspension of
disbelief, they can be seen as analogous to tasks given to
600 The Hippocampus Book
Test battery and different MTL lesions
1.0
0.5
mean z score
0.0
-0.5
-1.0
-1.5
NC H H H PR H ability is computed using the entire data set. However, an
(RF) (ISC) + + +
EC PH EC
+ +
type of lesion PH PR
+
PH
Figure 13–7. Pooling data across the test battery. The results
of several tasks of the San Diego test battery were pooled using
z-scores to reveal a graded deficit proportional to the extent of
the medial temporal lobe lesion and the way it was made. EC,
entorhinal cortex; H, hippocampal lesions; ISC, ischemic; NC,
normal control; PH, parahippocampal cortex; PR, perirhinal
cortex; RF, radiofrequency. (Source: Zola-Morgan et al., 1994.)
people. For example, the concurrent object discrimination
task can be thought of as analogous to learning a list of words,
the delayed retention of object discriminations to remember-
ing a passage of prose, and so on. Moreover, when the data
from all of tasks used in the San Diego test battery are com-
bined across experiments by averaging z-scores (reflecting rel-
ative statistical variability rather than absolute values), a clear
deficit is observed after lesions of the medial temporal lobe
that, with some exceptions (Baxter and Murray, 2001b), is
nicely graded with respect to extent of damage (Zola-Morgan
et al., 1994) (Fig. 13–7). However, the test battery has gradu-
ally fallen out of favor, partly because components of it are
now thought to depend on other brain structures. For exam-
ple, the deficit in the concurrent object discrimination task
after large medial temporal lobe lesions is probably due to
damage to area TE (Buffalo et al., 1998b), and this same task
involves habit learning in humans rather than declarative
memory (Bayley et al., 2005a).
Adequacy of These Tasks
as Models of Declarative Memory
One qualification of this ostensibly tidy picture was an argu-
ment first posed by Ringo to the effect that the delay depend-
ence of the DNMTS deficit may depend on the choice of
numerical scale with which performance is scored (Ringo,
1991, 1993). This is an important technical point. The out-
come of a single trial in DNMTS is the monkey choosing the
correct or incorrect object. It is therefore usual to plot per-
formance on a scale from 50% (chance) to 100% (perfect per-
formance) using a linear scale. Ringo argued that statistical
considerations require an arcsine or other transformation for
such data, particularly if scores come close to 90% correct as
they often do at short memory delays. His reanalysis of several
published DNMTS studies using such a transformation led
him to conclude that medial temporal lesions may not, in fact,
cause delay-dependent forgetting. He suggested that apparent
delay dependence in several studies could arise because lesion-
induced deficits are concealed at short memory-delay inter-
vals through use of the linear scale. For example, controls may
be scoring 92% correct and lesioned monkeys 90%—per-
formance levels that do not differ statistically when the vari-
arcsine transformation (which “stretches” the scale as values
approach either 0 or 100%) may show a different picture, as
Ringo’s reanalysis of several data sets indicated. This is a
cogent argument, although it did play down the overlap of
performance by lesioned and control monkeys at short delays
in some studies, as pointed out by Alvarez-Royo et al. (1992).
Its importance was nevertheless taken on board. It spurred the
development of, or at least the better appreciation of, auto-
mated versions of the WGTA that permit the presentation of
very short sample-choice intervals. Under these conditions,
no lesion-induced deficit is observed in the rate of acquisition
of the DNMTS task at “zero delay,” with control and lesioned
animals reaching the same performance asymptote. An unam-
biguous delay-dependent deficit still pops out at long delays
(Overman et al., 1990; Alvarez et al., 1994). Ringo’s concern,
however, should not be forgotten, as findings obtained more
recently with neurotoxic lesions of structures within the hip-
pocampal formation add a further note of caution. To claim a
differential effect on memory rather than perception, it is
essential to secure the statistical interaction between group
and memory delay. This is because a lesion that impaired per-
ception would be expected to have as large an effect at short
memory delays as at long ones—the animals being just as
poor at seeing the objects in front of them at both intervals.
This is the gold standard for establishing that the performance
of groups differs across the domain of long-term memory.
Many studies have aspired to this standard; the truth is that
few primate DNMTS studies have reached it.
A separate concern about the test battery is that the psy-
chological justification for regarding tasks in the test battery as
“declarative” is not always clear, and in some cases the argu-
ment borders on the circular. We have just noted that concur-
rent discrimination learning might be thought akin to a
person learning a list of words. The trouble is that there are
lists and there are lists. For example, Squire (1992) claimed
that the concurrent discrimination task in which eight pairs of
objects are presented a number of times during a single test-
ing session (a task developed originally by Moss et al., 1981) is
declarative, whereas a 20-pair concurrent task in which each
pair of objects is presented only once per day is not (Malamut
et al., 1984). The basis for these assertions was that the former
was impaired by medial temporal lobe lesions, whereas the
latter was not. Clearly this circularity, if that is all there is to it,

is unsatisfactory. In fairness, there is more to it, namely that
many behavioral tasks can be learned using different strate-
gies—even those of ostensibly the same logical structure (task
ambiguity being one of the key conceptual issues discussed in
Section 13.2). Varying the pattern and temporal spacing of the
trials may bias animals toward or away from using a “declara-
tive strategy”—an example of a quantitative change having a
qualitative effect. In the case of concurrent object discrimina-
tion tasks, normal monkeys might be able to remember the
“fact” that a particular object was paired with reward for a
period of time despite the interference produced by the other
eight interposed problems with different objects. Provided
another trial of the first problem comes along soon enough,
they could then use their fact-based declarative memory
to solve the problem. However, if the intertrial interval is long,
as with the 24-hour delay task, even the normal monkey
may find a declarative strategy less reliable. It might then fall
back on a stimulus-response strategy, by which is meant the
gradual strengthening of a disposition to reach for the
rewarded object and not for the nonrewarded object, rather
than explicit memory of a prior object and reward episode.
Because this nondeclarative strategy, according to the theory,
is also available to the lesioned monkey, no difference in per-
formance would be expected between groups. Interestingly,
people generally use a declarative strategy to learn lists, irre-
spective of how long the list and how many trials are presented
per day, although, as already noted, people also learn the con-
current object discrimination task in a habit-like manner
(Bayley et al., 2005a). Debate about the interpretation of tasks
first published 20 years ago may seem like a historical sideline,
but it is not. For the issue of “strategy used” versus “logical
structure’ is a key one and, as we shall see, has reemerged as an
issue with respect to the memory strategies mediating
DNMTS.
The way to avoid circularity is to come up with independ-
ent psychological evidence that a particular strategy is being
used or that performance on one task correlates with that
on another and allow this evidence to help guide definitive
predictions from the theory. In the case of concurrent object
discrimination learning, one way might be to use the “reward-
devaluation” procedure as originally developed in experi-
ments on rats (Adams and Dickinson, 1981). If the association
between reaching for an object and securing reward is learned
as a “fact,” pairing the reward with a toxin such as lithium
chloride should render it less palatable and so make the ani-
mal less inclined to reach for the object. This is because
the animal would, in some sense, infer from his “propositional
knowledge” that reaching for the object leads to the reward,
and thus seeing the object should retrieve a memory repre-
sentation of the reward and that it is now no longer worth
having. However, if the animal had merely learned the habit
of reaching for the object as a disposition (and learned
it because it was repeatedly paired with reward during train-
ing), reward devaluation would not have any immediate
effects on performance. The animal would reach for the
reward because reaching for rewards is what it had been
Theories of Hippocampal Function 601
trained to do, not because it has in any sense “remembered”
that doing so actually leads to anything. Murray and her col-
leagues at the NIH laboratory have used this reinforcer-deval-
uation procedure to dissociate fact and habit learning in
studies that have revealed that excitotoxic damage to the
amygdala disrupts the ability of animals to change their inter-
nal representations of reinforcement value (Malkova et al.,
1997). Stimulus–stimulus associations can therefore be repre-
sented in a manner that might be thought of as a “fact.”
However, neither the hippocampus nor the rhinal cortex
seems to be a brain structure mediating reinforcer devaluation
(Thornton et al., 1998).
As an aside, exclusion of the amygdala from a revised
medial temporal lobe memory system may also be mis-
taken because there are aspects of recognition memory that
do seem to involve stimulus value. Seeing old friends
again, such as colleagues at an annual scientific gathering,
invokes great pleasure. We all witness and experience this
on the first day of such a meeting—a curious pleasure as it is
all too often followed by heated debate in the lecture theater
with these self same people. Although this emotional charac-
teristic of stimulus familiarity may be mediated by the hip-
pocampus and/or the perirhinal cortex, it seems unlikely. It is
tempting to speculate that there may be role for the amygdala
as well.
13.3.3 Domains of Preserved Learning Following
Medial Temporal Lobe Lesions in Primates
When constructing an animal model of amnesia, it is impor-
tant to model spared memory as well as impaired memory, of
which Gaffan’s (1974) use of an associative task was an early
example. Historically, systematic analysis of the issue of mul-
tiple types of learning and memory and their differential sen-
sitivity to different lesions began in the rodent literature (see
Sections 13.4 and 13.5). It was, however, not long before the
idea was taken up in studies with primates beginning with
tasks involving motor skill. For example, Zola-Morgan and
Squire (1984) found that control and medial temporal lobe
(MTL)-lesioned monkeys could learn to thread a “lifesaver”
candy (which contains a central hole) along and then off a
thin wire. Eating the candy was the monkey’s reward. With
experience, normal and lesioned monkeys got steadily faster,
and they learned at an equivalent rate. Presumably, the
lesioned monkeys did not remember doing the task from one
day to the next or recognize the apparatus when it was shown
to them again, but they threaded the “lifesaver” nonetheless. If
only we could “ask” them about their mental experience in
such a situation!
Perceptual skills have also been studied in monkeys using
pattern discrimination tasks (Squire and Zola-Morgan, 1983).
These are slowly learned discriminations in which two similar
patterns (such as alpha-numeric characters) are differentially
paired with reward. Learning typically takes place slowly, over
hundreds of trials; and with one exception no deficit in learn-
ing is seen following even large MTL lesions or after
602 The Hippocampus Book
hippocampus-specific lesions. The interesting exception
noted by Squire is that performance is sometimes poorer over
the first few trials of the day. He suggested that this difference
between groups is due to controls remembering the first few
trials in an explicit or declarative way prior to the build-up of
within-session interference and the reliance on learned habits.
However, more recent experiments, discussed in the critique
later on, challenge the idea that components of the MTL
memory system are not involved in learning stimulus-reward
associations (Murray and Wise, 2004).
With respect to cognitive skills, that lesioned monkeys can
learn tasks such as DNMTS at a normal rate at zero delay
(Alvarez-Royo et al., 1992) seems to imply that they are able
to abstract, from the sequence of trial events, the appropriate
rule for performance. This finding is inconsistent with their
having any major deficit in “intelligence.” However, although
the investigation of primate cognitive skills, knowledge, and
“metaknowledge” (i.e., what they know they know) has
been considered by those working in a more ecological context
(Hauser, 2003), it has only recently been considered seriously
by behavioral neuroscientists (Hampton, 2001). Monkeys
have not yet been taught the subtle tasks of learning a finite
grammar or to make accurate weather forecasts—tasks that
have imaginatively extended the domain of preserved learning
in amnesics (Knowlton and Squire, 1993), but the range
of laboratory tasks of nondeclarative memory is steadily
expanding.
13.3.4 Selective Lesions of Distinct Components
of the Medial-temporal Lobe Reveal
Heterogeneity of Function
Squire and Zola-Morgan’s development of a test battery was
motivated by the reasonable ambition of developing tasks that
are analogues of the verbal and visual memory protocols
on which MTL amnesics are impaired. Combining data from
these tasks in the form of z-scores to create a single quantita-
tive index (Fig. 13–7) revealed a monotonic effect of lesions
within the MTL: the larger the lesion, the greater the deficit
(Zola-Morgan et al., 1994). However, mindful of the uncer-
tainties surrounding the classic but misleading concepts of
mass action and equipotentiality (Lashley, 1950), it is not sur-
prising that critics of declarative memory theory are wary of
the notion that the entire MTL functions as a single homoge-
neous unit. Indeed, further research has revealed that
restricted lesions of MTL structures cause little or no impair-
ment in some declarative tasks but do affect others. This is not
just a matter of graded task difficulty, as double dissociations
are seen (as we shall see shortly). Particularly problematic are
data suggesting that neither restricted hippocampal lesions
nor damage confined to the entorhinal cortex necessarily
cause an enduring deficit in DNMTS.
Absolute confidence in the “benchmark test” of DNMTS
began to unravel for a number of reasons. One problem has to
do with understanding what the test is really measuring psy-
chologically; another relates to identifying the necessary and
sufficient lesion that impairs it. These two issues are connected.
Mishkin’s “dual-circuit” theory relating to the putative role
of the “hippocampus” and “amygdala” in recognition memory
prompted the question of whether different types of informa-
tion processing occurred in the two routes of his dual pathway
(Mishkin, 1982). One prescient idea, based in part on theoriz-
ing by Mandler (1980) was that recognition memory could
be based on either a memory of the polymodal features of
an object or on memory of where the object had been seen
before. We may think of this as, on the one hand, distinguish-
ing between features that are intrinsically part of the object
and, on the other, between features that reflect the context in
which an object is seen. Whereas the former are part of (and so
would move with) an object when it is displaced, the latter
might not.
Two studies began the systematic exploration of this issue.
One examined cross-modal transfer of information, and the
other examined memory for the place where an object had
been presented. Murray and Mishkin (1985) found that
“amygdala” lesions (which included damage to the perirhinal
cortex) caused impaired cross-modal transfer, whereas “hip-
pocampal” lesions (which also damaged the parahippocampal
cortex) were with without effect. Monkeys were trained to
sample one of a restricted set of 40 objects in complete dark-
ness and then make a choice between the sample and another
member of the set in the light. Information acquired in
the tactile modality was sufficient to guide visually directed
choices accurately, but performance, at even short memory
delays, was selectively impaired by the “amygdala” lesions.
In contrast, Parkinson et al. (1988) found that their hip-
pocampal lesions selectively impaired both a place task and
an object-in-place task, whereas the amygdala lesions had
no effect. In a concurrent “place” and “object-in-place” tasks,
two objects were presented during the sample trial in two
of three possible locations on the tray in front of the monkey.
On the choice trial, these objects were again put in front
of the monkey. In the “place” choice trials, the monkey had
only to choose on the basis of the locations occupied by the
objects in the sample trial; whereas for the “object-in-place”
choice trials, the animals had to choose on the basis of
remembering the particular places occupied by particular
objects. It was later established that monkeys with “hip-
pocampal” lesions could remember one place but not two
(Angeli et al., 1993).
The sufficient lesions for seeing these deficits are now
known to be purely neocortical, but the historically still
important point to emerge from these studies is that recogni-
tion of novelty might be mediated by either “intrinsic” or
“contextual” components that are not disambiguated in the
standard DNMTS test. The distinction echoes the concept of
recognition being mediated either by a sense of familiarity
(intrinsic) or by recollection (contextual recall) (Mandler,
1980). Recognition by familiarity would involve the monkey
making its choice because of the two objects confronting him
in the choice test one evokes a sense of familiarity. It could do
that even if the monkey was unable to recollect the occasion
when or where it had seen it before. Recognition by contextual
recall, on the other hand, would involve the monkey utilizing

the cues of the WGTA surrounding him to “bring back to
mind” that of the two objects before him one had been pre-
sented before in this context. Although we cannot talk to the
monkey about it, we can nonetheless imagine the animal hav-
ing a private “recollective experience” in much the same way
that we would do in a similar situation.
Unfortunately, the important line of psychological think-
ing embedded in these ingenious studies was largely over-
shadowed by preoccupation with the anatomical implications
of the aspiration lesion technique. Damage to tissue lying
in the entorhinal, perirhinal, and/or parahippocampal cor-
tices might have been contributing to poor performance in
the “amygdala”- and “hippocampus”-lesioned animals rather
than damage to these target areas. Later studies using more
selective lesions have borne this out. For example, we now
know that cross-modal memory is affected principally by the
perirhinal but not the amygdala component of the original
lesions (Murray and Bussey, 1999). Similarly, the spatial task
appears to be unaffected by neurotoxic hippocampal lesions
(Malkova and Mishkin, 2003).
Studies during the late 1980s revealed, somewhat surpris-
ingly, that neurons in the perirhinal cortex were sensitive to
familiarity and relative recency (Brown et al., 1987; Brown and
Xiang, 1998). In parallel, careful anatomical studies began to
focus on the neocortical regions neighboring the hippocam-
pus, such as the entorhinal cortex (Amaral et al., 1987;
Insausti et al., 1987a,b) and both the perirhinal and parahip-
pocampal cortex (Suzuki and Amaral, 1994a, b). This work
was accompanied by further lesion studies. With few excep-
Figure 13–8. Conflicting findings in studies
of selective lesions of the hippocampus upon
delayed nonmatching to sample (DNMTS).
A. Murray and Mishkin (1998) found that con-
joint damage of both amygdala and hippocam-
pus without major damage to surrounding
perirhinal and parahippocampal cortex had no
effect on DNMTS with memory delays as long
as 40 minutes. B. Zola et al. (2000) reported
that damage restricted to the hippocampus
made using ischemia, radiofrequency heating,
or excitotoxins caused a delay-dependent
deficit in DNMTS over the same time delays.
C. Zola et al. (2000) observed a similar deficit
in another test of recognition memory—visual
paired comparison. B DNMTS; San Diego Lab
100
90
% correct
80
70
60
50
Theories of Hippocampal Function 603
tions, Gaffan’s laboratory in Oxford being one of them, these
studies remained focused on using the old warhorse DNMTS.
Lesions described as being in the “rhinal cortex” (part of the
anterior perirhinal cortex) were sufficient to cause severe,
enduring delay-dependent impairment of DNMTS irrespec-
tive of whether the hippocampus had also been damaged
(Meunier et al., 1993; Suzuki et al., 1993; Mishkin and Murray,
1994). Following these important discoveries, Murray and
Mishkin (1998) reinvestigated the role of the hippocampus
and amygdala using excitotoxic lesions. Not only does this
type of lesion leave the surrounding cortex unaffected, excito-
toxins should not damage fibers from the anterior perirhinal
cortex passing through the ventral amygdalo-fugal pathway to
the medial thalamus or the posterior perirhinal efferents pro-
jecting through the fimbria-fornix and posterior thalamus to
the same nucleus (see Chapter 3). The key finding was that
monkeys with average lesion sizes of 88% damage to the
amygdala and 73% damage to the hippocampus showed no
impairment in DNMTS (Fig. 13–8A). This comprehensive
study included varying list lengths and memory delays and,
following the protocols set by Alvarez et al. (1995), included
delays of up to 40 minutes in a subset of four monkeys. As no
deficit was observed, Murray and Mishkin concluded that, in
the MTL, “the rhinal cortex is not only necessary but also suf-
ficient to sustain visual recognition ability” (Murray and
Mishkin, 1998, p. 6579). This conclusion is entirely compati-
ble with the available electrophysiological and lesion data.
In passing it should be noted that these findings rescue part
of the memory circuit for memory originally proposed by
A Delayed nonmatching to sample (DNMTS; NIH Lab)
100
90
C
80 A+H
% correct
70
60
50
10 30 60 120 LL3 LL5 LL10
delays (s) list lengths
C Visual paired comparison (VPC)
70
% time viewing novel stimulus
65
C
H 60
55
50
8s 15s 1m 10m 40m 1s 10s 1m 10m
delays delays
604 The Hippocampus Book
Mishkin (1982) by showing that the critical pathway for a
judgment of familiarity emanates from the perirhinal cortex
to the medial nucleus of the thalamus, bypassing the hip-
pocampus itself.
However, in contrast and over memory delays of as short as
2 and 10 minutes, Beason-Held et al. (1999) found hippocam-
pus lesion-induced impairment. Zola et al. (2000) also
reported that restricted hippocampal damage can be sufficient
to cause impairments in both DNMTS and another test of
recognition memory called the visual paired-comparison
(VPC) task (Fig. 13–8B,C). The latter task, like certain tasks
developed by Gaffan much earlier (e.g., Gaffan et al., 1984),
involves no formal training—merely exposing the monkey to
pairs of black and white line drawings on a computer screen
(Bachevalier et al., 1993). After looking at these drawings for
about 30 seconds, monkeys spontaneously direct their eye
movements to a new drawing presented on the screen beside
one of the old ones (a phenomenon first observed experimen-
tally in studies of human infant perception (Fantz, 1964).
Many pairs of such stimuli can be presented, one after the
other, to build up a profile of the animal’s ability to detect
novelty spontaneously.
Drawing together data from experiments over 10 years
using a variety of lesion techniques and a large number of
monkeys, Zola et al. (2000) claimed that ischemic, radiofre-
quency, and excitotoxic lesions of the hippocampal region all
cause a modest but significant deficit in recognition memory
in both DNMTS and VPC. The deficit is apparent at delays as
short as 15 seconds, with larger effects seen (at least on
DNMS) at longer delays of 10 and 40 minutes. Consistent
with the declarative memory theory, they concluded that the
integrity of the hippocampal region is essential for recogni-
tion memory. The extent of hippocampal damage varied sub-
stantially among these three studies, ranging from 33% to
62% in the San Diego animals but, paradoxically, averaging
73% in Murray and Mishkin’s NIH study that found no
deficit. However, Nemanic et al. (2004) came to a somewhat
different conclusion after a similar comparison of DNMTS
and VPC. Their data point to a substantial deficit in both tasks
after perirhinal lesions but no deficit in VPC in hippocampus-
lesioned monkeys until longer delays are interposed and only
a slight deficit in DNMTS at a memory delay interval of 10
minutes. Bachevalier raised the important qualification about
her own study (Nemanic et al., 2004) that the average lesion
size in her hippocampal group was only 43.5% (this being
made with ibotenic acid), but this does not appear to be sub-
stantially different from the mean lesion size prevailing in the
San Diego studies that also used excitotoxins.
Comparison of Conflicting Studies Reveals Subtle
Differences in Lesion Size and Methodology
How is the discrepancy between these experiments to be
explained? Some reports have indicated that apparently
restricted hippocampal lesions did impair recognition
memory and others that they did not. There were several pro-
cedural differences. One of these was the use of extensive
pretraining by the NIH laboratory. Zola et al. commented
that: “Training on the rule provides the monkeys with
extended practice at holding novel objects in memory across
short delays, which might then make it easier to hold novel
objects in memory across the longer delays from which the
performance scores are derived” (Zola et al., 2000, p. 459). The
implication seems to be that avoiding preoperative training
makes for a more sensitive behavioral assay, a view consistent
with the apparently greater sensitivity to hippocampal lesions
of the VPC task in which there is no formal training at all. This
may be true empirically but is unsatisfying intellectually, as
there is nothing in declarative memory theory to explain this
dependence. Indeed, we should be wary of a logical inconsis-
tency with respect to mission of this enterprise, namely, mod-
eling amnesia. Amnesic patients have extensive experience
recognizing things prior to becoming amnesic but, according
to Squire, have a recognition deficit even if their brain damage
is restricted to the hippocampus. Why should the animal
model be any different? Furthermore, why did the San Diego
laboratory insist on using a protocol so different from what
they were doing with their patients?
A major of focus of attention has been on the extent
of damage to hippocampal and extrahippocampal regions.
Growing awareness of the importance of the extent of lesion
damage in individual animals led to the development of MRI-
based evaluations of the locus and extent of damage (Malkova
et al., 2001). Suitably calibrated, these evaluations offer the
opportunity of providing accurate, noninvasive estimates of
brain damage in advance of extensive postoperative testing
(an ethically desirable development apart from anything else).
The more usual postmortem histology has often—no doubt
to the dismay of experimenters (after years of training an
individual animal)—revealed wide variations in locus and
size after lesions made by various techniques. There was,
for example, an average of 18% extrahippocampal damage (to
the parahippocampal cortex) in the monkeys trained by Zola
et al. (2000), who claimed a specific hippocampus lesion-
induced deficit in DNMS; but there was only 4% extrahip-
pocampal damage in a study by Murray and Mishkin (1998),
who claimed the opposite. To be fair, despite incursion into
this neighboring brain area, there was no evidence of any
within-group correlation between performance and the
extent of parahippocampal damage in the Zola et al. (2000)
study (damage ranged from 0% to 46%). Not much comfort
should be drawn from this, however, as the study did not
report any correlation between the extent of hippocampal
damage and the recognition performance (ranging from 13%
to 76% across all 14 animals tested). In contrast, Murray and
Mishkin’s findings indicated, paradoxically, a positive correla-
tion between the extent of hippocampal damage and per-
formance at the longer delays: the greater the hippocampal
damage, the better the recognition performance.
Baxter and Murray (2001b) took this curious inverse rela-
tion further with a meta-analysis of these three studies of
DNMTS in monkeys with restricted hippocampal lesions.

Using an optimum d′ statistic (which takes into account dif-
ferences in the performance of control monkeys across stud-
ies), their analysis revealed an inverse correlation between the
loss in d′ and percent damage to the hippocampus (Fig. 13–9).
Zola and Squire (2001) argued that this meta-analysis was
invalid because it failed to partial out the potential influence
of various factors that differed across studies, other than
lesion size, such as whether pretraining had been given, the
way the lesions were made, and the delays used to assess mem-
ory—a caution that ironically did not prevent Zola himself
from generating a z-score statistic to characterize a wide range
of tasks (as in Fig. 13–7, p. 600). Baxter and Murray (2001a)
conceded the weakness of pooling data across studies that
used slightly different training protocols but asserted that
even when factors relating to lack of task identity are partialed
out a nonparametric analysis of the inverse relation between
loss of d′ and lesion size remains significant. As in Ringo’s ear-
lier analysis of delay dependence, the use of a d′ statistic was
helpful, though by no means critical, as the same result per-
tains when raw percent correct difference scores are used
(M.G. Baxter, personal communication). Moreover, each of
the pooled studies secured a trend or significant inverse rela-
tion on their own.
Baxter and Murray’s meta-analysis is an empirical observa-
tion, not an explanation about why there might be an inverse
correlation. One possibility is that residual hippocampal tissue
adjacent to the lesion produces aberrant neural activity that
disrupts neighboring brain regions. There is, however, another
possible reason for this paradoxical correlation that deserves
careful discussion. Once again, like the debate about concur-
rent object discrimination learning (above), it is the old prob-
lem of task ambiguity. We saw earlier that DNMTS is
amenable to two distinct strategies: a “familiarity” strategy and
a “recollection” strategy. If recollection is either exclusively or
Figure 13–9. Meta-analysis of DNMTS. Systematic comparison of
data from several laboratories reveal the paradoxical inverse relation
between DNMTS performance and hippocampal lesion size (Baxter
and Murray 2001). The various symbols represent data from differ-
ent studies: squares, Murray and Mishkin (1998); triangles, Beason-
Held et al. (1999); circles, Zola et al. (2000).
Paradoxical relationship between DNMTS and lesion size
-0.5 good
0.0
performance
loss in d’
0.5
1.0 poor
0 20 40 60 80 100
% damage
Theories of Hippocampal Function 605
most effectively mediated by the hippocampus, animals with
large hippocampal lesions would most likely use a familiarity
strategy and so rely on their intact perirhinal cortex to make
judgments of prior occurrence using the neuronal ensembles
identified in the pioneering work of Brown and colleagues
(Brown et al., 1987; Brown, 1996). Conversely, if recollection
remains feasible and is ordinarily preferred, animals with
small hippocampal lesions may automatically continue to use
their damaged hippocampus, attempt a recollection method
of solving the problem, but perform less well than control ani-
mals precisely because this structure is damaged. A negative
correlation between performance and size of lesion would
then emerge. It is not unreasonable to suppose that extensive
pretraining may also predispose animals to use a less effortful
familiarity strategy preferentially, thereby accounting, at least
in part, for the persistent failure to see restricted hippocampal
lesion effects on DNMTS in the NIH laboratory but their
presence in the San Diego laboratory (that generally did not
use preoperative training). The situation is laden with irony, as
Squire and colleagues have been uncertain of any simple map-
ping of the recollection/familiarity distinction onto structures
in the MTL (Squire et al., 2004) (a view endorsed by Stark in
Chapter 12), yet it could be precisely because the unoperated
monkeys in San Diego used recollection to perform DNMTS
that hippocampal deficits were seen.
The available data suggests that the VPC test may be a more
hippocampally sensitive test of recognition memory than
DNMTS (Zola et al., 2000; Bachevalier et al., 2002; Nemanic
et al., 2004). Unfortunately, we do not yet have a princi-
pled understanding of why. Manns et al. (2000b) have shown
in humans that performance on VPC is predictive of subse-
quent recognition memory performance in a standard two-
alternative forced-choice test, whereas performance in
perceptual priming (a measure of perceptual fluency) is unre-
lated to recognition performance. This is helpful because it
suggests that, in humans, VPC really is in the declarative
domain. Subjects do have some “awareness” of having seen
one of two stimuli before. However, this analytical study does
not directly address the familiarity versus recollection issue.
The problem we face is that monkeys may be using explicit
recollection to direct their gaze, or they may be using stimulus
familiarity (or perhaps both). If VPC is a pure “familiarity”
task uncontaminated by recollection, it is unlikely to prove to
be of continuing value for understanding the role of the pri-
mate hippocampus in more complex aspects of event memory
in which recollection is definitively engaged (e.g., cued recall
in which one stimulus brings to mind another with which it
has been associated). Bachevalier suspected that it entails little
more than familiarity but went on to suggest that its greater
sensitivity may reflect the greater role that novelty plays with
respect to all the stimuli in the task.
In VPC, animals are passively exploring two-dimensio-
nal back/white novel stimuli, not actually memorizing
the sample to select a future response (i.e., incidental
learning). It is presumably more ecological for
606 The Hippocampus Book
monkeys (and humans) passively witnessing a new
event to keep a trace (however weak it is) of the whole
event, because anything can later prove to be behav-
iorally relevant (i.e., the stimulus, its elements, and its
spatial and temporal contexts). This incidental encod-
ing could favor the formation of conjunctive represen-
tations not only of the different elements of the sample
but also of its location and contexts. (Nemanic et al.,
2004, p. 2025)
We come back to this thoughtful comment in Sections 13.5
and 13.6, where the argument is presented that incidental
encoding of stimuli does not necessarily commit an animal to
being able to make only familiarity judgments about them
and that the formation of conjunctive representations, even if
encoded automatically, is an essential building block of recol-
lection. Whatever the psychological basis of the VPC task, pro-
ponents of declarative memory theory do, nonetheless, like it.
However, there is a lurking suspicion of circularity in this
attraction; unlike DNMS, the task is reliably impaired by hip-
pocampal lesions at reasonably short memory delays.
Again, what is needed is an information-processing analy-
sis of specific tasks that is independent of any tests of their
sensitivity to brain damage. DNMTS and VPC both suffer
from the problem that there are at least two ways in which a
monkey might perform the test. We need either new, less
ambiguous tests of memory or new behavioral assays to estab-
lish when an animal is performing these ambiguous tasks one
way or the other. Assuming that these could be developed, we
could then return to the main task of mapping the cognitive
process onto underlying brain systems and networks. Given
that stimulus familiarity suffices to solve the DNMTS task,
instances where excitotoxic hippocampal lesions have no
effect (Murray and Mishkin, 1998) may be because the task
has been set up to encourage no more than a judgment of
familiarity (extensive pretraining?). By the same token,
instances where a deficit is seen (Zola et al., 2000) may occur
because the control animals enjoy the benefit, at least on some
trials, of explicit recollection. In such cases, the hippocampus
may provide the processing necessary for remembering the
object or its image on a computer screen in its spatiotemporal
context.
If this analysis is correct, a novel prediction is that context-
specific recognition memory is impaired by discrete hip-
pocampus damage in monkeys. A possible experiment might
be one in which the monkey would be shown object A once in
context 1 and a short while later object B in context 2. After a
memory delay, it would then be presented with different types
of recognition judgment. Numerous novel objects would be
used across of a series of trial triads. Some tests would require
no more than an absolute familiarity judgment: Has either of
the two objects been presented before? Others would involve
making a context-specific judgment: Can the monkey indicate
that object B in context 1 (or A in context 2) is a novel condi-
tion? On the analysis just presented, discrete hippocampal
lesions might affect context-specific recognition memory but
not judgments of absolute familiarity. However, such a task
might only work in “incidental” mode. Deliberate training
with multiple objects in the different contexts, or objects over
multiple trials, may engage conditioning processes that utilize
configural cues mediated by neocortical circuitry (see Section
13.5). However, if this or another appropriate protocol could
be developed, disagreement about whether the hippocampus
is or is not involved in “recognition” might then move forward
toward discussion about qualitatively different types of recog-
nition memory. Such experiments would start the process
of fractionating “declarative” into different kinds of proposi-
tional knowledge, just as the “visual system” has been frac-
tionated into different streams of processing. Precisely such
context-specific recognition experiments are already under-
way using rodents (Dix and Aggleton, 1999; Eacott and Nor-
man, 2004). In primates, context-specific discrimination tasks
have been explored (Dore et al., 1998), showing deficits after
neurotoxic hippocampal lesions but not yet recognition
memory.
The era of primate lesion experiments on recognition
memory using DNMTS has probably drawn to a close. This is
partly because of the ambiguities discussed above but also
because the individuals particularly interested in these issues
have moved on and a new generation of primate researchers is
tackling other issues. Some primate lesion experiments are
underway using new tasks, and others are focusing on single-
unit and multiple single-unit recording during memory tasks
(Suzuki and Eichenbaum, 2000; Squire et al., 2004). One com-
mon feature of these experiments is abandoning the notion
that recognition and association are likely to be fundamentally
different; both may be associative processes. Buckmaster et al.
(2004) have developed tests of paired-associate learning that
are more sophisticated than merely pairing an object with
a reward. The animal must learn that object A goes with object
B. Tests of transitive inference and delayed spatial recall are
also being added to the arsenal of tests with which the multi-
ple types of memory in the primate brain will eventually be
uncovered. Similarly, Suzuki and her colleagues have exam-
ined the hippocampal single-unit correlates of paired-associ-
ate learning in monkeys, for new pairs and for well-established
pairs (Wirth et al., 2003; Yanike et al., 2004). We consider the
application of new behavioral tests of declarative memory
later in our discussion of relational-processing theory (see
Section 13.5).
13.3.5 Remote Memory, Retrograde
Amnesia, and the Time Course of
Memory Consolidation in Primates
Given the uncertainties of testing remote memory in humans
(see Chapter 12), there has been interest in examining
retrograde amnesia in animals. The value of doing this is
that studies can be done prospectively. The training experience
of laboratory animals can be accurately controlled, with no
ambiguity about precisely what events—and when or
where—have occurred prior to a lesion. The experimenter

knows and does not have to rely on the uncertain testimony
of relatives. There are, however, looming difficulties in the
use of animals for such studies, one being the critical episodic/
semantic distinction.
Zola-Morgan and Squire (1990) sought evidence for grad-
ual memory consolidation by first teaching monkeys a series
of 100 object discrimination problems. They were divided
into five sets of 20 problems scheduled at intervals of 2, 4, 8,
12, and 16 weeks prior to creating aspiration lesions in the
animals’ hippocampus and surrounding parahippocampal
cortex. Two weeks later they were retested on each of the 100
problems by presenting pairs of discriminanda just once (to
examine retention uncontaminated by new learning). The
lesioned monkeys were impaired relative to controls on the
problems learned shortly before surgery, but the two groups
performed at a comparable, above-chance level for problems
learned 12 or 16 weeks earlier (Fig. 13–10).
As just described, the results do not necessarily require ref-
erence to a concept such as memory consolidation – they
could reflect no more than damage to a storage site and dif-
ferential rates of forgetting in control and lesioned animals
(Fig. 13–10, pattern in panel B1). Of greater significance are
the within-subject comparisons. The controls did best on the
most recent problems and worst on the problems learned ini-
tially, a pattern that reflects gradual forgetting over time. In
contrast, the lesioned monkeys did not fail at both training-
lesion intervals (Fig. 13–10, data pattern in panel B2), but
actually did worse on problems learned 2 weeks before surgery
than those learned 12 weeks earlier (Fig. 13–10, the true pat-
tern, in panel B3). This dual pattern of performance—forget-
ting in controls but gradual improvement in the lesion
group—is critical to the interpretation of the data. It strongly
Figure 13–10. Retrograde amnesia and declarative memory theory.
A. Average performance during a single postsurgery probe trial for
each of 100 object discrimination problems learned earlier. The data
are plotted as a function of the time interval between training and
the subsequent lesion. Problems learned 12 weeks before surgery are
remembered better than those only 2 weeks beforehand. (Source:
Zola-Morgan and Squire, 1990.) B. Data to be expected on various
models of how lesions might affect storage sites or memory consoli-
dation. B1. The lesion causes partial disruption of the site of storage.
A Memory for object discriminations
90
80
% correct
70
C
H
60
recent remote
50
2 4 8 12 16
learning-surgery interval (weeks)
Theories of Hippocampal Function 607
suggests that a memory consolidation process must have been
taking place in normal animals that was interrupted by the
hippocampal/parahippocampal lesion.
Other primate studies failed to obtain positive evidence for
consolidation (Dean and Weiskrantz, 1974; Salmon et al.,
1987; Gaffan, 1993), but there are several reasons why no
temporal gradient may have been observed. Inferotemporal
lesions, as studied by Dean and Weiskrantz, could have dam-
aged the actual site of memory storage in the neocortex, mak-
ing it impossible to see the effect on remote memory of any
putative process of consolidation orchestrated by the MTL.
The study by Salmon et al. (1987), also using large MTL
lesions, showed little forgetting in control animals and very
poor performance in the lesioned animals—concerns about
ceiling and floor effects that led directly to the design of the
later experiment with more restricted lesions. Gaffan (1993)
examined picture memory with various retention intervals
prior to the administration of fornix lesions, but it had several
curious features: First, a different number of pictures were
presented in the set just before surgery than at the earlier time
point; and second, a retention test was given just before
surgery that could have reminded the animals of the pictures
and so altered the extent to which they can be considered
as exclusively belonging to the “recent” or “remote” set.
Indeed, demonstrable improvement in the performance of the
control group between the two retention tests strongly sug-
gested that reminding altered the effective “age” of the memo-
ries. Although this weakens the force of Gaffan’s study, the
possibility that recall can induce “re-storage” of information
and so alter trace strength should not to be ignored.
The study by Zola-Morgan and Squire (1990) is therefore
widely considered the definitive study of gradual time-
Performance by the lesion group is always poorer than that of con-
trols, although they may display a shallower gradient of forgetting.
B2. The lesion causes disruption of retrieval. Performance is poor at
all time intervals. B3. The lesion causes selective disruption of long-
term consolidation. Performance shows an inverted-U shape, being
better for problems when there has been time for consolidation
prior to the lesion but, like normal controls, also forgetting over
time. The exact form of the memory gradients is critical to the
interpretation.
B Theoretical models
B1 B2 B3
partial
storage retrieval disruption of
failure consolidation
damage
performance
chance
e
re nt
ot
ce
m
re
608 The Hippocampus Book
dependent memory consolidation in primates, although there
are aspects of the results that are troubling. The use of lesions
encompassing both the hippocampus and the parahippocam-
pal cortex raises the question of whether damage restricted to
the hippocampus would cause a retrograde effect. This is
important given the lack of much retrograde amnesia in
patients R.B. and G.D. (see Chapter 12). A study using mon-
keys in whom ischemic and/or neurotoxic lesions were created
could offer a more exact model of the etiology and damage in
these patients. To our knowledge, this has not been done. The
use of a within-subjects design is a strength (with respect to
mimicking the human syndrome and allowing modest use of
animals), but it raises the question of whether the repeated act
of “reminding” the animal of the testing situation preopera-
tively could influence the memory strength of earlier trained
items. Putting a monkey back into the WGTA apparatus to
learn a new set of problems 4 weeks after it has learned an ear-
lier set could re-activate memory representations of the first
trained problems even though the specific stimulus items are
not presented again. This could happen because the training
context may be sufficient to activate hippocampus-specific
“pointers.” Put simply, being in a place you have been in pre-
viously reminds you of things that happened there and may
make these remembered events more memorable in the
future, a process that is more akin to a cyclical “reactivation
and re-storage” process than an inexorable time-dependent
consolidation process. Insofar as this did happen in the
within-subjects design, it would particularly affect the prob-
lems learned earliest before surgery—the very ones remem-
bered best by the lesioned animals (a different problem from
that in the Gaffan study where reminding occurred just before
surgery). This argument would be more convincing had there
been an improvement in memory for these same “early” prob-
lems in the control group, but the controls exhibited forget-
ting over time. here is also the further possibility of all animals
developing a “learning-set” over the succession of discrimina-
tion problems (Murray and Bussey, 2001). What is needed to
settle this issue is a systematic comparison of training condi-
tions that involve repeated reexposure to the context and of
training conditions that do not. Such a study is now most
unlikely to be conducted in primates, but fortunately this
issue has been followed up using rodents.
Studies of retrograde amnesia in nonprimate species have
been conducted using social transmission of food preferences,
cue and context fear conditioning, trace-eyeblink condition-
ing, spatial learning, and object discrimination learning tasks
(e.g., Winocur, 1990; Kim and Fanselow, 1992; Bolhuis et al.,
1994; Anagnostaras et al., 1999; see Squire et al., 2001 for
review). Some studies suffer from “ceiling” or “floor” effects in
the data (i.e., performance being so good or so bad across a
range of intervals that not all studies have both a forgetting
and a consolidation gradient). However, the general pattern is
of temporal gradients of retrograde amnesia consistent with a
gradual process of memory consolidation, with the sole
exception being certain spatial tasks that are discussed in
detail later (see Section 13.4). Data showing gradual time-
dependent changes in 2-deoxyglucose utilization in the hip-
pocampus after learning are also consistent with the idea that
increases in hippocampal activity occur for a limited period
time after learning (Bontempi et al., 1999; Frankland et al.,
2004; Maviel et al., 2004b). These changes may constitute a
physiological and molecular signature of consolidation.
There are numerous outstanding issues concerning mem-
ory consolidation as envisaged within the declarative memory
theory. A key issue is what determines whether new memories
are consolidated or allowed to fade. Most event memories that
humans make automatically during a typical day are lost;
some passive or active selection process must come into play
that determines what is retained or discarded (Morris et al.,
2003). Second, once set in motion, is consolidation a gradual,
inexorable, largely time-dependent process? Or is it a more
“quantal” process in which short consolidation episodes occur
repeatedly over a longer period of time, perhaps triggered by
contextual retrieval (Dudai and Morris, 2001)? Both gradual
and quantal ways of thinking about consolidation could result
in gradual temporal functions of retrograde amnesia when
average data are considered, but the underlying mechanism by
which individual traces are strengthened might be very differ-
ent. Third, how does consolidated information become inte-
grated or interleaved with information already stored in the
neocortex (McClelland et al., 1995)? Fourth, should a distinc-
tion be made between “consolidation” and “semanticization”?
The former would allow the persistent memory of discrete
events and enable the “recovered consciousness” and mental
time travel characteristic of episodic memory (Moscovitch,
1995). The latter would involve an interleaving process
through which the regularities that emerge from successive
similar episodes are abstracted and so be able to add to the
subject’s semantic knowledge. The later recall of such infor-
mation need not entail access to spatiotemporal tags
(Winocur et al., 2005). Fifth, is there just one type of memory
consolidation, or is there a family of distinct processes? Some
have argued for a distinction between “cellular” consolidation
that putatively operates rapidly within single neurons in a sin-
gle brain area, and “systems-level” consolidation involving the
interaction of different brain areas (Dudai and Morris, 2001).
Finally, what are the indices or pointers, what do they index or
point at, and how is neuronal and synaptic specificity in neo-
cortex realized by the signals emanating from hippocampus?
These and related issues constitute an important next step in
the development of declarative memory theory.
13.3.6 Critique
The declarative memory theory of hippocampal function has
been immensely influential and remains the most cited theory
of memory in neuroscience textbooks (Kandel et al., 2000;
Bear et al., 2001). Yet, although simplicity is desirable, one
should always remember Einstein’s dictum that scientists
should “keep things as simple as possible, but no simpler.”
Several critics empathize with this view, two describing the
theory bluntly as follows: “It is attractive, it is parsimonious, it

is extraordinarily popular, and it is wrong” (Murray and Wise,
2004, p. 194). Others, notably Gaffan (2002), go further in
opposing the whole concept of memory systems, particularly
that of a unitary declarative memory system encapsulated
entirely within the MTL. Whether these and other critiques are
well founded or merely idiosyncratic is a matter of opinion.
What is clear is that several problems with declarative
memory theory are now widely recognized. They range
from empirical issues to do with how well the data really do fit
the theory, many of which have already been discussed, to
conceptual issues such as the status of the taxonomy of mem-
ory systems, the role of conscious awareness in the encoding
and retrieval of declarative information in humans and ani-
mals, and problems surrounding the concept and time scale of
systems-level memory consolidation.
What Is the Status of the Memory Taxonomy?
The taxonomy depicted in Figure 13–3 is easy to grasp and
easy to remember. It is a joy to teach, and many a lecture in
research seminars begins with this diagram—but what does it
really mean? Is it meant to be an accurate depiction of evolu-
tionary or logical distinctions between different forms of
learning and their mapping onto specific structures in the
human brain? Or is it actually no more than an aide memoire?
Taking issue with the taxonomy may seem no more than a
semantic side show; but if the taxonomy is conceptually con-
fused, it is important to reflect on this. The term “taxonomy”
is used in evolutionary biology to denote “relatedness,” such as
between species, orders, and groups. Clearly, the use of the
term here is not intended to imply that the evolution of mem-
ory proceeded first through some ancient differentation
between short-term and long-term memory and then on
through all the binary divisions of the hierarchy. To the con-
trary, it is generally assumed that nondeclarative memory
evolved earliest (Sherry and Schacter, 1987). The term “taxon-
omy” is being used in a different sense—one expressing osten-
sibly qualitative distinctions between types of memory.
Nonetheless, some critics of declarative memory theory feel
the need to move beyond mere taxonomy to more precise,
noncircular statements of what is different (in information-
processing terms) about the identified forms of memory
depicted within the framework. Do the distinct nodes of the
taxonomy have different “inputs” such that they operate on
different types of information? Schacter and Tulving (1994),
for example, argued for the flow of information into memory
occurring in a serial, parallel, independent manner—the
information passing first via perceptual representational sys-
tems, then through semantic memory, and on into episodic
memory (Fig. 13–11). Activated subsets of semantic and
episodic memory constitute the inputs to working memory,
thereby creating a set of input-output relations between mem-
ory systems quite different from those envisaged in the declar-
ative memory theory. Continuing in this vein, we may ask of
the standard taxonomy (and of Schacter and Tulving’s alter-
native), whether the memory systems in it use different learn-
Theories of Hippocampal Function 609
Tulvings’s alternative SPI framework for propositional memory
Sensory inputs
Flow of Perceptual representation system
activated
subsets of
information
Working
Semantic memory
memory
Episodic memory
Figure 13–11. Tulving’s SPI model of propositional memory. An
alternative framework for the flow of information into and between
memory systems. Information flows into memory systems that
operate in a serial, parallel, independent (SPI) manner (Schacter
and Tulving, 1994). The flow is from a perceptual-representation
system into semantic memory and then into episodic memory.
Activated subsets of these systems interact with working memory
(Baddeley, 2001).
ing rules. Do they encode, store, and retrieve information
differently and so provide “outputs” different from those of
the rest of the brain? Do their storage mechanisms express
memory traces differently, with differing patterns of persist-
ence or susceptibility to consolidation? Others are concerned
that certain types of memory storage may start off in one way
(e.g., declarative) but then become another (e.g., nondeclara-
tive) through repetition in multiple contexts and/or the pas-
sage of time. That is, the ostensibly sharp boundaries of any
memory taxonomy may neither be immutable nor adequately
reflect dynamic changes in memory representation that occur
during the course of learning. However, to be fair to the archi-
tects of declarative memory theory, these are in part issues for
the future and precisely the sort of topics they have been
attempting to address, notably in the many studies character-
izing the nature of preserved learning in the presence of
amnesia (Squire et al., 1993; Cavaco et al., 2004).
These computational- and algorithmic-level questions
have to be addressed before we can proceed securely toward
any mapping onto the neural substrate in which they are
expressed—at the circuit, cellular, and even intracellular levels
of analysis. As currently drawn, the taxonomy implies that
types of learning and memory at the bottom of the hierarchy
can be mapped onto specific brain areas, ranging from regions
in the MTL to such structures as the striatum and cerebellum.
Matters, however, are unlikely to be so simple. One controver-
sial claim, made by Murray and Wise (2004), is that radically
new concepts of the embryology and anatomical organization
of major parts of the primate and rodent brain need to be
taken on board by memory researchers, notably ideas most
closely associated with those of Swanson (2000, 2004). One
aspect of this is a shift away from thinking of brain functions
as localized to discrete brain regions or to closely connected
brain areas such as MTL, toward thinking in terms of “recur-
610 The Hippocampus Book
rent loops.” It is too early to assess this idea securely. However,
one lesson of the last decade of functional imaging research in
humans is that distributed cerebral networks for memory
involving “top-down” interactions between the medial tempo-
ral and frontal lobes need to be considered and analyzsed
(Fletcher and Henson, 2001; Miyashita, 2004).
A last point about the taxonomy is the growing concern is
that it does not readily capture the sense that the “seamless”
control of behavior is almost certainly a matter of the coordi-
nated regulation of numerous brain networks. When learning
to drive a car for example, there is perceptual learning, motor
control, knowledge of facts about road signs, and memory for
previous similar traffic situations that one may have encoun-
tered. The taxonomy is suspiciously silent about how the out-
puts of the ostensibly independent forms of memory interact,
compete, or are coordinated in the brain. Experimental work
to date has been devoted largely to devising tasks that dissoci-
ate the psychological and anatomical components of memory
processing rather than investigating how these entities com-
pete or cooperate. As noted in Section 13.2, new developments
in memory research are turning toward this more synthetic
goal (Poldrack et al., 2001; White and McDonald, 2002). None
of these comments and criticisms on their own definitively
indicate that the standard memory taxonomy and its mapping
onto brain areas is wrong, but they are grounds for caution.
Inconsistencies Between the Animal Lesion
Data and the Declarative Memory Theory
Several lines of inquiry have turned up data that appear incon-
sistent with the declarative memory theory or, at best, are han-
dled awkwardly. Much of this is considered later in relation to
other theories of hippocampal function, notably spatial, con-
figural, and episodic theories. Studies of the neural basis of
recognition memory in rats and of other forms representa-
tional memory in primates have been conducted as explicit
tests of the declarative theory and so are considered here.
Hardly unique in science, there is conflict about the data.
Like the proverbial housewives living in medieval tenements
who always bickered at each other across the narrow alleyways
(they were always arguing from different “premises”), there is
a persisting conflict about the way studies of recognition
memory in rats are perceived. Mumby summarized the situa-
tion fairly as one in which “most investigators are looking out-
side the hippocampus” (Mumby, 2001, p. 159) to explain the
neurobiological basis of recognition memory. In contrast,
Squire et al. (2004) continued to defend the view that damage
to the hippocampus does impair recognition memory at long
memory delays.
A range of techniques have been developed to study recog-
nition memory in rodents that complement the DNMTS and
VPC paradigms for monkeys. There are several DNMTS pro-
tocols in which rats are explicitly trained to learn a matching
or nonmatching rule and selectively rewarded on correct
choice trials. They include a Y maze nonmatching goal-boxes
procedure (Aggleton, 1985; Aggleton et al., 1986), tasks in
which rats traverse runways to displace goal objects that are
either familiar or unfamiliar (Rothblat and Hayes, 1987;
Mumby et al., 1990; Kesner et al., 1993), and even procedures
for rats using images on computer screens (Gaffan and Eacott,
1995). A continuous delayed nonmatching paradigm, teas-
ingly called cDNM, uses a “go/no-go” digging response for the
recognition of odors (Otto and Eichenbaum, 1992).
Spontaneous novel object recognition (NOR), analogous to
the VPC protocol for primates, is an increasingly popular par-
adigm for recognition memory first developed by Ennaceur
and Delacour (1988). After prior habituation to the context of
testing, the animal is confronted by two objects that it investi-
gates and explores, though never formally rewarded for doing
so. The objects are generally identical in a sample phase but
differ in the memory test phase. Trial-unique goal boxes,
objects, or smells are used for rodent DNMS, cDNM, and
NOR, with the relative probability of the choice or the extent
of spontaneous investigation of the novel cue serving as the
index of memory of prior occurrence. One issue of concern
has been that not all of the rodent studies have routinely var-
ied the memory delay within the protocol as has been the cus-
tom in the primate work; this is important, as use of only
short memory delays can yield inappropriate conclusions
(Clark and Martin, 2005). There are also subtle but important
differences of protocol across ostensibly similar paradigms.
For example, some studies of NOR require all subjects to
accumulate some minimum period of object exploration dur-
ing each trial (e.g., 30 seconds), these trials of necessity then
being of indeterminate duration; others have a set trial dura-
tion but then average the normalized exploration scores across
subjects. Clark et al. (2000) presented data that favor the for-
mer approach, and they thereby set new standards for these
types of experiment in rats (a move by the San Diego group
into rodent work that may have been prompted by a certain
frustration about the way so many rodent experiments were
being conducted). Mumby (2001) provided a perceptive dis-
cussion of many other differences between these and the pri-
mate tasks discussed earlier.
Most of the data reported on rats indicates that restricted
hippocampal dysfunction (lesions of various kinds, fornix
section, intrahippocampal drug infusions) has minimal effect
on these tasks (see Table 1 in Mumby, 2001). In contrast,
perirhinal and postrhinal cortex lesions, including reversible
inactivation using glutamatergic ligands and disruption of
cholinergic neuromodulation, cause clear deficits (Bussey et
al., 1999; Warburton et al., 2003; Winters et al., 2004; Winters
and Bussey, 2005). There are exceptions to this pattern, but
when taken as a whole it is inconsistent with the prediction
from declarative memory theory that damage anywhere
within this group of MTL structures, including the hip-
pocampus, should cause a proportionate deficit in recognition
memory. Starting with a study by Aggleton et al. (1986) using
novel and familiar goal-boxes in a Y maze, a series of analyti-
cal studies led by Rawlins resolved why hippocampal dysfunc-
tion impaired nonspatial working memory in a radial maze
but appeared to have no reliable effect on DNMTS (Rawlins et

al., 1993; Steele and Rawlins, 1993; Cassaday and Rawlins,
1995, 1997). It turned out that the size of the goal box is crit-
ical, with small compact goal boxes being treated by rats as
discrete “objects,” whereas larger boxes are treated as “spaces”
and so engage spatial and relational encoding (see Sections
13.4 and 13.5). Hippocampal lesions only affect memory for
large goal boxes. These findings have been complemented by
concern that certain protocols for testing nonspatial recogni-
tion memory may have cryptic spatial or contextual compo-
nents. For example, Nadel (1995) was the first to point out
that the deficit in DNMTS at long memory delays in certain
primate studies was confounded by the monkeys being, for
practical reasons, removed from and then later returned to the
WGTA testing apparatus at the longer but not the shorter
delays. Control but not hippocampus-lesioned monkeys
might then benefit from contextual cues aiding recall.
However, cryptic spatial processing is unlikely to be an
issue for a different DNMTS protocol developed by Mumby et
al. (1992a). They trained rats with memory delays varying
from 4 to 600 seconds with the sample and choice compo-
nents at opposite ends of a runway during a trial but sched-
uled equally often at both ends across trials. Subsequent to
surgery, no impairment was seen in rats with aspiration
lesions of the hippocampus, or of hippocampus and the
amygdala, at any delay except the longest (10 minutes) at
which the hippocampal lesion groups then performed more
poorly than controls. Although sometimes cited as support for
the idea that hippocampal lesions can affect recognition
memory in rats, there is a caveat here also. Specifically, the
within-subject comparisons showed that the control group
got paradoxically better at the 10-minute delay rather than the
lesion group getting worse. This improvement by controls was
not seen in several later studies, and thus Mumby’s (2001) rea-
sonable conclusion was that this single statistical difference
was the exception to the rule that hippocampal lesions have
minimal if any effect on rodent DNMTS. However, using a
large test battery in the manner of Zola-Morgan and Squire
(1985), Mumby et al. (1995) again found a small but signifi-
cant hippocampal lesion deficit in DNMTS at the longest
delay tested (120 seconds) relative to both no surgery and par-
tial parietal cortex lesion control groups. Clark et al. (2001)
later also secured a deficit in rats with excitotoxic hippocam-
pal lesions at the longest delay tested (also 120 seconds). Thus,
another equally reasonable way of summarizing the data is
that when hippocampal lesion-associated deficits have been
seen they have tended to be at long memory delays or with
long list lengths (Mumby et al., 1992b; Steele and Rawlins,
1993; Mumby et al., 1995; Wiig and Bilkey, 1995; Clark et al.,
2001). This would be consistent with the declarative memory
theory were it not for the much greater sensitivity of these
tasks to perirhinal lesions of comparable or even smaller size.
Either the hippocampus is at the “apex” of declarative mem-
ory processing or it is not.
What might be the key difference between studies that find
a deficit after hippocampal lesions (Clark et al., 2000;
Broadbent et al., 2004) and those that do not (Ennaceur and
Theories of Hippocampal Function 611
Aggleton, 1994; Winters et al., 2004; Forwood et al., 2005)?
Lesion size or, more generally, the degree of hippocampal dys-
function is one possibility. For example, Ennaceur and
Aggleton (1994) found no effect of fornix lesions on NOR at
delays varying from 1 to 15 minutes but dropping to chance
levels over 4 hours. Other studies using fornix lesions reported
similar results, but it should be remembered that fornix
lesions in monkeys have little impact on DNMTS and may
leave many aspects of MTL function intact. Perirhinal lesions
have consistently been observed to impair NOR (Bussey et al.,
1999; Murray and Bussey, 1999). Consistent with these find-
ings but drawing a very different conclusion, Clark et al.
(2000) conducted a comprehensive study of NOR using
groups of rats subjected to sham surgery, fornix lesioning, or
radiofrequency and excitotoxic hippocampal lesioning.
Importantly, they required the absolute accumulation of 30
seconds of total object exploration time by all subjects.
Fornix-lesioned subjects did not show a deficit, consistent
with earlier findings. However, rats with either type of hip-
pocampal lesion showed impairments at a 1-hour memory
delay. In a follow-up study by Broadbent et al. (2004), a deficit
in NOR was observed as a function of hippocampal lesion
size. Small lesions caused no impairment, whereas larger
lesions did. Given the parallel, distributed nature of hip-
pocampal processing (see Chapter 14), it is reasonable to sup-
pose that many functions of the MTL that require
hippocampal processing could continue relatively normally
after partial lesions.
Finally, turning back to primates, Murray and Wise’s
(2004) critique of the declarative memory theory offers sev-
eral other examples of instances where the two sides in the
debate draw different conclusions from a common set of data.
One example has to do with whether the perirhinal cortex has
only memory functions (Buffalo et al., 1998a; Teng et al.,
2000) or also participates in aspects of perceptual processing.
For example, Teng et al. (2000) reported that rapid learning of
a discrimination between simple three-dimensional objects
(e.g., a red versus a green peanut shell) is only slightly
impaired by hippocampal lesions, notably over the first
few trials of the day (which were described as more “declara-
tive” in nature), whereas the much slower learning of two-
dimensional pattern discrimination (e.g., N versus W) is
unaffected by hippocampal lesions unless they extend to
include damage to the tail of the caudate nucleus. In contrast,
Bussey et al. (2002) reported that discrimination of com-
pound visual stimuli with the ambiguity between the features
maximized is strikingly impaired by perirhinal lesions,
whereas the discrimination of those with minimal common
features is unaffected. A follow-up study by Bussey et al.
(2003) used morphed images to produce a single-pair dis-
crimination task in which the two stimuli to be discriminated
had either very high or very low feature ambiguity (Fig.
13–12). The results show a clear perirhinal deficit on the
slowly learned, maximal feature ambiguity task. These find-
ings are problematic for the declarative memory theory for
two reasons. First, the theory asserts that the MTL receives
612 The Hippocampus Book

A Low feature ambiguity

100 100
90 90

mean % correct
80 80
% correct

C C
70 PRh 70 PRh

60 60
50 50
40 40
1 2 3 4 5 6 7 8 9 10 11 12

B High feature ambiguity

100 100 **
90 ** ** ** 90
** ** *

mean % correct
80 ** 80
C
% correct

* C
70 PRh 70 PRh

60 60

50 50

40 40
1 2 3 4 5 6 7 8 9 10 11 12
Sessions
Figure 13–12. Perirhinal cortex and perception. Lesions of the perirhinal cortex, ostensibly
part of a medial temporal lobe memory system, cause a deficit in the perceptual discrimination
of stimuli with high-feature ambiguity (B) but not low-feature ambiguity (A). The symbols in
the bar graphs on the right indicate scores of individual monkeys. (Source: After Bussey et al.,
2003.)

perceptually processed visual inputs from the inferotemporal
cortex and is not itself involved in making perceptual dis-
criminations. Second, it asserts that rapidly learned visual dis-
crimination tasks, other things being equal, are most likely to
be learned in a declarative manner whereas slowly learned dis-
criminations are acquired in a habit-like way, as in the Teng et
al. (2000) study. Yet here it is the slowly learned discrimination
that is most sensitive to lesions of the perirhinal cortex.
Indeed, a body of data from rats, monkeys, and humans as
well as relevant computational modeling support the notion
that the perirhinal cortex has a role in object perception
(Murray et al., 2005). Data from amnesic patients with
restricted hippocampal damage or damage extending into
neocortical structures of the MTL (identified radiographi-
cally) indicate that the discrimination of faces and scenes of
ever greater similarity can pose a particular challenge for
patients (Lee et al., 2005).
Certain Comparative Problems When Asserting
that Declarative Memory Must Be Conscious
The insistence on MTL structures mediating memory that can
be consciously declared is clearly problematic for a theory that
seeks to encompass both humans and animals. In numerous
laboratory tasks, animals have been shown to remember
things that people would describe as facts (e.g., that a certain
food is safe to eat) and that they can remember events (e.g.,
that an initially novel object has been seen before). However,
will we ever know whether an animal is conscious of its mem-
ories? Moreover, even if we did, how could an animal ever
“consciously declare” that it knows or remembers something
from the past? Does it have a sense of its own life in the way
that humans do—of the state of mental awareness that
Tulving (1983) called “autonoetic consciousness”? If the
answers to these questions are negative, one is tempted to
wonder whether a central feature of the theory lies more in the
realm of metaphysics than in empirical science. In fact, a com-
prehensive theory of brain and mind should identify under
what circumstances animals and humans are conscious, what
they are conscious of, and perhaps why consciousness is
advantageous for some but not other forms of memory
(Zeman, 2002).
This issue is usually finessed in what was referred to earlier
as the “folk psychology” approach to consciousness that the
declarative memory theory has taken to date. Nonetheless, in
the context of work on “blindsight,” animal experiments are
making inroads into the issue of mental awareness. For exam-
ple, Cowey and Stoerig (1995) devised an ingenious procedure
 Theories of Hippocampal Function 613

in which monkeys that had been trained to reach accurately
toward one of several targets were also required to “report”
whether they were visually aware of the targets. Monkeys were
placed in front of a touch screen. Each trial began with them
looking at a fixation point and ended with their reaching out
to touch various images presented a little while later in the left
or right visual field. Operated monkeys (with large unilateral
striate cortex lesions) were observed to reach as accurately in
their blind hemifield as normal monkeys (Fig. 13–13A). The
“reporting” of awareness was achieved in a separate training
condition in which the monkeys had to touch a specific area
of the screen if a target failed to occur at a time when they
might have expected to see one (the possibility of a target was
indicated by an auditory cue). The lesioned monkeys correctly
reported that they could not see a target when targets were
deliberately not presented on selected trials to their intact
hemifield, but the monkeys incorrectly reported not seeing tar-
gets that were actually presented to their blind hemifield (Fig.
13–13B). These “unseen” targets were the very ones to which,
in the first training condition, the animals had reached accu-
rately. To all intents and purposes, these monkeys lack “phe-
nomenal” vision; that is, they lacked the ability to “comment
on” seeing things to which they could reach accurately
(Weiskrantz, 1997).
Analogous procedures might be developed for studies of
memory in primates, the point being to devise a way of disso-
ciating between the successful performance of a memory task
and the quite separate display of awareness that one is (or is
not) remembering. In an important step, Hampton (2001)
exploited techniques developed in parallel for pigeons (Sole et
al., 2003) to examine whether rhesus monkeys know when
they remember. Using computer-controlled techniques, each
of two monkeys was briefly shown an image on a touchscreen
(Fig. 13–14). The image then disappeared for a delay interval
during which the animals may have sometimes remembered it
and other times forgotten it. They were later tested for their
memory of the image. What made this experiment different
from conventional delayed match-to-sample testing was that
the monkeys were allowed, for two-thirds of the trials, to
choose between progressing to the memory test or declining
to do it. Declining resulted in a guaranteed but less preferred
reward than could be obtained by accepting to do the memory
test and choosing correctly. For the remaining one-third of the
trial sequences, the monkeys were not given a choice and were
forced to take the memory test. The results showed that both
monkeys performed more accurately on memory tests they
had opted to take than on enforced tests. A control experiment
included trial sequences that began without an image to
remember during the delay interval; the monkeys routinely
reacted to these sequences by declining to do the memory test
when given the option to do so. Moreover, as the memory
delay interval was extended (a procedure likely to promote
forgetting), the monkeys showed a temporally graded reluc-
tance to take the memory tests. Taken together, these results
suggest that the monkeys’ decision whether to take the mem-
ory test at the end of a trial sequence was likely to have been
based on their own self-generated “awareness” of whether they
could remember the image. This sort of procedure might be
adapted to establish whether lesions of the hippocampal for-
mation (or, better still, reversible inactivation) disrupt aware-

Figure 13–13. Blindsight in primates. Cowey

and Stoerig’s (1995) “commentary” procedure
for studying visual awareness in primates. A. The
animals watch the fixation spot (center) and
then upon hearing a tone reach toward one of
four peripherally located and briefly illuminated
targets. Monkeys with striate cortex lesions can
do this as well as control animals. B. The animals
again watch the fixation spot and, upon hearing
the auditory cue, reach for one of the five briefly
illuminated lights on the left, the light on the
right or, when no flash occurs, for the panel at
the top left. Normal monkeys correctly identified
all trial conditions. Monkeys with unilateral stri-
ate cortex lesions identified lights flashed in
their good hemisphere but treated the right-
hand illumination as if it had not occurred.
Thus, monkeys display “blindsight” in being able
to reach toward objects they “report” being
unable to see. Can an analogous “reporting” pro-
cedure be developed in the domain of memory?
(Source: Cowey and Stoerig, 1995).
614 The Hippocampus Book
A Metamemory - experimental design
Study

phase
Delay
Delay
interval
p = 0.33 p = 0.67
Choice
phase
Test phase
or small
reward  
Preferred Primate
Peanut Pellet
if Correct
Figure 13–14. Metamemory. A. Monkeys were allowed, during two-thirds of the trials of a
recognition test, to choose whether to take a memory test on the basis of their covert judgment
of whether they could remember. B. Memory performance on trials when they chose to take
the recognition test was significantly better than on the remaining one-third of trials for which
they were forced to take the test. (Source: Cowey and Stoerig, 2001).
ness of memory to the same extent that they disrupt accurate
performance. Might monkeys with hippocampal disruption
show above-chance familiarity for a previously presented
stimulus even though they “declare,” in the optional choice
test, that they would prefer not to take the test? Might they
know but not remember? A human imaging study of this issue
(Kao et al., 2005) suggests that such a task, even if it could be
developed for monkeys, would be likely to involve an interac-
tion between the MTL, mediating successful memory, and the
ventrolateral prefrontal lobe, mediating memory awareness.
Nondeclarative Memory: Are Spared Learning
Abilities Nonpropositional or Learned Tasks
That Can Be Performed Without Awareness?
A critical claim of the theory is the independence of declara-
tive and nondeclarative memory. The two key differences
claimed by Squire (1992) for these superordinate categories of
memory relate to the “propositional” status of the informa-
tion being processed and the necessity of “awareness” during
the encoding and recall of such information.
The existence of spared learning in the presence of amne-
sia has been the subject of comment for about a century
(Claparede, 1911; Weiskrantz, 1997). It is a particularly
intriguing feature of the syndrome to discuss with students or
present to a public audience. How can it be that H.M. could
learn to draw in a mirror yet not remember doing so? How
could he and other patients have this curious disconnection in
B Recognition Memory
100
Choose
Proportion correct
Memory
Test
50
Forced
Memory
Test
0
C.
 
Proportions


their conscious awareness of the past? Anecdotes about spared
“unconscious” memory function abound. Each new genera-
tion of neuropsychologists add their own; an excellent exam-
ple is an endearingly personal article entitled “Memories of
H.M.” (Ogden and Corkin, 1991). Patient E.P.—one of the
most profoundly amnesic patients to have been subject to
detailed neuropsychological and neuroradiological examina-
tion in the San Diego neuropsychological studies—is another
case in point.
During the first 2 or 3 years in which we visited his
house, he was wary and slow to accept the idea that we
wished to talk with him and administer tests. After
some conversation, and with encouragement from his
wife, E.P. would after a number of minutes seat himself
at a table for testing. During the subsequent years, the
same tester has visited his house more than 150 times.
Now when she arrives he greets her in a friendly man-
ner and moves readily and promptly to the table even
when his wife is not present. Yet, his pattern of greet-
ing and acceptance occurs without any recognition of
who the tester is, and he will repeatedly deny that he
has seen her before. (Stefanacci et al., 2000, p. 731.)
The declarative memory theory, perhaps more than any
other theory of memory, has put forward this disconnection
of consciousness as a centerpiece. It makes the distinctive
claim that such “nondeclarative” learning is not merely
spared, as others had claimed previously, but that such learn-

ing occurs in amnesics at a normal rate and in a manner indis-
tinguishable from that of controls. To emphasize this key
point, it is not only that amnesic patients can display classic
conditioning, as Warrington and Weizkrantz (1968) were the
first to describe, it is that they do so at a normal rate despite a
lack of awareness of the fact of learning (Gabrieli et al., 1995).
However, because certain ostensibly “nondeclarative” tasks
may be subject to contamination of learning using a declara-
tive strategy, findings from patients with mild amnesia who
have residual declarative memory can be misleading. The
importance of drawing firm theoretical conclusions only from
those rare individuals with severe amnesia was a theme of an
early critique of the declarative memory theory (Weiskrantz,
1997) and, in a paradoxical twist, one now taken up by Squire
et al. (2004) in their careful comparisons of performance by
people and animals with partial versus complete damage to
the MTL (e.g., patient R.B. versus patient E.P.).
Nondeclarative learning does not, of course, usually take
place outside of consciousness. Learning to ride a bicycle, for
example, is a motor skill for which most of us are acutely con-
scious of what we are doing as we attempt to learn. The declar-
ative memory theory makes the radical claim, however, that
although aware of the act of learning this awareness does not
play any causal role in the encoding or expression of learned
motor acts. This is a rather interesting idea but one that is dif-
ficult to test rigorously. It cannot be tested by seeing what
learning capacity remains while we are unconscious—not
because we cannot then ride bicycles (though that too) but
because the lack of “awareness” to which the theory makes ref-
erence is not a lack of consciousness as such but an absence of
awareness of the information relevant for learning at the
moment of motor recall.
This absence of the referent of awareness is highly specific
in the nondeclarative domain. For word-stem completion
priming, for example, people are asked to think of a word—
any word—that begins with the stem given as a cue. Subjects
are aware they are being tested, aware that the test involves
words, and attentive to the task in hand. Amnesic subjects,
however, are unaware that the words they come up with are
often words they were shown earlier. What they fail to
“declare” is not the words themselves, but that they have the
phenomenological attribute of being words they remember
having seen earlier. This simultaneous sense of being tested in
the present and bringing to mind events from the past is what
is meant by “mental time travel” (Schacter and Tulving, 1994;
Tulving and Markowitsch, 1998), and it is an attribute of mind
that amnesics lack (Weiskrantz, 1997).
In a similar vein, studies of incidental learning suggest that
unilateral MTL lesions can disrupt memory for information
to which a person’s attention is not overtly directed until the
point of recall. In a famous study, Smith and Milner (1981)
gave subjects a series of small toys, one by one, asking them
the likely price of the real thing (e.g., a car). Their attention at
encoding was directed to the value of the toys. One by one, the
toys were put down by the experimenter in various places on
the table in front of the subjects, without the subject’s atten-
Theories of Hippocampal Function 615
tion being drawn to the matter of location. Later, without
prior warning, the experimenters asked subjects to recall
where the objects had been placed. The subjects were never
asked to pay attention to and, in that sense, be attentively
aware of the location of the objects until the retention test.
That normal subjects could do this task suggests that they may
automatically encode attributes of a stimulus to which their
attention is not directed despite paying attention to other
attributes of the stimulus, such as its value. In contrast, people
with large right MTL lesions were impaired. In this case, the
experimental manipulation of “awareness” occurred at the
point of encoding, but manifestations of this differential
awareness only came to light at the point of recall. We return
to the issue of differentiating “automatic” and “intentional”
encoding in Sections 13.5 and 13.6.
This discussion leads us to suggest that the nondeclarative
nature of nondeclarative memory may be less to do with
whether the skill being performed can ever be “declared” than
with whether subjects are consciously unaware of the prior
occurrence of the stimulus at the time of recall. The impor-
tance of awareness rather than propositional status is particu-
larly well revealed in studies of eyeblink conditioning, an
advantageous paradigm for the present purposes as it can be
studied in both humans and animals. In this form of condi-
tioning, an initially neutral stimulus such as a tone is pre-
sented for a few hundred milliseconds before a puff of air to
the eye. Through repeated pairings of these two stimuli, the
subject develops a conditioned response (CR) to the tone (the
conditioned stimulus, or CS). Eyeblink conditioning has been
extensively studied in humans and animals and is known to
obey the basis principles of classic (i.e., Pavlovian) condition-
ing (see Section 13.5). Learning depends on the contiguity and
contingency between the CS and the puff of air (the uncondi-
tioned stimulus, or US). It is a form of conditioning for which
the cerebellum was found to be essential (Clark et al., 1984).
Following the observation that amnesics could acquire classic
conditioning (Warrington and Weizkrantz, 1968) and that it
occurs at a normal rate (Gabrieli et al., 1995), it had been
widely assumed that this is not a form of learning in which the
MTL is ever engaged. This turns out to be not quite right.
Two frequently studied forms of human eyeblink condi-
tioning are named (somewhat inappropriately) the “delay”
paradigm (in which, confusingly, the US follows the CS with
no delay between the end of the CS and the onset of the US);
the “trace” paradigm in which CS offset occurs 300 to 1000 ms
prior to US onset (Fig. 13–15). As the CS and US are not con-
tiguous in time, Pavlov suggested (and others since) that some
“trace” of the CS must linger in memory if an association is to
be formed with the US to allow effective conditioning (hence
the name). Although no deficit is seen in amnesics using the
delay paradigm, McGlinchey-Berroth et al. (1997) have shown
that such patients have impaired trace eyeblink conditioning
across a range of trace intervals. The contrast between these
two protocols is instructive as what subjects learn to do in
each case—to make automatic, appropriately timed CRs—can
hardly be said to be nonpropositional in the former case but
616 The Hippocampus Book
Delay conditioning Trace conditioning
CS
US
time
propositional in the latter. It is therefore an important test for
the theory to identify what might be “declarative” about trace
conditioning.
Clark and Squire (1998) used a differential trace condi-
tioning paradigm in which the CS (a tone) was followed by
the air-puff US, whereas a second stimulus, the CS- (a noise),
was presented on its own. Sessions consisted of multiple pre-
sentations of each stimulus to human subjects, some of whom
were subject to a delay paradigm and others to a trace para-
digm. All were required to watch, attend to, and try to remem-
ber a silent movie (Charlie Chaplin’s “The Gold Rush”) while
these CS and US events were happening, thereby command-
ing a great deal of their conscious awareness. At the end of the
experiment, in addition to being quizzed about the movie, the
subjects were asked a set of questions with true/false answers
designed to explore their awareness of the stimulus contin-
gencies. The key finding was that successful trace conditioning
was correlated with levels of awareness of these contingencies.
A subset of control subjects who showed good awareness of
the CS predicting the US after the trace delay conditioned
well. Other control subjects, and all amnesic subjects tested,
were less aware or even unaware of the stimulus contingen-
cies—they conditioned poorly. Awareness was unrelated to
levels of conditioning in the delay paradigm, a result that is
not universally obtained (e.g., Knuttinen et al., 2001).
Furthermore, to explore whether the correlation between level
of awareness and degree of trace conditioning was causal,
Clark and Squire (1998) manipulated awareness directly.
Manipulations that facilitated awareness facilitated eyeblink
conditioning; manipulations that reduced it blocked condi-
tioning. A subsequent critique (LaBar and Disterhof, 1998)
queried Clark and Squire’s use of a differential conditioning
protocol somewhat different from those used in many previ-
ous studies of human eyeblink conditioning; they also won-
dered whether a better “online” measure of awareness than a
postexperimental questionnaire might be developed. Manns
et al. (2000a) provided both in a study of normal subjects who
showed a close predictive relation between developing aware-
ness of the CS-US trace contingency and subsequent levels of
conditioning. These investigators also showed that the critical
feature of this awareness is the subject’s awareness of the CS-
US contingency rather than their awareness of the likelihood
of blinking. Thus, it appears that what is “declarative” about
trace conditioning is an awareness of the fact that the tone
predicts that an air-puff will come shortly—an awareness that
can be expressed propositionally. This awareness may arise
Figure 13–15. Protocols for delay and
trace conditioning. Time lines showing the
onset and offset of the conditional stimu-
lus (CS) and unconditional stimulus (US)
in two distinct forms of classic condition-
ing. Appropriately timed CRs develop in
300-1000 msec both protocols, yet only one is thought to
be hippocampus-dependent.
through some interaction between hippocampus and neocor-
tex during the course of conditioning and is somehow fos-
tered by using the trace conditoning paradigm. By virtue of
this, a suitably timed signal can be sent to the cerebellum
where the eyeblink response component of the conditioning
takes place. The eyeblink CR, a learned but automatic defen-
sive response, can then be executed at the appropriate time.
To summarize, what is nondeclarative about nondeclara-
tive learning may actually be the lack of awareness of what is
being remembered at the point of recall, rather than any
intrinsic inflexibility in what has been learned or a lack of
propositional status of the content of learning.
Are Fact-Memory and Event-Memory
Processed by a Common Brain System?
Superficially, amnesics present with a deficit restricted to
episodic memory. They cannot remember events for any
length of time, but their factual knowledge about the world
and their knowledge of language are both intact. Semantic
memory shows all the appearances of being preserved. The
declarative memory theory holds that this dissociation is pro-
foundly deceptive. It asserts that there is a “unitary” process
underlying the formation of both event and fact memories.
The reason amnesic patients display intact semantic memory
is because so much of a person’s factual knowledge was
acquired years earlier, extending from the years of childhood
on through life. Consolidation of such memory traces would
be long completed. Conversely, a patient’s failure of event
memory often relates to relatively recent events such as a for-
gotten conversation of the day before.
Although most amnesic subjects have some residual
episodic memory function, a few are claimed to have little or
none. In Chapter 12, we noted the striking case of patient E.P.,
who has essentially no measurable recent episodic memory
but who, becoming amnesic when adult, displays good factual
knowledge of the world acquired earlier in life and remarkably
good memory for spatial and other episodic-like information
acquired during childhood. However, if this interpretation is
correct, people who have sustained damage to the MTL at a
much younger age should also show impaired semantic mem-
ory. Data concerning the effects of bilateral hippocampal
pathology sustained early in life are relevant to this issue
(Vargha-Khadem et al., 1997). Three young developmental
amnesics who sustained brain injuries at birth, 4 years, and 9
years of age, respectively, were aged 14, 19, and 22 at the time

of first testing. The anoxic-ischemic episodes they had are
likely to have affected the hippocampus preferentially. MRI
measurements revealed hippocampal volumes ranging from
43% to 61% of normal. T2-weighted relaxometry measure-
ments also revealed hippocampal abnormalities, but MR spec-
troscopy values were within the normal range. Additional
MRI measurements in other brain areas suggested that bilat-
eral hippocampal changes are not only the primary pathology
but might be essentially the only pathology in all three cases
(Mishkin et al., 1997).
The striking neuropsychological feature is that their amne-
sia is largely exclusive to episodic memory. They are tempo-
rally and spatially disoriented: forgetting the date and
appointments, getting lost, mislaying their belongings.
Despite severe problems of episodic memory, these children
all attained levels of speech and language competence and of
literacy and factual knowledge within the low average to aver-
age range. They all attended mainstream schools, and their
acquisition of semantic information appears to be normal (or
at least near-normal) as assessed by performance on the
Wechsler Adult Intelligence Scale (WAIS) verbal intelligence
tests. Their performance on a Basic Reading and Reading
Comprehension subtest was normal in all three cases, as was
their Spelling subtest (with the exception of one subject whose
spelling is poor).
The pattern of memory preservation and memory loss
shown by these children is in striking contrast to what would
have been expected on the basis of the declarative memory
theory. Vargha-Khadem and her colleagues argued that these
three patients raise the possibility that the “basic sensory
memory functions of the perirhinal and entorhinal cortices
may be largely sufficient to support the formation of context-
free semantic memories but not context-rich episodic memo-
ries, which must therefore require the additional processing
provided by the hippocampal circuit” (Vargha-Khadem et al.,
1997, p. 379). However, further research is required before
such conclusions can be drawn with any certainty as later
commentaries on these cases and parallel primate studies have
recognized (Mishkin et al., 1998; Bachevalier and Vargha-
Khadem, 2005). For one thing, the children appear to have
some residual episodic memory, and this may have been suffi-
cient for them to learn well in school, albeit more slowly than
other children and perhaps only after frequent repetition.
Moreover, although Vargha-Khadem and her colleagues inter-
viewed the children’s parents, they did not report on any dis-
cussions with the childrens’ schoolteachers. It is difficult to
believe that their teachers would have failed to notice their
severe episodic memory difficulties and thus compensated for
these problems with their teaching. The early damage to the
brain may also have resulted in morphological reorganization
of MTL structures in a way that does not occur when adults
become amnesic. Certainly the MRI measurements do not all
point to a completely destroyed hippocampus. Taken together,
although fascinating cases, these concerns blunt the critical
impact on declarative memory theory that they might other-
wise have had.
Theories of Hippocampal Function 617
Does it make sense to suppose that the memory of facts
and events could depend on a single brain system? According
to the theory, both types of information require the integrity
of the hippocampal formation for their formation (encod-
ing); and both result, after memory consolidation, in stable
long-term traces in neocortex. Success in recalling either a fact
or an event depends only on the “strength” of the traces estab-
lished through consolidation. In contrast, Nadel and col-
leagures (Nadel and Moscovitch, 1997, 1998; Nadel et al.,
2000) have argued that memory for an event requires access to
a hippocampally based “contextual” trace (where the event
happened) and a prefrontal “temporal” trace (when the event
happened). Retrieval of event memory is a reconstruction
based on both of these traces and other information about
the event itself stored elsewhere in the neocortex. Retrieval of
factual information, on the other hand, is not thought to
require either a contextual or a temporal trace; it is context-
independent.
In conclusion, we still do not understand the precise role of
the hippocampus in episodic and semantic memory or, within
the domain of episodic memory, in familiarity and recollec-
tion. The immediately preceding discussion, complementing
Chapter 12, points to the need to develop new animal models
of these forms of memory and of retrieval to help resolve the
issues. We return to these important ideas in Section 13.6 after
we have discussed the other major theories of the hippocam-
pus: cognitive mapping and predictable ambiguity.
Á
13.4 Hippocampus and Space: Cognitive
Map Theory of Hippocampal Function
“Space,” wrote O’Keefe and Nadel (1978, p. 5), “plays a role in
all our behavior. We live in it, move through it, explore it,
defend it . . . yet we find it extraordinarily difficult to come to
grips with space.” This strident rhetoric, together with discus-
sion of the philosophical concept of space, introduced a book
that outlined the then new “cognitive map” theory of hip-
pocampal function. The theory has developed substantially
over the past quarter century.
The primary stimulus for its development was the discov-
ery of “place cells” in the hippocampus of freely moving rats
(O’Keefe and Dostrovsky, 1971). These cells are neurons that,
as discussed in Chapter 11, are characterized by patterns of fir-
ing that increase from a generally low level to higher rates as a
freely moving animal moves through a particular region of
space. As different place cells are responsive in different loca-
tions and are found throughout the septo-temporal axis of
the hippocampus, it was natural to suppose that the collective
firing of different place-cell ensembles in different environ-
ments might be the neural substrate of distinct “maps” of
space. Observations on place cells were quickly followed by
the discovery that lesions of a major fiber tract into the hip-
pocampal formation, the fimbria/fornix, resulted in an appar-
ently selective deficit of spatial, but not cue, learning by rats
618 The Hippocampus Book
in a circular maze task (O’Keefe et al., 1975). These two find-
ings were the trigger for a comprehensive review of the litera-
ture and one that offered an imaginative reassessment of
findings hitherto explained in other ways (O’Keefe and Nadel,
1978).
The cognitive map theory was the first of several spatial
theories of hippocampal function. Its authors have since out-
lined modifications of the theory, beginning with Nadel’s
work on “context” (Nadel and Willner, 1980), through ideas
about multiple memory traces stored in the hippocampus
(Nadel and Moscovitch, 1997), and now incorporating neu-
roimaging findings from humans performing virtual naviga-
tion tasks (Burgess et al., 2002). O’Keefe has also explored the
relevance of the theory to the neural basis of language, in par-
ticular spatial prepositions (O’Keefe, 1996). Other spatial
memory theories include the “scene memory” hypothesis of
Gaffan (1991), the “fragment assembly” hypothesis (Worden,
1992), and the closely related idea of the hippocampal forma-
tion being involved in “path integration” (McNaughton et al.,
1996; Whishaw, 1998; Redish, 1999). There have also been
ideas about the “metrics” in which spatial information is
processed (Poucet, 1993) and a “dual,” or “parallel-map,” the-
ory (Jacobs and Schenk, 2003). This latest development argues
for a synergistic interaction of a directional “bearing map” and
an allocentric “sketch map”—each mediated by distinct but
overlapping subregions of the hippocampal formation. Like
the ideas about path integration, it incorporates the discovery
of another category of spatially tuned cells, the head-direction
units (Taube and Schwartzkroin, 1987; Taube, 1998). These
are cells, described in Chapter 11, that fire when a rat’s head is
facing a particular direction in a familiar environment irre-
spective of the animal’s location. In parallel with this neurobi-
ological theorizing, there has been continuing interest in the
psychological literature concerning the concept of “cognitive
mapping” (Gallistel, 1980) that dates back to Tolman’s classic
paper (Tolman, 1948). Although the term cognitive mapping
was used in a slightly different sense by Gallistel, this renewal
of interest has prompted debate that, in an unexpected turn,
has cast doubt on the explanatory power of the concept
(Bennett, 1996; Wang and Spelke, 2002). Mackintosh has also
queried its adequacy as an account of how animals get from
one place to another but expressed the view that the 1978
book “deserves its status as a modern classic” (Mackintosh,
2002. p. 181). Numerous reviews have been published
(Poucet, 1993; Muller, 1996; Taube, 1999), as have two com-
prehensive books focusing on the neurobiology of spatial,
context, and episodic memory (Redish, 1999; Jeffery, 2003).
13.4.1 Outline of the Theory
The five main suppositions of the original (1978) cognitive
map theory relate to representation of the environment, nav-
igation through it, the evolution of an ostensibly specialized
spatial memory system, storage of spatial information, and
the relevance of a specifically spatial system in animals to the
mediation of episodic and semantic memory in humans.
Box 13–4
Cognitive Map Theory
1. Representation: The vertebrate brain has a neural system—
the “locale” system—that organizes the encoding and rep-
resentation of perceived stimuli with respect to an
“allocentric” spatial framework, or “cognitive map.” The
spatial locations of landmarks are stored in this map dur-
ing exploration. This locale system is in the hippocampus.
2. Navigation: The locale system is used for spatial naviga-
tion. The various anatomical components of the hip-
pocampal formation are involved in mediating different
aspects of spatial information processing, with distinct
classes of neurons responsive to an animal’s place, head-
direction, view and its movement through space.
3. Evolution and the laws of learning: Spatial mapping
evolved as one of the multiple memory systems of the ver-
tebrate brain with its own distinctive learning rules. Other
types of learning and memory include simpler associative
mechanisms and certain geometrical tasks that can be
acquired using “taxon” strategies. These other systems
obey laws of learning that are different from those of the
locale system.
4. Sites of storage: Spatial maps are stored in the hippocam-
pus. They are not consolidated or stored elsewhere in the
brain, although information stored in the hippocampus
does interact with information stored elsewhere for the
purpose of guiding navigational behavior. An intrahip-
pocampal consolidation mechanism may nonetheless exist.
5. Extension of the theory to humans: Whereas the cognitive
map is purely spatial in animals, it subserves the storage
and recall of linguistic and episodic memories in humans.
Lateralization enables the right hippocampus to maintain
a primarily spatial function, whereas the left hippocampus
incorporates a temporal sense and linguistic entities that
together provide the basis for mediating episodic memory.
As already noted, numerous developments of the original
theory have occurred since 1978, leading to slightly different
ideas about spatial memory than those in the list in Box 13–4.
In the discussion that follows, there are references to these
developments, as appropriate. Although distinct, they are in
the same genre as the original cognitive map theory.
The first of O’Keefe and Nadel’s (1978) ideas—the concept
of a “locale,” or mapping, system—implies that when forming
memories of the world neural activity in the brain automati-
cally encodes stimuli in a connected mental framework rather
than as isolated stimulus traces or independent pairwise asso-
ciations. This encoding happens, for example, as laboratory
rats explore a novel environment (Fig. 13–16A). It occurs
quickly, sometimes in one trial; and it relies on information
from the place cells indicating the animal’s location, particu-
larly the distance from the boundaries of the test arena. Head-
direction cells provide a representation of direction based on
what are described as infinitely long vectors. This curiosity-
driven exploration involves something more than perceptual
learning about the identity of individual cues (objects, land-

Figure 13–16. Place cells, boundary vectors, and the puzzle of place
cell-guided navigation. A. As rats explore a simple environment in a
test arena, they encode information about location, such as distance
from the side walls, into a map-like internal representation. Place
cells emerge as the interaction of two or more Gaussians (inset)
whose centers are determined by the distance from prominent fea-
tures such as walls. (Source: After O’Keefe and Burgess, 1996.) B. A
marks) or the development of familiarity that later enables
recognition. The putative hippocampal mapping system also
encodes where landmarks are located in relation to each other
in some kind of geometrical framework.
The idea of an organized structure to memory is, of course,
hardly unique to the cognitive map theory, as many other the-
ories of human memory embody the same idea, including
artificial intelligence-derived theories of semantic memory
(Collins and Quillian, 1969) and connectionist theories of
memory consolidation (McClelland et al., 1995; McClelland
and Goddard, 1996; O’Reilly and Norman, 2002). Some form
of memory organization would also be helpful for encoding
one-time events in relation to static landmarks; and, partly
for this reason, the hippocampus in humans could be help-
ful for encoding information into episodic memory. The recall
of a specific event, such as what one did during a recent holi-
day, generally involves first remembering where one was
before proceeding to remembering with whom one spent the
holiday or what unusual things happened. Gaffan (1991)
was the first to make this point explicitly, later building it into
a “scene-specific” theory of episodic memory (see Section
13.6).
The second supposition is that, once stored, locale informa-
tion can be used for spatial navigation. Whereas the first propo-
sition was that information is stored in a map-like framework,
the second relates to how that information is used. How does
an animal navigate using its “cognitive map”? This question
prompted the supposition that the unusual anatomical cir-
cuitry of the allocortex (see Chapters 3, 4, and 8), so unlike the
Theories of Hippocampal Function 619
rat at place A wants to navigate to B but not to C. Place cells (gray
circles and ellipses) have firing fields in these three areas. A subset of
the cells with fields at place A fire with a certain probability with the
animal there, but it is not clear how at place A the rat can activate
the subset of place cells for place B but not those for place C to
direct its choice behavior appropriately. Additional “goal cells” seem
to be required to achieve navigation. (Source: After Morris, 1991.)
canonical local circuits of neocortex (Douglas and Martin,
2004), could be precisely what is needed to perform the geo-
metrical computations involved in finding the way around.
This might include computations for encoding the distance
and direction between landmarks, rotating the map for appro-
priate alignment with perceived cues, identifying the location
and heading direction of the animal within it, and performing
various other geometrical functions essential for effective allo-
centric navigation through space. In short: Where am I? Where
am I going? How do I get there?
O’Keefe and Nadel (1978) offered several speculations
about how such spatial operations could be implemented
in the hippocampal formation. Considering the original ideas
on representation and navigation together (Box 13–4, propo-
sitions 1 and 2), they supposed that sensory information
entered the hippocampal formation via the entorhinal cortex,
from which it projected into the dentate gyrus and then to
succeeding stages of the trisynaptic circuit as it was then
understood (but see the discussion of hippocampal connec-
tivity in Chapter 3, Section 3.7). Different pyramidal cells were
thought to be driven by different proportions of excitatory or
inhibitory inputs. The group with mainly excitatory inputs
(“place cells”) were defined as those that fired when an animal
occupied a particular position in relation to a constellation of
sensory cues (e.g., visual, auditory, olfactory). Because of the
convergent nature of hippocampal circuitry, it was suggested
that the probability (or rate) of place-cell firing should decline
in a monotonic but nonlinear fashion as individual cues
become obscured or are removed but be relatively insensitive
620 The Hippocampus Book
to the loss of any one cue. This prediction about place cells
was upheld (O’Keefe and Conway, 1978). Other pyramidal
cells (“misplace cells”) were presumed to have, in addition,
sensory-driven inputs relayed via feedforward inhibition such
that they would increase their rate of firing when sensory cues
were removed. This increased firing was held to be the neural
event that triggered exploratory behavior, perhaps via the
subicular output through the fornix to the nucleus accumbens
(see Chapter 3, Section 3.4.2). The original theory also stated
that the firing of “theta cells” is coupled to the translational
movement of the animal from one position in the map to
another, an idea supported by studies in which the different
frequencies of theta activity observed during jumping
through space correlated directly with the distance moved
rather than the force exerted (Morris, 1983). More recent neu-
ral modeling studies have modified these ideas substantially
(see Chapter 14). The unfolding picture about the wide range
of distinct inhibitory interneurons and their axonal connec-
tions (see Chapter 8) has also raised the specter of yet deeper
subtlety at the level of local circuits.
Since 1978, there have been many technical and conceptual
developments in hippocampal single-unit recording in freely
moving rats. The development of multiple single-unit record-
ing (“ensemble recording”) with “stereotrodes” and “tetrodes”
has shed fresh light on a range of theoretical issues: multiple
place fields, local views, gating by the potential for movement,
memory properties, multiple reference frames, pattern com-
pletion, and pattern separation (as discussed in Chapter 11).
Numerous uncertainties have been clarified. It is now known,
for example, that pyramidal cells acquire their place fields rap-
idly, fire in proportion to an animal’s speed of motion, tend to
be directionally sensitive only in directionally constrained
environments (e.g., linear tracks), and show temporal preces-
sion in their phase relation to hippocampal theta as an animal
moves through the cell’s place field. Improved recording tech-
niques have also established that single cells do not, in general,
have place fields in different parts of a single environment,
and that physically adjacent pyramidal cells may have place
fields in quite different parts of the environment. A boundary
vector model (BVC) has been developed (Hartley et al., 2000)
according to which places can be modeled as the sum of two
or more Gaussians, with the center of each Gaussian located
by its distance from a specific environmental feature such as a
wall (as in Fig. 13–16A). With this view, place fields represent
probability distributions and do not specify precise locations.
Certain theoretically unanticipated properties of CA1 place
cells have also been discovered, such as their sensitivity to
physical restraint, their apparent insensitivity to lesions of the
dentate gyrus or CA3, and alterations in receptive field size
and stability with aging (see Chapter 11). The discovery of
head-direction cells in the presubiculum and thalamus (Taube
et al., 1990a, b) and of “grid cells” in the dorsomedial entorhi-
nal cortex (Fyhn et al., 2004; Hafting et al., 2005) are impor-
tant additions to the idea that the hippocampal formation and
interconnecting structures constitute, in animals, an essen-
tially spatial system.
Notwithstanding these developments, we still do not fully
understand how a map of space could be represented (Box
13–4, proposition 1) or used (Box 13–4, proposition 2)
by means of hippocampal circuitry. The relation between
location firing (in pyramidal cells of CA1 and CA3), head-
directional firing (in the presubiculum and anterior thala-
mus), and movement firing (in the interneurons of the
hippocampus and dentate gyrus) is also poorly understood.
Those concerned with specific details might reasonably won-
der, for example, whether the head-direction firing of pre-
subiculum cells is a component of “locale” or “taxon”
processing (or both). Various logical puzzles about the link
between mapping and navigation have also been pointed out.
For example, as place cells are defined as cells that fire only
when an animal occupies a specific position in space, how can
the mapping system access information pertaining to places
other than the one presently occupied by the animal (Morris,
1991)? That is, if a rat is at place A and wants to navigate to
place B but not to place C, how does it at place A access infor-
mation relevant to going to B rather than going to C (Fig.
13–16B)? Place-cell firing on its own does not seem to help
because the place cells corresponding to B and C cannot, by
definition, fire until the animal gets to B or C, respectively. This
kind of logical conundrum is clearly problematic for the the-
ory, but it may not be fatal if goal cells can be identified.
Perhaps these are in other brain areas (Hok et al., 2005). Other
models of the neural implementation of goal representations
are also being developed (Redish, 1999; Biegler, 2003). Several
neural network and temporal difference learning models that
grapple with these issues are outlined in Chapter 14.
The third feature of the theory is that spatial mapping has
evolved in response to specific environmental demands, and that
it constitutes one of the multiple memory systems of the verte-
brate brain (Fig. 13-17). The emphasis on evolution, lacking
prominence in the declarative memory theory, has been
informed by studies of cache recovery in passerine birds,
homing in pigeons, territory size and mating systems in small
mammals, and other aspects of naturalistic spatial behavior
(Sherry and Schacter, 1987b; Clayton and Krebs, 1995; Dyer,
1998; Healy, 1998; Shettleworth, 1998; Jacobs and Schenk,
2003). The “neuroethological” component of the thinking
behind the cognitive map theory has been explicit from the
outset (O’Keefe and Nadel, 1979; Nadel, 1991), fostering stud-
ies of sex and seasonal differences in spatial memory that
might not have happened had all hippocampal research
remained anchored solely in a paramedical context. Work in
both vertebrate and invertebrate species has revealed both
convergent and divergent evolution with respect to how to
find food, water or sex—and then get safely back home.
Studies of animal navigation have, however, revealed mecha-
nisms that appear to have nothing to do with cognitive maps,
such as ideothetic “path integration” by insects (Wehner et al.,
1996) and rodents (Etienne and Jeffery, 2004) and the “snap-
shot” processing of landmarks by bees (Collett, 1992). The rel-
evance of the neuroethological approach to hippocampal
function and mechanisms has been the subject of cogent crit-

Types of learning
Spatial Associative
Locale Taxon Classical Instrumental
Conditioning Conditioning
Orientation Cue guidance
Hippocampal Formation
Figure 13–17. Multiple forms of memory in the cognitive map the-
ory. Only the locale system, requiring cognitive mapping, is medi-
ated by the hippocampus. The theory has less to say about other
forms of learning and memory.
icism (Squire, 1993; Bolhuis and Macphail, 2001), but many
see it as an important adjunct to the development of cognitive
mapping theory.
However, the bulk of the work addressing the cogni-
tive map theory has been conducted with the laboratory rat
and, more recently, various laboratory strains of mice. A veri-
table “hippocampal industry” emerged, beginning with
lesion/behavior studies and single-unit recording work on
animals, that now finds modern expression in a variety of
behavioral tasks used in conjunction with the full spectrum of
contemporary neuroscience techniques. Particularly exciting
are the technical innovations that complement classic lesion
techniques, with the new tasks including ensemble single-unit
recording from large numbers of hippocampal neurons
(Wilson and McNaughton, 1993), monitoring the expression
of immediate early genes (Guzowski et al., 1999; Vann et al.,
2000), and reversible pharmacological interventions that are
sufficiently sustained in time to investigate memory processes,
such as systems-level consolidation (Riedel et al., 1999). Mice
have nibbled their way onto the hippocampal stage with tar-
geted gene deletions, hippocampus-specific deletions, and
even inducible gene knockouts (Tonegawa et al., 1995; May-
ford et al., 1996; Tsien et al., 1996; Abel and Lattal, 2001). The
hippocampal research industry, which has seen the number of
papers on rodent spatial representation and navigation rise
from around 10 per year in 1980 to several hundred per year
(Sutherland and Hamilton, 2004), owes a great debt to the
cognitive map theory.
With respect to multiple memory systems, the scope of
O’Keefe and Nadel’s (1978) original analysis was narrower
than that encompassed in the declarative framework dis-
cussed earlier (Figure 13.4). Their book distinguishes only
Theories of Hippocampal Function 621
between spatial memory and nonspatial associative learning
(such as classical and instrumental conditioning). Whereas
allocentric “locale” learning was held to be motivated only by
curiosity and to be extremely fast, associative learning was
thought to be motivated by rewards related to basic motiva-
tional needs (such as food for a hungry animal) and to be
slower, cumulative, and governed by quite different learning
rules. However, this spatial versus associative distinction has
come under close scrutiny (just as it did in connection with
recognition memory in the previous section of this chapter).
Certain classic phenomena in the associative domain (such as
“overshadowing” and “blocking,” to be explained later) have
also been found to operate within the domain of spatial learn-
ing (Biegler and Morris, 1999; Chamizo, 2003) although sub-
ject to certain constraints associated with the perception and
local processing of geometry (McGregor et al., 2004a).
Associative learning can also be very fast, and in phenomena
such as sensory preconditioning, is not obviously motivated
by biological rewards. Although O’Keefe and Nadel’s original
dichotomy now feels incomplete, understanding the bound-
aries between cognitive mapping and other forms of learning
and memory is proving to be a fruitful area of research.
Also of interest is how spatial and nonspatial systems compete
or work synergistically to ensure the seamless control of
behavior; and it remains a fundamental issue (White and
McDonald, 2002).
The fourth proposition of the 1978 theory is that allocen-
tric spatial information is stored in the hippocampus. Unlike the
time-limited role for the hippocampus in the declarative
memory theory, the original version of the cognitive map the-
ory has both initial encoding and long-term storage in the one
brain structure. Nadel and Moscovitch (1997) extended this
idea in a “multiple-trace theory” of episodic memory in which
memory retrieval episodes can initiate re-storage. The possi-
bility of multiple traces—even for single episodes—leads to a
different account of the temporal gradients of retrograde
amnesia that are sometimes seen for information acquired
prior to brain damage. The key difference from the account
offered by the declarative memory theory lies in the impact of
partial damage. Complete lesions at various time points after
training should always result in temporally extensive, or “flat
gradients,” of remote memory loss. In contrast, partial lesions
might result in a reverse gradient if older information, recol-
lected previously, had laid down additional traces in areas of
the hippocampus spared by the lesion. The declarative mem-
ory theory makes no differential prediction as a function of
lesion completeness.
Although the original cognitive map theory of hippocam-
pal function was based largely on observations in rats, its
extension and likely modification to account for primate and
human hippocampal function was recognized at the outset.
Proposition 5 in Box 13–4 recognizes that later chapters of
O’Keefe and Nadel’s monograph addressed the relevance
of their ideas to human semantic memory (e.g., Chapters 14
and 15) and, specifically, the way in which spatial relations
are fundamental to certain linguistic prepositions that reflect
622 The Hippocampus Book
the knowledge of relations (e.g., “beside,” “near,” “above”).
Although not originally influential in the human neuropsy-
chology community, these ideas have nonetheless been
extended to such domains as reasoning and abstract thought
and are now an active area of research. For example, Gattis
(2001) identified the sharp distinction between a “generalist”
school of thought, in which space is merely a special case of
“relational processing,” and a Kantian perspective, in which
space (and time) are special. Nadel and Hardt (2004) strongly
reasserted the latter position, whereas O’Keefe (2005) has con-
tinued his exploration of “vector grammar” and how it might
explain the essentially various meanings of spatial propo-
sitions.
Cognitive science apart, one obstacle to pursuing these
lines of thought has been in securing definitive evidence for
causally linking hippocampal cell activity in humans to any
such processes. Hippocampal damage in humans does not
obviously disrupt a person’s ability to use spatial prepositions.
Invasive single-unit recording in humans is rarely feasible, but
some intriguing examples of “place cells” have been recorded
from the human hippocampus with depth electrodes placed
to monitor epileptic seizure activity (Fried et al., 1998;
Ekstrom et al., 2003). Place cells have also been observed in
the hippocampus of monkeys free to drive themselves around
slowly in a motorized “cab” (Matsumura et al., 1999). Spatial
correlates are therefore by no means unique to rodents, but
securing single-unit evidence for vector grammar has not yet
been achieved.
Functional brain imaging has also been used to study
hemodynamic activation during human navigation. The
problem here is that fMRI is a technique that does not obvi-
ously lend itself to the study of a mental process—naviga-
tion—that ostensibly requires subjects to be mobile. One
solution is to use “virtual reality.” Whether virtual reality in
subjects lying prone in a brain scanner really engages the same
neural mechanisms as “true” navigation can be debated—
place cells in rats being sensitive to whether animals are
restrained or allowed to move (Foster et al., 1989)—but if one
accepts that a virtual reality approach might be viable, mod-
ern computing software has certainly made it possible.
Numerous studies have now examined whether the hip-
pocampus and adjacent structures of humans are activated
during apparent “navigation” through imaginary towns
(Maguire et al., 1998), featureless arenas, and swimming pools
(Thomas et al., 2001; Parslow et al., 2004) as well as other tasks
that distinguish mapping from merely following a sequential
route (Morris and Mayes, 2004). Burgess et al. (2002) made a
strong case that virtual reality is informative and specifically
discussed patterns of activation in relation to spatial frame-
works, the dimensionality of space, and both orientation and
self-motion. In contrast, a more cautious stance about the
findings to have emerged from functional imaging of spatial
navigation is presented in Chapter 12.
Finally, by way of introduction, is there anything special
about space? Declarative memory theorists think not. In their
view, a memory system exists in the MTL that processes all
forms of propositional knowledge, whether it comes to us via
visual, tactile, olfactory, or other sensory modalities. Why,
their concern continues, should space be any different? The
“no different” perspective appears to look on memory pro-
cessing of “space” as involving multimodal sensory informa-
tion, with cues that may be familiar or unfamiliar and with the
capacity to realize “declarations” of spatial knowledge such “A
is near B.” Given this, why would encoding or storage be any
different from that of other object-orientated information
that is processed via the ventral stream of the perceptual
processing pathways of the neocortex terminating in the
inferotemporal, perirhinal, and parahippocampal cortices
(Ungerleider and Mishkin, 1982)? Indeed, so absurd may it
seem to think that space could be different, a major review of
the functions of the MTL considered the adequacy of the spa-
tial theory of hippocampal function in little more than a cou-
ple of paragraphs (Squire et al., 2004).
Not surprisingly, the opposing camp is resistant to being so
subsumed. For those prepared to contemplate a neural system
for cognitive mapping (including critics of O’Keefe and
Nadel’s theory, such as the present author), thinking about
space in this declarative way reflects a conceptual misunder-
standing: Space is not a sensory modality. We do not have sen-
sory organs for space or a cranial nerve devoted to it. Space
is a construct of mental processing. Moreover, space has
a structure that requires but is logically independent of the
multimodal sensory information used to identify it. For
example, many objects whose location we know are often hid-
den, whether it is acorns cached by a squirrel or food in a
kitchen cupboard. It is precisely in this situation that spatial
memory becomes so useful—the exact opposite of the situa-
tion that applies to normal recognition memory. Similarly,
the appearance of a specific area of space can change across
the seasons despite its containing inanimate or sessile living
things that do not ordinarily move around. What does
not change when a cupboard door is closed, a tree loses its
leaves, or a rock becomes covered in snow is the location of
the remembered item. There is, in short, a geometric “stabil-
ity” about remembered space that is independent of
the appearance of the stationary objects and landmarks that,
paradoxically, define that stability (Biegler and Morris, 1993).
This paradox is at the heart of why pattern separation and
pattern completion are network operations at the heart
of hippocampal processing (see Chapter 14). The closely
related concept of “context” is also not necessarily something
that is “out there” waiting for a sensory/perceptual system to
process it; it is a neural construct (Jeffery, 2003). These and
other features of “space” could have been major driving forces
in the evolution of the distinctive cognitive systems that medi-
ate it.
This completes the initial overview of the five key features
of the cognitive map theory. A selective presentation of rele-
vant data follows, in four successive sections, followed by a
concluding critique. The relevance of the theory to humans is
not considered further, and the reader referred to the very dif-
ferent perspective offered in Chapter 12.

13.4.2 Representing Spatial Information, Locale
Processing, and the Hippocampal Formation
A key concept with respect to the representation and storage
of spatial information is the distinction between egocentric
and allocentric space. Egocentric space refers to the locations
of objects relative to the viewer—to the left or the right—
directions that necessarily alter as the viewer moves around.
Representing this kind of space is not thought to require
a cognitive mapping system. Allocentric space, on the
other hand, is a representation of spatial relations within some
kind of absolute framework. For a freely moving subject,
an allocentric representation is required to identify whether
an object or landmark has actually moved in relation to
others. In contrast, objects are constantly “moving” in egocen-
tric representations. The distinction between egocentric
and allocentric space has not always been appreciated by
critics of the cognitive map theory, with some incorrectly
assuming that the theory predicts that lesions of the hip-
pocampal formation must impair any and all spatial learning
tasks (e.g., Nadel and Hardt, 2004). Many spatial tasks given to
animals, such as the simple T maze task described in Section
13.2 and certain touchscreen tasks, are ambiguous because
they can be solved using either an allocentric or an egocentric
strategy. The study of exploratory behavior has provided a key
test of this feature of the cognitive map theory together with
studies examining how animals localize objects in space.
Lesion studies and observations of immediate early gene acti-
vation have contributed to a detailed examination of this
issue.
Lesion Studies of Exploratory Behavior Reveal the
Importance of Allocentric Spatial Representation
It has long been known that rats explore spontaneously, and
the phenomenon has been systematically in the laboratory for
years (Berlyne, 1950, 1966; Halliday, 1968; Archer and Birke,
1983; Renner, 1990). Placed in a novel test arena, the rats
remain motionless for a period of time and then start to move
around, first at the periphery and then throughout the avail-
able space, where they explore objects and landmarks one by
one, often returning to an object just investigated until they
finally come to rest in one corner. This is not mere movement;
it is exploration to gain information. For example, renewed
exploration can be triggered easily after habituation by intro-
ducing a novel object into the arena, a phenomenon exploited
in the visual-paired comparison task (see Section 13.3).
Hippocampal lesions impair certain aspects of exploration.
They tend to make animals hyperactive yet also disinclined to
visit all regions of a space systematically and less likely to dis-
play habituation of activity over time (O’Keefe and Nadel,
1978; Whishaw et al., 1983). O’Keefe and Nadel (1978), Morris
(1983), and Renner (1990) have repeatedly made the point
that the hippocampus-dependent information-gathering
function of exploratory behavior should be clearly distin-
guished from any motor functions to which the hippocampus
Theories of Hippocampal Function 623
contributes (Vanderwolf and Cain, 1994). There remains,
however, the question of what information is gathered.
As the cognitive map theory asserts that this curiosity-
driven exploration is the basis for encoding a “map,” a critical
experimental prediction is that renewed exploration would be
triggered by alterations in the spatial arrangement of familiar
objects. Critically, hippocampus-lesioned subjects should dis-
play selective failure of renewed exploration to object dis-
placement but not to object novelty. Using a paradigm
developed by Thinus-Blanc et al. (1987), Save et al. (1992a)
compared the impact of small dorsal hippocampal lesions,
anterior and posterior parietal lesions, and sham lesions on
exploration in a circular arena containing five objects. In a sys-
tematic series of short exploration sessions, the animals
became familiar with the initial set of objects and their spatial
locations, after which the critical manipulations of spatial dis-
location and object novelty were introduced. The results
showed renewed exploration by controls and anterior parietal
cortex-lesioned animals in response to spatial dislocation;
there was no detectable reaction by hippocampus- or poste-
rior parietal-lesioned subjects. In contrast, renewed explo-
ration was shown by all groups in response to object novelty.
However, the effect was not large. In an ingenious follow-up of
this dissociation, Save et al. (1992b) examined exploration at
a particular point in space that was triggered by the absence of
an expected object (Fig. 13–18). Rats that had been subject
to preexperimental surgery first explored an open circular
arena that was slightly unusual in that it had a transparent
floor. Another rat, in a small box, was sometimes placed
underneath this floor in full view of the exploring animal.
In a sequential four-phase experiment, the experimental rats
first explored the open but empty arena. They were then given
two successive opportunities to explore the arena again with
the “stimulus rat” visible at one “zone” of space, and the study
ended with the final phase that, like the first, had no stimulus
rat available under the floor. A neat feature of this design
is that the overt visual appearance of the arena was identical
in sessions 1 and 4. As expected, introduction of the stimulus
rat in session 2 triggered extensive investigation centered
on the zone above the stimulus rat’s location, and this habitu-
ated between sessions 2 and 3. The key test was session 4
with the stimulus rat now removed. In control rats with sham
lesions, exploration was focused at the zone, but this did
not occur in animals with dorsal hippocampal or posterior
parietal lesions. The use of a recall paradigm may have
made the magnitude of the effect much larger. Taken together,
these findings imply three things: (1) exploration can be
triggered in normal rats by object novelty, spatial novelty,
and the absence of an expected object; (2) damage to hip-
pocampus and posterior parietal cortex selectively impairs
exploration triggered by spatial novelty, including the absence
of an expected object; (3) when it occurs, this exploration is
not merely activated but spatially directed in an appropriate
way. Such findings are consistent with the construction of a
“map” of space in the manner predicted by the cognitive map
theory.
624 The Hippocampus Book
A Response to spatial change in rats with HPC and parietal cortical (PPC) lesions
session S1 sessions S2 & S3
oriented exploration response to change
transparent floor
exploring rat stimulus-rat
z2
z1
NC
open field
z2
H or PPC z1
lesions
B Response to remembered places by C, HPC and PPC lesioned rats
200
mean % re-exploration
150
100
50
0 absent “cue” animal underneath. B. Data
-50
-100
zone1
An outstanding puzzle is why similar disruption of explo-
ration induced by spatial novelty was seen in the posterior
parietal cortex-lesioned group, a finding not predicted by the
theory if map storage is in the hippocampus. The absence of
direct anatomical connections between this structure and the
hippocampus led Parron and Save (2004) to wonder whether
there may be some larger neuronal network involving the
interconnected regions of the retrosplenial and entorhinal
cortices. In a follow-up study using the same behavioral pro-
cedures, they observed that lesions of these cortical structures
also caused disruption of renewed exploration in response to
spatial novelty. This finding adds to a growing body of evi-
dence from exploration and navigation tasks that the process-
ing of spatial information extends beyond the hippocampus
(Poucet and Benhamou, 1997; Kesner, 2000; Whishaw, 2004a;
Hafting et al., 2005).
For example, although some studies have suggested that
neurons in the entorhinal cortex are only weakly modulated
by the position of the animal (Quirk et al., 1992; Frank et al.,
2000), the dorsomedial region of the entorhinal cortex has
“grid cells” whose firing shows multipeak fields that are small,
stable, and directionally insensitive (Fyhn et al., 2004; Hafting
et al., 2005). These firing fields are unaffected by hippocampal
lesions, implying that their existence is not secondary to
“upstream” hippocampal processing that might be projected
back to the entorhinal cortex. Instead, they suggest that at the
border between the entorhinal and postrhinal cortex (a region
almost certainly lesioned in the Parron and Save, 2004, study
of exploration) there is an allocentric, view-independent
session S4
???
Figure 13–18. Exploration guided by the
absence of an expected stimulus. A. In a
multiple phase study, rats explored an
arena with a transparent floor that some-
times revealed the presence of another rat
under one zone (z1). Normal and both
C
hippocampus- and posterior parietal
PPC lesions
H lesions cortex-lesioned rats explored differently
when this second rat was removed: only
control animals spent more time in z1
than z2, reflecting their recall of the now
showing greater exploration of the area
where the now absent animal had been
located. See text for more details. (Source:
zone2 After Save et al., 1992b.)
representation of space. This implies that steps in the compu-
tation of allocentric space can occur upstream of the hip-
pocampus, although the multipeaked properties of entorhinal
“grid” cells indicates that an individual entorhinal cortex
cell provides a metric rather than a specific indication of
location. Perhaps ensembles of entorhinal cells are able to
“disambiguate” on their own, or they may provide a Fourier-
like input to the dentate gyrus and hippocampus, where
single-peaked cells are primarily observed. The discovery
of the entorhinal grid cells will have a big impact on the cog-
nitive map theory, but the full implications remain to be
worked out.
Immediate Early Gene Studies Reveal
Anatomical Dissociations with Respect
to Environmental Representations
Instead of using lesions, visualizing the expression of immedi-
ate early genes (IEGs), such as c-fos, Arc, and zif-268 (Morgan
et al., 1987; Link et al., 1995; Lyford et al., 1995; Tischmeyer
and Grimm, 1999), has been used to examine the patterns
of activation in multiple brain areas during both spatial and
nonspatial tasks. Some of these studies have focused on the
representation of the environment, others on spatial naviga-
tion, and yet others on memory consolidation (Dragunow,
1996; Guzowski, 2002). Different IEGs are used in different
studies. The choice can take advantage of the fact that some
IEGs are regulatory transcription factors that influence the
activity of other “downstream” genes (e.g., c-fos), whereas

other IEGs are effector genes that directly affect cellular phys-
iology (e.g., Arc).
There are several features of the IEG approach that are
noteworthy. First, from a low level of basal expression in ani-
mals, experience-dependent activation of IEGs is correlated
with increases in neuronal activity (Herrera and Robertson,
1996). There is, for example, a broadly similar activation pro-
file for c-fos, zif-268, and Arc in association with spatial learn-
ing, with some indication of differential sensitivity to
manipulation of task demands (Guzowski et al., 2001).
Second, the imaging observations made after sacrifice are
from animals whose brains are normal at the time of testing;
thus, IEG mapping links behavior to physiology in a manner
that complements the single-unit recording studies of
Chapter 11. This does not preclude combining the lesion and
IEG approaches, and studies using such a combined approach
have revealed alterations in activity in areas remote from the
site of the lesion (Aggleton and Pearce, 2001). Third, although
the analytical focus of a given study may be on activation of a
specific brain structure, the approach does not require a
“prospective choice,” as quantitative measures may be taken
of a large number of brain structures. The drawback of the
IEG approach is that, unlike single-unit recording for which
the measured output is a signal that is well understood in
information-processing terms, the “meaning” of an IEG signal
is unclear. There has been substantial discussion about the
possible functions of IEGs, ranging from metabolic replenish-
ment to regulation of synaptic plasticity, memory consolida-
tion, and even memory retrieval (Guzowski, 2002).
The technique was first used in relation to behavioral stud-
ies of exploration by Hess et al. (1995), who observed wide-
spread hippocampal IEG activation in various subfields. This
was a valuable first step, but it did not speak to whether the
hippocampus encodes information allocentrically. IEG activa-
tion may merely occur in the hippocampus in response to
novelty, stress, or the motor activity that accompanies explo-
ration. An important step forward was made in a study using
c-fos by Wan et al. (1999) who presented rats with computer-
displayed images in which novelty was introduced as either
new stimuli or novel spatial arrangements of hitherto familiar
stimuli, a comparison analogous to the original lesion studies
just discussed (Save et al., 1992a,b). A within-subjects proto-
col was used in which the images were presented bilaterally
such that familiar stimuli were presented to one eye and novel
stimuli (or novel rearrangements of familiar stimuli) to the
other eye (Fig. 13–19). As the projections of the rat visual sys-
tem are largely crossed, it was possible to compare normalized
patterns of activation in the two hemispheres within individ-
ual animals. Any nonspecific c-fos activation associated with
novelty, stress, or motor activity was effectively and elegantly
“subtracted” out. The key finding was that novel rearrange-
ments of familiar stimuli to one eye differentially activated the
CA1 region of the hippocampus. Novel stimuli activated the
perirhinal cortex, as previously shown by Zhu et al. (1995).
Brown and Aggleton (2001) summarized a body of work
using this approach.
Theories of Hippocampal Function 625
computer monitor
novel picture familiar picture
30cm
50
50°
juice tube perspex screen
45°
Rat observing hole
Figure 13–19. Bilateral viewing technique to image immediate
early genes such as c-fos. Novel and familiar visual patterns are pre-
sented simultaneously to each eye. The largely crossed visual system
of the rat enables the various c-fos patterns to be compared within
individual animals and then averaged. (Source: Courtesy of M.
Brown and J. Aggleton.)
A limitation of the bilateral viewing technique is that there
is no behavioral output of the simultaneous processing of
novel and familiar stimulus combinations in the two hemi-
spheres—indeed, no way of establishing whether the animals
even attend to the cues. Jenkins et al. (2004) therefore followed
up the Wan et al. (1999) study by a systematic analysis of nor-
malized c-fos patterns in animals running through and making
choices in a radial maze task (see below for a description of this
classic navigation task). Following a procedure first developed
by Suzuki et al. (1980), they first trained rats with eight dis-
tinctive cue cards placed between the arms of the maze. Then,
on the final day of training, the cue cards were left in their
standard arrangement for half the animals (Group Familiar)
or shifted to a novel spatial arrangement for the others (Group
Novel) (Fig. 13–20). Spatial novelty caused c-fos activation in
CA1, CA3, and the dentate gyrus, whereas no changes were
observed in the surrounding cortical regions such as the
perirhinal and parahippocampal cortex. This was not due to
differential sensitivity of the imaging method between hip-
pocampus and cortex because successful c-fos activation fol-
lowing spatial novelty was seen in the posterior parietal cortex.
In addition, animals in a companion study were shown to be
using the cue cards in their representation of the maze, as
indexed by behavioral sensitivity to cue rotation. These find-
ings nicely complement the lesion observations of Save et al.
(1992a,b). That the hippocampal formation can detect alter-
ations in the spatial arrangement of cues and respond by dif-
ferential activation of an IEG strongly suggests that the
information encoded about the familiar stimuli presented ear-
lier must have included some representation of their spatial
layout in the image or in the room. They support the notion
that the hippocampus is both necessary for and involved in the
representation of space. However, they also point to unex-
pected involvement of the posterior parietal cortex as well.
626 The Hippocampus Book
A Spatial arrangement of cues during training (left)
and c-fos measurement period (right)
Group Familiar
c-fos test day
Group Novel
Figure 13–20. Immediate early gene activation reveals regional
patterns of sensitivity to spatial novelty. A. The spatial arrangement
of the cues placed between the ends of the arms of the maze was
retained between training and the c-fos test session for Group
Familiar but was changed for Group Novel. B. The normalized
Guzowski et al. (1999) introduced a highly illuminating
new technique in which it is possible to identify time-locked
expression of IEGs using fluorescent in situ hybridization
(called catFISH after the cellular compartment in which time
of expression is observed). It was first applied to the imaging
of Arc on its own and later extended to enable simultaneous
imaging of Arc and Homer. The catFISH technique enables a
within-subject comparison of a different kind in which the
population of neurons activated by one experience (e.g.,
exploring a box) can be distinguished from a second popula-
tion activated by a later experience (e.g., exploring a different
box). Arc RNA first appears as discrete small intranuclear sig-
nals within 2 to 15 minutes of the cells being activated (e.g.,
the activation of a place cell as an animal occupies its place
field). After 20 to 45 minutes the nuclear signal disappears,
and a cytoplasmic mRNA signal develops that is translocated
to dendrites. As the time course of the nuclear signal is distinct
from the cytoplasmic signal, the activity history of individual
neurons at two time points can be inferred (see this chapter,
figures in color centerpiece). Two predictions of the cognitive
mapping theory were then tested. First, it was established that
a common population of place cells was activated when the
animals explored the same box (A) on two successive occa-
sions, with the same population of cells expressing Arc mRNA
in dendrites (signaling the first visit to the box) and in the
soma (more recent visit). Second, they observed Arc RNA
expression in two distinct but overlapping populations of cells
on sequential visits to two different boxes (A then B). In boxes
approximately 70 cm in diameter (about the same as those
used in many studies of place cell firing fields), they found
that approximately 38% of the CA1 pyramidal cells showed
both the dendritic and the somatic IEG signals, whereas oth-
ers displayed only one of the two signals. The standard c-fos
B Hippocampal (left bars) and cortical (right) c-fos
patterns following spatial cue re-arrangement
normalised counts of c-fos nuclei
100
Familiar
Novel
75
* * *
50
25
0
CA1 CA3 DG Peri EntL EntM
c-fos patterns show the differential sensitivity of structures in the
hippocampus (CA1, CA3, DG) versus neocortical regions such as
perirhinal and entorhinal cortex (medial and lateral). (Source: After
Jenkins et al., 2004.)
technique would not have revealed that the activation
response of the different environments was specific, whereas
Arc catFISH provides novel information about the informa-
tion content of neuronal activation.
In an extension of this work, Vazdarjanova and Guzowski
(2004) have compared the sequential IEG activation patterns
in CA3 and CA1 in an exploration task very like that used by
the Marseille group: a box containing objects that can be
moved about. Exploiting a further technical innovation, con-
focal images of Homer reflected neuronal activity that had
occurred approximately 30 minutes before sacrifice, whereas
nuclear images of Arc reflected more immediate neuronal
activation. The study revealed that when confronted by two
object arrangements successively (A then B), cells in area CA3
showed discontinuous activation patterns that were highly
suggestive of “pattern separation.” However, when confronted
with either identical or very similar boxes (A then A or A’), the
patterns overlapped very well, suggestive of “pattern comple-
tion.” The patterns in CA1 were more continuous across these
environmental manipulations. Integration of distinct CA3
and CA1 ensemble representations may provide a refined rep-
resentation of spatial context reflecting intrahippocampal
processing of information projected from the entorhinal cor-
tex. The observations of Leutgeb et al. (2004) on the firing
properties of CA3 and CA1 cells in various boxes and contexts
are consistent with this conclusion.
Distributed Representations, Pattern
Separation, and Pattern Completion
The lesion and IEG studies of exploratory behavior and reac-
tions to novelty clearly point to a role for the hippocampus in
representing spatial information, but on their own they say lit-

tle about how spatial information is represented. Single-cell
and ensemble recording is the way forward to understand and
quantify the representational structure of information (see
Chapter 11) guided by neural network models of how the sys-
tem might work (see Chapter 14). There is also a role for
behavioral studies in identifying neural network processes
such as pattern separation and pattern completion that act on
such representations (Rolls and Treves, 1998).
With one approach to “pattern separation,” Gilbert et al.
(1998) trained hungry rats to forage for food on a “cheese-
board” apparatus with a series of food wells arranged in a row
such that the separation between the wells varied from 15 to
105 cm. Training took the form of pairs of trials, a sample trial
followed by a choice, with as many as 16 trial-pairs per day.
During each sample trial, the rats ran from a start box at the
periphery of the cheeseboard to displace a single object over
one of the wells to secure food. During the choice trial that fol-
lowed a few seconds later, two identical copies of this same
object were placed to cover two wells, with one object covering
the same baited well, the other an empty one farther down the
row (i.e., delayed matching to place). This task provided a way
to measure how easily the rats could discriminate spatial loca-
tions in memory as a function of their distance apart. Prior to
being lesioned, choice performance was very good, as high as
80% on even the smallest distance and rising to 90% at the
largest separation. Large electrolytic lesions of the hippocam-
pus caused performance at the smallest separation (15 cm) to
fall to chance, but there was a graded recovery, as this distance
increased reaching nearly 90% at the largest distance (105 cm).
Performance was also unaffected by changing the start loca-
tion by 180 between sample and choice trials. This pattern of
performance indicates that one function of the hippocampus
may be to “separate” patterns of spatial information, this being
possible without the hippocampus for widely separated cues
but not for cues close together. Findings consistent with this
interpretation using a radial maze task had also been reported
by McDonald and White (1995). In a follow-up study using
the cheeseboard, Gilbert et al. (2001) showed that discrete
neurotoxic lesions of the dentate gyrus (DG) caused a compa-
rable pattern of impairment, whereas CA1 lesions had no
effect. This dissociation is intriguing, but it raises the puzzle of
how spatial information, orthogonalized at the level of the
granule cells in the DG, is able to influence target structures of
the hippocampus in animals that have been subjected to CA1
lesioning. The usual “output” for information, however effec-
tively it is processed in DG, would no longer be available. An
output route via the ventral hippocampus is one possibility as
the CA1 lesions were restricted to the dorsal hippocampus.
There may also be the possibility of output directly from the
CA3 to the lateral septal nucleus, but the route to the neocor-
tex is then indirect. It would be interesting to explore this
route, as the finding hints at a radical new conception of infor-
mation flow within the hippocampal formation.
Another innovative approach to pattern completion has
involved molecular-genetic techniques to address what is, in
effect, a systems-level issue. Nakazawa et al. (2002) engineered
a mutant mouse in which the NR1 subunit of the N-methyl-
Theories of Hippocampal Function 627
D-aspartate (NMDA) receptor (see Chapters 6, 7, and 10) was
excised selectively in the CA3 region of the hippocampus.
Using the Cre-Lox technique to realize this selective ablation,
the resulting progeny (CA3-NR1 knockout mice) were shown
to have normal physiology and behavior on a range of tests
but certain highly specific impairments. For example, whereas
LTP was normal in the mossy fiber input to CA3 and in the
CA3 output to CA1, it was absent in the longitudinal associa-
tion pathway of CA3. This pathway is likely to be important
for building associative spatial relations between cues in an
environment (McNaughton and Morris, 1987; Treves and
Rolls, 1994). When trained in a watermaze (see Fig. 13–23,
below, for a description of this task), the mice learned nor-
mally with all cues present but showed selective deficit in
recall when tested with a subset of the training cues.
Interestingly, place-specific firing of CA1 complex-spike cells
was also “fuzzier” in the mutants when recording sessions
took place using partial cues. Nakazawa et al. (2002) inter-
preted this observation as implying a role for NMDA recep-
tors in area CA3 of the hippocampus in pattern completion,
and they outlined a model of the way areas CA3 and CA1 may
interact during the storage of information (Fig. 13–21). This
study also nicely illustrated the power of adult-onset, cell
type-restricted genetic manipulations in the integrative study
of the molecular, cellular, and neuronal circuitry mechanisms
underlying spatial cognition, a considerable advance on first-
generation transgenic techniques (Nakazawa et al., 2004).
13.4.3 Using Spatial Information: Spatial
Navigation and the Hippocampal Formation
The second key idea in cognitive map theory is that neural
activity in the hippocampal formation occurs during and is
required for spatial navigation. Allocentric navigation utilizes
the information stored as a “map” during earlier locale pro-
cessing, and reading information out of maps requires hip-
pocampal place cells to fire. Other cognitive processes related
to navigation involving the hippocampus include the ability to
mentally rotate a stored representation to align it with a cur-
rent view, to discover and use shortcuts, and so on. A key pre-
diction is that the integrity of the hippocampal formation is
absolutely required for locale-based navigation. This predic-
tion led to the development of a number of novel navigational
tasks for rats, the first of which was the radial maze.
Radial Maze
The radial maze is an apparatus in which rats (or mice) are
given the opportunity to collect food hidden at the end of
arms of the maze that radiate out from a central choice area
(Olton and Samuelson, 1976). In its simplest version, a hun-
gry rat is placed on the maze and allowed to run around until
it has collected all the food. Once the food at the end of one
maze arm has been eaten, it makes sense to avoid that arm,
which normal rats gradually learn to do. Once mastered, the
rats run to each arm without revisiting arms a second time
(Fig. 13–22A). They have, in effect, solved a special case of the
628 The Hippocampus Book
Control Mice
A Full cue condition
EC input DG/EC input
SC
CA1 CA3
firing
rate RC
position
C Partial cue condition
SC
CA1 CA3
firing
rate RC
position
Figure 13–21. Pattern completion in the storage and retrieval of
spatial memory using areas CA3 and CA1. Note the interconnecting
circuitry of CA3 and CA1, emphasizing pathways that form NR-
dependent modifiable synapses in CA3. A, B. Basic wiring of CA3
and CA1 illustrates a possible encoding mechanism for pattern com-
pletion. In normal and CA3-NR1 knockout (KO) mice (left and
right columns, respectively), a full cue input (top row) is provided to
CA3 from the dentate gyrus (DG) or the entorhinal cortex (EC) and
to CA1 from the EC (downward-facing arrows). During learning,
“traveling-salesman’s problem” of getting everywhere with the
least effort. Olton (1977) reported that rats with various
lesions that deafferented or deefferented the hippocampus
(e.g., fornix or entorhinal cortex lesions) were severely
impaired in their initial learning. Whereas normal animals
collected all available food from the ends of the maze arms
with little or no retracing, those with lesions moved around
the maze haphazardly, making numerous repeat entries into
arms from which they had already obtained food. The initial
interpretation of these data was that these surgical disconnec-
tions of the hippocampus disrupted spatial navigation and so
supported the cognitive map theory.
However, Olton began to doubt his interpretation follow-
ing a study suggesting that the deficit was really in trial-
specific “working memory.” This term was introduced into the
animal literature by Honig (1978) to refer to the memory of
information useful for only a single trial in a behavioral task.*
*Working memory” has a different meaning in human memory research,
where it refers to a limited capacity consisting of several subsystems such
as a phonological store and visuospatial sketch pad (Baddeley, 2001).
CA3-NR1 mutant mice
B Full cue condition
EC input DG/EC input
SC
CA1 CA3
firing
rate
D Partial cue condition
SC
CA1 CA3
firing
rate
the normal mice make associative changes in synaptic efficacy in the
commissural-association pathway (small circles representing poten-
tiated synapses), whereas CA3-NR1 KO mice do not. C, D. In later
testing under partial cues (bottom row), a fraction of the original
input is provided (fewer downward arrows) to activate any memory
traces formed during training. Only control mice are able to benefit
from the associative plasticity in CA3 that “completes” the pattern
and so provides an output to CA1 that can then be used for spatial
recognition and navigation. (Source: After Nakazawa et al., 2002.)
Olton and Papas, 1979) adopted a modification of the task,
originally suggested by Jarrard (1978), in which both sham-
operated and fornix-lesioned rats were tested in a 17-arm
radial maze in which a fixed set of arms were baited before
each trial and others were never baited (Fig. 13–22B).
Following pretraining as normal animals, these investigators
found that fornix-lesioned rats were impaired in remember-
ing to avoid repeat entries into arms from which they had col-
lected the available food but were as successful as the
sham-lesioned animals in avoiding the arms that were never
baited. Essentially the same results were obtained whether the
baited arms were arranged adjacent to each other or were
mixed haphazardly with the never-baited arms. This finding
conflicts with the cognitive map theory, as neither a distur-
bance of spatial representation nor of navigation can explain
the dissociation between failure to avoid repeat entries into
the initially baited arms and successful avoidance of the never-
baited arms.
Olton et al. (1979) therefore argued for a distinction
between the anatomical mediation of “working memory” (i.e.,

A The 8 arm radial-maze
reward to be collected
reward collected
no reward
rat path
Figure 13–22. Radial maze. A. The eight-arm radial maze is a task
in which rats run from a central region to collect food reward from
the ends of the maze arms. The path shown is of a rat that has suc-
cessfully made five choices without retracing its path and has three
arms remaining where food is available (notional path data). B. A
17-arm radial maze with baited and unbaited arms. The path shown
memory required for only one trial) and “reference memory”
(i.e., memory used for a series of trials). They proposed that in
radial maze tasks in which some arms are baited and some
unbaited, the animal uses working memory to keep track of
which arms it had entered as the trial proceeds and reference
memory to avoid entries down arms that never contained
food. Olton’s insistence on thinking about reference and
working memory as independent rather than interacting
memory systems carries with it the implication that the ani-
mal is reading out from two memory systems simultaneously
as it works its way from choice point to choice point —a pre-
scient idea, as many behavioral tasks are likely to involve mul-
tiple memory systems. In any event, if the hippocampus is
involved exclusively in working memory, not only could this
account for the data it also implies that there may be a subset
of allocentric spatial tasks for which the integrity of the hip-
pocampal formation is not required—spatial reference mem-
ory tasks.
The results of the Olton et al. (1979) study were, on the face
of it, damaging to the cognitive map theory; but several prob-
lems soon became apparent. First, Olton examined a variety of
lesions but, strangely, never looked at discrete hippocampal
lesions. Second, there is nothing to stop rats choosing adjacent
arms in the standard radial maze, thereby performing per-
fectly without having to remember anything about the spatial
layout of the maze. The path shown in Figure 13–22A indi-
cates that our notional rat did not do this, but many radial
maze studies ran into this problem. It was solved by placing
doors between the central area and the maze arms that pre-
vented successive left or right turns by allowing the experi-
menter to control the timing of the animals’ choices. However,
Theories of Hippocampal Function 629
B A 17 arm radial-maze with some maze-
arms baited and others unbaited.
might be that of a fornix-lesioned rat, initially trained preopera-
tively, that is selectively visiting the maze arms that are always
baited, making occasional repeat errors at these arms but success-
fully avoiding the maze arms that are never baited. (Source: After
Olton and Papas, 1979.)
even with doors, it is unclear whether a spatial map is guiding
successive choices in this task (Brown, 1992; Brown et al.,
1993). Third, in the Olton et al. (1979) study, the animals were
initially trained as normal animals and were lesioned only
after asymptotic performance had been reached. This training
arrangement leaves open the possibility that the integrity of
the hippocampal formation is required for the initial learning
of which arms to avoid but perhaps not required for retention.
Using animals that had undergone surgery prior to training, a
series of studies by Jarrard (1978, 1986) established that the
initial learning of place reference-memory in an eight-arm
radial maze task in which four arms were never baited is con-
sistently impaired by both complete hippocampal aspiration
lesions and by more selective ibotenic acid lesions of the hip-
pocampus and dentate gyrus. These studies were important
for introducing the most selective type of hippocampal lesion
that has yet been developed. In the process, the cognitive map
theory was partially salvaged as the integrity of the hip-
pocampus is required for initial learning. However, the long-
term storage of information about which arms are baited or
unbaited may be outside the hippocampal formation (Barnes,
1988). The theme that the hippocampal formation may be
required for the initial learning of a spatial task with eventual
storage in the cortex is one that recurs in many tasks, includ-
ing the next one we consider.
Watermaze
Further evidence that the integrity of the hippocampal for-
mation is essential for allocentric spatial reference memory
came with the introduction of the watermaze (Morris, 1981).
 Box 13–5
Watermaze

The watermaze has become one of the most widely used tasks in behavioral neuroscience. It
was developed to study spatial learning and memory but is now often used as a general assay of
cognitive function for testing new drugs or other treatments of the nervous system.
The basic task is simple. Animals, usually rats and mice, are required to escape from water
onto a hidden platform whose location can normally be identified only using spatial memory.
There are no local cues indicating where the platform is located. Conceptually, the task derives
from “place cells,” as these cells also identify points in space that cannot be defined in relation
to local cues. This simple water escape task is then embedded into a variety of sometimes quite
complicated training and testing protocols to investigate specific theoretical issues.
APPARATUS

The apparatus consists of a large circular pool, generally 1.5 to 2.0 m in diameter, containing
water at around 25C made opaque by adding milk or other substance (e.g., latex) that helps
hide the submerged platform. The water in the pool is filled and drained daily via an automated
filling and draining system. The choice of water temperature, around 13C below body temper-
ature, is sufficiently stressful to motivate the animals to escape but insufficiently stressful to
inhibit learning. There is a stress reaction (corticosterone release) on day 1 of training, but this
habituates over days. If the pool temperature drops to 19C, performance improves, but when it
drops to 12C, it gets worse, reflecting the inverse U-shaped function relating stress to cognitive
function. The pool is located in a laboratory room with distinctive two- and three-dimensional
distal cues that aid orientation or surrounded with hanging curtains that occlude these room
cues. Cues may then be hung inside the curtains so they can be rotated relative to the room
when this degree of experimental control is required. A video camera is placed above the center
of the pool to capture images of the swimming animal and is connected to a video or DVD
recorder and an online computer system running specialized tracking software. The top surface
of the hidden platform, usually about 10 to 15 cm in diameter and thus between 1/50th and
1/100th of the surface area of the pool, is 1.5 cm below the water surface (Fig.13–23A).
PROTOCOLS

“Reference memory” protocols have been widely used in which the platform is in a fixed loca-
tion relative to the room cues across trials and days. The animals are placed in the water at and
facing the side walls of the pool at different start positions across trials, and they quickly learn
to swim to the correct location with decreasing escape latencies and more direct swim paths
(Fig. 13–23B). The tracking system measures the gradually declining escape latency across trials
and parameters such as path length, swim speed, directionality in relation to platform location,
and so on. Observation of the animals reveals that, having climbed onto the escape platform,
they often rear up and look around, as if trying to identify their location in space. Rearing
habituates over trials but then dishabituates if the hidden platform is moved to a new location.

Figure 13–23. Watermaze. A. Axonometric drawing of a typical watermaze setup with overhead
video-camera and rat swimming to find the hidden platform. B. Representative escape latency
graph and swim paths across various stages of training: initial swimming at the side walls, then
circuitous paths across the area of the pool, and finally directed path navigation. C. The hidden
platform is removed for posttraining probe tests. Whereas normal or sham-lesioned controls
swim to the target quadrant (NE, within dotted gray lines), rats with hippocampus, subiculum,
or combined lesions do not. (Source: After Morris et al., 1990.) D. Overtraining of hippocam-
pus-lesioned rats can result in quite focused search patterns during a probe test. (Source: After
Morris et al., 1990.) E. Atlantis platform. The hidden platform is at the bottom of the pool
where the swimming rat cannot bump into it by chance. Online automated data capture of
swim paths is used to determine whether the rat swims within a virtual zone around the plat-
form’s location, raising the platform to within 1.5 cm of the water surface when a criterion is
reached. This protocol trains highly focused search patterns. (Source: After Spooner et al., 1994.)
F. Reversible hippocampal inactivation with a glutamate antagonist during training (encoding)
or at retention (retrieval) results in poor probe test performance compared to the controls,
which have the hippocampus working continuously. Analysis of the search patterns show that
rats trained with the hippocampus “on” during encoding, but “off ” at retrieval display searching
at the wrong location in the pool, as quantified using zone analysis. (Source: After Riedel et al.,
1999.)

630
 Theories of Hippocampal Function 631

A The watermaze B Paths and latency during place navigation

video
camera start mid end

Escape latency (sec)

hidden
platform
pool

base board

C Post-training probe tests (no platform) D Overtraining

control hippocampus subiculum hippocampus & hippocampus
lesioned lesioned subiculum lesioned lesioned

E Atlantis platform F Dissociating spatial memory and search strategies

by inactivation at encoding or retrieval.
on/on on/off off/on

Water surface
% time in trained quadrant

50

40

30 chance
20
zone
10 analysis
Platform Platform
down up 0
on/on on/off off/on off/off

G Delayed matching to place (DMP)

2h TI
I
IT
I
s
15

Day T1 T2 T3 T4 80 sham
ITI lesion
latency (sec)

N 60
short
N+x 40
medium 20
N+y 2h
long 0
T1 T2 T3 T4 T1 T2 T3 T4
Trials within a day, averaged across days

Figure 13–23. (Continued) G. Delayed match-to-place training involves four trials per day with
the location of the hidden platform moved between days. Training can continue indefinitely
with this protocol, enabling within-subject drug manipulations throughout the life-span and
averaging across days. Acquisition typically takes 8 to 10 days. For trial 1 of each day, the ani-
mals search for the platform, typically taking 60 seconds to find it, and encode its location; they
then show fast escape latencies during trials 2 to 4. Hippocampus-lesioned rats cannot learn
this task irrespective of the intertrial interval (ITI) between trials 1 and 2. The shaded zone
shows the ITI between T1 and T2 extended to 2 hours. (Source: After Steele and Morris, 1999.)
(Continued)
 Box 13–5
Watermaze (Continued)

During or after training is complete, the experimenter conducts a probe trial in which the
escape platform is removed from the pool, and the animal is allowed to swim for 60 seconds.
Typically, a well trained rat swims to the target quadrant of the pool and repeatedly across the
former location of the platform until it starts to search elsewhere (Fig. 13–23C). This spatial
bias, measured in various ways, constitutes evidence for spatial memory. Rats with lesions of
the hippocampus and dentate gyrus, subiculum, or combined lesions, do poorly in posttraining
probe tests.
However, if rats with hippocampal lesions are given “overtraining” (typically consisting of a
large number of trials over many days), performance can improve in probe tests. Even rats with
hippocampal lesions can show quite localized searching (Fig. 13–23D).
Numerous other protocols have been developed to test specific hypotheses. Many involve
cryptically moving the hidden platform. This might be a “reversal” procedure in which, after
one location has been thoroughly trained the platform is moved to a different quadrant of the
pool. Because it is hidden, it is not apparent that anything has changed until the animal fails to
find the platform in its usual place. The focus is on how the animal reacts to this change and
how quickly it learns the new location. The relearning that occurs with the reversal protocol has
been used in a major “factor analysis” of the determinants of watermaze behavior across strains
of mice.
As the animals sometimes “bump” into the submerged platform by chance, one useful inno-
vation is an “on-demand” or “Atlantis” platform that is initially at the bottom of the pool and
becomes available only when the animal swims in its vicinity for some predetermined time. An
automatic release system allows the platform to rise gently to near the surface of the water (it
remains hidden). This procedure results in acquisition of a highly focused searching strategy
that focused on the target location during training. Reversible inactivation of the hippocampus
with a drug that blocks excitatory neurotransmission after the training is completed results in
animals displaying localized searching at inappropriate places in the pool. Pharmacological
inactivation during training results in failure to develop this search strategy because the ani-
mals cannot learn the place at which to execute the strategy in the pool (Fig. 13–23E). The
accuracy of the searching can also be measured using a zone analysis that measures time spent
in a virtual zone around the place where the platform is located.
With other protocols, the platform can be moved to a new location each day, creating what is
called the “delayed matching to place” (DMP) procedure (Fig. 13–23F). The animal cannot
know, with this procedure, where the platform is hidden on trial 1 of each day; it can only
search for it. However, if it can encode this new location in one trial, the animal finds the plat-
form much faster during trial 2 and subsequent trials of that day. The memory delay interval
between trials 1 and 2 (ITI) can then be systematically varied to explore how well one-trial spa-
tial memory is remembered, a procedure with some similarities to delayed matching and non-
matching tasks used to examine recognition memory. Rats with complete hippocampal lesions
cannot show the rapid, one-trial learning required for the DMP task and are just as poor at a
short ITI between trials 1 and 2 as a long one (Fig. 13–23G). Treatment with an NMDA antago-
nist such as D-AP5 results in a selective deficit in memory at a long ITI, but the animals can
learn with short memory intervals between trials (see Chapter 10, Section 10.9).
Other variants include alterations to the apparatus, such as constraining the path of the
swimming animal to minimize navigational demands (e.g., an annular watermaze), decreasing
the number of available extramaze cues between training and testing to look at pattern comple-
tion, the use of floating platforms, and yet other manipulations. A radial watermaze has also
been introduced, combining the virtues of the radial maze with the ease of training to escape
from water. This has proved invaluable for testing transgenic mice expressing familial
Alzheimer mutations.
TREATMENTS AND CONTROLS

A wide variety of treatments have been explored including lesions, drugs, and molecular-
genetic alterations. They alter watermaze “performancez” in various ways, but experimenters
must be cautious, as such alterations need not be specific deficits (or improvements) in spatial
learning or memory processes per se. Lesions or drugs may have a direct effect on learning
mechanisms, and many seem to do so, but they may also affect an animal’s ability to see the
extramaze cues or their motivation to escape from the water rather than learning per se. Factor
analytical studies reveal that many molecular-genetic alterations influence the probability of
mice to stay at the side walls (thigmotaxis) instead of swimming into the center of the pool and
that this effect is statistically independent of spatial memory.

632
 Theories of Hippocampal Function 633

Accordingly, treatments must be accompanied by relevant control conditions. A common

protocol is to include trials in which the escape platform is made visible, the idea being that
treatments that merely affect motivation to escape should impair performance in this task as
well as the basic task. It is unclear how sensitive this assay is, but it does provide a first pass at
detecting gross sensorimotor abnormalities. As blind rats have been claimed to do surprisingly
well in the watermaze (except in probe trials), other psychophysical techniques have been intro-
duced offering precise control of the spatial frequency of cues that are more sensitive to subtle
visual deficits. The use of sham-lesioned animals, vehicle-infusion conditions, “floxed” mice,
and other pertinent manipulations has also become widespread to ensure that any alterations
in spatial learning in the experimental group are not an unintended by-product of achieving
the treatment.

WATERMAZE AS AN ASSAY

As our understanding of the impact of various treatments has developed, the watermaze has
gradually been subsumed into test batteries and assays for investigating aspects of brain func-
tion other than just hippocampal function. This is a strength of the assay but also an analytical
weakness. The task is so sensitive to manipulations of normal brain function in many brain
areas, not just the hippocampus, that it can be used almost like a “litmus test” of the “normal-
ity” of brain function. This is valuable, as it brings behavioral observations of function into
fields of neuroscience that have historically relied exclusively on stress (corticosterone release),
neuropathology (stroke research), biochemical analyses (Alzheimer’s disease), or electrophysiol-
ogy (development of cognitive enhancing drugs). The analytical weakness is that a task affected
by such a wide variety of treatments is gradually being revealed as having less “specificity” than
was once believed.
Innovation in the development of new training protocols is enabling the simplicity and
speed of learning to escape from water to be retained in the arsenal of behavioral tools at the
disposal of neuroscientists. Arguably, the watermaze illustrates both themes of this book: (1) a
tool to study hippocampal function and (2) an illustration of how studies of hippocampal
function have had a wide impact on neuroscience in general.

SELECTED REFERENCES

Morris, 1981, 1984; Sutherland and Dyck, 1984; Brandeis et al., 1989; Good and Morris, 1994;
Nunn et al., 1994; Whishaw and Jarrard, 1996; Gallagher and Rapp, 1997; Lindner et al., 1997;
Sandi et al., 1997; Lipp and Wolfer, 1998; Morgan et al., 2000; Prusky et al., 2000a; D’Hooge
and De Deyn, 2001.

Morris et al. (1982) trained rats to find a hidden escape plat-
form in a circular pool of opaque water. Some had been sub-
jected to “hippocampal” aspiration lesioning (damaging the
hippocampus, dentate gyrus, and subiculum), others cortical
control lesioning (damaging the same small amount of corti-
cal tissue in the region of the parietal cortex as was necessar-
ily damaged when making the hippocampal lesion), or no
surgery (unoperated control). The cortical “control” lesion
was included in this and many other early rat studies (in con-
trast to the monkey work discussed in Section 13.3). In the
watermaze, both the normal and the cortical tissue-lesioned
groups quickly learned to swim toward the hidden platform
from any starting location; only the hippocampus-lesioned
group was impaired in executing directed swim-paths.
Performance during a posttraining probe test showed that
both control groups swam persistently across the exact spot
where the platform had been located during training, a pat-
tern of behavior that could not occur during training and
was arguably reminiscent of “free recall.” In contrast, the
hippocampus-lesioned group swam all over the pool. The
deficit in escape latency disappeared when the platform was
made visible, indicating that the impaired place navigation
was unlikely to be secondary to any change in motivation to
escape from the water and reappeared when it was subse-
quently hidden again. Similar results were obtained in the first
study to use neurotoxic lesions in the watermaze by
Sutherland et al. (1983), investigators who were primarily
responsible for popularizing the task (Box 13–5).†
Over the years since the watermaze was developed, there
have been numerous replications of the deficit in place-
navigation learning caused by hippocampal lesions. Other
pertinent findings concern the impact of lesions to structures
afferent or efferent to the hippocampus and of fiber pathways
interconnecting these structures, as comprehensively reviewed
†
There have been differences in opinion about the name “watermaze” for
an apparatus that is not obviously a maze. Some describe it as a “water
task,” others as a “milk bath,” and many use an acronym “MWM.” The use
of the term “maze” is appropriate, as it is an apparatus in which the ani-
mal has to find its way around—and that is what one does in a maze. By
analogy, golfers play on a “golf course” not “golf fields.”
634 The Hippocampus Book
by D’Hooge and De Deyn (2001). For example, deficits in
place navigation have been reported after selective lesions of
the entorhinal cortex (Schenk and Morris, 1985), perforant
path (Skelton and McNamara, 1992), dentate gyrus
(Sutherland et al., 1983), area CA1 (Nunn et al., 1994), medial
septum (Hagan et al., 1988; Kelsey and Landry, 1988), fornix
(Morris, 1983), and subiculum (Taube et al., 1992). Deficits
are generally seen only after bilateral lesions. The overwhelm-
ing consensus is that the integrity of the hippocampal forma-
tion and its afferent and efferent connections is essential for
the acquisition and/or consolidation of normal, appropriately
directed spatial navigation.
However, as with the radial maze, further study has revealed
lesion dissociations that are not predicted by the cognitive map
theory and point to a functional heterogeneity that underlies
normal performance. Dysfunction in a variety of structures
has an impact on performance, notably damage to neocortical
structures such as the retrosplenial and parietal cortices,
frontal cortex, basal forebrain, striatum, and cerebellum—an
issue addressed in the critique below. A problem with the
watermaze is that its procedural simplicity belies an underly-
ing complexity with respect to the numerous sensorimotor
and cognitive processes that, in practice, it engages. That it has
been classified as a “spatial memory task” should not lull users
into believing that other sensorimotor and cognitive processes
are not involved. It would be tidy if the impact of lesions or
other dysfunctions of distinct brain areas or neurotransmitter
systems is isomorphic to the spatial mapping or navigation
processes identified within the theory, but the data point to a
more complex picture in which the hippocampus engages with
other brain structures to mediate effective performance.
Before considering navigation in the watermaze in detail, it
should be noted that many dry-land place-finding tasks have
also been developed, including the earlier Barnes arena task
first developed in the context of studies linking hippocampal
LTP to spatial learning (Barnes, 1979), a place preference task
in which rats are systematically rewarded by intracranial brain
stimulation for visiting a particular location (Fukuda et al.,
1992), similar appetitive place preference tasks (Rossier et al.,
2000), and a place avoidance task. These tasks are conceptually
similar but, being on dry land, the animal can stop moving
and is also more likely to use olfactory and somatosensory
(vibrissa) cues. This is likely to make them less dependent on
visual cues than the watermaze. With place avoidance, the
mirror image of place navigation, animals are trained stay
away from a location to avoid the punishment of mild foot-
shock (Bures et al., 1998). Place avoidance has been used
extensively to examine the differential role of allocentric ver-
sus ideothetic cues in location learning by allowing the circu-
lar arena to rotate slowly in relation to extramaze cues. Rats
can be successfully trained to avoid a sector defined in relation
to room cues or in relation to its location on the rotating
arena. An interesting commercial twist on place navigation
and place avoidance is incorporated into modern robot vac-
uum cleaners, a good example of a biologically inspired artifi-
cial navigation system (Trullier et al., 1997). These machines
clean the floor as they work their way around a living space
and typically incorporate two key features of spatial memory.
They “remember” where in the room to go to get their battery
recharged and to avoid recleaning places that they have
already moved across.
Varied Training Protocols Reveal the Importance
of the Dorsal Hippocampus for Spatial Navigation
A key distinction regarding the watermaze is between thinking
of it as an apparatus and as a task. Sometimes described as the
latter, it is really an apparatus in which a wide variety of train-
ing protocols can be carried out to address specific issues.
Some modifications have come into widespread use, others
not. One was a modification of the original spatial reference
memory protocol (Morris et al., 1986a) that for various rea-
sons was not taken up by other laboratories. It involves train-
ing rats to choose between two visible platforms, one of which
is rigid and offers escape from the water, the other floating and
so offers insufficient buoyancy for escape. Local-cue and spa-
tial versions of this task were compared. Hippocampal lesions
impair the spatial task but have no effect on the procedurally
similar cue task. In many respects this is a more rigorous test
of the dissociation between locale and taxon strategies pro-
posed in the cognitive map theory than the comparison
between the hidden and visible platforms of the “standard”
watermaze task because it compares two tasks of comparable
difficulty. It also removes the “recall” element of the standard
procedure because the navigational demands are minimized
and the animal need only approach either platform and then
use spatial recognition to decide whether it or the other plat-
form is associated with escape. Interestingly, the hippocam-
pus-lesioned group displayed first-choice performance that,
starting at 50%, gradually rose above chance. The effect was
not large, but it was both detectable and statistically signifi-
cant—a first suggestion that allocentric place navigation is
possible after hippocampal dysfunction. One possibility is
that it is easier in this task to bridge temporally discontiguous
gaps in the processing of relevant sensory information
(Rawlins, 1985).
Although a consensus emerged that the standard hidden
platform reference memory version of watermaze spatial nav-
igation is a “hippocampal task”—being easy to train and
producing reliable, replicable results in different laborato-
ries—the impact of such lesions is more variable than many
may appreciate. First, the sensitivity of the “cue task” (which is
unimpaired by hippocampal lesions) has been called into
question following the observation that rats with experimen-
tally reduced visual acuity display impaired place navigation
but are unimpaired in the cue task (Prusky et al., 2000b). This
dissociation falsely masquerades as a deficit in place learning.
Although a nice “proof-of-principle,” it is unlikely that hip-
pocampal lesions cause any change in visual acuity. However,
regionally nonselective transgenic manipulations and many

peripherally administered drugs may affect visual acuity.
Psychophysical procedures have therefore been introduced to
provide a more rigorous test of sensory acuity than the cue
task (Prusky et al., 2000a; Robinson et al., 2001), and it would
be valuable to see these used with selected transgenic lines.
Those with cell- and region-specific mutations are also
unlikely to be impaired, but the important lesson that needs to
be taken on board from these sensory studies is that caution
must be exercised about treating the watermaze as an ostensi-
bly selective assay of memory function. Interestingly, the cue
task is impaired by dorsal striatum lesions (McDonald and
White, 1994), a sensitivity that is more likely due to the role of
this structure in habit learning than any sensory impairment.
Second, as in the two-platform task, some place navigation
does take place in rats with hippocampal dysfunction. Using
overtraining in the standard, reference-memory single plat-
form procedure, Morris et al. (1990) found that rats with
lesions in the hippocampal formation could learn to escape to
the platform with short escape latencies after about 80 trials of
training. Overtraining never obviated slightly elevated escape
latencies relative to those of the controls; but after a mixed
series of hidden and visible platform trials, probe test per-
formance was spatially focused in the correct training quad-
rant and quantitatively indistinguishable from that of the
controls (Fig. 13–23D). This slow, incremental spatial learning
was apparent in animals with hippocampal or subiculum
lesions, whereas those with combined hippocampus and
subiculum lesions failed to learn despite extensive overtrain-
ing. Eichenbaum et al. (1990) also showed incremental learn-
ing in rats given fornix lesions when the animals were
consistently trained from a single location. Probe test per-
formance was, however, sensitive to the use of novel starting
locations around the pool. Animals subjected to retrohip-
pocampal lesioning that encompassed the region of the
entorhinal cortex containing the entorhinal grid cells also
showed gradually improving probe test performance, but
analysis of the paths taken revealed accuracy in swimming
across the correct platform location but little indication of
“recognizing” that this was the correct place to stop and search
(Schenk and Morris, 1985). However, not all studies have
observed effects of entorhinal lesions on spatial reference
memory in the watermaze (Bannerman et al., 2001), raising
the possibility of functional heterogeneity across the medial,
intermediate, and lateral regions of the entorhinal cortex. A
key issue may be whether such lesions encroach on the dorsal
region containing the grid cells (Hafting et al., 2005).
A third important issue, first raised by Andersen in relation
to the “lamella theory” of hippocampal information process-
ing (Andersen et al., 1971, 2000), is how much hippocampal
tissue is actually required for learning or retention. This was a
fresh way of looking at the impact of lesions on function—
focusing less on whether a lesion causes a deficit than if nor-
mal performance is still possible with a partial lesion of a
specific size. Moser et al. (1993) discovered that aspiration
lesions of the ventral hippocampus that leave only small “min-
Theories of Hippocampal Function 635
islabs” of dorsal hippocampus intact allow normal learning. In
contrast, lesions of the dorsal hippocampus that left the ven-
tral hippocampus apparently intact did not. They suggested
that a relatively small number of lamellae in the dorsal hip-
pocampus, if adequately connected to entorhinal cortex, the
septum, and other brain regions, were sufficient for learning.
A follow-up study using ibotenic acid lesions confirmed these
findings (Moser et al., 1995). The second study was conducted
because the aspiration lesions of the original study may have
inadvertently left the ventral hippocampus deafferented, and
thus the apparent dissociation between dorsal and ventral hip-
pocampus could have been an artifact. In fact, similar results
were obtained with as little as 27% of dorsal hippocampus
being required for normal spatial learning that was indistin-
guishable from that shown by sham-lesioned controls (Fig.
13–24). In parallel, Jung et al. (1994) observed that the firing
fields of place cells were less spatially selective in the ventral
hippocampus. Taken together, these observations fit well with
anatomical studies of the connectivity between the entorhinal
cortex and hippocampus, with particular zones of the dorso-
medial entorhinal cortex preferentially projecting to the dor-
sal (septal) hippocampus of the rat. Detailed anatomical
analyses have suggested the concept of “parallel closed loops”
as the substrate for both reverberation and the processing of
functionally distinct sets of cortical information (Amaral and
Witter, 1989; Witter et al., 2000) and of “functional differenti-
ation” along the long axis of the hippocampus (Moser and
Moser, 1998b). (For further discussion, see Chapter 3.)
Other studies using the watermaze, T maze alternation,
and contextual fear conditioning are consistent with this
dorsal-ventral (i.e., septo-temporal) gradation of function
(Bannerman et al., 1999; Richmond et al., 1999). The search
for a specific function for the ventral hippocampus has
included a return to J.A. Gray’s ideas about a possible role in
fear and anxiety (Bannerman et al., 2004). For example,
Kjelstrup et al. (2002) found that rats with selective lesions
located in the ventral pole of the hippocampus were less fear-
ful on an elevated cross maze and showed decreased neuroen-
docrine stress responses. These animals did, however, display
normal contextual fear conditioning. Bannerman et al. (2002)
have also implicated the ventral hippocampus in anxiety as
ventrally lesioned animals freeze less to unsignaled foot-shock
in a test chamber. They also show less neophagia (the fear of
novel foods). However, the notion that the ventral hippocam-
pus is not involved in spatial learning at all can be overstated.
De Hoz et al. (2003) observed that changing the reference
memory training protocol to one involving fewer trials but
over a larger number of days enabled rats with large dorsal
hippocampus lesions (i.e., only ventral tissue intact) to learn
just as effectively as rats with only spared dorsal tissue, as
measured by posttraining probe tests. Similarly, Broadbent et
al. (2004) found that ventral hippocampal lesions encompass-
ing approximately 50% of total hippocampal volume can
impair spatial memory in a watermaze reference memory
task. Thus, the dorsal/ventral distinction is not clear-cut.
636 The Hippocampus Book

A Search patterns during probe tests of representative animals with partial or complete lesions

intact HPC dorsal 20-40% intact ventral 40-60% intact absent HPC

B Relation between size of residual tissue and performance during the probe test

dorsal HPC intact

70
intact HPC
60
absent HPC
time in quadrant (%)

50
40

30

20

10
ventral HPC intact
0
0-20 20-40 40-60 60-80 100
residual HPC tissue (%)
Figure 13–24. Partial lesions reveal a dissociation of function along performance in animals with dorsal tissue. B. Quantitative assess-
the longitudinal axis of the hippocampus. A. After training by ani- ment of the probe tests show near-normal performance of rats with
mals with spared tissue volumes of different sizes that began at dif- as little as 20% to 40% of the dorsal hippocampus intact. Inset.
ferent ends of the longitudinal axis of the hippocampus, posttraining Parasagittal diagrams of the hippocampus show the locations of
probe tests (hidden platform absent) revealed a graded change in intact (white) and lesioned (black) tissue along the longitudinal axis.
animals in which only ventrally located tissue was intact but good (Source: After Moser et al., 1995.)

A fourth important issue is asymmetry in the impact of
lesion size on initial acquisition versus retention. Retention is
more sensitive. If normal rats are trained in the watermaze
and then given small lesions of the mid- or ventral hippocam-
pus immediately after training, memory retention is impaired.
However, such lesions have no effect on learning or later
retention when given before training (Moser and Moser,
1998a). This dissociation is paradoxical from the perspective
of thinking about the hippocampus as a distributed associa-
tive memory system (Rolls and Treves, 1998) because such
systems are generally robust in the face of partial damage.
However, the finding is not necessarily incompatible with
such a perspective if encoding and trace storage takes place
across an extended region of the longitudinal axis in normal
rats. Animals given lesions before training would use the
remaining tissue, and normal learning should occur provided
there was enough of it. Small lesions created after training
may indeed damage only a small proportion of the synaptic
storage sites but may nonetheless interrupt inputs to and out-
puts from such sites, even when neurotoxic lesions are used.
To test this idea more rigorously, Steffenach et al. (2002) inves-
tigated whether the integrity of the longitudinal-association
pathway of CA3 pyramidal cells is critical for integrating such
distributed information. In animals previously trained as nor-
mal rats in the standard reference memory task, a single trans-
versely oriented cut of no more than 3% of the hippocampal
volume was made bilaterally through the dorsal CA3 region
(the lesion shown in Figure 13–1, panel 5B). Memory reten-
tion was impaired. These knife cuts inevitably damage cells as
well and possibly some Schaffer collaterals coursing to CA1,
but no impairment was observed in controls deliberately
given small, equivalent-sized neurotoxic lesions of CA3 cells
but leaving the longitudinal fibers intact. This study is unusual
amongst behavioral studies in using FluoroJade tract-tracing
to observe degenerating fibres on either side of the cut.
Taken together, these studies led by the Trondheim group
have definitively established that the efficient acquisition and
retention of spatial memory requires only a small “minislab”
of hippocampal tissue, preferentially although not exclusively
in the dorsal hippocampus. The integrity of longitudinally

oriented, translamellar connections of CA3 pyramidal cells in
this minislab is essential for normal functioning. It is not yet
known if this is a deficit in memory retrieval or consolidation,
but varying the interval after initial training before the longi-
tudinal cut is made would be one way of investigating this
question (see discussion of memory consolidation below).
Spatial Recognition Does Not Require CA3 Cells
If cognitive mapping involves both spatial representation and
navigation, the question arises of whether different parts of
the hippocampal formation are involved in recognizing a
place versus navigating to it. Using a modification of the
watermaze that involved changing it to no more than a circu-
lar swimming track, Brun et al. (2002) have shown that CA3
lesions can leave CA1 place cell firing fields and spatial recog-
nition intact. Rats were intensively trained to swim in an
annular watermaze for a hidden platform that could rise from
the bottom of the pool at a specific location once the animals
had completed two or three laps of the pool. They displayed
“knowledge” of this potentially safe place in the pool by hesi-
tating or by swimming more slowly there during the succes-
sive laps. Complete CA3 lesions leave this “spatial recognition”
intact, while complete lesions of the hippocampus impair this
ability (just as they impair navigation in the open pool). This
finding complements the further finding that CA1 place cells
can be normal in rats with CA3 lesions, which points to the
importance of the direct pathway from layer III of the entorhi-
nal cortex into CA1 (Brun et al., 2002). Taken together with
the study of Steffenach et al. (2002), there is therefore a dou-
ble dissociation, reflecting the importance of the longitudinal
connections in CA3 for navigation and the direct perforant
path input to CA1 for spatial recognition. It would, nonethe-
less, also be valuable to damage the cells of origin of the layer
III input to investigate the impact on CA1 place cell firing and,
if the lesion could made be large enough while remaining
selective (a tall order), on behavior as well.
Place Cells, Head-direction Cells, and Navigation
Although these lesion studies indicate that the integrity of at
least part of the hippocampus is necessary for spatial naviga-
tion, it does not require that place and head-direction units
are the critical neural mediators of navigation. Indirect evi-
dence that these cells are important has come from a number
of studies in which pharmacological or genetic manipulations
cause alterations in both place cell firing and spatial learning
(Mizumori et al., 1994; Wilson and Tonegawa, 1997; Kentros
et al., 1998; Rotenberg et al., 2000; Cooper and Mizumori,
2001). However, an alternative possibility is that these cells
only provide information about current location and direc-
tion, and other still unidentified cells mediate the representa-
tion of goals and carry out the information processing
responsible for navigational choices—the problem identified
in Figure 13–16B. Such cells may or may not be located in the
hippocampal formation.
Theories of Hippocampal Function 637
Prompted by tantalizing observations on the relation
between place cells and navigation by O’Keefe and Speakman
(1987) and their dependence on the spatial reference frames
used by the animals for specific tasks (Gothard et al., 1996;
Zinyuk et al., 2000), Poucet and colleagues embarked on a sys-
tematic research program to identify whether the firing of
place cells is critical for navigational decisions and goal iden-
tification (Lenck-Santini et al., 2001, 2002). As behavioral and
cell firing had to be obtained simultaneously, they began by
returning to the use of traditional mazes. The principle guid-
ing this work was that altering place cell firing fields by rotat-
ing or otherwise changing environmental cues should also be
associated with systematic changes in navigational behavior if
place cell firing and navigation are causally related.
Lenck-Santini et al. (2001) trained rats in a continuous
spatial alternation task on a Y maze. One arm of the maze (G)
served as the goal where food could be obtained. The animal
had to alternate its visits to the other arms (A, B) to secure
food in G. Thus, the correct sequence would be A → G* → B
→ G* → A → G* → B and so on that secured reward (*) upon
every visit to G, whereas an incorrect sequence might be A →
G* → A → G, which did not (Fig. 13–25). This incorrect
sequence was classified as an “alternation error” because the
animal returned to A when it should have gone to B. A differ-
ent, and initially much rarer, sequence might be A → G* → B
→ A in which, having correctly alternated between A and B
after a visit to the goal, the animal then returned to A instead
of going back to the goal. This was called an “orientation
error.” A prominent cue card was located on one side of the
maze. Once all the animals were trained and performing well
(around 80% correct choices), a total of 47 single units were
recorded while the animals continued to run on the maze.
Various manipulations of the cue card (rotation, removal)
were systematically made to observe the impact on both place
cell firing and behavior. As expected, the probability of correct
alternation sequences declined during these manipulations;
but more importantly, they did so in a systematic way. Pairs of
recording/behavior sessions were compared. The second ses-
sion of each pair was categorized as a “consistent” session
when the location of a place field on the Y maze relative to the
goal arm was the same as during the first session; it was cate-
gorized as “inconsistent” when the cue manipulation between
the pairs of sessions caused some shift in the relative location
of the place field to the goal. Behavioral choice performance
was significantly poorer on inconsistent sessions. Moreover, an
analysis of the kinds of errors made by the animals indicated
that alternation errors occurred occasionally on consistent
sessions, whereas inconsistent sessions were characterized by a
striking increase in orientation errors. These findings indicate
that the locations of place fields in a learned spatial navigation
task are relevant to performance. Inconsistent place fields were
statistically associated with disorganized spatial behavior.
Similar observations were also made by Dudchenko and
Taube (1997) with respect to head-direction cells.
In a follow-up study, Lenck-Santini et al. (2002) attempted
to compare the value of correct place cell firing in a “locale”
638 The Hippocampus Book
The relationship between the spatial firing of CA1 cells and spatial navigation
A B A B
correct alternation error
120°
cue card cue card
goal
A G B G A
Figure 13–25. Relation between place cell firing and spatial naviga-
tion. Rats from whom CA1 cell recordings were made simultane-
ously were trained to alternate on a Y maze such that the correct
order of arms would be A to Goal, then B, then back to Goal, and so
on (A → G → B → G, and so on) (left panel). With the cue card in
its training position, errors occurred rarely but when they did occur
task that required spatial memory with that in a “taxon” task
that did not. They reasoned, in keeping with the cognitive
map theory, that inappropriate place fields induced by
cue manipulations should have a greater effect on behavior in
the locale task. They used a place preference task, one of the
appetitive analogues of the watermaze (Rossier et al., 2000), in
which rats search around a large circular arena for
food they receive if they enter a virtual circular zone located at
one particular position in the arena (the food drops into the
zone from a hidden feeder above—a scientific version of
“manna from heaven”). In animals from which place cell
recordings were being taken simultaneously, this task was
conducted in one of three ways: with the virtual zone defined
as at a distance from a cue card at the periphery of the
arena, with it placed adjacent to the cue card, and with
the zone defined by a local cue explicitly outlining its area. The
key finding was that certain manipulations, such as visibly
rotating the cue card while the animal was in the apparatus,
created shifts between the expected location of the virtual
zone as defined by the animal’s place fields and by the
now rotated cue card. In brief situations of ambiguity (a few
minutes), most rats searched in a region defined by the
various place fields that were being recorded rather than as
determined by the now rotated cue card. This did not occur in
the taxon task when the zone was adjacent to the cue card
or visibly identifiable by local cues. Poucet et al. (2003) argued
that spatial searching behavior correlates well with place firing
fields only when the rat is likely to be using a place navigation
strategy. Data showing a correlation between the ability of
an animal to find a goal on a linear track and the alignment
of place fields with salient room cues is consistent with
Poucet’s interpretation (Rosenzweig et al., 2003). Place firing
fields, however, may be maintained in a taxon task when ani-
cue card
A B
orientation error
goal goal
G A A G B A
tended to be alternation errors (middle panel). However, when the
cue card was rotated (as shown) or removed or the goal location
was rotated, orientation errors increased in frequency. They
occurred when the cells were induced to display place fields that
became out of register to their standard positions. (Source: After
Lenck-Santini et al., 2001.)
mals alternate between locale and taxon tasks (Trullier et al.,
1999).
However, as noted by Jeffery (2003), the findings of these
studies are only correlational. Experiments in which neural
representations of location (e.g., place fields) are manipulated
directly, rather than indirectly via environmental manipula-
tions, are required to provide a yet more rigorous test of the
role of place cells in mediating behavior. It is not obvious,
however, how this might be done. Dudchenko (2003) took the
same view with respect to head-direction cells and suggested
that selective disconnection lesions might be a way of teasing
apart cause and effect.
Knowing Where, Getting There, and the Puzzle
of Hippocampal Involvement in Path Integration
The term “spatial navigation” has been unpacked in various
ways. The original theory made only a binary distinction
between locale and taxon processing during navigation, as
if there are only two ways to get from A to B. However, evolu-
tion has discovered numerous other ways to navigate, ranging
from “path integration” to “local views” (Benhamou, 1997;
Biegler, 2000; Rodrigo, 2002). Inspired in part by ideas relating
the hippocampal theta rhythm to aspects of movement mon-
itoring and control (Vanderwolf et al., 1973), a series of stud-
ies by Whishaw explored the possibility of functional
dissociation in navigation between “getting there” and “know-
ing where” (Whishaw et al., 1995). They recognized that spa-
tial learning in the watermaze involves both learning the
layout of distal cues and the development of navigational
strategies to use this learned information. Following the sev-
eral observations that hippocampus-lesioned rats may be able
to acquire spatial information slowly, Whishaw et al (1995)

wondered if fimbria-fornix lesions might disrupt only the
“getting there” component and not the “knowing where.” This
line of reasoning was to lead to unraveling the different senses
of “getting there.”
Although rats with fimbria-fornix lesions normally have a
profound deficit in watermaze place navigation when trained
to a hidden platform from several starting points, they display
several indications of having learned the platform location
when trained using a careful shaping procedure. They learn to
head in the correct direction, slow their speed of swimming
when approaching the vicinity of the platform, and search in
the correct quadrant when the platform is removed. Although
not as effective as control animals on all measures, there is
clear evidence that fimbria/fornix-lesioned animals can learn
something about location. However, a substantial deficit reap-
pears when the animals are switched to the delayed match-
ing-to-place (DMP) protocol in which the hidden platform is
moved from one location to another between days. Whishaw
et al. (1995) suggested that this profile is consistent with dis-
ruption of a process associated with getting to a hidden plat-
form, not one of learning its location. This interpretation was
backed up by a later study that, following visible platform
training from three starting locations around the pool,
included probe tests from a novel start location. Once again,
fimbria/fornix-lesioned rats swam in the correct direction
toward the now hidden platform, implying some knowledge
of where it was located. Neither they nor the control rats could
do this if vision was temporarily occluded (Whishaw, 1998).
In both studies, section of the fimbria-fornix was complete,
and the findings were in no way due to partial damage.
Whishaw (1998) also drew attention to similar data demon-
strating some indication of effective place navigation after
prior visible cue training in the studies of Morris et al. (1982).
Morris et al. (1990), in contrast, had argued that acquiring a
spatial representation that encoded a single escape location
might occur slowly in rats with hippocampal lesions, with the
“map” stored in the neocortex, provided interference could be
minimized. Visible platform training may help in this respect,
particularly if many trials are undertaken. Both the “getting
there” and the “slow learning” ideas can account for the failure
of rapid learning in the DMP protocol in both Whishaw’s and
Morris’s studies. Moreover, the data concerning a novel start
location are ambiguous because the experimenter cannot con-
strain the animals’ swimming patterns in the open pool dur-
ing earlier training. The view from the ostensibly novel start
location is likely to be familiar.
The focus by Whishaw (1998) on the “getting there” com-
ponent of spatial navigation is important. One the many ways
of getting from A to B is “path integration” (often called dead-
reckoning), a process that has been extensively studied in a
variety of animal species (Mittelstaedt and Mittelstaedt, 1982;
Wehner et al., 1996; Etienne et al., 1998). Path integration was
identified as the evolutionary building block of the “frag-
ments” within the fragment-fitting extension of the cognitive
mapping theory in which small parts of space (the fragments)
are assembled by an annealing process into a larger map of an
Theories of Hippocampal Function 639
entire area (Worden, 1992). Path integration operates using
ideothetic cues (i.e., cues generated during the course of
movement) that may be vestibular, kinesthetic, or visual flow
in origin. McNaughton et al. (1991) and Bures et al. (1998)
have joined Whishaw in emphasizing the importance of path
integration to hippocampal function. Redish (1999) shares
this view of its importance but places the path integrator out-
side the hippocampus.
A special case of path integration, referred to as “homing,”
has been extensively studied in hamsters (Etienne and Jeffery,
2004). It occurs when an animal, having left its nest site to for-
age for food, somehow keeps a continuous record of the dis-
tance and direction it has moved and uses this information to
take a relatively direct path back to the nest. This is not a mat-
ter of retracing its steps as the paths out may be circuitous,
whereas those back are direct. Field studies with desert ants
have also revealed the accuracy of this form of home-base
navigation and imply that it is an evolutionarily old form of
navigation. It solves the problem of getting food and then get-
ting home again without recourse to any “map.” Laboratory
studies of mammals using rotating arenas, movable nest
boxes, and foraging trips in the dark have established that
homing can function in total darkness, that it accumulates
errors over both time and the circuitous movements of the
animal, and that it typically returns the animal to a point near
but just short of the home-base by a vector likely to intersect
the initial part of the outward path. The animal can then
switch to searching using local cues (e.g., olfaction) to find the
exact spot of a nest or burrow entrance. As path integration
accumulates errors over time, it is not surprising that the vec-
tor processor can be “reset” by visual or olfactory cues, such as
familiar landmarks or territory-defining odors. In addition, as
many rodents are nocturnal and live in burrow systems, a self-
motion monitoring system of this kind is likely to be helpful
in many situations. Etienne’s view is that the neural algo-
rithms used to yield directional and positional information in
rodents remain elusive. Path integration capacity develops
postnatally, presumably as specific local circuits that process
the relevant sensory information become functional, and it is
first observed as young pups abandon their sole use of olfac-
tory cues to find the dam in the nest box in favor of forays in
the world beyond. Right from the start, the very rapid and
very direct return paths they display appear to be guided by
path integration.
The idea that the hippocampal formation might be
involved in path navigation has been offered by a number of
investigators (Wiener, 1993; McNaughton et al., 1996; Bures et
al., 1998; Whishaw, 1998). Whishaw and colleagues (Whishaw
and Tomie, 1997; Whishaw et al., 2001) trained rats on an
arena similar to that used by Etienne in which they sponta-
neously carried large food pellets back to their next box that
was either visible or hidden at the edge of the arena. Search
paths from the home-base were circuitous and marked by sev-
eral pauses as the rats tried to find the cryptic food pellet, but
the return path was much more direct. In normal animals, the
return paths continued to be direct in the dark, suggesting
640 The Hippocampus Book
guidance by a homing vector. However, return paths were
clearly disrupted in fimbria/fornix-lesioned animals, which
often took circuitous routes around the circumference of the
arena and quantitative changes with respect to both direction
and velocity. However, data from Alyan and McNaughton
(1999) have cast doubt on this finding because rats with hip-
pocampal lesions seem to be able to home accurately in the
dark, a finding that led Redish to place the path integration
machinery outside the hippocampal formation (Redish,
1999). The precise role of the hippocampus in path integra-
tion remains unclear, but the topic has been of considerable
interest for neural network modeling (see Chapter 14).
13.4.4 Comparative Studies of Spatial
Memory and the Distinction Between
Spatial and Associative Learning
The third key feature to highlight about the cognitive map
theory is the dual claim (1) that allocentric spatial mapping
evolved as a distinct, independent memory system of the
vertebrate brain mediated by a specific brain area and (2) that
spatial learning is qualitatively distinct from associative learn-
ing in that it is rapid, flexible, and uniquely encodes informa-
tion into map-like representations. These two strands of
an “adaptive specialization” hypothesis of hippocampal func-
tion (Sherry and Schacter, 1987; Papini, 2002) have been inves-
tigated in two contrasting research constituencies: by students
of comparative cognition (who are broadly sympathetic) and
by general process learning theorists investigating possible dif-
ferences between spatial and associative learning (who
are not).
Comparative Studies Using Avian
Species and in Naturalistic Situations
The original cognitive map theory put value on adopting a
“neuroethological” as well as a neuropsychological approach
to structure/function relations, an aspect reaffirmed by
(Nadel, 1991. p. 224). “It is important,” he wrote, “to adopt a
perspective on brain and behavioral organisation that takes
into account ecological, evolutionary and ethological aspects
of the animal under study.” The argument is that if space is
important because “we live in it, move through it, explore it,
defend it,” there will have been selection pressures of various
kinds favoring the evolution of effective cognitive mecha-
nisms for representing and navigating through space. Since
1978, study of the hippocampus in relation to “naturalistic”
spatial behavior, in both field settings and the laboratory, has
thrown up some striking brain/behavior correlations in sev-
eral avian and mammalian species.
One finding is that the relative volume of the hippocampus
in certain species of passerine birds (e.g., tit species in Britain,
chickadees in North America) is larger in the subspecies that
store seeds than in those that do not (Krebs et al., 1989; Sherry
et al., 1989) (Fig. 13–26A). For example, the hippocampal for-
mation of the food-storing marsh tit is 31% larger than the
non-food-storing great tit, even though the animal and the
rest of its forebrain are much smaller. The importance of this
correlation between the relative size of the hippocampus and
cache recovery derives from the argument of Andersson and
Krebs (1978) that food-storing is an evolutionarily stable
strategy provided the animal that stores seeds utilizes the pri-
vacy of memory to relocate seed caches. Not surprisingly, the
alternative strategy of tagging sites with some visible marker
to avoid having to use memory is vulnerable to intruders,
which then steal the caches. Ingenious field studies of tits in a
wood near Oxford using radiolabeled seeds to enable the relo-
cation of nonrecovered caches provided evidence that some
kind of location memory is involved (Shettleworth and Krebs,
1982; Shettleworth, 1998). Similar observations were made in
corvids that cache in the autumn but do not retrieve until the
winter or early spring (Balda and Kamil, 1989). Effective food
caching by scatter hoarders therefore relies on memory. A nice
feature of this fieldwork is the implied reminder that memory
can be allied to “privacy.” Like us, there are things animals do
and remember and keep to themselves.
Food caching is not an isolated case. Sherry et al. (1992)
pointed out that artificial selection among pigeon breeds
appears to have had an effect on hippocampal volume similar
to that of natural selection among passerine birds—the
hippocampus of homing pigeons being larger than that
of nonhoming domestic breeds. Studies of brood parasitism
are also interesting: Female cowbirds that lay their eggs in
the nests of other unsuspecting birds have a larger hippocam-
pus than male cowbirds (Lee et al., 1998). Cowbirds doing
this have to seek out and remember the locations of possible
nests and then time their egg laying to coincide with that of
the host bird. The occurrence of a common neuroanatomical
feature in unrelated species ostensibly exposed to similar
selection pressures raises the possibility that convergent evo-
lution is responsible—the development of an adaptive spe-
cialization.
A potential problem with naturalistic studies is that they
can be overly suggestive. As Shettleworth somewhat guardedly
put it, “animals show many behaviors for which explanations
in terms of human-like understanding readily come to
mind: the animal has a cognitive map, a concept, a theory of
mind . . . a leap of imagination may be necessary to grasp that
behaviors so readily explicable by such intuitively appealing
mechanisms are accomplished in completely different ways”
(Shettleworth, 2001, p. 279). She and Krebs were among the
first to appreciate that field studies had to be followed up,
where possible, by rigorous laboratory analogues. They went
on to create an avian laboratory setting in which food storing
and retrieval could indeed be studied under controlled condi-
tions, confirming the observations of the role of spatial mem-
ory in cache recovery first discovered in the wild. A typical
avian food-caching laboratory is an indoor room with “trees”
where seeds can be hidden (Fig. 13–26B). Such rooms are
often described as “large,” but they are of course much smaller
than the areas over which the birds normally forage in their
natural habitat, an issue always to be borne in mind when a

A Relative HPC volume in storers/non-stores
0.4 non-storers
storers
hippocampus residuals
0.3
0.2
0.1
0.0
-0.1 4
6
-0.2 8 7 5
-0.3
-0.35 -0.25 -0.15 -0.05
telencephalon residuals
C Spatial memory error over time
125
non-storers
storers
distance error (mm)
100
75
50
25
0 -0.15
2.5 5 10
retention interval (seconds)
Figure 13–26. Relation between relative hippocampal size and spa-
tial memory in birds. A. Study of 35 species or subspecies of passer-
ine birds revealed that hippocampal volume, relative to the rest of
the telencephalon (residuals), is larger in storing species. (Source:
After Krebs et al (1989). B. Laboratory aviary used to study storing,
window-shopping (see text), and other food-caching protocols in
birds. (Source: Courtesy of Nicola Clayton.) C. Spatial memory is
laboratory study yields a null result. Researchers using this
kind of facility were faced with the challenge of how to com-
pare cache memory by a species that does store seeds with the
spatial memory displayed by a species that does not.
Shettleworth and Krebs did this by creating a paradigm called
“window shopping” in which seeds that could be retrieved
later were placed “on display” behind Perspex screens in holes
drilled into the trees. Thus, during the initial sampling trials,
the birds were allowed to fly into the aviary without storing or
retrieving—merely to see what was on offer and where. Later,
the Perspex screens were removed, and the birds (presumably
armed with their credit cards) could fly back and collect what
they had seen earlier. In this and other one-trial associative
paradigms (Clayton and Krebs, 1995), the act of caching was
finessed while still allowing comparisons of spatial memory
between storers and nonstorers. Other approaches include
operant paradigms with “peck screens” (an avian touch-
screen) to cast light on what qualitative aspects of memory
might be better with a large hippocampus. In this way, Biegler
et al. (2001) found that coal tits (a food-storing species) per-
Theories of Hippocampal Function 641
B Laboratory aviary to study food storing
2
9
1
3 9
1
0.05 0.15 0.25
D Development of HPC size with experience
pre-experimental baseline
0.10
residual hippocanmpal volume
inexperienced storers
experienced storers
0.05
0.00
-0.05
-0.01
20 35-59 60-83 115-138
stage of experiment
(days posthatch)
more persistent over time in coal tits (a storing species) than great
tits (a nonstoring species). (Source: After Biegler et al., 2001). D.
Relative hippocampal volume at different stages of development.
The structure becomes larger if the young marsh tits are allowed to
store sunflower seeds (black bars), but it regresses and shows more
apoptotic cells if the birds are given powdered seeds that cannot be
cached. (Source: After Clayton and Krebs, 1995.)
form better than great tits (a nonstoring species) on a task
assessing the persistence of memory over time (from 2.5 to 25
seconds) but not on tasks that assess the fine-grained resolu-
tion or, somewhat surprisingly, the capacity of memory (Fig.
13–26C). This is an important reminder that not all parame-
ters of spatial memory are better in storers than nonstorers.
Another question that seminaturalistic studies have made
possible concerns the ontogenetic development of the
caching-associated difference in hippocampal size. How does
experience interact with any genetic predisposition? In a clas-
sic study, Clayton (1995) reared juvenile marsh tits (a food-
storing species) under conditions in which they were either
allowed to store seeds or prevented from doing so (by giving
them only powdered food). Tits allowed to store had the usual
large hippocampal volume (Fig. 13–26D). However, birds that
were prevented from storing showed a gradual decline in hip-
pocampal size and the highest proportion of apoptotic neu-
rons. These findings suggest that hippocampal volume, and
perhaps neurogenesis also (Patel et al., 1997), may be partially
regulated by activity-dependent factors such as the behavior
642 The Hippocampus Book
of the animal, a concept captured by the slogan “use it or lose
it.” If so, it might be expected that relative volume in a species
might also be seasonal. Specifically, the onset of hoarding
behavior might the trigger for an increase in volume, with
regression later. This has been confirmed in black-capped
chickadees (Smulders et al., 1995) and appears to be due to a
net increase in cell number (Smulders et al., 2000). The possi-
bility that it is merely due to a general seasonal mechanism
operating on both food-storing and nonstoring species is less
likely given that seasonal changes in volume have been shown
not to occur in at least one nonstoring species (Lee et al.,
2001). Males of this same species do, however, show changes
in the size of song-related nuclei, such as the higher vocal cen-
ter (HVC) and archistriatum (RA) across the seasons that cor-
related with song stereotopy. It therefore seems reasonable to
infer that changes in the relative volume of the hippocampus
of food-storing avian species results from a species-specific
and regionally specific form of structural plasticity that occurs
in response to seasonal pressures to engage in food-caching
and later retrieval. One slightly puzzling finding is that the
seasonal variation in size is, in black-capped chickadees, asso-
ciated with the time of year when caching occurs, but hip-
pocampal volume regresses in the winter—the very time
when caches are recovered. Why retrieval places less of a
demand on the need for a large hippocampal volume than
memory encoding is unclear.
Allometric brain/behavior relations in the avian brain are
intriguing, but they are a “chicken and egg” problem! What
causes what? Lesion studies were a first step in investigating
causal relations, it being essential to check that lesions of
the avian hippocampus impaired location memory, just
as such lesions do in mammals. These studies confirmed
this prediction (Sherry and Vaccarino, 1989). Hampton and
Shettleworth (1996) went on to show specificity to location
rather than color memory after hippocampal lesions, and
work using reversible inactivation in chickadees has both
borne out this claim and established that it operates at the
time of memory encoding (Shiflett et al., 2003). These and
other studies (Columbo et al., 1997) have established that the
functional integrity of the avian hippocampus is required for
spatial memory (Macphail, 2002), but they do not really nail
down whether the variation in the size of the hippocampus is
causally related to the very changes in behavior with which
they are correlated. Nor is it clear to what aspect of spatial or
other information processing the avian hippocampus con-
tributes. Work on homing pigeons has shed light on the sec-
ond of these issues.
Homing Pigeons Reveal Hippocampus-
dependent and Hippocampus-independent
Components of Navigation
Homing pigeons have a remarkable ability to fly back to their
lofts over long distances from both familiar and unfamiliar
release sites. The possible contribution of the hippocampus to
homing behavior has been the subject of a comprehensive
series of visual-sighting and radio-tracking experiments by a
group based in Pisa in Italy over the past two decades. Early
studies suggested that the integrity of the hippocampal for-
mation was critical for aspects of the navigation required, par-
ticularly near the home loft, but more recent work has pointed
to a more complex picture.
Pigeon homing seems to be guided by a multiplicity of
strategies and mechanisms, in part because of their need to
cope with differing environments and potential variation in
weather conditions. When released from an unfamiliar loca-
tion, experienced homing pigeons fly off with a “vanishing
bearing” in the correct direction of home (this being the com-
pass bearing that the bird takes measured from the release site
to last sighting after release). To do this, it was suggested by
Kramer (cited in Rodrigo, 2002) that they use a “long-range
navigational map” that may, at least in Italy, be partly reliant
on odors (Papi, 1990). A problem with this concept is that it is
far from clear how the pigeons learn such a map. In any event,
they also keep flying in the correct direction and have long
been thought to have some kind of a “compass” to do it
(Schmidt-Koenig, 1961, cited in Gagliardo et al., 1999). Once
they approach the vicinity of the loft, they recognize the land-
marks and the layout of the area they are used to flying around
at home and so find their way back to their perch in the loft.
It is not just the loft area that may be spatially familiar; the
birds can also learn about the landmarks at familiar release
sites and even the roads over which they may fly. Indeed,
Italian pigeons stylishly display a classic superiority in matters
navigational by following old Roman roads such as the SS
Aurelia (Lipp et al., 2004).
The early lesion studies established that pigeons with telen-
cephalic lesions, including the hippocampal formation, have
vanishing bearings from an unfamiliar release site that are as
accurate as controls (Bingman et al., 1984). Their long-range
navigational map is therefore intact. They also reach the vicin-
ity of the loft but sometimes have difficulty finding it—unless
they have been given sufficient postoperative experience of the
loft vicinity (Bingman et al., 1987). This implies that the “sun
compass” can work as well, this having been learned through
sightings of the sun at different times of day when outdoors at
the loft site. More recent studies have confirmed that the hip-
pocampus plays only a small role in the navigational map.
Acquisition and retention is independent of the hippocampus
if the birds acquire their navigational map during free flights
(Ioale et al., 2000). However, young birds can also learn the
navigational map when held in an outdoor aviary. Under these
conditions, learning seems to be hippocampus-dependent; but
perhaps surprisingly, its retention is not (Gagliardo et al.,
2004). The reason for this difference is that hippocampal
lesions disrupt the process by which sun compass information
is used only in the context of new learning (Bingman and
Jones, 1994). Use of this compass requires that the animals
keep track of time and orient in an appropriate direction to
the angle of the sun given the time of day. When “clock-
shifted” by being kept on artificial light–dark schedules, the
orientational bearing is shifted from the correct homeward

direction. Experienced pigeons later subjected to hippocam-
pal lesioning show the usual clock-shift effects, indicating
their successful use of the sun compass. However, the hip-
pocampus is necessary for learning to use this sun-compass
information.
Landmarks they have learned about at a familiar release site
can be used for homing in two distinct ways: “pilotage” in
which they use the perceived layout of the landmarks to head
toward the loft or “site-specific compass orientation” in which
they use the landmarks to recall the sun compass direction in
which to fly. Whereas pilotage requires quite a sophisticated
spatial representation, site-specific compass orientation does
not. Although pigeons with lesions of the hippocampal for-
mation can learn to use familiar landmarks to navigate, it is
possible that they do it in only one of these two ways.
Gagliardo et al. (1999) studied experienced pigeons rendered
anosmic by means of nasal zinc sulfate (a procedure that pre-
vented the use of their navigational map) to compare the man-
ner in which control and lesioned birds used familiar
landmarks. As shown in Figure 13–27A, the birds were
released on a number of occasions from La Costanza, north of
Pisa, and from Livorno, situated southeast of Pisa. Prior to the
critical test days, the birds were clock-shifted in their lofts by 6
hours. This should have no effect on the vanishing bearings
taken from a release site if the birds use a representation of the
spatial layout of the landmarks to find their way home.
However, if they use site of the landmarks around each now
familiar release site to work out a site-specific compass orien-
tation, a 6-hour clock shift should result in an approximately
120 error in the initial flight direction. The results of test
releases from both sites indicated a striking difference between
the controls and the birds with hippocampal lesions (Fig.
13–27B). The controls headed in the correct direction, with
only a small deviation of 34 in the direction expected given
the clock shift; the lesioned birds showed a mean bearing error
of 154 in the counterclockwise direction. Thus, the controls
were using pilotage, whereas the lesioned birds were not. There
is no conflict here with the earlier findings demonstrating that
learning to use the sun compass requires the hippocampus
because these were experienced homing pigeons that had
Figure 13–27. Avian hippocampus and landmark learning. A Home and two release sites
A. The home loft of the pigeons was near Pisa. The animals
were extensively trained to fly home (to Pisa) from either
of two release sites, La Costanza and Livorno. B. When ren-
dered anosmic and clock-shifted by 6 hours, the controls
headed in the correct direction whereas the hippocampus-
lesioned birds showed greater variability around a mean,
reflecting the expected shift of 120. The reason a clock
shift of 6 hours is expected to produce a deviation of 120
rather than 90 is because at around midday in summer
the azimuth of the sun moves about 20 per hour. (Source:
After Gagliardo et al., 1999.)
Theories of Hippocampal Function 643
learned about the sun compass long before attaining hip-
pocampal lesions. Given the multiplicity of ways that pigeons
can navigate, these data represent an unambiguous demon-
stration of the critical role of the hippocampal formation in
landmark learning in a naturalistic setting.
Although ostensibly supporting a close link between spa-
tial memory and hippocampal function, one must recognize
that the cytoarchitectonic appearance and intrinsic circuitry
of the avian hippocampus is very different from that of the
typical mammalian hippocampus described in Chapter 3.
There are both similarities and differences between the avian
and mammalian brain (Jarvis et al., 2005), but both Lee et al.
(1998) and Macphail (2002) make the case—largely on
embryological grounds—for the avian hippocampal forma-
tion being considered homologous to the hippocampus and
dentate gyrus of mammals. Certain arguments seem uncon-
vincing or circumstantial, such as the presence of synaptic
plasticity (this also occurs outside the hippocampus) and the
claimed similarity in neurotransmitters used (but similar neu-
rotransmitters are used all over the brain in numerous orders
of animals). Electrophysiological studies are also equivocal.
For example, Bingman et al. (2003) reported some evidence of
“space-specific” firing of hippocampal cells in homing
pigeons trained in a small plus maze, but the correlate was
transitory and less clearly tied to space than in recordings
from rats. The possibility remains that greater spatial speci-
ficity would be observed in a task in which the pigeons were
allowed to engage in homing, but this is not yet technically
feasible. One prominent avian behavioral biologist, writing
about the pigeon brain, noted that: “the pigeon and rat hip-
pocampal formation reside in different forebrain environ-
ments characterized by a wulst and neocortex, respectively.
Differences in the forebrain organization of pigeons and rats,
and of birds and mammals in general, must be considered in
making sense of the possible species differences in how the
hippocampal formation participates in the representation of
space” (Bingman et al., 2003, p. 117). Clearly, there are differ-
ences of opinion on the similarity of the avian and mam-
malian brain, and it is tempting to wonder if some of those
who work on birds are making a virtue of necessity in sup-
B Vanishing bearings of anosmic
birds released from Livorno
La Costanza
16.7 km anosmic controls
birds
Home (Pisa)
HPC anosmic
13.6 km lesioned birds
Livorno
644 The Hippocampus Book
porting a claim for homology, aware of wider interest in the
brains of mammals. Although understandable, the issue has to
be resolved on the basis of all the evidence.
Comparative Studies of Spatial
Memory in Rodents
Hippocampal-oriented neuroethology studies have been con-
ducted in mammals as well, but the research is less well devel-
oped. One idea about mammalian hippocampal evolution is
that sexual selection may have led to a relative increase in hip-
pocampal volume of certain polygymous male rodents, an
important case because sex differences evolve slowly and thus
any correlated differences in cognitive function and neu-
roanatomy are especially strong evidence of an adaptive mod-
ification of the brain. In support of this idea, Jacobs et al.
(1990) found that polygymous male meadow voles, which
have an area over which they range about five times larger
than that of females, show a corresponding sex difference in
hippocampal volume (Fig. 13–28). In contrast, in monoga-
mous prairie and pine voles, male and female hippocampal
size is equivalent. Earlier laboratory studies revealed that
males of the polygymous meadow vole are superior to females
in place learning in Tolman’s sunburst and other mazes
(Gaulin and Fitzgerald, 1989; Gaulin et al., 1990). The inter-
action between natural and sexual selection has also been
revealed in studies of subspecies of kangaroo rats, some of
which store food and some of which do not, some polygy-
mous and some monogamous. In comparisons across species,
hippocampal size was found to be largest in males and scatter
hoarding in polygymous Merriam’s kangaroo rats (Sherry et
al., 1992; Jacobs and Spencer, 1994).
One proposed modification of the cognitive map theory,
inspired by the allometric studies, pigeon homing work and
by the discovery of head-direction cells, is the parallel-map
theory of Jacobs and Schenk (2003). This explicitly identifies
a “bearing map” (in which olfactory gradients or distal cues
Figure 13–28. Sexual selection influences hippocampal size. A.
Space use by mammalian species with different mating systems.
Under polygamy, male home ranges encompass several smaller
female ranges; under monogamy, males and females share a joint
home territory. (Source: After Jacobs, 1995.) B. Relative hippocam-
A Male and female range size
polygamy: meadow vole monogamy: pine vole
are used to establish a sense of direction) and a “sketch map”
(which is an allocentric representation of local landmarks).
The two mapping systems are held to be mediated by and
to interact at different subregions of the hippocampal forma-
tion; the primary evidence for this is the consistent patterns
of distinctive search strategies seen in spatial tasks after
selective lesions or other pharmacological forms of interven-
tion. It is too early to assess the status of this proposed modi-
fication of the original cognitive map theory, but Jacobs
(2003) made the case that it is pertinent to issues of evolution
and ontogeny across a range of species, including other
vertebrates. Jacobs and Schenk also used it to offer novel
explanations of certain hitherto puzzling lesion and pharma-
cologically induced dissociations in task performance by lab-
oratory rats.
The relative volume of a brain area is a limited measure. It
is important to know whether this reflects variation in cell
number (and, if so, of what cell type), cell proliferation or cell
death, dendritic arborization, synapse number or density, or
other neuronal processes. Finer grain neuroanatomical or bio-
chemical measures could offer insights into precisely what co-
varies with the type of information stored—capacity or
temporal persistence. Given that neurogenesis may be under
hormonal control (Galea and McEwen, 1999), there has been
considerable interest in the possibility that seasonal changes in
hippocampal volume reflect neurogenesis (Barnea and
Nottebohm, 1994). Neurogenesis is influenced by environ-
mental enrichment in some strains of mice (Kempermannet
al., 1998) and by voluntary activity in running wheels (van
Praag et al., 1999). There is evidence for seasonal changes in
spatial learning ability and fine-grained aspects of hippocam-
pal anatomy (Jacobs, 1995). However, Jacobs’ own studies of
possible seasonal variation in wild caught squirrels revealed
no difference in cell proliferation or total neuron number in
the hippocampus that co-varied with seasonal variation in
demands on spatial memory (Lavenex et al., 2000). Chapter 9
discusses neurogenesis in the dentate gyrus in more detail.
pal volume in polygamous meadow voles and monogamous pine
voles. The argument is that the hippocampus needs to be larger in a
male meadow vole, which has to patrol a much larger territory.
(Source: After Jacobs et al., 1990.)
B Relative hippocampal volume
0.06 meadow vole
pine vole
hippocampal volume
brain volume
0.05
0.04
 Theories of Hippocampal Function 645

Neuroanatomical studies have nonetheless revealed consis-
tent genetically stable variations in the intrinsic circuitry
of the hippocampal formation across strains of mice (Lipp et
al., 1999). Schwegler and Lipp (1981) discovered that the
length of the infrapyramidal mossy fibers (IIP-MFs) is
negatively correlated with two-way avoidance learning (a task
impaired by hippocampal lesions) (Fig.13–29A,B), a finding
that holds across a number of mouse (and rat) strains
(Schwegler and Lipp, 1983). Later work showed that it is
positively correlated with performance in spatial tasks, includ-
ing the radial maze and reversal learning in the water-
maze (Crusio et al., 1987; Bernasconi-Guastalla et al., 1994).
Mouse strains, such as C57/BL6, known to be good at spatial
learning have an extensive infrapyramidal mossy fiber system.
Other strains, such as DBAs and BALB/C, have a limited
infrapyramidal mossy fiber system. However, such correla-
tions, although striking, should be treated cautiously not
least because there may be other covariates. For example,
BALB/C mice have retinal problems and limited vision;
and DBA mice are reported to have deficiencies in pro-
tein kinase C (PKC), an enzyme implicated in synaptic plas-
ticity.
To try to unravel the reported correlations and take a step
toward causation, postnatal injections of thyroxine (Fig.
13–29C) have been successfully used to induce variations in
infrapyramidal mossy fibers within a single strain and corre-
sponding variations in the learning rate (Lipp et al., 1988;
Schwegler et al., 1991). Still, the anatomical disquiet remains:
These injections may be having other unmeasured effects.
Some measure of behavioral specificity has been realized by
the finding that the correlations with infrapyramidal mossy
fiber length holds for spatial working memory on a radial
maze but not cue working memory (Crusio et al., 1987;
Schwegler et al., 1990). However, later studies revealed strain-
specific covariations between the extent of the infrapyramidal
mossy fiber distribution and both intermale aggression
(Guillot et al., 1994) and paw preference (Lipp et al., 1996),
which are clearly unrelated to spatial learning and memory;
and there is a report of failure to observe the correlation with
radial maze performance (Roullet and Lassalle, 1992).
Lipp et al. (1999) recognized that these “noncognitive”
findings may not fit well with some prevailing theories of hip-
pocampal function but are relevant to a “multifunctional”
view in which the dorsal hippocampus mediates spatial

Figure 13–29. Avoidance learning and infrapyramidal mossy fibers. A. Relative location of the
supra- and infrapyramidal mossy fibers (IIP-MF) projecting from the dentate gyrus (DG) to
area CA3. B. Correlation between two-way avoidance learning and IIP-MF extent across mouse
strains. C. Thyroxine induces alterations in both two-way avoidance learning and extent of IIP-
MF. (Source: After Lipp et al., 1999.)

A Mossy fibre pathway (black)

showing supra- and infra- CA1
pyramidal parts of the
pathway. PYR LM
FIM
RAD
SP GC
OR
CA4
CA3

SGL
MOL
Infrapyramidal mossy fibres (IIP-MF)

B Strain Comparisons C DBA/2 Thyroxine Postnatal

80 100
DBA/2 90
70
80
% avoidance

60 70
BALB/c
60
50
50
40 40
NMRI C57BL/6
30 30
SM/J
20 FEMALE
20
MEAN
± ± SEM ICR C3H 10 MALE
10 0
0 1 2 3 4 5 .5 1 1.5 2 2.5 3 3.5
% IIP-MF in CA3/CA4 % IIP-MF in CA3/CA4
646 The Hippocampus Book
memory functions whereas the more ventral region is more
relevant to emotional and aggressive behavior. They also sug-
gested, partly on the basis of studies of the survival of various
mouse strains through successive harsh winters at a Russian
field station that “behavioral reactivity” rather than spatial
memory may be the parameter of greatest adaptive signifi-
cance in relation to the infrapyramidal mossy fibers.
To conclude this subsection, allometric brain/behavior
correlations in birds and mammals have revealed several find-
ings consistent with spatial interpretation of hippocampal
function, but they have not definitively established that it is
only spatial, rather than other kinds of memory, with which
changes in hippocampal volume or structure are correlated.
More analytical studies are required to unravel precisely what
features of memory correlate with changes in hippocampal
volume or circuitry. Identifying evolutionary adaptations,
attributing them definitively to particular selective pressures,
and then relating them to specific mosaic variations in brain
size or circuitry is a slow, painstaking task (Lipp et al., 1999).
The criticism has been made that studies of evolutionary
function not only do not, but cannot, provide information
about neurobiological mechanisms (Bolhuis and Macphail,
2001; Macphail and Bolhuis, 2001). However, even if this log-
ical point were to be accepted, the implications are hotly
debated (Hampton et al., 2002). Shettleworth (2003) offered a
thoughtful defense of studies of the “neuroecology” of learn-
ing. She highlighted, by way of example, striking data from
two populations of chickadees by Pravosudov and Clayton
(2002). Instead of comparing different strains or species, they
tested the “adaptive specialization” hypothesis in a single
species, black-capped chickadees. One population was living
in Alaska, where winters are fierce, and the other in Colorado.
The different latitudes at which these distinct populations of a
common species live might have resulted in a quantitative dif-
ference in the degree of selection pressure. Birds were caught
in the autumn, around the time that seasonal volume changes
should be maximal, and brought to a laboratory in California.
Once there, the Alaska population was observed to be better at
food storing and retrieval than the Colorado birds, better on
spatial but not color memory (an allometric correlation anal-
ogous to earlier lesion data (Hampton and Shettleworth,
1996), and had relatively larger hippocampal volumes. It is not
unreasonable to conclude that “mosaic” evolution of brain
structure driven by natural selection on particular behavioral
capacities (Barton and Harvey, 2000) will continue to con-
tribute to understanding qualitative and quantitative differ-
ences in the hippocampus within and across species.
Maps, Spatial Representations,
and Geometry: Is Spatial Learning Qualitatively
Different from Associative Learning?
The other strand of the adaptive specialization idea is the issue
of whether hippocampus-dependent spatial learning is quali-
tatively distinct from associative learning. The cognitive map
theory asserts that spatial learning is different from “taxon”
learning in being rapid, flexible, and uniquely encoding infor-
mation into maps. As noted above, the animal learning con-
stituency is suspicious of this idea.
The claim that spatial learning is qualitatively different
would be stronger if the psychological properties of either
type of learning could be shown to be specific to that type of
learning. A particular sticking point has been whether certain
phenomena known to occur in classical conditioning, such as
“blocking” and “overshadowing,” do or do not occur in the
spatial domain. Blocking refers to the ability of one stimulus
to block learning about another, and overshadowing is a
differential sharing of associative strength when two stimuli
are conditioned together. Modern animal learning theories,
discussed in more detail in Section 13.5, rely on “error-cor-
recting” learning rules in which changes in the associative
strength of a stimulus occur only when the expected outcome
of that stimulus differs from the actual outcome. Although
such rules enable learning to be rapid and flexible, stimuli
compete with each other for associative strength rather than
become linked together in anything like a “map.” The condi-
tioning phenomenon that most directly established the idea of
error-correcting learning and captured everyone’s attention
when it was first discovered is blocking (Kamin, 1968). A stim-
ulus, call it A, is arranged to predict reinforcement (R*). After
several pairings of A and R* have occurred and the associative
strength of A has reached an asymptotic value, a second stim-
ulus is added (B). Stimulus B now also predicts R*; and one
might expect, given the contiguous pairing of B and R*, that
this association would be readily learned also. Often, however,
little appears to be encoded about the relation between B and
R*. The prior learning of the A → R* association is said to
block learning about B. An important control is one in which
stimuli A and B are introduced together for the first time in
the second phase of a blocking experiment and arranged to
predict R*. Under these conditions, both the A → R* and the
B → R* associations are learned (although overshadowing
sometimes occurs and is determined by such factors as stimu-
lus salience or proximity). Many other control procedures
have been conducted among a vast array of experiments
investigating blocking and the processes responsible for it
(Mackintosh, 1983).
One reason hippocampus-dependent learning might be
thought to be qualitatively different from conventional asso-
ciative conditioning is because hippocampal lesions do not
reliably affect blocking. An early conditioning study indicated
that there may be an effect of partial hippocampal lesions on
blocking (Ross et al., 1984), but at later more definitive exper-
iments indicated that blocking can occur normally in animals
with complete hippocampal lesions (Garrud et al., 1984;
Holland and Fox, 2003). Of greater concern for the cognitive
map theory is whether blocking occurs between locale and
taxon learning or even within the spatial domain. A compre-
hensive program of work led by Chamizo has shown that
blocking can occur in the radial maze between intra- and
extramaze cues (Chamizo et al., 1985) and entirely within the
spatial domain in the watermaze (Rodrigo et al., 1997). In the

later experiment, analogous to the studies using classical con-
ditioning, the pool is surrounded by heavy black curtains and
two-dimensional cues hung at specific locations within it (A,
B, C, and so on). This spatial array of cues can then be rotated
collectively from trial to trial; in this way, the experimenter
can be confident that a rat’s directed navigation in the pool is
determined by this array rather than by other, experimenter-
uncontrolled cues. In addition, although swimming is permit-
ted at certain phases of the experiments (to enable learning
and later testing of what has been learned), the animals can
be either allowed to swim or placed on the platform during
the critical blocking phase when another cue (X) is added.
Blocking occurs normally. Chamizo (2003) summarized a
wide range of such experiments and concluded that there is
as yet no basis to suppose that spatial learning differs from
conventional conditioning with respect to the use of an error-
correcting learning rule. Mackintosh (2002) recognized cer-
tain tensions in the associative account between the
occurrence of blocking, as required by animal learning theory,
and other evidence that rats often learn about the locations of
all available landmarks in simple exploratory tasks, as we saw
in the work of Save et al. (1992a,b) earlier. However, he shared
Chamizo’s skepticism and asserted that “the behavioral evi-
dence does not seem to have supported O’Keefe and Nadel’s
original hypothesis that true spatial learning or locale learning
is quite distinct from associative learning” (Mackintosh, 2002,
p. 165).
There remain some difficulties in accepting these conclu-
sions without qualification. First, no lesion studies of blocking
in the spatial domain have been conducted, and we must take
on trust that the spatial memory seen in these studies is
hippocampus-dependent. This seems likely, but we have
already noted instances in which multitrial spatial reference
memory tasks in the watermaze can be learned by rats with
hippocampal lesions. The definitive hippocampus-dependent
task in the watermaze is now recognized as being the DMP
paradigm, and one-trial blocking experiments have not been
conducted. Second, placement on the hidden platform may
be insufficient for much learning with the kind of hanging
cues used in Chamizo’s laboratory. This is because two-
dimensional cues at the periphery are known to be much less
effective than three-dimensional cues, the animals cannot
explore them proximally (e.g., using vibrissa movements), and
the constellation of cues in the “blocking” phase of training
(A, B, X) may be sufficiently similar to those in the earlier
training phase (A,B) that the animals do not notice the differ-
ence. The outcome is then apparent rather than true blocking.
Although various controls argue against this, the data meas-
ures reported are sometimes the proportion or ranking of ani-
mals showing an effect rather than absolute tracking data
(such as time spent in a quadrant or zone).
Partly for these reasons, Biegler and Morris (1999) devel-
oped a food reward task in a large arena in which to explore
blocking under conditions in which (1) the local cues were
three-dimensional objects that could be explored by the ani-
mals in a multisensory manner, and (2) there was explicit
Theories of Hippocampal Function 647
directional polarization provided by a specific arrangement of
distal extramaze curtains. The task involved learning the loca-
tion of food in relation to two identical landmarks, with a
third and later fourth cue added to disambiguate two possible
places where food might be found. The paths taken by the ani-
mals were tracked automatically. Blocking was again seen in
the spatial domain, and this occurred despite the animals
noticing and exploring the added landmark during the block-
ing phase (as shown by the tracking data) but then choosing
not to learn about its association with reward. The use of an
“instrumental” rather than a classic conditioning paradigm
introduced complications, as the conditions across groups
could not be exactly matched on the first trial of the blocking
phase, though Biegler and Morris (1999) argued that this did
not invalidate their conclusion. Unpublished lesion studies
were also conducted, but performance was too poor in ani-
mals given lesions before training for any useful information
to be obtained about the role of the hippocampus in blocking
using this paradigm.
One point noted by Biegler and Morris (1999) concerns a
potentially interesting difference between blocking in the
spatial and conditional domains. With conventional condi-
tioning, the relations learned, such as A → R*, are condi-
tional (‘if–then’) associations (and not necessarily temporal
although often described as such). Relations must be more
than that in spatial learning because the Landmark → R*
association must provide some information about the dis-
tance and direction from a landmark (L) that food is to be
found. Under circumstances in which a memory representa-
tion must be formed of the nature of the relation between two
cues, blocking may obey different rules. For example, when
the relation between two spatially dispersed landmarks L1 and
L2 and food has been learned, the addition of a third land-
mark, L3, may not always result in blocking simply because
the location of R* is already predicted. The relative spatial
location of the new cue L3 to the old cues L1 and L2 may be
critical to its incorporation into an associative or “map-like”
representation. Blocking may be more or less likely as a func-
tion of whether the vector (argument and direction) from L3
→R* is similar to or different from the vectors from L1 and L2
to R*. The conclusion of Chamizo (2002, 2003) that spatial
information interacts during learning in apparently the same
way as it does during conventional conditioning is probably
secure, but there are issues to do with vectors and geometry
that deserve further investigation.
Swimming against the tide of this developing experimental
skepticism in the animal learning constituency, Cheng and
Gallistel have gone farther than O’Keefe and Nadel (1978)
in suggesting that the brain possesses a “geometric module”
for space that operates according to distinctive learning rules
(Gallistel, 1980; Cheng, 1986; Margules and Gallistel, 1988).
This followed observations of the behavior of rats searching
for food hidden at one corner of a rectangular enclosure.
Despite the presence of disambiguating cues located at
the four corners (visual or olfactory), the rats ignored them
and searched systematically at both the correct corner and
648 The Hippocampus Book
the diagonally opposite corner. These two corners are geomet-
rically equivalent in a symmetrical rectangle. These findings
seemed to imply that rats were more sensitive to environ-
mental shape than to local features, such as the distinc-
tive cues—hence the proposal of a geometric module.
Observations and models of place cells—specifically
their determination by boundary vectors—are consistent
with both this and O’Keefe’s preferred local feature view
(O’Keefe and Burgess, 1996; Hartley et al., 2000; Lever et al.,
2002).
In an ingenious test of the geometric module idea, Pearce
et al. (2004) wondered if the symmetrical search patterns seen
in a rectangular array reflected the operation of a specialized
geometric module or something simpler. The animal learning
constituency enjoys wielding Occam’s razor, and this issue
represented another opportunity. Might the rats have instead
learned only the “rule” of searching at a corner defined by one
long wall on the right and one short wall on the left? To test
this, they first trained rats to search for a hidden escape plat-
form in a rectangular watermaze and observed, in a conven-
tional probe test with the platform absent, focused searching
at both the “correct” corner and the geometrically equivalent
corner (Fig. 13–30A). This result is identical to that found by
Cheng and Gallistel in the food-rewarded task. The animals
were then transferred to training in a kite-shaped maze (Fig.
Figure 13–30. Geometry and the shape of spatial learning to come.
A. Rats trained to find an escape platform in one corner of a rectan-
gular watermaze also search in the geometrically equivalent corner.
B. The rats were then transferred to swimming in a “kite-shaped”
arena, with the platform again located at the corner with a long wall
on the left and a short wall on the right. C. During posttraining
probe tests, the rats now searched at the correct location and at the
apex. D. Hippocampus-lesioned rats (gray shading) are poorer than
A Hidden platform in basic rectangle
A B
geometrically incorrect
correct
incorrect correct
D C
C Searching at left-hand
end of a long wall.
Performance
ct
e
ct
ex
us
re
rre
Ap
or
bt
co
O
C
In
13–30B). Training continued to the one corner with a long
wall on the right and short wall on the left. In posttraining
probe tests, preferential searching was now observed at the
correct location but also at the apex of the maze (Fig.
13–30C). This finding makes no sense if the animals are using
a geometric module representing the shape of the kite but is
readily explicable if the animals have learned to search prefer-
entially at the left-hand end of any long wall. Interestingly, rats
with hippocampal lesions were poorer at learning to search at
the correct corner unless prominent disambiguating local
cues were added to walls on either side of the apex during
training (Fig. 13–30D,E). Once these local cues were added,
the training corner and the apex were easily distinguished
visually and hippocampus-lesioned animals then learned as
well as sham-lesioned controls. The implication of this ingen-
ious study is that there need be no recourse to an abstract geo-
metric module to learn these kinds of tasks and no need to
suppose that the hippocampus mediates such a module. The
data are, however, consistent with the boundary vector model
of place cells in which their location is fixed by intersecting
vectors from local cues.
Yet more abstract geometry is implied by the observation
that rats trained to forage for food in the center of a square
arena can transfer to searching at the “geometric” centre of
rectangular and triangular arenas (Tommasi and Thinus-
sham-lesioned controls when learning to search correctly in the kite
maze. E. When disambiguating local cues were added to the long
walls of the rectangle and on either side of the apex, hippocampus-
lesioned animals could learn as rapidly as controls. The bar graph
data have been simplified to the long walls of the rectangle and
either side aid the description of this study. (Source: After Pearce et
al., 2004.)
B Transfer to kite-shaped arena
G
incorrect
F apex obtuse H
locally
correct
E
D HPC lesion deficit E Successful local cue learning
Performance
Performance
Chance
Sham HPC Sham HPC

Blanc, 2004). This result is not, however, always observed. One
classic series experiments contrasted a map-like representa-
tion of space to one based on vectors (Collett et al., 1986).
Vectors encode distance and direction. In a directionally
polarized environment, a vector representation would be one
in which object A may be represented as a vector with a spe-
cific “argument” (i.e., length) and “angle” from landmark P.
Many objects (A, B, C, and so on) both visible and hidden,
may be represented in such a way, but it does not require there
to be any “map” of space. Specifically, there may be no encod-
ing of the relative locations of A, B, and C to each other. To test
this idea, gerbils were trained to find sunflower seeds at par-
ticular distances and directions from landmarks in a room
that provided cues enabling directional polarization but not
localization. The landmarks were then manipulated in various
ways. In one manipulation, animals were first trained to find
food hidden halfway between two identical landmarks. The
two landmarks were then moved farther apart. If the animals
had learned to find reward at some abstract “centre” of the
array, they would be expected to search preferentially at a
common point, now a bit farther from each of the landmarks.
Instead, searching occurred at two locations, each at the
appropriate distance from each landmark along the line
between them (Fig. 13–31A). In a yet more striking example
of vector-based representation, the gerbils were trained to find
sunflower seeds at the center of a triangular array of three dis-
tinctive landmarks (Fig. 13–31B). The landmark array was
then rotated by 60 such that the vectors representing reward
location from each landmark were now outside the array
Figure 13–31. Maps or vectors? A. Gerbils were trained to find
food halfway between two landmarks (large circles). When the land-
marks were moved farther apart in a posttraining probe test, the
animals searched at two distinct locations (small squares denote
time spent searching). B. Gerbils were trained to find food in the
A Searching for food between B Searching for food at the centre of 3 distinctive
2 landmarks landmarks
training non-reward
probe tests
Food
Theories of Hippocampal Function 649
rather than inside. Astonishingly, a significant proportion of
the search visits during a posttraining probe test with the
rotated array were outside the landmark array. This striking
pattern is not to be expected had the animals learned a “map”
of space with food at the center of the landmark array. It is
consistent with the encoding of three independent vectors
specifying the distance and direction of food from each of the
three landmarks. During training, these vectors meet at the
single, central point; after rotation they do not. Collett et al.
(1986) did not go on to conduct lesion studies, so we do not
know the impact of the hippocampal dysfunction. However,
in a conceptually similar lesion study using the watermaze,
Pearce et al. (1998) found that hippocampus-lesioned rats
could be trained to locate a hidden platform at a specific dis-
tance and direction from a visible landmark that moved places
within the pool between days. It seems that the hippocampus
is not required to encode vectors across multiple trials but
may be necessary to assemble multiple vectors into a map-like
representation.
With respect to multiple types of memory, there are good
grounds to believe that there has been selection pressure
to develop map-like spatial representation and navigation sys-
tems in vertebrates. However, the studies comparing spatial
learning with conventional associative conditioning have
revealed fewer differences than might have been expected
on the basis of the original cognitive map theory. The possi-
bility that spatial information is fundamentally associative,
in contrast to the position adopted within cognitive mapping
theory, has to be considered. Section 13.5 takes up this
center of an array of three distinctive landmarks in an environment
in which directionally polarization is determined by other distal
cues. When the landmark array was rotated by 60 in a probe test, a
substantial proportion of the searching was now outside the arena.
(Source: After Collett et al., 1986.)
training non-reward
probe tests
Food
Directionally polarised
environment
Landmark
Locations of searching in
non-reward probe tests
650 The Hippocampus Book
issue further; but before turning to that, we discuss the
issue of where allocentric spatial information is stored in the
brain.
13.4.5 Storage and Consolidation
of Spatial Memory
A fourth major proposition of cognitive map theory is that the
memory traces required for allocentric representations
through a familiar place are stored in the hippocampal forma-
tion itself. This storage is held to be distributed, possibly along
both the longitudinal and transverse axes of at least the dorsal
hippocampus for even a single memory, such that damage to
the structure should cause a loss of both processing capacity
(encoding, retrieval, and navigation) and storage sites (local
alterations in synaptic weights). Hippocampal damage should
therefore cause a “flat” gradient of retrograde amnesia for spa-
tial information, as first shown by Bolhuis et al. (1994). In
contrast to the retrieval of factual information by humans or
contextual fear by rats, where gradients are observed (see
Section 13.3), the integrity of structures in the hippocampal
formation is also claimed to be essential for successful recall of
episodic memory in humans (Viskontas et al., 2000; Cipolotti
et al., 2001; Maguire, 2001). Moreover, very remote episodic
memories may still elicit neural activation of the hippocam-
pus long after they were first acquired (Nadel et al., 2000),
although the use of “reminders” prior to brain scanning has
raised a query about the true of age of ostensibly old memo-
ries in imaging work. Given the very different perspective on
these matters in declarative memory theory, it is no surprise
that memory consolidation has become one of the battlefields
of contemporary cognitive neuroscience.
We have seen that the “standard model” of memory per-
sistence (see Section 13.3) is that some systems-level process
of hippocampus-dependent memory consolidation takes
place after learning that, by enabling an interaction of hip-
pocampal and neocortical ensembles, secures the stabilization
of memory traces outside the hippocampus (Squire and
Alvarez, 1995a; Kapur and Brooks, 1999; Manns et al., 2003).
The neurological function of such systems-level consolidation
presumably has nothing to do with the protection of memory
against brain damage. Exploiting the unfortunate occurrence
of relevant brain damage in humans, or its experimental cre-
ation in animals, is just a technique for revealing this neuro-
logical process. Rather, consolidation is thought to have two
functions. First, it engages mechanisms that “gate” whether
memory traces are to be retained. Second, once consolidation
is complete, a lasting neocortical memory site exists that
would enable much faster retrieval of information than would
be possible were it always necessary for there to be a neural
“loop: through the hippocampus. A consolidation process
might also allow retrieval of information from neocortical
traces when it is required without recourse to consciousness,
but the activation of this neural loop may be a signature of
conscious awareness in the specific domain of memory
(Moscovitch, 1995). One situation where conscious awareness
is usually engaged is in the spatial domain, where it may be
helpful in updating during navigation.
Reflecting on these putative functions of systems-level con-
solidation, Moscovitch wondered if memory retrieval with
recovered consciousness might be disentangled from other
features of the standard model. This led on to a series of arti-
cles in which he and Nadel proposed an alternative “multiple
trace” model of memory persistence (Nadel and Moscovitch,
1997, 1998; Moscovitch and Nadel, 1998). In this model, a
subset of memories is permanently mediated by corticohip-
pocampal circuitry. Specifically, the theory holds that each
reactivation of memory leads to—or at least can lead to—the
formation of additional traces that are located at different
points along the longitudinal axis of the hippocampus
(anterior-posterior in humans). Temporal information is
thought to be stored in the prefrontal lobe, at least in humans.
Consolidation is thus a “reactivation–restorage” cycle that
occurs in a punctuated manner over time, rather than as a
gradual, inexorable biological process—an analogy might be
drawn with the concept of “punctuated equilibrium” in evolu-
tionary theory (Gould and Eldredge, 1977). There would then
be multiple traces of a common past event, but these traces
could entail different associations with other information that
reflect their distinct provenance. Temporally graded retro-
grade amnesia for spatial information can then be explained
by the proliferation of memory traces in the hippocampus
without recourse to a neocortical consolidation process, a pre-
diction that has been successfully modeled with a connection-
ist network (Nadel et al., 2000). Remote spatial memories are
represented by a larger number of traces and so be more
accessible than recently acquired memories. According to a
later version of this model (Nadel and Bohbot, 2001;
Rosenbaum et al., 2001), this multiple-trace theory applies
particularly to “contextually rich” memories but not those that
are context-free or semantic in character. Whether these pro-
posals should be seen as “extensions” of the cognitive map the-
ory or as independent ideas inspired by it is unclear. They are
probably more the latter.
The declarative memory and the multiple trace theories
make contrasting predictions about retrograde amnesia in the
particular case of partial lesions. Unfortunately for theories of
memory, the capricious nature of human brain injury results
in most MTL amnesic subjects having only partial damage to
the hippocampus (Rempel-Clower et al., 1996; Bayley et al.,
2005b). Such patients do not discriminate the two theories.
One neurological case does exist where the damage is appar-
ently extensive yet there remains clear evidence of sparing of
spatial memory from a much earlier time period in the
patient’s life—patient E.P. (Teng and Squire, 1999). However,
we have seen that the testing of remote memory in humans
and the identification of an explicit temporal gradient is
fraught with difficulty (see Chapter 12). Can animal studies of
spatial memory shed light on this important theoretical issue?
Studies using the watermaze have consistently shown a flat
gradient of retrograde amnesia for spatial memory when it is
first tested postoperatively. In such studies, rats have typically
 Theories of Hippocampal Function 651

first been trained on a reference memory version of the task
(i.e., to find a single hidden platform); then, at periods rang-
ing from a few days to up to 3 months later, they have been
subjected to large excitotoxic or radiofrequency lesioning of
the hippocampal formation (Bolhuis et al., 1994; Mumby et
al., 1999; Sutherland et al., 2001; Clark et al., 2004, 2005b).
Subsequent to surgery, the animals have been retested in the
watermaze either with the platform in its original training
location, with it moved to a new location, or using probe tests
(platform absent). Martin et al. (2005) used such a design
with animals given partial hippocampal lesions (circa 35%
sparing of either the dorsal or the ventral hippocampus)
and others complete lesions ( 5% sparing). They used an
Atlantis platform during training to encourage highly focused
searching and better memory. An initial posttraining probe
test provided no evidence of better retention of remote spatial
memory, even in the group with a partial lesion of either the
dorsal or ventral hippocampus (Fig. 13–32A). This first find-
ing of Martin et al.’s (2005) study is inconsistent with both the
declarative memory and the multiple-trace theories but con-
sistent with the original statement of the cognitive map the-
ory, subject to the proviso of there being some threshold of
intact residual tissue ( 35%) necessary to mediate effective
navigation.
However, in these and other studies, the performance of
sham-lesioned rats in the watermaze tends to be poor after
surgery, possibly reflecting a “floor” effect that masks any con-
solidation or multiple trace process. To address this problem,
Bolhuis et al. (1994) followed up the initial probe test with
retraining to unmask cryptic spatial memory traces. If any-
thing, their “savings” data favored better relearning by recently
lesioned groups (made 3 days after original training) than the
remote groups. Rats given subiculum lesions that left the hip-
pocampus intact were also capable of relearning to a high

Figure 13–32. Retrograde amnesia for spatial memory. A. An initial
“pre-reminder” probe test conducted 2 weeks after sham (white),
and either partial (gray) or complete (black) hippocampal lesions
revealed no evidence of spared spatial memory recall in either
lesioned group. This study confirms the earlier findings of Bolhuis
et al. (1994) but under conditions in which a floor effect in the
sham control group did not occur. (Source: After Martin et al.,
2005.) B. Reminding given in the form of successive probe tests
revealed some sparing of spatial memory recall in the partially
lesioned group. The reminding procedure was shown not to induce
relearning. (Source: After Martin et al., 2005.) C. Extensive training
from an early age failed to reveal spared spatial memory in animals
with complete hippocampal lesions. (Source: After Clark et al.,
2005a.) D. Spatial recognition memory tested using an annular
watermaze that obviates the need for hippocampal updating during
navigation also failed to reveal spared spatial memory after com-
plete hippocampal lesions. (Source: After Clark et al., 2005b.)

A Spatial recall before reminding B Spatial recall with reminding

first probe test acts as a reminder post-reminder probe test

100 100
relative memory

relative memory
sparing (%)

sparing (%)

60 60
20 20
-20 -20
recent remote recent remote

C Spatial recall with extensive training

probe test 60
correct quadrant
% time in the

Age (days) 40
21 90
20 chance Lesion groups
NC and sham
0 H (partial)
v.remote H (complete)
D Spatial recognition in annular watermaze
probe test 30 Training-surgery intervals
% time in the
correct zone

recent: 0-3 days

20 remote: 6-9 weeks
v.remote: 14-16 weeks
10

0
remote v.remote
652 The Hippocampus Book
degree of proficiency. This is consistent with the integrity of
hippocampal circuitry being important for some aspects of
spatial memory storage. Mumby et al. (1999) also used
retraining and a within-subjects protocol with two watermaze
tasks that were trained at different time points prior to sur-
gery. They found little evidence of spatial memory in their
complete hippocampal lesion group for either of the two
tasks. Sutherland et al. (2001) used both retraining and probe
tests and observed a significant trend for longer training–
surgery intervals to be associated with poorer retention—once
again the opposite of what is predicted by the declarative
memory theory. These studies reflect a consistent pattern, but
the use of retraining after surgery confounded two factors: the
impact of the lesion on consolidation (the focus of interest)
and its impact on new learning (a separate matter).
In their comparison of partial and complete lesions, Martin
et al. (2005) used a different approach: a series of rewarded
probe trials at 1-hour intervals. This gradually revealed above-
chance performance by the partial hippocampus-lesioned
group that had undergone surgery immediately after training
(i.e., their recent group). No recovery of remote memory was
observed in the group for which there was a much longer 6-
week interval between the end of training and surgery (Fig.
13–32B). This apparent “unmasking” of latent spatial memory
was not due to any relearning during the series of probe trials,
as de Hoz et al. (2004) had separately shown that the conduct
of a probe trial with an Atlantis platform serves, in animals
with partial lesions, only as a reminder cue. The result vindi-
cates the interpretation drawn by Bolhuis et al. (1994) much
earlier.
Noting the same problem of weak postsurgery memory,
Clark et al. (2004) extended the presurgery training to as
many as 80 trials, finding that it improved the performance of
the sham/control group. The point of doing this is to increase
the opportunity of seeing an upward consolidation function
in the data (along the lines of Fig. 13–10B3). However, using
large radiofrequency lesions restricted to the hippocampus,
they again observed both poor memory and a completely flat
gradient of retrograde amnesia in postsurgery probe tests.
Another attack on the problem was suggested by the fact that
patient E.P. had learned his home neighborhood as a young
man. Clark et al. (2005a) therefore trained normal rats from
very early in life (21 days) for a total of 392 trials over 69 days.
Hippocampal and sham lesions were made at 90 days of age,
and the animals were tested 2 weeks later. Once again, the
complete lesion group failed to display above-chance per-
formance and that despite the use of a reminding procedure a
few days before testing (Fig. 13–32C). Undeterred, Clark et al.
(2004) wondered if there was something unusual about the
standard watermaze task, as other nonspatial tasks had found
evidence of a temporal gradient, albeit only using a savings or
choice measure (Cho et al., 1993, 1995; Ramos, 1998; Maviel
et al., 2004). Accordingly, they also trained rats in an Oasis
maze, in which they have to search for water in a small well
located on a “desert-like” table top, and in an annular water-
maze (Hollup et al., 2001) in which the animals swim in a cir-
cular corridor until they find an escape platform. The use of
the annular watermaze is interesting as the need for naviga-
tional updating is minimized; the swimming rats have only to
recognize a potentially safe place along the ring and slow
down when they get there. However, using a single probe test,
both tasks again showed a flat gradient (Fig. 13–32D).
The animals’ joy in eventually finding water in the Oasis
maze cannot have been matched by any exhilaration on the
part of proponents of the declarative memory theory by the
outcome of this series of experiments. Clark et al. (2004) sug-
gested that a feature that may distinguish tasks where remote
spatial memory is spared (cross and radial maze tasks) com-
pared to those where it is impaired (watermaze tasks) is that
only in the latter does the expression of memory require the
now-lesioned animal to recall by navigating through open
space. That the hippocampus might be essential for expressing
spatial memory, rather than storing or retrieving it, is remi-
niscent of Whishaw’s distinction between “getting there”
(impaired by lesions) and “knowing where” (which may not
always be). A processing deficit of this kind conveniently pre-
dicts that any underlying hippocampal-neocortical consolida-
tion process would remain cryptic.
There are, however, problems with this account. First, it
incorrectly predicts that remote memory should be intact in
the annular watermaze because its navigational demands are
minimized. The animal does not have to learn or access any
navigational strategies for “getting there”—it has merely to
swim in a constrained circle. Second, remember that reminder
cues revealed latent spatial memory in partially lesioned ani-
mals (Martin et al., 2005). Presumably. sufficient hippocampal
tissue was spared for the expression of memory. However, this
latent spatial memory was in the recent-memory group, not
the remote. If cryptic consolidation does occur, the effects on
trace strength must therefore be substantially smaller than
those of forgetting. Third, remember that different results are
obtained after small partial lesions as a function of whether
they are made before or after training (Moser et al., 1995;
Moser and Moser, 1998b). When spatial memory is acquired
by normal animals and then partial lesions are made in the
hippocampus, performance is seriously compromised and
needs to be relearned. This makes sense if trace storage of spa-
tial memory is within the hippocampus, it being fully distrib-
uted along the longitudinal axis when created in an intact
hippocampus. Relearning would be possible with a partially
intact hippocampus, and this would also be enough to express
the memory once retrieved.
The last twist of the consolidation story comes from the
use of novel techniques. Using a behavioral training proto-
col similar to that of Martin et al. (2005), indirect evidence for
the presence of a systems level consolidation process after
watermaze training was obtained using a pharmacological
approach. Riedel et al. (1999) observed that chronic reversible
inactivation of the hippocampus for 1 week after training
through bilateral infusion of a GluR1–5 antagonist impaired
spatial memory when the animals were tested 16 days later
(i.e., long after the effects of the drug had worn off). The drug

demonstrably blocked fast synaptic transmission in the hip-
pocampus, implicating the necessity for hippocampal neural
activity for some period after training if spatial memory was
to persist. Riedel et al. (1999) explored the impact of delaying
the “shut-down” by a few days and got the same results, but
they did not examine a longer time course after training. This
reversible but chronic inactivation approach is promising ana-
lytically; but even if hippocampal inactivation long after
training was found to have minimal impact on remote mem-
ory, it would not mean that the neural activity required for
memory persistence is necessarily a hippocampal-neocortical
dialogue of the kind presumed with the “standard model.” It
could also be the creation of multiple traces in the hippocam-
pus, or some other intrahippocampal consolidation process,
that requires neural activity. Reversible inactivation may
not be quite the help for distinguishing the rival theories it
first promised to be. A similar ambiguity pertains to the strik-
ing evidence that sectioning the direct input from layer III
of the entorhinal cortex to area CA1 of the hippocampus
(the temporo-ammonic tract) is essential for memory consol-
idation (Remondes and Schuman, 2004). When lesions were
made before 7 days of watermaze training, the animals showed
good probe test performance 1 day after training but poor
memory 28 days later; sham-lesioned animals showed no
decline. The same result was obtained when the lesions were
made 1 day after the end of training. However, if the lesions
were delayed until 21 days after the end of training, both the
sham and temporo-ammonic tract-lesioned animals showed a
continued, albeit weaker, tendency to swim in the correct
quadrant 7 days later. Collectively, these three experiments are
consistent with the idea that “ongoing cortical input conveyed
by the temporo-ammonic path is required to consolidate
long-term spatial memory” (Remondes and Schuman, 2004,
p. 702), the implication being that this pathway is necessary for
the hippocampal-cortical dialogue implicated in the declara-
tive memory theory. However, this is not absolutely required
by the data. It could still be that temporo-ammonic input is
essential for ongoing consolidation in the hippocampus.
A final expression theory, as discussed by Martin et al.
(2005), relates to the role of the hippocampus in the retrieval
and expression of consolidated memory traces that reside in
the neocortex but for which the hippocampus plays a part in
their retrieval. There are a variety of possibilities. One is that
the hippocampus does store information but stores only the
“indices” (Teyler and DiScenna, 1986) or the “cartoons”
needed to retrieve a consolidated cortical memory trace rather
than detailed sensory-perceptual information (this remains in
the cortex). With this view, the hippocampus might retrieve a
cortical memory in a manner analogous to conducting a key-
word search on an electronic document. Partial lesions would
compromise the “indices” rather than the detailed sensory-
perceptual memory traces and so limit the ability to retrieve
cortical memory traces when a recall process is necessary. The
circumstances surrounding memory retrieval and the very
character of the information retrieved (Nadel and Bohbot,
2001) may determine whether the hippocampus plays a nec-
Theories of Hippocampal Function 653
essary role in this process. Memories are widely thought to be
stored in a distributed manner (i.e., their component traces
are stored across different brain areas). Retrieval of the right
combination of traces might be possible under circumstances
in which the retrieval cues are either particularly apposite or
sufficiently rich to disambiguate cortically based traces with
overlapping components. However, under circumstances in
which the retrieval cues do not permit easy disambiguation or
where they have to be generated indirectly through a process
of recovered consciousness (Moscovitch, 1995), the process-
ing capacity of the hippocampus and its stored indices would
remain essential. Disambiguation would generally be critical
for context-dependent episodic memory in which a unique
combination of traces corresponds to the particular event
from the past that needs to be reactivated. Consistent with this
idea, the expression of human episodic memory is sometimes
found to be permanently affected by hippocampal damage, as
noted earlier. Notwithstanding our focus on memory retrieval
processes, this account is consistent with Rosenbaum et al.’s
(2001) revision of the multiple trace theory. It distinguishes
between context-dependent and context-free memories and
suggests that only the former remain dependent on the hip-
pocampus for our lifetime. Winocur et al. (2005) have
reported data consistent with this view in a study of rats that
lived in a “rodent village” for 3 months prior to explicit train-
ing on a spatial problem in the village. Posttraining hip-
pocampal lesions had no effect on effective navigation to the
correct location in the village, which they interpreted as
implying the existence of a detailed “semantic” map stored in
the neocortex. Rats with less experience of the village were
impaired by posttraining hippocampal lesions because they
presumably lacked a context-free memory of the village.
Unfortunately, the lesions used in this study were partial
(around 50%), allowing some use of spared hippocampal tis-
sue in memory retrieval. Winocur et al. (2005) argued against
this being of significance on the grounds of finding no corre-
lation between lesion size and retrieval, but this is a weak
argument based on only a small number of subjects with dif-
ferent sized lesions.
Studies of interactions and correlations between single-unit
and field-potential recordings in hippocampus and neocortex
during and after sleep are highly suggestive of consolidation-
like processes (Siapas and Wilson, 1998; Sirota et al., 2003).
Targeted molecular studies can help resolve some of the ambi-
guities at the mechanistic level, although not necessarily all of
them, as a daunting combination of novel behavioral proto-
cols, lesions, and molecular interventions will probably prove
necessary to unravel the complexities. Frankland et al. (2001)
made the intriguing observation that heterozygous CAMKII
mutant mice show normal LTP in the hippocampus, decaying
LTP in the cortex, and failure to consolidate both spatial and
contextual information. They suggested that the instability of
synaptic potentiation in the cortex could be the basis of the
consolidation failure. This is an interesting idea, but it is
equally consistent with the idea that effective retrieval relies on
the synergy of different kinds of information permanently
654 The Hippocampus Book
stored in both hippocampus and neocortex. Frankland and
Bontempi (2005) summarized these developing new lines of
research using transgenic animals and IEG markers to plot the
time course and regional contributions of neocortex and hip-
pocampus to memory consolidation.
13.4.6 Critique
The cognitive map theory has been highly influential. The role
of the hippocampus in spatial learning and memory is
accepted, particularly in rodents; and such learning is widely
used as an “assay” for investigating the cellular and molecular
mechanisms of learning and its dysfunction in models of
neurodegenerative disease. “Spatial memory” has become a
keyword at international meetings, such as those of the
Society for Neuroscience and the Federation of European
Neuroscience Societies, and there are now laboratories all over
the world investigating the neurobiological basis of allocentric
spatial representation and spatial navigation. Although much
of this reflects a general interest in spatial learning and mem-
ory rather than commitment to one particular theory, it is
perhaps not surprising that it has been criticized as the theory
“that wouldn’t die.”
However, the theory clearly has major difficulties, both
conceptual and empirical. This chapter’s presentation of the
theory was built around four themes—spatial representation,
spatial navigation, comparative work, and studies of memory
storage and consolidation—and has already included a meas-
ure of critical discussion. A few additional points should be
noted in relation to these same four themes.
Spatial Representation, Maps,
and the Geometric Module
With respect to representation, one set of problems center
around whether the concept of a “cognitive map” is an expla-
nation of anything or merely a beguiling metaphor (Healy,
1998). A map is an easily understood concept, but maps are
things that people look at to extract information. Adopting
this term for the neural activity of a region of the brain seems
to carry with it the mental baggage that there must be some
cryptic homunculus that is “looking at” the map to do like-
wise. In the absence of a mechanistic account of how informa-
tion is extracted from the map, the explanation it offers is
incomplete. With respect to the concept of cognitive mapping
itself, Bennett (1996) argued that the now widespread use of
the term in many different ways by investigators in ethology,
psychology, and brain science is confusing and that the con-
cept has outlived its usefulness. This feels unpersuasive. That
some people use a scientific term loosely is not grounds for
castigating those who use it more precisely. As Bennett (1996)
himself conceded, Tolman (1948) introduced it and O’Keefe
and Nadel (1978) used it explicitly in relation to the ability to
represent the environment in a viewer-independent allocen-
tric manner and to solve such problems as novel shortcuts.
This definition is concise. Time and again, O’Keefe’s group
have come back to a central prediction of the theory in rela-
tion to brain activity associated with exploration and spatial
novelty, most recently in showing differential activation of the
human hippocampus proper when people take shortcuts in a
virtual reality environment (Maguire et al., 1998; Burgess et
al., 2002). Quantitative models are precise and testable, such as
the boundary vector model of place cells (Hartley et al., 2000).
However, the concept of a geometric module is undergoing
something of a forensic examination with experiments in
rodents (Golob and Taube, 2002; McGregor et al., 2004b;
Pearce et al., 2004) and humans (Wang and Spelke, 2002),
raising questions about whether there is any need for a
“global” representation of environmental shape, as in the
study of the kite-shaped watermaze discussed earlier. Pearce’s
study, part of a larger series, breaks new ground because it
suggests that earlier data pointing to the existence of a global
geometric module centered on the hippocampus is also con-
sistent with a different account that sees no need for such a
representation. Learning may require only a perceptual system
that can distinguish long barriers from short ones (i.e., per-
ceptual features of local cues in general) and bridge the tem-
poral delay between turning appropriately as the animal
approaches the long wall and the subsequent receipt of
delayed reward. It remains to be seen where this research on
local features versus boundaries and geometry will lead.
Successful Spatial Navigation
Without the Hippocampus
A second set of difficulties for the theory is that rodents with
nearly complete cell loss in the hippocampal formation can,
with certain training procedures, learn allocentric spatial
tasks. This is the longstanding claim of the “working memory”
theory of Olton et al. (1979), who argued that spatial reference
memory tasks can be learned by lesioned rats. We have seen
that this claim is correct, but the rate of learning is much
slower. Regardless of whether it is slower, the inescapable
implication is that extrahippocampal structures can support
spatial learning. This finding that the integrity of the hip-
pocampus, or of hippocampal synaptic plasticity, is not essen-
tial for spatial reference memory is not necessarily fatal for the
cognitive map theory because recent experimental and theo-
retical studies have established that effective spatial navigation
can be realized in many ways, including strategies other than
map-guided navigation. However, certain observations, such
as normal probe test performance in the watermaze by rats
with complete hippocampal lesions (Morris et al., 1990) must
be accepted as problematic unless the theory is to descend into
irrefutability. That rapid, one-trial spatial learning in the
DMP task is always impaired (Steele and Morris, 1999) sug-
gests that the integrity of the hippocampus is essential for
rapid encoding of spatial information (spatial events?), but
the gradual accumulation of information about spatial regu-
larities of an environment might be information that can,
albeit slowly, be encoded, stored, and retrieved in the neocor-
tex without recourse to the hippocampus. That retrograde

amnesia is apparently always seen after both partial and com-
plete hippocampal lesions in rats appears to be contradictory,
but the initial learning in such tasks always takes place in nor-
mal animals.
In passing, it is relevant to note that knocking out the
capacity to express a particular form of NMDA receptor-
dependent LTP through mutation of the GluR-A receptor
selectively impairs one-trial spatial working memory but not
incremental spatial reference memory (Zamanillo et al., 1999;
Reisel et al., 2002). The knockout is not hippocampus-
specific, but a genetic rescue of the receptor mutation that
appears to be largely expressed in the dorsal hippocampus
(Schmitt et al., 2005) is sufficient for partially rescue of effi-
cient spatial working memory performance. This body of
work is discussed in Chapter 7. For the present, note only that
it illustrates how new techniques are shedding light on what it
means to classify a task as hippocampus-dependent. A mouse
in which GluR-A is absent shows apparently normal fast
synaptic transmission in the hippocampus but fails to show a
rapidly induced form of LTP. Such an animal is like a lesioned
animal in that it cannot display types of memory encoding
that occur in the hippocampus; but it is more subtle than a
lesioned animal, as it allows information throughput in the
hippocampus, which may be essential for neocortical encod-
ing of incremental learning.
If this is correct, one would expect that damage to certain
neocortical brain areas (and not just the entorhinal cortex)
would have effects on spatial navigation that are clearly not
predicted by the theory. These have indeed been observed.
Deficits in spatial navigation in a variety of tasks follows
lesions in the posterior parietal cortex (DiMattia and Kesner,
1988a; Kesner, 1998) and the midline retrosplenial/cingulate
cortex (Sutherland et al., 1988). Harker and Whishaw (2004)
summarized a body of conflicting data, including strain differ-
ences, regarding involvement of the retrosplenial cortex in var-
ious spatial tasks including navigation. Aggleton and Vann
(2004) concurred that retrosplenial lesions do impair allocen-
tric spatial navigation in the watermaze, though lesion size can
be a critical factor (Vann and Aggleton, 2004). Focusing on a
different part of the neocortex, Kesner (2000) argued that there
is parallel processing of spatial information in the hippocam-
pus and parietal cortex, with the former more important for
“spatial events” and the latter part of a neocortical “knowl-
edge” system. This distinction echoes the episodic/semantic
distinction to which we return later. It is also is relevant to the
issue of whether memory storage is in the hippocampus or
there is some hippocampally guided consolidation process.
Kesner (1998) is not alone in regarding a strictly spatial view of
hippocampal function as too narrowly drawn.
Multiple Types of Memory?
Another conceptual problem is that the simple division of
learning processes into spatial versus nonspatial now appears
dated. Although there have been many developments of the
cognitive map theory, including its application to various
Theories of Hippocampal Function 655
aspects of human memory, its adequacy as a general theory of
multiple types of memory is limited. It has little to say about
the relations between propositional (declarative) and non-
propositional forms of long-term memory (e.g., skill learn-
ing), between working and long-term memory, or, more
curiously, about the nature of perceptual representations. Its
proponents would argue that the theory was never intended to
be a general theory of memory and cognition and that Figure
13–17 reflects their primary focus. It is a theory built around
the discovery of place cells and the sensitivity of spatial navi-
gation by rodents to various kinds of brain damage. It has
imaginatively extended from these beginnings, but its place as
a building block of a more general neurobiological theory
remains a task for the future.
However, even within the domain of the cognitive analysis
of space, there are problems with respect to the claim that spa-
tial learning is fundamentally different from associative learn-
ing. In 1978, it was valuable to make the point that associative
conditioning is generally slow and inflexible, whereas place
learning can occur during one trial; but strictly behavioral
studies of allocentric place learning have now revealed that
spatial learning displays several qualitative characteristics of
associative learning, such as “blocking” (Chamizo, 2003).
Spatial learning may be associative after all.
Integrity of the Hippocampus Is Required
for Many Nonspatial Learning Tasks
If rats with hippocampal lesions can learn certain spatial
tasks, a separate concern is that such lesions are now also
known to affect a range of nonspatial tasks, including classic
operant tasks with a temporal component, such as differential
reinforcement of low rates of response (DRL). Although it has
always been possible to construct a spatial account of DRL
performance, a temporal one seems more reasonable
(Rawlins, 1985). However, there a number of other tasks, such
as nonlinear discrimination learning and socially transmitted
food preferences, for which a spatial account would strain
credibility. Section 13.5 is a detailed examination of such tasks
in the context of the “predictable ambiguity” theories for
which they are relevant.
Coda
As a last word, it is worth reflecting on the fact that the cogni-
tive map theory—the theory that refuses to die—enjoys the
security of a number of key physiological findings that are not
enjoyed by the other main theory discussed so far, the declar-
ative memory theory. The theoretical battlefield between these
two theories would be very different if the discovery of other
spatially responsive neurons (e.g., head-direction units, view
cells, grid cells) had not occurred. All of these discoveries were
made since 1978. Hippocampal neurons are not differentially
active as a function of stimulus familiarity, but they clearly are
responsive to an animal’s location or view of the world. This
is a major if somewhat neglected obstacle to the credibility of
656 The Hippocampus Book
declarative memory theory. We have also seen a number of
studies in which interventions in the induction and expres-
sion of LTP have an effect on spatial learning and retention
(see Chapter 10), although the extent to which these effects are
cognitively selective is not well understood. Some comfort for
the declarative memory theory in the physiological domain
can be derived from glucose uptake and IEG studies indicat-
ing greater neuronal activity in the hippocampus soon after
learning than later, perhaps reflecting the dynamic process of
memory consolidation. These observations are pertinent to
the task of trying to build a neurobiological account of hip-
pocampal function to which we return in Section 13.6.
Á after a delay, the presentation of reinforcement (US). The
13.5 Predictable Ambiguity: Configural,
Relational, and Contextual Theories of
Hippocampal Function
The two major theories discussed so far command strong sup-
port and continue to be refined in the light of new findings.
Neither will be discarded until a cogent alternative is widely
seen to account for a larger body of data. However, numerous
other theories of hippocampus function have been proposed.
They include hypotheses that build on modern associative
learning theories whose origins, unlike the concept of multi-
ple memory systems, lie in the phenomenon of classical and
instrumental conditioning rather than human amnesia. In
one way or another, their applications to thinking about hip-
pocampus function all have to do with dealing with the prob-
lem of “predictable ambiguity.” This phrase is intended to
capture the sense that a stimulus can consistently mean one
thing in one situation but something else in a different one.
Learning theory research is characterized by exacting train-
ing protocols that explore conditioning phenomena far
removed from the classic account of learned salivation to the
sound of a bell. As we touched on briefly in the preceding sec-
tion, examples include protocols for examining whether learn-
ing occurs when an animal’s “expectations” are violated or that
it alters the “associability” of stimuli or the contexts in which
they occur (Rescorla and Wagner, 1972; Pearce and Hall,
1980). Such learning could include both excitatory and
inhibitory associations (i.e., that a predicted event is more or
less likely to happen). Beyond simple associations, there are
also experience-dependent “attentional” phenomena and
“occasion-setting” paradigms in which a stimulus (or more
broadly an entire context) influences the retrieval of other
memories quite apart from any associations in which it may
also become engaged. Such phenomena have led to the sugges-
tion that associative conditioning, far from deserving its low-
level status as a “nondeclarative” form of stimulus-response
“habit” learning (within declarative memory theory) or of
mere “taxon” learning (in the cognitive map theory), is more
complex and might offer a conceptually economical account
of a variety of ostensibly cognitive phenomena. Proponents of
this approach have introduced the idea that conditioning
offers insights into the way in which animals learn about “the
causal texture of their environment” (Dickinson and
Mackintosh, 1978). This implies that conditioning could have
to do with the acquisition of knowledge that could be used
inferentially (Mackintosh, 1983; Rescorla, 1988; Pearce and
Bouton, 2001; Pickens and Holland, 2004) as first discussed
explicitly by Dickinson (1980). These intellectual develop-
ments are relevant here because psychological processes have
been identified in modern “learning theory” for which the hip-
pocampus is a potential anatomical substrate.
Paradoxically, it has long been known that animals with
hippocampal lesions are generally normal regarding simple
associative conditioning tasks. These tasks involve no more
than the pairing of an initially neutral stimulus (CS) with,
result is the gradual development of associative strength and
its expression as a conditioned response (CR). In general, sim-
ple delay conditioning depends on other brain circuits. For
example, hippocampal lesions are without effect in nictitating
membrane (NMR) delay conditioning in which a CS starts
several hundred milliseconds before and then overlaps with
the presentation of a strong puff of air to the eye (US). NMR
learning was shown independently by several groups (using
lesion, inactivation, and unit-recording techniques) to be
mediated by circuits in the spinal cord and cerebellum (Krupa
et al., 1993; Hardiman et al., 1996; Beggs et al., 1999). As noted
in Chapter 11, hippocampal pyramidal cells nonetheless show
firing patterns during NMR conditioning that are correlated
in time with the learned response (Berger et al., 1983), but this
neural activity is not usually necessary for normal perform-
ance. The exception to this is when a “trace interval” is
inserted between the end of the CS and the onset of US, as we
saw in the discussion of the declarative memory theory; alter-
ations in excitability are then seen in both young animals
(Moyer et al., 1996) and aged animals (Moyer et al., 2000).
Simple fear conditioning to a punctate CS such as a tone (in
which it is sounded for a period prior to and overlapping the
delivery of an aversive stimulus such as a mild electric shock)
is thought to involve storage of traces in the amygdala (Davis
et al., 1993; LeDoux, 2000), although others have argued that
the role of the amygdala is primarily neuromodulatory
(Vazdarjanova and McGaugh, 1999; McGaugh, 2000). This
alternative is important because, as we saw in the discussion of
late LTP (see Chapter 10), activity in the amygdala can modu-
late persistence of LTP in the hippocampus. These and many
other studies have broadened our understanding of the role of
various brain areas in learning and memory.
The hippocampus may, however, become definitively
engaged in conditioning situations in which the associations
into which the initially neutral CS enters are predictably incon-
sistent. In Section 13.2 we saw how a specific behavioral task is
said to be ambiguous if there are two or more ways to solve it.
A different sense of ambiguity is when the outcome of a spe-
cific CS is sometimes one US but on other occasions either
another US or the absence of the first US (often abbreviated as
NoUS). For example, a light and a tone may each predict the

US when presented on their own but not when presented
together. How can an organism cope with this and other forms
of predictable ambiguity and so behave appropriately, as
many behavioral studies have shown that they can? This class
of problem is called the XOR problem in the artificial intelli-
gence and cognitive science communities (Rumelhart and
McClelland, 1986) and a nonlinear task by students of the ani-
mal learning theory. The nonlinearity arises because it had
been thought that CSs acquire associative strength as a func-
tion of the reinforcement with which they are linked. However,
if this were all there is to learning, the net associative strength
of the combined light and tone should always be higher than
that acquired by either CS on its own despite the combination
of CSs always being followed by nonreward. A nonlinear solu-
tion to the problem seems to be required.
Suppose, however, that the brain has a learning device that
can resolve such ambiguity. It would then be possible to learn
the differential outcome of single and combined stimulus pre-
sentations and so react appropriately on every occasion. The
configural-association theory of Sutherland and Rudy (1989b;
Rudy and Sutherland, 1995b) was the first of several hypothe-
ses about hippocampal (and cortical) function that specifi-
cally addressed this kind of problem. The relational-processing
theory (Cohen and Eichenbaum, 1993; Eichenbaum and
Cohen, 2001), a development of Squire’s original declarative
memory theory, also tackled the issue of ambiguity. It took
things further in focusing on “relationships” being encoded
between stimuli (e.g., that A is better than B and that B is bet-
ter than C), rather than stimulus configurations (AB, BC). It
Figure 13–33. Coping with predictable ambiguity. A. Hirsh’s model
recognized a distinction between a “performance line,” which stores
simple CS–US relations and expresses them as conditioned
responses, and a “context memory,” which modulates the perform-
ance line when a CS predicts a US in one context but something
A Associative encoding and retrieval
Stimulus Motor
Input Output
Performance Line
Stimulus
Input Context Memory
Theories of Hippocampal Function 657
also emphasized the possibility that such stimulus relations
can be “flexibly” retrieved in an appropriate way in novel situ-
ations (e.g., allowing inferences such as that A is better than
C). Ideas about the involvement of the hippocampus in con-
textual retrieval (Hirsh, 1974; Good and Honey, 1991) are
closely related. What we remember about a stimulus and its
associations depends critically on the context in which it
occurs, including the absence of an expected stimulus as
occurs in the “extinction” of conditioning when a reinforcing
US no longer follows a CS. Wide-ranging ideas about the con-
tribution that context cues make to encoding and retrieval,
including the notion that contexts “contain and predict rather
than simply compete with explicit CSs” (Nadel and Willner,
1980, p. 218), helped spur the development of now widely
used learning paradigms such as context fear conditioning.
Although intellectually distinct and occasionally at logger-
heads with each other, the configural-association, relational-
processing, and contextual-retrieval theories have two
common threads: They all set out to solve problems involving
“predictable ambiguity,” and they all argue that the hip-
pocampal formation is central to achieving it.
One of the early theorists to recognize the problem was
Hirsh (1974). He looked upon conditioning as a process in
which stimulus input came to trigger motor output along a
“performance line” as a function of simple CS-US associations
(Fig. 13–33A). He argued that it is independent of the hip-
pocampus. On the other hand, he also supposed that a sepa-
rate “context memory” could somehow interact with the
performance line, particularly during memory retrieval, and
different in another. (Source: After Hirsh, 1974.) B. Modern learning
theory has developed an elaborate framework that is relevant to
thinking about the ways in which the hippocampus, functioning as
a context memory, could modulate internal representations of asso-
ciations. See text for discussion.
B Contemporary representational framework for
associative memory
CSrep CSrep
Stimulus Action Motor
Input Representations Output
USrep USrep NoUS
rep
Context Memory
Context - CS associations
Stimulus
Input Context - US associations
Context modulation
CS - US associations
658 The Hippocampus Book
so ensure appropriate motor output when a specific stimulus
predicted reinforcement in one context but the absence of
reinforcement in another. Both context memory and the
process of contextual retrieval were identified as mediated by
the hippocampus. Contemporary research has elaborated this
framework in numerous ways (Fig. 13–33B). First, stimulus
input is said to give rise to internal representations of CSs and
USs that can enter into a variety of associative relations rang-
ing from first-order associations (CSrep–Usrep where “rep”
refers to an internal representation of these stimulus events),
to second-order relations (CS1–CS2), and even predictions of
the absence of a previously expected stimulus (CS–NoUS).
These associative relations constitute “knowledge” and (some-
how) give rise to goal-oriented action representations that
constitute appropriate behavior (e.g., approach food) and
then to motor output (the particular movements produced to
fulfill an action). The concept of contextual modulation
remains, but the manner in which contexts are represented
and enter into associations is now recognized as quite compli-
cated. The representations (lower part of Fig. 13–33B) may be
“elemental” or “configural,” and these in turn may become
associated with specific CSs or USs that occur in their pres-
ence (context-event associations). In addition to direct associ-
ations between contexts and other stimuli, contexts may also
modulate the expression of other CS-US associations, such as
those expressed within the knowledge system.
Several theories addressing the problems of predictable
ambiguity suppose that the hippocampus is a form of config-
ural, relational, or context memory; but they differ with
respect to the precise set of processes by which such stimuli
enter into or modulate other associations. A consequence of
this is that the field has become somewhat baroque and
impenetrable. However, it is a serious endeavor that attempts
to go beyond descriptive taxonomy to a more detailed and
predictive understanding of process and mechanism. The sec-
tion that follows is a selective representation of this field of
research.
13.5.1 Configural Association Theory
Learning is often complicated by ambiguity. A stimulus may
signify one event in one situation and another event elsewhere.
It is then a “fact” that a tone predicts food in one situation and
a “fact” that it does not in another—a complication that the
declarative memory theory may have recognized but offered
no mechanistic way of addressing. Animals can interpret
ambiguous associations such as these provided they are given
sufficient additional information and a learning device sophis-
ticated enough to do it. In a series of papers, Rudy and
Sutherland (Sutherland and Rudy, 1989; Rudy and Sutherland,
1995; O’Reilly and Rudy, 2001) proposed a theory of hip-
pocampal function, building upon earlier work with animal
learning theory in which the configuring of stimuli together as
fused compound CSs is the way that animals solve such prob-
lems (Rescorla, 1973). The essence of Sutherland and Rudy’s
and later O’Reilly’s proposals are as follows (Box 13–6).
Box 13–6
Configural Association Theory
1. A class of “nonlinear” associative problems can be solved
by the process of stimulus “configuring.”
2. The hippocampus builds and stores “configural associa-
tions” between elemental stimuli (1989), serves an
enabling role in their encoding and eventual storage in
neocortex (1995), or helps “pattern-separate” stimuli rap-
idly to aid slow cortical learning (2001).
Ambiguity, Conditioning, and the Hippocampus
Consider the case of two auditory CSs such as tones (call them
A1 and A2) that, through repeated pairings, become selec-
tively associated with two types of food (F1 and F2). The
experimenter arranges matters such that when a light (L) is
illuminated in the test chamber A1 signals F1 and A2 predicts
F2. However, when the animals are in the dark (D), A1 is
arranged to signal F2 (rather than F1) and A2 predicts F1. Rats
and other animals can resolve this predictive ambiguity and
learn “what leads to what” in each situation.
Rescorla’s proposal was that animals (mentally) create
stimuli that are “configural” associations between L and A1, L
and A2, D and A1, and D and A2. These would be the config-
ural cues (LA1), (LA2), and so on. If these configural cues
are discriminable, they could enter into unambiguous associ-
ations with reinforcement without interference. Learning
would otherwise obey established principles of associative
learning, namely the gradual accumulation of associative
strength to each configural stimulus as a function of how well
the sum total of stimuli available at any one time predicts the
outcome (Rescorla and Wagner, 1972; Mackintosh, 1983). The
animal would have solved the problem and could express its
“knowledge” of associative relations in performance.
In its original form, Sutherland and Rudy (1989) proposed
that the hippocampus was the physical substrate for encoding
and storing configural associations. They also proposed that
the configurations could be quite arbitrary combinations of
individual stimulus events (e.g., of a light and a tone) occur-
ring together in time, one after another in sequence, or in spa-
tial proximity. The latter aspect of this proposal neatly
extended the domain of the theory to tasks such as place
learning where multiple stimuli in a given context might be
said to define a place. This represented something of a volte
face for Sutherland, who had earlier argued, in keeping with
cognitive map theory, that the learning of a “mapping strat-
egy” in the watermaze was distinct from associative learning
(Sutherland and Dyck, 1984). His new idea, with Rudy, was
that place learning might actually be a special case of config-
ural learning in which the individual extramaze cues around a
watermaze or radial maze enter into stimulus configurations
and so (somehow) guide the animal to the correct target. Data
that had hitherto offered support for the spatial mapping the-
ory could then be radically reconsidered as potentially sup-

porting the new hypothesis instead. Similarly, the supposition
that the hippocampus encodes “configurations” takes the the-
ory into territory not considered in the original version of the
declarative memory theory. With that theory, the consolidated
“trace strengths” of the associations between A1 and F1 and
between A1 and F2 (in the example above) would be equal,
and the animal would have no way of resolving the predictive
ambiguity of its factual knowledge.
An important feature of the original 1989 configural asso-
ciation theory was that it made a number of clear-cut predic-
tions about when deleterious effects of hippocampal lesions
would and would not be observed in associative learning. This
is because Sutherland and Rudy recognized that not all kinds
of associative learning require configurations to be formed
(simple delay conditioning being such a case), whereas others
do (nonlinear problems). In short, the theory could be
tested—an attractive, widely respected but always precarious
feature of any good theory.
Tests of the Original Configural
Association Theory
The behavioral tasks Sutherland and Rudy considered to be
valid tests are our first taste of the somewhat baroque nature
of this field of research. Arguably they are problems that an
animal is unlikely to face in its natural habitat, a worrying fea-
ture because it is not clear what might have been the evolu-
tionary selection pressure to develop an “ambiguity learning
module,” However, the apparent artificiality of these tasks
reflects a creditable desire on the part of the theory’s architects
for their hypothesis to be tested rigorously. Four types of task
were studied intensively. They are called negative patterning,
transverse patterning, biconditional discrimination, and
feature-neutral discrimination learning (Fig. 13–34). An
explanation of each follows, with reference to exemplar exper-
iments.
With negative patterning, two stimuli are each paired indi-
vidually with reinforcement, but the combination of the two
stimuli is nonrewarded. This is annotated in Figure 13–34 as
A, B, AB , where A and B are individual stimulus ele-
ments,  refers to reward, and is the absence of reward. As
summarized by Rudy and Sutherland (1995), several studies
from their laboratory from 1989 onward indicated that neu-
rotoxic hippocampal lesions made in either of two ways
(kainic acid plus colchicine or ibotenic acid alone) cause a
deficit in negative patterning. These tasks included operant
Figure 13–34. Four tasks used to test critical pre-
dictions of the configural-association theory. The A Negative patterning
letters A, B, and so on refer to arbitrary stimuli; and
the symbols  and refer to the availability of
A+ , B+ , AB-
reinforcement.
C Bidirectional discrimination
AB+ , CD+ , AC- , DB-
Theories of Hippocampal Function 659
lever-pressing tasks and a swimming task with cues hanging
above potential escape locations (Alvarado and Rudy, 1995a).
This finding did not, however, prove replicable by others.
Davidson et al. (1993) observed no deficit in negative pattern-
ing in animals with two types of hippocampal lesion. In
Davidson et al.’s version of the task, the rats were required to
press a lever several times during a tone (T) or light (L)
for occasional reward (). This requirement provided a more
effective baseline against which to observe a decrease in the
rate of responding during the nonreinforced TL– compound.
They also used a posttraining transfer test with a novel clicker
stimulus to check that learning of the original task did depend
on the formation of a true TL configural cue. Although we
cannot be certain why different outcomes were obtained in
the two studies, Davidson et al. (1993) made the further
observation that animals with the same kainic acid 
colchicine lesions used by Alvarado and Rudy, but not those
with ibotenic acid lesions, displayed disinhibited rates of oper-
ant responding during the intertrial interval between tone and
light presentations. Both types of lesion are excitotoxic and
should therefore spare fibers of passage, unlike older lesion
techniques. However, kainic acid also produces lesions distant
to the hippocampal formation, such as in the piriform and
periamygdaloid cortex. This additional damage could have led
to disinhibited responding, and its potential presence raises
the possibility that altered motor performance rather than
impaired learning contributed to the positive result in some
of the earlier Rudy and Sutherland studies. An operant
lever-pressing task used by McDonald et al. (1997) indicated
that hippocampal lesions slowed down but did not ultimately
prevent the rate of learning of a negative patterning task,
and fornix lesions had no detectable effect on learning at
all. Lesions of the avian hippocampus were also ineffective
in impairing negative patterning in pigeons (Broadbent et
al., 1999; Papadimitriou and Wynne, 1999). Bussey et al.
(2000) found no deficit in a configural conditioning task
(conditioned approach to a place where food was available)
after radiofrequency fornix lesions or neurotoxic perirhi-
nal/postrhinal (PRPH) lesions. All groups learned to
approach in response to either an auditory tone or a brief
period of darkness and to withhold approach when these two
CSs were presented together. They also used posttraining
transfer tests to establish that a configural solution had been
used for the task, and “behavioral histology” that took the
form of showing that the fornix-lesioned rats were impaired
in T maze alternation and the PRPH-lesioned rats in object
B Transverse patterning
A+B- , B+C- , C+A-
D Feature-neutral discrimination
AC+ , C- , AB- , B+
660 The Hippocampus Book

recognition memory. Overall, the available data from these
and other studies on negative patterning represents, at best,
only mixed support for the theory.
Transverse patterning fares better. It involves a series of
simultaneous discrimination tasks of the form A versus B ,
B/C , and C/A in which the stimuli A, B, and C are
equally often rewarded () and nonrewarded ( ). This set of
problems, as noted by Moses and Ryan (2006) when compar-
ing predictions of the configural and relational theories, is
akin to the children’s game of “rock, paper, scissors.” Alvarado
and Rudy 1995b) reported that rats with hippocampal lesions
could readily learn to swim toward A and avoid B and to
learn B/C when either task was presented over successive
trials on its own (Fig. 13–35). However, they failed to learn the
full transverse pattern when all three problems were presented
together. This result is beautifully consistent with the theory
provided it is the AB configural stimulus that triggers
approaches to stimulus A, the BC configuration to B, and the
CA configuration to C. Unfortunately, it is also a task that may
be amenable to solution without recourse to explicit configur-
ing. This would be a “conditional solution” in which the ani-
mals learn that one stimulus (e.g., A) is correct given the
presence of B, that B is correct given C, and that C is correct
given A. The distinction between configural learning and rule
learning may seem subtle, but the acquisition of an “if–then”
conditional rule obviates the need to form stimulus configu-
rations and may be a form of learning that depends on other
brain structures (e.g., the caudate). Conditional rules are not
easy for rats to learn; but with sufficient training and provided
the information with which they are provided is consistent,
they can do it.
Later work on transverse patterning failed to replicate the
deleterious effects of hippocampal dysfunction. For example,
using an operant task that made it less likely that the animals
would process the cues as configural “scenes, Bussey et al.
(1998) found that fornix lesions significantly facilitated trans-
verse patterning. An identified weakness of a first experiment
was the relatively poor performance of sham-lesioned rats on
the critical third phase when all three problems were trained
together. This was addressed in a second study with all three
phases trained concurrently. The sham-lesioned group was
now above chance, but the trend toward a fornix lesion-
induced facilitation remained. We come back to transverse
patterning in the discussion (below) of relational processing
theory.
Biconditional discrimination tasks can also be solved using
configural cues, but unfortunately they may also be learned
through the application of a conditional rule. Whishaw and
Tomie (1991) trained rats to pull on strings of different widths
(A, B) and different odors (C, D) in combinations that led to
reward (e.g., AB, CD) or no reward (e.g., AC , BD ).
These pairings meant that any one elemental cue predicted

Figure 13–35. Transverse patterning. A. Rats were trained to swim toward hanging cue cards that
were either placed in front of a possible escape location (e.g., A) or merely hanging over the pool
(e.g., B ). The spatial locations of A and B would be randomized across trials. B. The stimu-
lus patterns used for the full transverse patterning problem. C. Deficit in hippocampus-lesioned
rats (H) compared to sham lesion controls. (Source: After Alvarado and Rudy, 1995b.)

A Apparatus B Stimuli C Performance

one pair at a time all pairs together

100 100
90 90
% correct

% correct

80 80
A B 70 70
60 60
50 50
+ _ 40 40

100 100
90 90
% correct

% correct

80 80
70 70
60 60
50 50
+ _ 40 40
NC H 100
90
% correct

80
70
60
50
+ _ 40

reward and nonreward equally often, but the creation of four
configural cues would enable a solution. Here the configural-
association theory fared less well: Rats with hippocampal
damage and sham-lesioned animals were both able to learn to
pull the correct string to get the reward. Similar negative
results had also been found in a similar earlier study using pri-
mates (Saunders and Weiskrantz, 1989).
The fourth category of ambiguous problem that Suther-
land and Rudy (1989) considered is the feature-neutral dis-
crimination task. An explanation of the name for this protocol
may help clarify what is being examined. The “feature” of the
task is a particular stimulus; let us call it stimulus A (as in Fig.
13–34). In conjunction with stimulus C, the feature A signifies
that a reward is available. Conditioning should lead to stimu-
lus A acquiring some associative strength. However, in con-
junction with stimulus B, feature A now signifies that a reward
is unavailable. Stimulus A should therefore become
“inhibitory.” By scheduling all four types of trial, stimulus A
should remain “neutral” with respect to its scheduled associa-
tion with reward (it is as often paired with reward as not)
while still being the critical feature that enables the rat to learn
the discrimination. Hence, it is called a feature-neutral dis-
crimination task. As in negative patterning, no combination
of elementary associations between stimuli and a reward can
solve the problem, and the configural association theory
uniquely predicts that rats with hippocampal lesions should
fail. Unfortunately, two studies using this design have both
found that rats with hippocampal lesions can learn it. In fact,
Gallagher and Holland (1992) found a facilitation of some
aspects of feature-neutral discrimination learning by rats with
neurotoxic hippocampal lesions, a result that echoes Bussey et
al.’s (1998) results on transverse patterning after fornix
lesions.
Taken together, this pattern of results from the four sets of
tasks provided only qualified support for the configural asso-
ciation theory. The results of the negative and transverse pat-
terning tasks sometimes upheld predictions of the theory,
whereas the outcome of the biconditional and feature-neutral
discrimination tasks did not. It is perhaps worth reflecting
further on the curious paradox that whereas a normal rat can
solve these problems we ostensibly smarter humans often
struggle to understand them. The struggle is nonetheless
worthwhile, as understanding ambiguity and its role in learn-
ing is important. All these tasks are ones in which an implicit
learning device is assumed to encode a stimulus configuration
that acquires associative strength through its consistent asso-
ciation with reward. Ostensibly, no new principles of associa-
tive learning are involved—only a configuring device. The
question therefore arises of whether this device is likely to be
in the hippocampal formation. One reason to be skeptical is
that the anatomical and electrophysiological data discussed in
Chapters 3, 4, and 8 strongly suggest that the sensory infor-
mation to which the hippocampal formation has access is
already polymodal and already highly processed. Therefore, a
neurobiological concern about the theory, quite apart from
the inconsistencies in the behavioral data, is not whether stim-
Theories of Hippocampal Function 661
ulus configurations are ever formed but that their construc-
tion is more likely to have taken place in the neocortex.
Revised Theory
In response to the mixed picture of initial results, Rudy and
Sutherland presented a revised theory that wisely put the con-
figuring part of the system outside the hippocampus in neo-
cortex. However, they still asserted that the hippocampus
plays a role in learning configural associations by “selectively
enhancing the salience of configural representations” (Rudy
and Sutherland, 1995, p. 382). A yet later change in the theory
emphasizes, in a similar vein, the role that the hippocampus
could play in rapidly separating similar patterns for effective
processing in the cortex (O’Reilly and Rudy, 2001). These
changes were in response to observations that ambiguous
tasks that could eventually be learned by animals with hip-
pocampal damage were often learned more slowly than by
controls. It was a set of changes that would rescue aspects of
the theory but at the cost of making it more difficult to test
definitively. In some respects, this is unfortunate because,
arguably more than other theorists, Rudy and Sutherland have
been rigorously honorable in demanding explicit predictions
and definitive tests.
Going down with the ship is all very well, but a key issue is
whether the logical structure of tasks, such as those laid out in
Figure 13–34, actually constrains the way in which an animal
solves any one task. The original version of the configural
association theory implied that this was the case. However,
when comparing negative patterning with feature-neutral dis-
criminations and biconditional tasks, Rudy and Sutherland
(1995) noted a parametric gradation across the tasks. This
gradation was from negative patterning that absolutely
required animals to learn to withhold responding to a config-
ural stimulus composed of two elements (i.e., AB ) that are
also presented alone and reinforced (A, B) to bicondi-
tional tasks in which none of the elements are ever presented
alone and reinforced (i.e., always X with Y and either  or ).
They argued that this parametric variation is important
because it relates to the extent to which individual stimuli
acquire associative strength during conditioning that is likely
to conflict or oppose the strength that should be shown by
configural stimuli. If this gradation is significant, it might be
that the task with the greatest conflict would be definitively
impaired by hippocampal damage, whereas that with less con-
flict would be least affected. To some extent, the data available
in 1995 did fit this pattern.
However, one aspect not considered was the possibility that
distinct CSs may not interact only by forming configural cues
but in more subtle ways as well. For example, under some cir-
cumstances, one CS may modulate or “set the occasion” for
the other CS to enter into a conditioned association. A good
illustration of this is the feature negative discrimination task in
which a stimulus A is reinforced on its own A but not rein-
forced when B is presented before A (i.e., B → A ). Under
these circumstances, instead of an AB configural stimulus
662 The Hippocampus Book
being created (which may happen when all the stimuli are pre-
sented concurrently), B acts in a different manner, modulating
whether the presentation of A retrieves a memory of the US or
the absence of the US. Occasion-setting is an associative
process in which one CS serves as a signal for an association
between other stimuli in a hierarchical manner. Work by
Holland and his colleagues indicates that hippocampal lesions
can disrupt one type of negative occasion-setting (Holland et
al., 1999), raising the possibility that the memory traces
responsible for conditioned associations between stimuli are
encoded and stored in the cortex (including configural stim-
uli) but can be subject to hierarchical control by stimuli
processed in the hippocampus. However, an alternative inter-
pretation remains feasible. This is that although the hip-
pocampus is not involved in forming excitatory CS-US
connections it is involved in inhibitory learning (i.e., that a
particular CS does not lead to an otherwise expected US).
Chan et al. (2001) had somewhat unfashionably defended an
inhibitory view of hippocampus function. They looked on the
feature negative occasion-setting task as one in which, in nor-
mal rats, a CS forms both excitatory and inhibitory associa-
tions, and the occasion-setter helps to choose between them.
The disruption caused by hippocampal lesions may then be a
deficit in inhibitory, rather than hierarchical, occasion-setting.
The problem with this view is that there are many occasions in
which hippocampal lesions have no effect on the extinction of
CS-US associations—the classic test of an inhibition theory
(Wilson et al., 1995).
In the further refinement of the configural framework,
O’Reilly and Rudy (2001) have semantically recast the theory
to refer to “conjunctive representations” rather than “configu-
rations” and suggested that there are two kinds of conjunctive
learning. One type is thought to occur incidentally or auto-
matically, irrespective of task demands; this is the type that is
rapidly learned and involves the hippocampus. The other type
is slower and more deliberate, and the kind of learning that
emerges as a consequence of a deliberate type of problem-
solving. This new incidental versus deliberate framework is
also explicit about the contribution that local circuits in the
hippocampal formation make to algorithms such as “pattern
separation” and “pattern completion” (in the dentate gyrus
and area CA3 respectively—see Chapter 14), and to the simi-
larities between tasks involving configurations of individual
CSs with context-dependent associations. We return to issues
to which these revisions of the theory are relevant in Section
13.6, together with reference to the important issue of “auto-
maticity” in hippocampus-dependent learning.
13.5.2 Relational Processing Theory: Refinement
of the Declarative Memory Theory
A significant feature of memory is the ability to recall facts
and events in circumstances different from those in which the
information was acquired in the first place. The inherent “flex-
ibility” of this form of memory was particularly emphasized
in a revision of the declarative memory theory advanced by
Cohen and Eichenbaum (1993) and later revised by them
(Eichenbaum and Cohen, 2001). It has been developed further
in relation to episodic memory processing (Eichenbaum,
2004), which we address later. The theory also places empha-
sis on the encoding of “relations” between stimuli. This is
quite different from forming configurations (as just dis-
cussed), and more akin to the idea of associations between
facts, such as that one fact can remind us of another. The
information processing necessary for “flexible relations” is
held to occur at the time of encoding but only becomes appar-
ent when retrieval is required in circumstances different from
those of the original learning. Like the original version of the
declarative memory theory, this theory holds that relational
processing is within the capacity of (at least) “higher” mam-
mals, that it depends on activity in the MTL, and that the
engagement of the hippocampal formation is time-limited.
“Relational processing” is held to be implemented by the hip-
pocampal formation, whereas anatomically distinct neocorti-
cal regions of the MTL mediate the processing and storage of
individual stimuli. This theory led to a number of distinctive
lines of experimentation that require discussion separate from
those already outlined in the discussion of the declarative
memory theory but to which the configural association theory
is, as we shall see, also relevant. The distinctive propositions of
the theory are in Box 13–7.
“Relational representations” were defined as memory rep-
resentations that are “created by and can be used for compar-
ing and contrasting individual items in memory, and weaving
new items into the existing organization of memories. This
form of representation maintains the compositionality of the
items, that is, the encoding of items both as perceptually dis-
tinct objects and as parts of larger scale scenes and events that
capture the relevant relations between them” (Cohen and
Eichenbaum, 1993).
The gist of the first proposition is that the relations that
humans and animals are able to store and recall go beyond the
mere fact that two stimuli were experienced together in tem-
poral contiguity—the “CS predicts US” type of association so
extensively studied in conditioning paradigms. “Comparing
and contrasting” implies that relations between stimuli can be
more sophisticated. They may be causal, but they may also
Box 13–7
Relational Processing Theory
1. Declarative memory generally involves processing the
relations between different items. Relational processing at
encoding enables flexible access to information in situa-
tions quite different from those of the original learning.
2. Relational processing is carried out by the hippocampal
formation, but storage of individual items in intermediate
memory takes place in the perirhinal and parahippocam-
pal cortex.
3. The role of the hippocampus in memory is temporary, as
in the declarative memory theory.

relate to physical relations (e.g., that A is near to, or bigger
than, B; as in “Versailles is near Paris”) or to abstract relations
(as in “the ugly sisters were unfriendly to Cinderella”).
Moreover, relations can extend beyond one-to-one pairings to
elaborate networks of interconnections between facts and
events—the very networks that constitute our personal
understanding of the world. The theory’s proponents asserted
that their view echoes William James, who described memory
as an elaborated network of associations that can be applied
across a broad range of situations (James, 1890). Although the
circumstances or conditions of learning are not well specified
in the theory, in the manner of contemporary learning theory,
the emphasis on relations is clearly a step beyond reference to
the “facts and events” in Squire’s (1992) original version of the
declarative-memory theory. Cohen and Eichenbaum (1993)
have put their finger on an important issue; and, indeed,
Squire has now incorporated some of their ideas into his
modern writings (e.g., Squire et al., 2004).
Whereas some types of factual information tend to be
recalled in a rote-like fashion, as in a child’s recitation of
arithmetic (“two-plus-two is four, four-plus-four is eight . . .”),
remembering facts and events is often deductive. They may be
recalled in different contexts from those in which they were
originally experienced and in a manner that permits access to
other perceptual features of a current situation. Indeed, cer-
tain “facts” may not have been stored at all (such as knowing
the number of windows in your house, which most people
work out inferentially by mentally walking through the build-
ing and counting them). The overall network of relations does
not need to be present or to be recalled to access a smaller sub-
set of facts. Flexibility also allows deductive inference. An
illustrative example is inferring geographical relations.
Suppose an American schoolboy living in South Carolina has
learned in school that “Paris is the capital of France,” that
“Madrid is the capital of Spain,” and that “France is farther
north than Spain.” Based on these facts, the child should be
able to infer the additional fact that “Paris is north of
Madrid.” If then told that Madrid is on the same latitude as
Washington, DC, he might, assuming he knows his American
geography, also draw the further inference that “Paris must be
north of where I live because Madrid and Washington are on
the same latitude.” This illustrates how new spatial facts can be
inferred even though never explicitly learned. Of course, flex-
ibility is a hallmark of the cognitive map theory as well, but
there it is held to be a specific, unique property of spatial
memory. The supposition now is that flexibility includes the
spatial domain but extends well beyond it to size, social rela-
tionships, and yet others. The information processing pro-
vided by the hippocampus is not domain-specific.
The second and third propositions of the theory concern
the relation between the hippocampal formation and neocor-
tex. Like Squire’s theory, Cohen and Eichenbaum argued that
damage to the hippocampal formation causes loss of the
capacity to store new explicit memories without having an
effect on implicit memory (i.e., for stimulus relations of which
we are not consciously aware. They agreed that hippocampal
Theories of Hippocampal Function 663
lesions spare the acquisition and expression of motor, proce-
dural, and cognitive skills. The refinement they offer is what
amounts to fractionation of declarative memory into two sub-
components: (1) the processing and storage in intermediate
memory (ITM) of stimulus items in isolation carried out in
the perirhinal and parahippocampal gyrus; and (2) the pro-
cessing of stimulus relations in such a way as to enable flexible
access, carried out by the hippocampal formation (Fig.
13–36). As long-term storage is held to be in other areas of
association cortex, the stimulus representations in the MTL
are asserted to be “compressed.” This is a metaphor taken from
modern computer science; but a claim with no direct evidence
to support it requires a much deeper understanding of neural
representations in the hippocampal formation than we
presently possess.
There are three other features to highlight. First, because
spatial relations are only one of the many types of flexible rela-
tions that humans and other mammals can encode, the theory
embraces some of the positive evidence hitherto held to sup-
port the cognitive map theory in an ostensibly more general
theoretical position. In fact, Eichenbaum (1996) argued that
spatial learning fractionates into flexible and inflexible forms,
with only the former sensitive to hippocampal disruption
(reminiscent of Olton’s “working memory” theory in which
spatial reference memory is insensitive to such lesions). This
may account for successful spatial learning by hippocampus-
Figure 13–36. Three functional components of the relational pro-
cessing memory system: cortical-hippocampal connections showing
the three main components. The cortical areas store short-term
memory (STM) and long-term memory (LTM) traces of specific
items; the parahippocampal region serves as intermediate memory
(ITM) for specific items and does the job of cue compression. The
hippocampal formation computes relational representations in a
manner that enables representational flexibility.
Relational processing theory - processes
and anatomical mediation
Cortical areas
STM & LTM storage
Parahippocampal region
ITM
cue-compression
Hippocampus
relational representations &
representational flexibility
664 The Hippocampus Book
lesioned rats during overtraining (Morris et al., 1990) because
after extended training the animals can learn discrete
approach responses from each of several starting locations
in the pool. However, this finding is also consistent with the
hippocampus-independent “taxon” approach strategies of the
cognitive map theory. Second, the theory states that the role of
the hippocampus in the formation of flexible representations
is time-limited. Such representations are then consolidated in
the neocortex, where they can be accessed flexibly without
further hippocampal input. Thus, the machinery for creating
flexibility is in the hippocampus, but (somewhat clumsily) the
machinery for flexible recall is in the neocortex. Third, if the
hippocampal formation is damaged, stimulus items that are
presented together may become fused into configurations.
Associations can still be created, but the theory supposes that
they are not relational but always configural. This idea capi-
talizes on the same kind of data that led Rudy and Sutherland
(1995a) to propose their revised theory.
Evidence for Hippocampus-dependent
Relational Representations and Flexible Access
The first study directly investigating relational processing in
humans involved brain imaging using positron emission
tomography (PET) (Henke et al., 1997). This study required
subjects to look at slides containing a person or a house and
asked them to judge the gender of the person and whether an
inside or outside view of the house was being displayed or to
guess whether the person was likely to live in the house or be
a regular visitor to it. Only the latter instructions encouraged
“comparing and contrasting” of the two pictures and so trig-
gered relational processing. Comparison of the relative brain
activation in the two conditions revealed greater activation in
the right MTL during the relational condition. This finding
supports the relational processing account, all the more so
because the associations formed were nonspatial. Other
human studies, using fMRI, have partially supported the idea
that relational processing can drive hippocampal activation
(Cohen et al., 1999). However, in an explicit comparison of
two matched spatial and social relationship tasks, Kumaran
and Maguire (2005) saw activation of the human hippocam-
pal formation only during the spatial relation task. A network
of real people, some of whom were friends (or friends of
friends) and all of whom lived at various places in the same
city, was identified in an initial briefing session. The main
tasks to be conducted in the scanner included getting a crate
of wine from one person to another (together with various
control tasks). In the social version of the main task, the crate
could be passed from friend to friend. In the spatial version, it
was passed to the nearest neighbor. Hippocampus activation
was observed when subjects were required to focus on and
navigate spatially (e.g., their friend’s houses). A quite different
network of brain areas was activated when subjects wandered
mentally within their social network—retrieving knowledge
about their friends and their relationships to each other—
including the medial prefrontal cortex, insula, superior tem-
poral sulcus, and other areas implicated in a proposed
circuitry for social cognition (Adolphs, 2003). In a further set-
back for the relational processing account of human hip-
pocampal function, a sharp distinction between item and
associative memory is also not fully supported by other con-
temporary fMRI studies, as reviewed in Chapter 12.
Hippocampal activation can sometimes be seen with individ-
ual memory items, particularly when they are novel, though
the novelty-driven nature of such activations is often the
unusual spatial context in which a stimulus is presented and
thus cryptically relational.
Animal studies have played a bigger part in driving the the-
ory. Bunsey and Eichenbaum (1996) developed an odor-
guided paired-associate learning task for rats to examine
whether learned information could be used inferentially in
novel testing situations. The animals could dig through food
cups containing a mixture of ground rat chow and sand to
secure a fruit-loop cereal reward. One of several odors was
added to the sand/chow mixture so the rats’ experience of dig-
ging was associated with the odor. The animals took to the
task quickly, perhaps because digging for food is as much a
part of their natural foraging strategy as remembering where
they are (Fig. 13–37).
Paired-associate learning tasks are most extensively used in
human cognitive psychology and typically involve presenta-
tion of the first member of each pair (the so-called cue item),
followed by either free recall of the second item or a two-alter-
native forced-choice test between two potential associates of
which only one is correct. The pairs may be word pairs, words
and faces, or other combinations of stimuli. Bunsey and
Eichenbaum’s paired-associate training protocol for rats used
the two-alternative forced-choice test. The initial training
phase consisted of digging through the sample item, in which
a single digging cup with a single odor was presented (e.g.,
cocoa  cue A), followed immediately by the two-alternative
choice test in which two cups were presented and the animal’s
task was to dig in the cup whose odor was to be associated
with the initial cue item (e.g., coffee  cue B). That is, follow-
ing cue A the animal was rewarded during the choice phase if
it dug through cue B but was unrewarded if it dug through the
other cup (e.g., salt  cue Y). Conversely, if the animal was
presented with turmeric as a first-item cue (cue X), digging in
the salt cup (cue Y) in the choice phase secured reward
whereas digging through cue B (coffee) did not. Having
learned in this way that A goes with B and that X goes with Y
(over many training trials), a second set of paired associates
was then taught in which the rats learned that odor B goes
with C (not Z) and that odor Y goes with Z (not C). Sham and
ibotenate hippocampus-lesioned rats learned both sets of
“premise” paired associates in this way quite rapidly and at the
same rate (Fig. 13–37B). An ingenious feature of the experi-
ment was inherent in the design. The two sets of paired asso-
ciates deliberately contained common elements: odors B
(coffee) and Y (salt). This commonality afforded the opportu-
nity of exploring whether rats that had learned that A
goes with B (and B goes with C) could access these stimu-
lus–stimulus representations flexibly and so reveal transitive
knowledge about the relation between A and C. In a probe test
 Theories of Hippocampal Function 665

A Odor paired associates B Normal learning of paired associates C Hippocampal deficit in transitivity

sample
90 0.45

80 0.35
70 0.25 C
C

errors to criterion
H

preference index
60 H 0.15

50 0.05
choice 40 -0.05

30 -0.15

20 -0.25

10 -0.35

0 -0.45
AB BC
XY YZ training set
training set 1: AB & XY
sample A X
D Primate study of inference in paired-associate learning
100
relational discontinuous
choice B vs Y B vs Y
training set 2: BC & YZ 90 C
Entorhinal
% correct

sample B Y
80

choice C vs Z C vs Z
70
training set 2: AC & XZ
sample A X 60
? ?
choice C vs Z C vs Z 50
premise probe premise probe

Figure 13–37. Flexible retrieval of learned paired-associate infor-
mation. A. The rat approaches and digs through the scented
sand/chow mixture during a “sample” trial until it gets its fruit-loop
reward. A short while later, it is presented by two other scented food
cups in a “two-alternative forced choice” test. The scent associated
with the previous sample indicates in which of the choice cups the
animal should dig to get another reward. B. Sham-lesioned and
lesioned rats were equally capable of learning the odor-paired asso-
ciations. The question at issue is the nature of the relations that
were encoded during this training. C. The test for transitivity
showed that the sham rats were above chance in choosing C given
A, and Z given X, whereas the hippocampus-lesioned animals could
not draw these inferences. (Source: After Bunsey and Eichenbaum,
1996.) D. Findings from an analogous primate study revealed selec-
tive impairment on the probe but not premise choice trials after
entorhinal cortex lesions. (Source: After Buckmaster et al., 2004.)

for this, the rats were first given either odor A or X as the ini-
tial single-cup cue item and then presented with a choice
between cups containing odors C and Z. The striking result
was that the sham-lesioned animals chose appropriately (i.e.,
C after A and Z after X), but hippocampus-lesioned rats did
not (Fig. 13–37C).
In further experiments also using the same sand-digging
method, Dusek and Eichenbaum (1997) considered the issue
of hippocampal involvement in building ordered representa-
tions using a classic Piagetian seriation paradigm. It was
adapted for use in work with primates by McGonigle and
Chalmers (1977). Later work established that pigeons and rats
could also behave “logically,” but Eichenbaum’s group was the
first to explore the anatomical substrate. Sham-, fornix-, and
perirhinal/entorhinal-lesioned rats were trained in a “transi-
tive inference” task to choose odor A over odor B and then
B C, C D, and D E using a training protocol that began
with successive trials of each discrete problem and ended with
the individual problems in random sequence (see Box 13–8).
After they had mastered these four problems, the animals were
presented with two other novel problems: choosing between
odors B and D and between A and E. Both involved a choice
between odors that had never previously been presented
together, but only the former pair (B D) required knowl-
edge of a serial order. This is because the “probe” odors B and
D had each served equally often as rewarded or as nonre-
warded odors during training, whereas the “end-anchor”
odors, A and E, were always positive or negative odors, respec-
tively. Sham-lesioned rats chose both B over D and A over E,
whereas rats with fornix or perirhinal/entorhinal lesions
could only do the A versus E comparison (Fig. 13–38).
Work in species other than the rat has revealed mixed sup-
port for the relational processing idea. A study in pigeons
revealed that they could learn the paired-associate task and
successfully solve the inferential probe tests, but hippocampal
lesions were without effect (Strasser et al., 2004). Using objects
in a WGTA, Saunders and Weiskrantz (1989) trained monkeys
on a biconditional task (like those described in the previous
666 The Hippocampus Book

Transitive inference task trols succeeded in doing this. This finding suggests that the
normal controls had acquired an abstract representation of
the object pairings that was more “flexible” than that acquired
by the lesioned monkeys. The successful learning of the bicon-
100 C ditional task with poor inferential probe test performance is
90 FX
arguably more consistent with the relational processing
Percent correct

PRER
80 account than the configural-association theory.
Buckmaster et al. (2004) trained monkeys on a series of
70
procedures intended to emphasize relational representations:
60 a paired associate task analogous to that of the Bunsey and
chance Eichenbaum (1996) study, a transitive inference task based on
50
the original primate work of McGonigle and Chalmers
40
(1977), and a novel spatial delayed recognition span proce-
BD AE dure. They were also received training in object discrimina-
relational end-anchored
probe pair probe pair tion learning and DNMS (simple associations and recognition
memory). Monkeys with large ( 90%) entorhinal lesions
Figure 13–38. Hippocampal participation in transitive inference. acquired and performed these latter tasks normally but were
Choice performance by controls and fornix-lesioned (FX) and
impaired on each of the three tasks that required relational
perirhinal/entorhinal-lesioned (PRER) groups during the two
processing. The entorhinal cortex-lesioned monkeys gradually
choice tests after seriation training. Both lesioned groups are at
chance on the B versus D comparison, but both perform as well as
acquired the difficult discriminations required in the paired-
controls on the A versus E end-anchored choice test. (Source: After associate task, but acquisition was slower than normal,
Dusek and Eichenbaum, 1997.) suggesting that a different, less flexible configural learning
strategy was being used. Performance by normal monkeys
in both the premise and probe trials was more than 80% cor-
rect (Buckmaster et al., 2004), but there was selective impair-
section) followed by an interrogation of what the monkeys
ment on the probe trials after entorhinal cortex lesions (Fig.
“knew” about which object was related to which other object.
13–38D). The results on the transitive inference tasks were
The animals were trained, through a series of stages, with pair-
consistent with those obtained in rats, but aberrant poor
ings in which each object was equally often rewarded or non-
performance by one normal control monkey in this task
rewarded, but specific pairings were consistently rewarded
make it difficult to draw firm conclusions. The perennial
(e.g., AB, CD) or consistently nonrewarded (AC , BD ).
problem of the numbers of subjects used in primate studies is
Hippocampal disruption (fornix lesions or combined fornix,
often quite low. The spatial span task could not be learned
hippocampal, and mammillary body damage) did not affect
by the lesioned monkeys, a finding that would be consistent
acquisition of the task but did impair performance in the
with the cognitive map theory were it not for the deficits in
posttraining probe test of “what goes with what.” In this
the other relational tasks that were explicitly nonspatial.
probe, the monkeys were confronted with three objects (A, B,
These are important results, as studies of entorhinal cortex
C) and, having displaced A to receive a reward, got a second
lesions had previously failed to reveal any lasting deficit in
reward on the same trial only if they then displaced B (AB
DNMTS (Leonard et al., 1995), but the commonality of
having been a rewarded pair during training). Only the con-
the deficit here with hippocampal lesions is consistent with
the proposal made in Chapter 3 that the entorhinal cortex
should be viewed as a major component of the hippocampal
Box 13–8 formation.
Transitive Inference Premise Training A potential difficulty is that the transitive inference task is
a classic example of a procedure in which the devil is in the
A B Individual premise problems detail, both conceptually and procedurally (McGonigle and
B C trained separately Chalmers, 2003). Like other ambiguous tasks, it is a task that
C D
can be solved in several ways. Van Elzakker et al. (2003) noted
D E
at least four theoretical accounts of successful task perform-
ance and argued that a simple “excitatory stimulus value”
Ordered mental representation of relations
account cannot be ruled out. This may seem surprising as, on
the face of it, the novel B D probe compares two stimuli that
A B C D E
should be of equal excitatory stimulus value as they are
equally often paired with reward and nonreward during train-
Probe choice tests
ing. However, there are some cryptic asymmetries to the train-
B vs. D Test of transitivity
ing protocol that may render this assumption false. When
A vs. E Nontransitive novel pairing learning D E in the five-problem series, the pairing is with
a stimulus E that is never reinforced. D may therefore not have

to acquire a very high net excitatory stimulus value before the
difference between D and E crosses the threshold necessary for
criterion performance. Similarly, A is always reinforced, and
thus the A B difference might also be realized despite B
actually having a fairly high associative strength. Given this,
the relative values of B and D may be different despite the
ostensibly equal pairing with reward and nonreward. In addi-
tion, one might then expect a B versus D probe to be easier in
a four-problem series (A B, B C, C D, D E) than in
a five-problem series in which an E F problem is added.
Exactly these results were found by Van Elzakker et al. (2003),
who therefore queried whether relational processing and
inferential memory retrieval were being used to perform the
task. Computational modeling by Frank et al. (2003) has taken
this further, with a detailed treatment of the impact of the
“end-anchors” on the associative strength of stimuli with
which they are routinely paired, the role of configural repre-
sentations in the performance of normal (unlesioned) ani-
mals, and the balance between effects of “pattern separation”
and “pattern completion” in the task. They also queried the
adequacy of a relational processing account of transitive infer-
ence and made the interesting prediction that hippocampus
or dentate gyrus lesions created after training should have no
effect on performance of the BD probe, in contrast to the dele-
terious effect on choice when made before training. This is an
important prediction because, in contrast, the relational pro-
cessing theory would require the hippocampus to be intact at
the time of a probe test. This valuable pretraining versus post-
training comparison of the configural and relational process-
ing accounts has not yet been reported.
The lesion studies that have at least partially supported the
relational processing theory of hippocampus function have
been complemented by unit-recording data using similar par-
adigms (Wood et al., 1999). Multiple single-unit recordings
were taken of CA1 pyramidal cells firing during a 1-second
period just prior to a rat deciding whether to dig in a contin-
uous recognition variant of the paradigms just described.
Food cups containing sand were repeatedly presented to ani-
mals in different locations of an open arena. For trials in
which the cups were scented with a novel odor (half the tri-
als), the sand had a fruit-loop reward buried at the bottom; for
trials in which the odor was repeated (i.e., no longer novel),
no fruit loop was present. The rats quickly learned the logic of
the situation: to dig when it was worth it and refrain from
doing so when it was not. The unit-recording data taken dur-
ing the Shakespearean “to dig or not to dig” decision period
indicated that approximately one-third of the hippocampal
cells had task-related nonspatial correlates. Place-specific fir-
ing was certainly seen in some cells (about another third). Of
the nonspatial cells, some were associated with approach
behavior, some with approaching a particular odor irrespec-
tive of its location in space, and some with whether the trial
was with an odor that was familiar or novel (match/non-
match)—all relations of a nonspatial character. These data
indicate that the functional correlates of CA1 pyramidal cell
firing may depend in part on the task a rat is being trained to
perform, not just on its location in space. This is a deep idea
Theories of Hippocampal Function 667
that requires further investigation in analytically exacting
tasks. However, a follow-up study (Dudchenko et al., 2000)
revealed this continuous recognition task to be insensitive to
hippocampal lesions, again raising doubts about the extent to
which a “relational” description of the cellular correlates is
justified. The parallel to what is seen in NMR conditioning is
beguiling—as if the hippocampus mischeviously likes to “lis-
ten in” on things it does not have anything to do with. We later
argue that this is exactly what an incidental learning system
may have to do.
These experiments are taken by Eichenbaum and Cohen
(2001) to indicate that: (1) rats develop a knowledge of stim-
ulus relations that can be retrieved flexibly to guide inferences
from memory; (2) hippocampal cells are responsive to
such task-related attributes; and (3) the integrity of the fornix,
hippocampus, and perirhinal/entorhinal cortices are essential
for representing these attributes flexibly. This interpretation
clearly goes beyond the domain of both the original
declarative memory and cognitive map theories but leaves
open whether the hippocampus is engaged in relational pro-
cessing at encoding, retrieval, or both. As just noted in the
contrast with a configural association account of transitive
inference, making lesions before training does not distinguish
these two alternatives. Reversible inactivation, using drugs
that inhibit synaptic transmission cell firing, would also pro-
vide an opportunity to explore whether the hippocampus
needs be active at encoding and retrieval, as it does for spatial
reference memory (Riedel et al., 1999), and whether it plays a
time-limited role in systems-level memory consolidation as
the relational processing theory predicts. If time-limited, it
would imply that “flexibility” can be displayed by neocortical
circuits also, provided information is subject to both hip-
pocampus-dependent encoding and a period of hippocam-
pal-neocortical consolidation.
The notion of representational flexibility and its depend-
ence on the hippocampus is also illustrated by experiments on
the social transmission of food preferences using a paradigm
first developed by Galef and Wigmore (1983). Two rats are put
together for a short period. One of these rats, the demonstra-
tor, is arranged to have recently eaten a novel foodstuff. The
animals are then separated and, after a memory delay, the
other animal, the observer, is tested with two novel foods. One
novel food is the same as the one the demonstrator has eaten,
and the other novel food is not. In this two-alternative forced-
choice test, the observer rat shows a preference for the novel
food eaten by the demonstrator over the other (Fig. 13–39A).
Some social feature of the situation conquers the animal’s
usual neophobia. An early series of analytical studies (Galef,
1990) had established that this social transmission of food
preferences is mediated by carbon disulfide (CS2) in the breath
of the demonstrator rat that acts a “carrier” of the smell of the
food that the demonstrator has eaten. Although this social-
transmission task is not obviously relational in quite the same
way as the paired-associate task, it is flexible with respect to
how the learned information is expressed. Information about
the novel food is acquired by the observer on the breath of the
demonstrator rather than through actual consumption, and
668 The Hippocampus Book
A Odor-guided, paired associate learning
Phase I
demonstrator
Phase II
CS2+food odour
demonstrator observer
Phase III observer
?
Figure 13–39. Social transmission of food preferences. A. Odor-
guided paired associate learning in which information about a food
odor that is safe to eat (gray food) is acquired by an “observer” rat
smelling the breath of a “demonstrator” rat that has previously
eaten that food. This acquired knowledge is later expressed in the
this “knowledge” is later expressed by the observer rat.
Procedural learning systems in which knowledge is embedded
in performance would not support such learning.
Winocur (1990) discovered that this task is sensitive to
hippocampal lesions created either before or shortly after the
demonstrator–observer interaction. Faster forgetting, over 6
days, was observed in the animals lesioned before training. In
animals subjected to lesioning after the social transmission
phase but before the preference test, a consolidation gradient
was observed: Those with hippocampal lesions displayed ret-
rograde amnesia 1 day after lesion creation, whereas those
whose lesions were created 5 to 10 days later performed sig-
nificantly better (Fig. 13–39B). Bunsey and Eichenbaum
(1995) confirmed Winocur’s finding of faster forgetting using
rats with ibotenate lesions that encompassed the hippocam-
pus, dentate gyrus, and subiculum. Rats with more selective
hippocampus  dentate lesions showed little forgetting over
24 hours, but the longer 6-day interval used in Winocur’s
study was not examined and these more selective lesions were
incomplete at the temporal pole. In the light of data concern-
ing differential functions of the dorsal and ventral hippocam-
pus, it is possible that social transmission of food preference
could be one that is specifically sensitive to ventral hippocam-
pal lesions.
Evidence for Functional Dissociations Between
Anatomical Components of Declarative Memory:
Role of the Hippocampus
One point of departure between this and Squire’s (1992) ver-
sion of declarative memory theory has to do with functional
dissociations between components of the MTL memory sys-
tem. Squire argued against functional distinctions, except at
the input stage where some differentiation between the infor-
B Retrograde amnesia after hippocampal lesions
100 C
% correct (diet sample eaten)
H
90
80
70
60
50
0 2 5 10
consolidation period (days from training to test)
form of a preference in a two-alternative choice test over another
food (white). (Source: After Galef and Wigmore, 1983.) B.
Retrograde amnesia for social transmission after hippocampal
lesions. (Source: After Winocur, 1990.)
mation projected to the perirhinal and parahippocampal cor-
tex is recognized (Suzuki and Amaral, 1994a,b). In contrast,
the relational processing theory requires the hippocampal for-
mation to perform a role in declarative memory distinct from
that performed by the surrounding neocortical structures.
The latter structures are held to be an intermediate memory
store (ITM) for individual items, whereas the hippocampal
formation is the “relational processor.” This modification
accommodates the finding that DNMTS for trial-unique cues
is impaired by neocortical damage to the perirhinal and
parahippocampal gyrus but is unaffected (or only very mildly
affected) by hippocampal lesions. The new findings, discussed
above, support the relational processing modifications of the
declarative memory theory.
One point to note in passing is that, whereas most studies
of DNMTS have been conducted using visual cues and
objects, olfactory cues have been used extensively in
Eichenbaum’s rodent studies. Faster forgetting of olfactory
cues after entorhinal lesions was first observed by Staubli et al.
(1984). Otto and Eichenbaum (1992) confirmed this finding
using a continuous delayed nonmatching task (cDNM) with
odors. Fornix lesions had no effect, whereas perirhinal and
parahippocampal cortex lesions caused faster forgetting.
Thus, for the type of information used most extensively in the
studies on which the theory has been built (odors) an inter-
mediate memory does appear to exist in the neocortical
regions of the medial temporal area.
13.5.3 Contextual Encoding and Retrieval
The original version of spatial mapping theory focused on
how environmental landmarks of a specific context are
explored and integrated into spatial maps for the purposes of
navigation (O’Keefe and Nadel, 1978). However, “context” can

be important in ways other than as a space to be moved
around. Memories of discrete events could include associa-
tions between an event and its context, such as the place or
time at which the event happened (Nadel and Willner, 1980).
Generally, a context has also to be learned about before the
representation of it can enter into other associations
(Fanselow, 1990), but although this representation is often
spatial it need not be (Jeffery et al., 2004). It could include
temporal cues, particularly during extinction of condi-
tioned associations (Bouton, 2004); it may include feelings
associated with administered drugs (Overton, 1964) or natu-
rally occurring motivational states such as hunger or thirst
(Davidson, 1993). A context may also be “cognitive” as in the
mental frameworks or schema associated with particular
classes of problem (Morris, 2006). The importance of context-
dependent memory is well demonstrated by occasions on
which it fails, such as when we embarrassingly fail to recog-
nize a person we have recently met in a different context.
Contextual encoding is clearly an effective, often automatic
associative mechanism for binding events into memory.
As shown in Figure 13–33B, the modulation of memory by
contexts can be mediated by distinct associations and associa-
tive processes (Balsam and Tomie, 1985; Bouton and Moody,
2004). As with the formation of stimulus configurations or
relational associations, contexts can influence the attention
paid to a stimulus, “set the occasion” for a stimulus to predict
one outcome or another, disambiguate stimuli and their pre-
dictive significance in other ways, or provide an incidental or
deliberate way of organizing attended information. In general,
the ability to form context-event associations is parasitic upon
animals having previously developed a memory representa-
tion of the context if that is necessary (which may take time)
and upon the occurrence of rapid one-trial learning (because
unique events do not happen twice).
It turns out that some of the distinct aspects of contextual
encoding, modulation, and recall depend on the hippocam-
pus, but others do not. Accordingly, a simple formula such as
“context-learning  hippocampus-dependent learning” is
inadequate. Experiments on the role of the hippocampal for-
mation in context-associated information processing have
now developed a number of subtleties, and their discussion
leads us on toward a more “episodic” account of hippocampal
function.
Context Fear Conditioning
A learning paradigm called “context fear conditioning” is
now widely used to study context-event learning. It is rap-
idly learned (Fanselow, 1990), generally though not always
hippocampus-dependent (Winocur et al., 1987; Selden et al.,
1991; Gisquet-Verrier et al., 1999), and has proved ideal as a
rapid assay of lesion, drug, and forebrain-specific genetic
manipulations that might differentially affect short- and long-
term memory for fear and the signal-transduction mecha-
nisms engaged in its consolidation (Aiba et al., 1994;
Frankland et al., 2001; Ohno et al., 2001).
Theories of Hippocampal Function 669
Fear conditioning to a discrete stimulus is mediated by the
amygdala (Davis et al., 1993; LeDoux, 2000; Maren, 2005).
Whether shown as an altered startle response or somatic
immobility (“freezing”), circuits interconnecting the lateral,
basolateral, and central nuclei in the amygdala have been
implicated in fear conditioning using lesion, pharmacological,
single-unit, and LTP studies. However, the amygdala interacts
with the hippocampus in certain forms of fear conditioning.
The most widely used “context-freezing” task, as developed
by Fanselow, involves three conceptually separate but often
overlapping phases (Fig. 13–40A). First, the rat or mouse is
placed into a distinctive box with a grid floor (the context)
and allowed to explore it with a view to forming an integrated
representation of its shape, appearance, odor, and somatosen-
sory characteristics. This learning takes a minute or two, a
fact that points to a somewhat neglected link between this
form of associative conditioning and exploratory behavior.
Second, a weak or intermediate-intensity electric shock is
delivered to the animal through the grid floor. This may be
during the same session as the initial exploration or a later one
(protocols differ slightly across laboratories). In either case,
the animal’s reaction to the shock may be barely detectable, it
may run around for a bit, or it may immediately remain
immobile. Whatever its immediate or “unconditioned” reac-
tion to the shock, this behavior gradually gives way to sus-
tained periods of immobility in which practically the only
observable movement is breathing. This conditioned reaction
(CR) is a “species-specific defense reaction” (Bolles, 1970)
shown by various rodent species and is generally called “freez-
ing.” The probability of freezing varies as a function of the
number and intensity of the shocks delivered. If the initial
exploratory period is not included, contextual fear condition-
ing does not occur (Kiernan and Westbrook, 1993), a phe-
nomenon that Fanselow (1990) referred to as the “immediate
shock effect” (Fig. 13–40B). However, when sufficient
exploratory time in a context is allowed, subsequent fear con-
ditioning occurs successfully. This learning phase ends with
the animal’s return to its home cage. The third phase of test-
ing involves returning the animal to the context at a pre-
scribed memory interval after training (e.g., the next day).
The experimenter monitors the level of freezing (Fig.
13–40C). Little freezing is seen during a postoperative mem-
ory test in rats subjected to hippocampal lesioning soon after
context fear conditioning. However, the extent of freezing
increases systematically as a function of the interval between
context fear conditioning and the time when a hippocampal
lesion is created (Kim and Fanselow, 1992). Animals that have
conditioned well typically freeze for as much as 60% of a short
testing session. This freezing response is only slowly forgotten
over time (weeks or longer) and only extinguished by repeated
context exposures in the absence of shock. Studies of the
extinction of conditioned fear, including contextual fear, are
currently a focus of considerable interest for their transla-
tional implications for the treatment of posttraumatic stress
disorder (Cahill and McGaugh, 1996; Ressler et al., 2002;
Nader, 2003).
670 The Hippocampus Book
A Contextual fear conditioning
context
exploration
conditioning
electric shock
B Immediate shock effect C Context freezing
30 5 100
freezing %
4 80
defecation (#)
defecation (#)
freezing %
20
freezing %
3 60
2 40
10
1 20
0 0
1 3 9 27 81
placement-shock interval (s)
Figure 13–40. Context fear conditioning. A. The three phases of
training: context exploration, fear conditioning, and posttraining
memory testing (Source: After Fanselow, 1990.) B. The “immediate
shock effect” that reflects the absence of context fear conditioning
unless sufficient time is allowed to explore the context (Source: After
Although by far the most popular, freezing is not the only
measure used to assess fear conditioning. Other measures
used include the relative preference for a safe unshocked
chamber compared to the conditioning chamber to which it is
connected or the extent to which fear, conditioned to a dis-
crete CS, inhibits appetitive or consummatory behavior.
Selden et al. (1991) examined both the extent to which a
clicker stimulus (the discrete CS) would evoke fear after pair-
ings with shock and the extent to which an initially nonpre-
ferred and brightly lit white chamber would be favored over a
black one in which the clicker and shock had been presented
(the safe and conditioning contexts, respectively). They
observed a double dissociation between the effects of excito-
toxic amygdala and hippocampal lesions on fear conditioning
to explicit and contextual aversive cues. Amygdala lesions
selectively impaired conditioning to the clicker measured by
its capacity to reduce drinking (i.e., inhibition of consumma-
tory behavior). Hippocampal lesions selectively impaired con-
ditioning to the context measured by the proportion of time
spent in the black and white chambers in a preference test
(Fig. 13–40D). Independent work by other groups confirmed
the double dissociation between the dependence of contextual
post-training
memory test
!!!!!! !!??
D Preference for a safe context
freezing level place-preference
250
C C
H
seconds on safe side
200 H
150
100
50
0 0
1 7 14 28 days
Fanselow, 2000.) C. The extent of freezing as a function of the inter-
val after context conditioning before a hippocampal lesion is made.
(Source: Kim and Fanselow, 1992.) D. The impact of hippocampal
lesions on place-preference as a measure context fear. (Source:
Selden et al., 1991.)
freezing on the integrity of the hippocampus and of tone-
elicited freezing on the amygdala (Kim and Fanselow, 1992;
Phillips and LeDoux, 1992). The hippocampal dependence of
contextual freezing seems to be unrelated to the level of activ-
ity the animal displays in the conditioning chamber prior to
shock delivery (Gewirtz et al., 2000). This lack of correlation
is important, as it is known that hippocampal lesions can
sometimes cause hyperactivity (e.g., during exploration of a
novel environment), and the deficit in context freezing might
therefore have been no more than a secondary, nonmnemonic
consequence of a lesion-induced alteration in motor behavior.
Two arguments marshaled by Fanselow (2000) against this
possibility include the selectivity of the hippocampal lesion
deficit to context but not discrete CS-associated fear and the
presence of a within-subject temporal gradient of retrograde
amnesia for discriminable contexts (see below). Thus,
although other measures have been developed that would not
be subject to the hyperactivity objection, freezing has retained
its popularity, and the research field has stuck to it.
The difference between fear conditioned to a discrete
CS (amygdala-dependent) and to a context (hippocampus-
dependent) reflects the prior learning of the context. Fanselow

suggested that the rat must first form “an integrated
mnemonic representation of the many features of the context
and the rat must have this representation in active memory at
the time of the shock” (Fanselow, 2000, p. 75). He likened it to
a “Gestalt” memory of the environment, analogous to the spa-
tial maps that O’Keefe and Nadel (1978) had suggested are
formed during active exploration. It is therefore noteworthy
that this necessary time period of exposure to a novel condi-
tioning chamber is similar to that which Bostock et al. (1991)
first reported to be necessary for rats to form a stable hip-
pocampal firing field in a novel recording chamber. This “uni-
fied representation” of the environment then functions as an
“configural” stimulus that, through classical conditioning, is
associated with shock—in a manner essential identical to that
implied by the configural association theory. Rudy and
O’Reilly (1999) provide direct support for this view in exper-
iments with normal rats in which they showed that preexpo-
sure to the conditioning context, but not to the individual
stimulus elements that made up the context presented one
after the other, facilitated later contextual fear conditioning.
They also established that the extent to which fear condition-
ing generalizes to other contexts that had not been used for
conditioning is influenced by preexposure in a manner sug-
gestive of “pattern completion” during conditioning (see
Chapter 14). Specifically, if contexts X and Y contain common
elements and context Z is quite different, preexposure to con-
text X but not to context Z facilitates generalization to context
X after fear conditioning in context Y. These observations,
reminiscent of Guzowski’s data obtained using the immediate
early gene Arc (see Section 13.4) suggest that contextual freez-
ing reflects fear elicited by a “stimulus” that is a mental entity:
It is a conjunctive memory representation of the environment
based on polymodal sensory information. This led Rudy to an
ingenious test of his context preexposure facilitation (CPF)
concept. Rats were led to expect, over a series of exposure
days, that being transported in a distinctive container around
the laboratory would always end in them being taken to a par-
ticular test chamber (context A). On the critical conditioning
day, they were transported in the container but unexpectedly
put into a distinctively different test chamber (context B) and
given an immediate shock. Remember that immediate shock
does not ordinarily result in context fear conditioning; and,
consistent with this, no contextual fear conditioning occurred
to context B. However, the rats displayed freezing when placed
in context A, the context they had expected to arrive at after
transport in the distinctive container but not the one in which
they actually received the shock. This “CPF effect” was not
observed in rats with hippocampal lesions (Rudy et al., 2002)
A complication to the story that the hippocampus is criti-
cal only for contextual fear conditioning is that several studies
have shown that disruption of the ventral hippocampus (with
lesions or drugs) can disrupt fear conditioning to a discrete CS
such as a tone (Bast et al., 2001; Zhang et al., 2001). This may,
in part, be due to the presence of direct afferents from the ven-
tral hippocampus to the amygdala (Pitkanen et al., 2000) that
ordinarily mediates CS fear conditioning. Whatever the rea-
Theories of Hippocampal Function 671
son, a one-to-one association between hippocampal process-
ing and context freezing is an oversimplification.
Differential Effects of Hippocampal Lesions
Before and After Fear Conditioning
The importance of exploratory learning about the context is
relevant to a puzzling feature of studies examining the impact
of hippocampal dysfunction. Hippocampal lesions made
shortly after conditioning are effective in blocking contextual
freezing (Kim and Fanselow, 1992; Phillips and LeDoux, 1992,
1994; Maren et al., 1997; Frankland et al., 1998), whereas those
made prior to conditioning have varied effects (Winocur et al.,
1987; Maren et al., 1997; Gisquet-Verrier et al., 1999; Maren,
2001). This dissociation with respect to the relative timing of
training and the lesion is not because the lesions are incom-
plete (although often they are). Clearly, were they incomplete,
there may be sufficient spared tissue for hippocampus-
dependent learning to occur when the lesion is made prior to
conditioning. That this is not the explanation for the differen-
tial effects of lesions made pretraining versus posttraining is
because if an identical incomplete lesion is made after train-
ing a substantial deficit in context fear conditioning is still
observed.
The likely explanation for the differential effect, as outlined
by Maren et al. (1997), Rudy and O’Reilly (1999), and
Anagnostaras et al. (2001), is that an animal with a hip-
pocampal lesion prior to conditioning is predisposed to asso-
ciate the fearful shock with some “elemental” feature of the
conditioning chamber (e.g., its smell) rather than the “unified
representation” that because of the lesion it is unable to form.
As this type of single-stimulus classical conditioning is unaf-
fected by hippocampal lesions, less disruption of contextual
freezing is to be expected in the anterograde domain. This
possibility alerts us to the importance of the exact design of
the chambers in which these sorts of experiments are con-
ducted. Different contexts are sometimes readily distinguish-
able in terms of an elemental feature; in other cases, they may
only (or more readily) be distinguished by the set of cues that
forms a single unified representation. The devil is again in the
detail with respect to understanding the pattern of results,
particularly in relation to the impact of genetic manipulations
that, unless inducible, occur long before conditioning. As
inducible, region-specific mutations are rare, we do not
include a discussion here of this developing and important
field but do recognize its likely importance in the years ahead.
Creating lesions after training offers the opportunity to use
contextual freezing as a way of investigating the temporal gra-
dient of retrograde amnesia. Rapid learning (during a single
session) and the fact that retention is extremely robust over
time are features that render the task highly suitable. Kim and
Fanselow (1992) produced electrolytic hippocampal lesions in
rats 1, 7, 14, or 28 days after cued and context fear condition-
ing. As noted above (Fig. 13–40C), a graded deficit in contex-
tual freezing is seen as a function of the training–lesion
interval, with animals given lesions 28 days after training
672 The Hippocampus Book
being indistinguishable from those that underwent sham
surgery. This finding was interpreted in terms of a time-
dependent consolidation process for contextual fear, consis-
tent with declarative memory theory. Maren and Fanselow
(1997) obtained similar results, but only over much a longer
time interval, in animals with excitoxic hippocampal lesions.
An elegant within-subjects study by Anagnostaras et al. (1999)
examined the effects of electrolytic hippocampal lesions on
animals conditioned in two highly distinctive contexts with
training scheduled 50 days apart. Conducted at UCLA, the
two contexts were named after two famous streets in
Hollywood. The “Sunset Room” was a darkened laboratory
containing conditioning chambers that were triangular, had
white plastic walls, and were scented with acetic acid. The
“Wilshire Room” was a distinctive, well lit laboratory with
conditioning chambers that were rectangular, made of alu-
minum and plexiglass, and scented with ammonium hydrox-
ide. Previous work had indicated that there is little
generalization between these two environments; the animals
would have readily distinguished them during training and
testing. The key finding was that lesions made shortly after
context fear conditioning in the Sunset Room but 7 weeks
after conditioning in the Wilshire Room (or vice versa)
impaired later expression of the recent memory but not
remote memory. “Location, location, location” is, not surpris-
ingly—as important in Hollywood as anywhere else.
Contextual Control of Stimulus Significance?
There are ways in which context can affect behavior other
than by entering into associations with the US as in context
freezing. One way is to modulate the effectiveness of other
associations. As noted long ago by Nadel and Willner (1980),
contexts can be said to “contain” events that occur within
them. The relation between cue and context is then “hierar-
chical” rather than associative in the traditional sense. Work
by Moita et al. (2003) nicely illustrates the impact of contex-
tual cues on CA1 pyramidal cell firing during auditory fear
conditioning (see Chapter 11). Place fields were first observed
in the conditioning chamber prior to fear conditioning. Rapid
fear conditioning then took place over a few trials in which
shock was presented at the end of a white-noise CS, with a
control group receiving temporally random presentations of
the white noise and shock. Freezing was observed in response
to the CS. In parallel, there was an increase in CA1 cell respon-
siveness that was specific to the place fields in the chamber
where a specific cell normally fired. This suggests that the spa-
tial representation encoded and stored during the earlier
exploration can incorporate elements of the task-related fear
conditioning events into its representation.
Contexts can also “set the occasion” for a particular cue
having a particular significance or meaning. Good and Honey
(1991) developed the first satisfactory experimental design to
investigate the role of the hippocampus in such a process.
Their protocol ensured that subjects became familiar with two
contexts (operant conditioning chambers) and two distinct
cues (tones and clickers) and were equally reinforced in rela-
tion to both sets of cues. Rats were trained in an appetitive
paradigm (collecting food pellets) in which one discrete stim-
ulus (A) was reinforced in one context (X), whereas a second
stimulus (B) was reinforced in Y. The critical test of the ability
of either context to retrieve the appropriate significance of
these cues was to present, for the first time, A in context Y and
B in context X. The phenomenon of “context specificity” was
displayed as a reduction in the appetitive responses elicited by
A and B in their now “inappropriate” contexts. Rats subjected
to electrolytic (Good and Honey, 1991) or neurotoxic (Honey
and Good, 1993) hippocampal lesioning failed to show this
reduction of conditioned responding. This failure was not
because the subjects with lesions could not discriminate the
two contexts; a separate experiment established that they had
no difficulty in doing this when only one of the two contexts
predicted the availability of reinforcement.
The implication appeared to be that the integrity of the
hippocampus is necessary for a process associated with the
contextual control of associative conditioning—associative
retrieval—rather than learning a unified representation, as
described by Fanselow. However, this implication is insecure
for two reasons. First, remember that even the ostensibly
robust deficit in contextual freezing induced by hippocampal
lesions does not always occur if the lesions are produced
before training as they were in the Good and Honey (1991)
experiment. Second, Good and Honey’s deficit in contextual
retrieval has not proven easy to replicate in other laboratories.
A series of experiments by Hall et al. (1996) failed to do so, and
later work showed that hippocampal lesions have no effect on
the rate of learning a conditional context discrimination test
over a series of test sessions (McDonald et al., 1997).
Good et al. (1998) put forward the suggestion that there
may be a subtle but important difference between the “inci-
dental” processing of contextual cues (when their processing
is not formally necessary to learn a task) and the “intentional”
or “contingent” processing of contextual cues (when their pro-
cessing is essential for learning that task). Their results indi-
cated that when rats with hippocampal lesions were required
to use context cues over several sessions to differentiate the
stimulus significance of discrete cues, there was no deficit.
However, when the animals were presented with a single
unexpected context specificity test, with context novelty con-
trolled, the hippocampus-lesioned group was impaired (Fig.
13–41). The possibility that there may be something critical
about the hippocampus in relation to the distinction between
automatic and deliberate processing is directly relevant to the
conjunctive representations theory (O’Reilly and Rudy, 2001),
an issue we revisit in the section of episodic memory, below.
Hunger and Thirst as Contexts
Patient H.M. is unable to report whether he is hungry. He does
not ask for meals and, quite soon after eating, attempts to eat
again if a plate of food is placed before him (Hebben et al.,
1985). That he may have forgotten that it is a long time since
 Theories of Hippocampal Function 673

A Incidental context specificity B Contingent context-specificity

15 15
Control
H
mean response rates

mean response rates

12 12

9 9 H S+
H S-
6 6 control S+
control S-
3 3

0 0
ffe e

ffe e
nt

nt 1 2 3 4 5 6
m

m
re

re
sa

sa

2 days blocks
di

di

Figure 13–41. Incidental versus contingent processing of context presented in both contexts and differentially rewarded in each of
cues. A. When the discrete CSs predicting the availability of food them, both control and hippocampus-lesioned rats readily learned
in the conditioning chamber were unexpectedly presented in the to discriminate their significance appropriately. (Source: After Good
other context, controls but not hippocampus-lesioned rats showed et al., 1998.)
a decrease in responding. B. When the discrete CSs were repeatedly

he last ate or that he has just had a recent meal is unsurpris- which placement in an operant chamber was associated with
ing—yet another indication of his memory deficit—but that footshock when the rats were hungry but not when nonde-
he has apparently no awareness of the internal state of hunger prived (or vice versa). An A/(AB) type of discrimination
or its absence is an additional deficit. Motivational states ratio was used to measure the relative degree of conditioned
could contribute to the contextual control of behavior. freezing in the two situations, and the data in Figure 13–42A
A number of animal experiments have addressed this ques- indicate that normal animals could learn this discrimination,
tion, with early studies using spatial tasks in which a rat was whereas rats with neurotoxic lesions of the hippocampus and
required to perform one spatial response when hungry and a dentate gyrus could not. A later study by Kennedy and Shapiro
different one when thirsty. Rats could learn this interesting (2004) sought to distinguish involvement of the hippocampus
contextual-dependent task, but any lesion-associated deficit in learning such a discrimination and a role in contextual
could arise because of a failure to discriminate the motiva- retrieval. Three visual distinct goal-boxes (shown graphically
tional states, failure to use this information to retrieve the as black, gray, and white in Fig. 13–42B) were, across succes-
appropriate response, or failure of spatial memory itself. sive trials, randomly placed at the ends of the arms of a three-
Davidson and Jarrard (1993) examined a nonspatial task in choice “pitchfork”-type maze and the animals’ deprivation

Figure 13–42. Motivational states as hippocampus-dependent con- fornix or hippocampal lesions were no longer able to use their dep-
textual cues. A. Control but not hippocampus-lesioned rats learned rivation state to retrieve a memory of the appropriate response.
to discriminate whether to be afraid in a place as a function of their They did, however, continue to avoid the never-rewarded goal box.
state of hunger. (Source: After Davidson and Jarrard, 1993.) B. After Goal boxes were moved randomly across trials. H, food available
preoperative training to choose a visually distinctive goal box to to a hungry rat; NoR, never-rewarded box; Th, water available to a
secure an appropriate reward when hungry or thirsty, rats with thirsty rat. (Source: After Kennedy and Shapiro, 2004.)

A Hunger as a context for fear B Hunger and thirst as contexts for discrimination

0.7 100 NoR

Th- H+
mean percent correct (1st trial)
discrimination ratio (A/A + B)

80
C
C F
0.6 H Trial N
H 60
H-
NoR Th+
40
chance
0.5
20
Trial N+1

Blocks of hungry and sated states pre-lesion post-lesion

674 The Hippocampus Book
state switched between hunger and thirst on alternate days.
One goal-box provided food if the animals were hungry (but
not otherwise), another supplied water if they were thirsty,
and the third never provided reward. Before any lesions were
made, training showed that the animals could learn to make
the appropriate choice as they ran the maze. After fornix
lesioning or neurotoxic hippocampal lesioning, the ability to
choose between the hunger and thirst boxes fell to chance,
although the animals continued to avoid the goal-box that
never provided reward. The sham-lesioned rats continued to
perform effectively. Occasional probe trials were scheduled in
which hunger was repeated across two successive days; choice
behavior continued to be controlled by motivational state
rather than any alternation pattern in the sham-lesioned con-
trols. Different results in a similar task were reported by
Deacon et al. (2001), but all training was postoperative and a
reward was consistently available independent of motivational
state. The rats with hippocampal lesions may therefore have
solved the task by updating their predicted “value” of the
reward (i.e., food when hungry and food when sated) using an
amygdala-dependent learning process. Accordingly, it seems
secure to conclude that the hippocampus is required for using
internal motivational states in a contextual manner for
memory retrieval.
Contextual Control of Extinction
The phenomenon of “extinction” refers to changes in learned
behavior that occur when a stimulus or a response is no longer
followed by the reinforcement with which it was previously
associated. In the extinction of classical fear conditioning, a
tone (CS) that has previously predicted the imminent arrival
of shock (US) is now followed by nothing. What then hap-
pens, in terms of performance, is that the overt expression of
the prior conditioning becomes weaker over successive extinc-
tion trials. An old-fashioned view of this change is that extinc-
tion is some kind of “unlearning” process, a breaking of the
associative bonds between cue and consequence. However,
various phenomena discovered about the determinants of
extinction have led many to doubt that unlearning is what
normally occurs. Extinction is more often a learning process
in which contextual factors play a particularly important
modulatory role—the extinction context altering what US
memory is retrieved when the previously trained CS is pre-
sented (Bouton and Moody, 2004). During training the CS
comes to evoke a memory representation of the US; during
extinction it may acquire a memory representation of “no-
US” (as in Fig. 13–33), but importantly there is no overwrit-
ing of the previous CS–US association. Which memory is
retrieved in a given memory test depends on the context of
testing, with the nature of the environment and how it differs
from other environments, the passage of time, or other factors
influencing the operative “context” as it is perceived by the
animal. As Bouton and Moody put it, “the fact that extinction
is more context-dependent than conditioning is consistent
with the idea that the animal codes extinction as a kind of
conditional exception to the rule—one that depends on the
current context” (Bouton and Moody, 2004, p. 665).
Studies by Bouton have implicated the hippocampus in
certain extinction phenomena but not others. Wilson et al.
(1995) using fornix lesions and Frohardt et al. (2000) using
neurotoxic hippocampal lesions have provided evidence that
the integrity of hippocampal function is needed for a context-
processing phenomenon called “reinstatement” but is not
required for another phenomenon called “renewal.” The dis-
tinction between them is important. Both are instances in
which an initially conditioned and then extinguished CR
recovers. Reinstatement is said to occur after independent pre-
sentations of the US in the same context as original training,
with memory test consisting of CS presentations in this same
context (reinstatement of the CR does not occur if the US pre-
sentations occur in a different context). In renewal of the CR,
the previously trained CS is extinguished in a different context
and then tested in the original training context (renewal does
not occur if CS extinction occurs in the same context
throughout). These experiments used the classic paradigm of
conditioned suppression in which rats were first trained to
press a lever for food reward in each of two contexts. They
were then presented with a CS (lights turning off) on a num-
ber of occasions, this stimulus being paired with weak electric
shock (US). Rats reduce their rate of appetitive responding
during the CS (“conditioned suppression”), and this decline
in operant responding served as a quantifiable measure of fear
in much the same way as freezing. Once trained in this way,
the CS was then “extinguished” by being presented several
times without shock over a series of lever-pressing sessions.
Rates of operant responding recovered until they were as high
during the CS as in its absence (this is expressed as a “sup-
pression ratio” calculated as CSrate/(CSrate  preCSrate), a
measure that tends toward 0.5 when rates of responding have
equilibrated). Reinstatement was observed as the selective
recovery of the fear response to the CS (i.e., somewhat con-
fusingly, this is a return to a low suppression ratio) after rats
had been exposed several times to the US alone in the training
context. Critically, this reinstatement effect occurred in con-
trol animals but not in subjected to fornix or neurotoxic hip-
pocampal lesioning (Fig. 13–43A). Conversely, renewal of the
CR occurred just as much in control as in lesioned rats using
an ABA design in which the CS was extinguished in a differ-
ent context (B) from that in which it was initially conditioned
and later tested (i.e., context A) (Fig. 13–43B).
Both phenomena are context-dependent, but Bouton
argued that the reason reinstatement is hippocampus-
dependent, but renewal is not, is because the hippocampus is
important for context–US associations but not for all forms of
contextual modulation of retrieval such as positive occasion-
setting or conditional rules. A strong context–US association
rapidly created by the unsignaled US presentations retrieves a
memory representation of the shock and, consequently,
reminds the animals of the original CS–US association. That
is, the re-presentation of the ostensibly extinguished CS then
retrieves a memory of the US (with which it was associated

A Reinstatement (Hippocampus-dependent)
CS-US training and then Extinction;
US-only in same or different contexts;
Test CS in original (same) context
radiofrequency lesions
0.5 0.5
suppression ratio
0.4 0.4
0.3 0.3
0.2 control different 0.2
fornix different
0.1 control same 0.1
fornix same
0.0 0.0
after 1 2 3 4
extinction test trial
Figure 13–43. Reinstatement and renewal after the extinction of
classic fear conditioning. A. Using conditioned suppression of oper-
ant responding as a measure of conditioning, presentation of the
US only in the same context as that in which CS-US conditioning
had taken place “reinstated” conditioned fear in control rats but not
during training) rather than of the absence of the US (as had
developed during extinction). This inferential feature of
context-dependent modulation of conditioning critically
involves context-US associations and thus the hippocampus.
In renewal, the subject also learns a CS–US association during
training and a CS–NoUS association during extinction.
However, in contrast to reinstatement, there is no strong
hippocampus-dependent context–US association formed.
Instead, the animal’s representations of the two contexts
(training and extinction) modulate whether the CS–US asso-
ciation or the CS–noUS association is evoked, and this modu-
lation does not require the hippocampus. It is most likely
mediated by neocortical circuits on their own. Frohardt et al.
(2000) offer several reasons from the animal learning litera-
ture for recognizing renewal and reinstatement as separate
context effects.
This tidy picture is complicated in two ways. First, a study
by Fox and Holland, (1998) that found no effects of neuro-
toxic hippocampus lesions on reinstatement in an appetitively
motivated task. It is possible that context–US associations in
appetitive conditioning are weaker than in aversive condition-
ing because food also occurs in the home cage, whereas shock
is specific to the testing chamber. However, a clear reinstate-
ment effect was observed in the appetitive task, the difference
being that it was just as great in the rats with hippocampal
lesions. An alternative and more parsimonious explanation
concerns the complication, noted earlier, of elemental versus
configural/conjunctive representations of contexts. In these
extinction studies, the hippocampal lesions are made prior to
conditioning. We have noted that contextual fear conditioning
sometimes occurs normally in such animals, as they have the
opportunity of representing the context with reference to
Theories of Hippocampal Function 675
B Renewal (Hippocampus-independent)
CS-US training in first context
Extinction in same or different contexts;
Test CS in original (same) context
neurotoxic
lesions
0.5
0.4
0.3
0.2
0.1
0.0
S D S D
after 1 2 3 4
Sham HPC extinction test trial
those that had fornix (left) or neurotoxic hippocampal (right)
lesions. B. Renewal of fear to a conditioned CS was observed in
both controls and fornix-lesioned groups after the CS was extin-
guished in a different context. (Source: After Wilson et al., 1995 and
Frohardt et al., 2000.)
some isolated or elemental feature of the testing chamber. One
way to reduce the likelihood of this happening in animals that
already have lesions is to ensure that two contexts are used in
the behavioral protocol and that the critical manipulations are
scheduled for only a short period (e.g., one session). The
Bouton studies used two contexts that differed in quite subtle
ways (e.g., the orientation of the grid bars), whereas the appet-
itive study by Fox and Holland did not (they used the control
procedure of no US reexposure). Thus, the animals in the
appetitive study of reinstatement after extinction would have
been able to form an “elemental” context–US association
using neocortical circuitry, and this may have been sufficient
to remind the animals of which CS–US association to recall as
effectively in lesioned as in normal rats.
Another problem is that the “renewal” phenomenon is sen-
sitive to muscimol-induced reversible inactivation of the hip-
pocampus (Corcoran and Maren, 2001, 2004; Corcoran et al.,
2005). In contrast to the argument just given, Maren argued
that Bouton’s use of pretraining lesions would have allowed
“elemental” representation of the two contexts and thus a
hippocampus-independent form of contextual-dependent
renewal of extinction in the lesioned rats. His use of post-
training muscimol infusions would have made it more likely
that a configural representation, acquired normally during
fear conditioning and extinction, would have been disrupted.
However, this explanation does not seem to account for the
dissociation between renewal and reinstatement observed in
the permanent lesion studies. Another concern is that the spa-
tial extent of the inhibition of hippocampal activity by the
muscimol infusions was revealed in autoradiographic studies
to be quite small and extended as much into cortex as it did
along the longitudinal axis of the dorsal hippocampus
676 The Hippocampus Book
(Corcoran et al., 2005). If renewal is mediated by a form of
contextual modulation controlled by extrahippocampal cir-
cuits (Bouton and Moody, 2004), these infusions may have
disrupted a cortex-mediated component of that process also.
As noted before, the use of reversible inactivation is analyti-
cally more powerful than permanent lesions, but there are
technical problems to its use that should be recognized
(Anagnostaras et al., 2002).
13.5.4 Critique
The valuable common feature of the three theories just dis-
cussed is their serious attempt to tackle the fundamental
problem of dealing with ambiguity in memory. However,
stepping back from the details of individual experiments,
aspects of each of the ideas are problematic.
First, although it is easy to understand the selection pres-
sure for memory systems for remembering facts and events or
finding one’s way around, it is not clear why mammals would
have evolved a specialized neural system for solving nonlinear
discrimination tasks. Ambiguity of cue significance may be a
task a scientist can invent, but is it rife in nature? To be sure,
there are observations indicating that monkeys execute differ-
ent behavioral actions in response to distinct fear-evoking
stimuli (e.g., snakes, eagles), but the stimuli from the animals
and the calls then given by troop members are in no way
ambiguous (Seyfarth et al., 1980). Where ambiguities do
occur, they generally arise as a function of the spatiotemporal
context of events and can therefore be solved without recourse
to configuring. Whether an animal should freeze or flee upon
hearing the sound or smell of a predator more likely depends
on the relative safety of the location it presently occupies.
Configuring stimuli might help solve tasks of a particular log-
ical structure; however, for example, the occasional failure of
rats with hippocampus lesions to learn transitive discrimina-
tion tasks could actually be secondary to deficits of a different
kind—the inability to form relational representations or to
allow a context to modulate retrieval.
Second, the relational processing theory is built around the
idea that associative relations may be more abstract and more
flexible than implied by traditional theories of conditioning,
with their emphasis on associative strength and unlabeled
associations. There are, however, aspects of the theory that
seem vague. Critics such as Mackintosh (2002), prefer the pre-
cision of quantitatively precise theories of associative learning,
which assert that a variety of qualitatively different associa-
tions can occur such as excitatory links, inhibitory links, and
stimulus representations that set the occasion for the retrieval
of others. This complexity of conditioning aside, consideration
of even the simplest of associations—the pairing of a tone with
shock—illustrates the problem of identifying the type of asso-
ciation. Is a tone–shock pairing relational or nonrelational?
Can it be either depending on the training situation? In terms
of the animal’s phenomenal experience, the tone may come to
evoke the memory representation of something painful about
to happen, or it may simply assume high associative strength
that somehow triggers freezing. The speed of learning a
tone–shock relationship and its ready expression in novel test-
ing situations points to it being, by Cohen and Eichenbaum’s
definition, a “flexible” representation. Conversely, its slow rate
of extinction suggests it is inflexible. In this and other exam-
ples, independent rules for identifying into which category an
association should be classified other than observed effects of
hippocampus lesions do not seem to exist. We are back to the
issue that has long haunted declarative memory theories:
irrefutability arising from logical circularity.
Third, both the configural and relational theories also lack
detail at the level of neurobiological implementation. We are
still at an early stage of thinking about how the circuitry of the
hippocampal formation is used for forming or storing config-
ural/relational associations, how it could implement the selec-
tive enhancement of the salience of configural units elsewhere
in the brain, or how it could realize memory representations
that would somehow afford flexibility at the time of retrieval.
What are the relative roles of the dentate gyrus, CA3, and CA1
in achieving any of these tasks? How might the organization
of the mossy fibers and the longitudinal associational pathway
of area CA3 help implement stimulus configuring (if it does)?
What role might associative LTP play in relational processing?
To be fair, the relational theory has taken an important step in
this direction in asserting an anatomically distinct ITM and
LTM for isolated stimuli compared to relations.
Finally, fourth, ideas about contextual processing are not
immune from some of the same criticisms. When is a stimu-
lus a context and when a “discrete stimulus”? In the animal lit-
erature, a context tends to be a box. In the human literature, it
is as often the color of the ink in which a word is presented.
This is deeply confusing. Furthermore, how precisely can a
contextual representation function in the several ways we have
discussed? Why is the hippocampal formation involved in
some of these aspects of contextual processing but not others?
In short, how can a contextual account avoid the same charge
of occasional circularity that we have leveled against the
declarative memory theory? There are two strands to any
answer to these questions. One is that there is a formal body
of theoretical work on associative conditioning that has delin-
eated distinct ways in which contexts can enter into associa-
tions with CSs and USs or modulate other associations, as
represented in Figure 13.33B. Reinstatement and renewal are
good exemplars, as their associative basis is operationally quite
different, and a specific prediction could then be made about
the outcome of a hippocampal lesion. Neuroscientists trying
to understand mechanisms at the neural level can capitalize
on this prior psychological research—as they have done with
such phenomena as the immediate shock effect, the context
preexposure facilitation effect, and the distinction between
renewal and reinstatement.
The second strand of defense for the context ideas is the
sheer plausibility and simplicity of the idea that context plays
a critical role in coping with ambiguity. Events happen—at
specific moments in time and space and in specific social
contexts.Their meaning and significance is often context-

dependent, including semantic knowledge, just as so many
aspects of our behavior. In addition, events often happen
unpredictably. This simple fact requires that any system whose
role is to keep track of the important events of our lives must
be able to respond instantly, diverting attentional resources as
required and encoding information rapidly as it happens and
with reference to the context in which it occurs. This is why
the distinction between incidental and contingent processing
of contextual information may be important. Context pro-
cessing, as O’Keefe and Nadel (1978) recognized long ago, has
the potential to be one building block of a true episodic mem-
ory system. It is to that issue which we now turn.
Á 3. Differential role of subcomponents: Components of the
13.6 Episodic Memory, Hippocampus,
and Neurobiology of Rapid Context-
specific Memory
The last hippocampal theory to be considered, and one in
which there is growing interest, is that the hippocampal for-
mation participates in certain aspects of episodic memory (in
humans) and episodic-like memory (in animals). This idea has
emerged in an integrative manner from neuropsychological
studies of developmental amnesia (Vargha-Khadem et al.,
1997) and work concerning scene memory, object–scene asso-
ciations, and top-down control of memory processing in
monkeys (Gaffan, 1994a; Miyashita, 2004). In other animals,
behavioral studies of food caching by avian species (Clayton
and Dickinson, 1998), studies of sequence learning in rodents
(Fortin et al., 2002), single-unit correlates of prospective and
retrospective memory (Wood et al., 2000; Ferbinteanu and
Shapiro, 2003), pharmacological interventions affecting the
encoding and retrieval of rapid paired-associate learning
(Day et al., 2003), and genetic interventions targeted condi-
tionally at specific components of hippocampal circuitry
(Nakazawa et al., 2003) are all contributing to a developing
perspective. That a range of approaches are converging on the
same general idea indicates that the theory may be on the
right lines (Box 13–9).
13.6.1 Concept of Episodic Memory
First introduced by Tulving (1972) and elaborated in a num-
ber of ways since (Tulving, 1983, 2004; Schacter and Tulving,
1994), episodic memory refers to the memory of a unique
event and/or a temporal sequence of events that collectively
comprise an episode. As originally defined, the content of
episodic memory is a mental representation of “what” hap-
pened during an event, “where” it happened, and “when.”
Writing in 1983, Tulving asserted that episodic memory
“receives and stores information about temporally-dated
episodes or events, and temporal-spatial relations among
these events” (Tulving, 1983, p. 385). Thus, remembering what
you did on holiday, where you went, and with whom would be
an instance of episodic memory. It is contrasted with seman-
Theories of Hippocampal Function 677
Box 13–9
Hippocampus: Episodic and Episodic-like Memory
1. Conceptual issues: Episodic memory is the recall, by
humans, of discrete events that happened at a particular
place and a particular time. Such recall entails mental
time travel. Episodic-like memory in animals is the mem-
ory of “what, where, and when” with respect to events.
2. Hippocampal involvement: The hippocampus is one of a
network of brain structures that mediate the automatic
encoding and retrieval of attended events and the contexts
in which they occur (episodic-like memory). Distinct
brain structures, including the prefrontal lobe, mediate
different and more volitional aspects of this form of
memory.
hippocampal formation (e.g., CA1, CA3, dentate gyrus)
are differentially involved in dissociable components of
episodic-like memory, such as pattern separation and pat-
tern completion.
4. Awareness: Episodic memory and episodic-like memory
are distinguishable, as only the former requires “auto-
noetic” consciousness. This cannot, at present, be studied
neurobiologically. Awareness may also be an attribute of
episodic-like memory in animals, but this need not be
autonoetic.
tic memory—factual knowledge—which is not bound to any
temporospatial context.
Episodic memory has several important properties. First,
the encoding of attended events often occurs automatically;
and by virtue of this the “temporal-spatial relations” of
Tulving’s definition cannot be deliberately selected or ignored.
“Automaticity” implies that humans cannot normally “turn
the system off ” and voluntarily decide which attended events
or what features of the context in which they occur are
encoded (Martin et al., 1997; Morris and Frey, 1997). This par-
adoxically “reflexive” feature can be illustrated with reference
to an example. Imagine that during a routine activity such as a
shopping trip you witness a car crash. You were not expecting
it to happen nor planning ahead of time to remember it. Still,
what happens in front of you automatically commands your
attention and it then becomes impossible not to encode and
later remember that it happened, where it happened, and
when. This inevitability of the associations formed at the time
seems to be a characteristic of the brain system that processes
this information. You get home and tell your family that you
saw a car crash. They ask about what happened. You describe
the scene of the crash, the crash itself, what happened next,
and anything else that was distinctive or unusual. This is
straightforward. But how strange it would be to tell your fam-
ily of having seen a car crash, but then to say that you cannot
remember anything about where it happened, stating only that
it was “half-an-hour ago and involved two cars and a bicycle.”
During ordinary discourse, the “What happened?” enquiry
carries with it the implicit trilogy of “what, where, and when.”
678 The Hippocampus Book
This automatic “binding” of event, place, and time infor-
mation is a characteristic of episodic memory that has neuro-
biological implications. It points to the need for a memory
system that has anatomical access to highly processed infor-
mation from all sensory modalities, is downstream of the
attentional filters of the neocortex and the perceptual mecha-
nisms for object-centered representations, and is downstream
of but reciprocally connected with semantic memory. It must
also be a system that stays online, can encode information rap-
idly, and is generally independent of volitional control. This
last qualification of the automaticity property arises because
some aspects of episodic memory can be volitional. For exam-
ple, subjects in a neuropsychology experiment asked to learn a
set of novel paired-associates or to remember a passage of
prose are deliberately cooperating in a learning exercise. Their
later memory of what happened is clearly an instance of
episodic memory, but it occurs in the context of an intentional
course of action, unlike the event of witnessing a car crash.
The “what-where-when” definition also captures a second
important feature of episodic memory that marks it off from
semantic memory, the other major category of declarative
memory. This is the need for “mental time travel” as a part of
our phenomenological experience of remembering. This con-
cept may feel a bit fuzzy at first, but it is crucial. A person
remembering an event, as distinct from merely declaring they
know something, travels in their mind through time to that
moment in the past when the event happened. Their experi-
ence is then one of being simultaneously in the present while
also reexperiencing the past, a mental coexistence that tran-
scends time. This is one of those psychological juxtapositions
that may never have seemed paradoxical until it is pointed
out: Mental time travel is a capacity of mind that we all take
for granted, is apparently effortless, and yet is extraordinarily
complex. Efforts to get a handle on this experimentally is
sometimes done by asking subjects to make a subjective
assessment of whether they “remember” or merely “know”
something. We do not yet have the ability to pose such a ques-
tion to animals in a neurobiological study.
Tulving (2004) made the strong claim that the capacity for
mental time travel is unique to humans. Suddendorf and
Corballis (1997) shared this view, arguing that animals are
forever mentally trapped in the present or, as Roberts (2000)
so nicely put it, they are “stuck in time.” This is a controversial
claim, as biological scientists are rarely at ease with unsup-
ported claims of human uniqueness. Like Pinker’s arguments
for the uniqueness of human language (Pinker, 1994), Tulving
asserted the human uniqueness of mental time travel in an
arresting way. There are, he wrote, “things that bees, birds and
humans all do. But there are also things that bees do but birds
and humans do not. There are things that birds do that bees
and humans do not. And there are things that humans do that
bees and birds do not.”
A third feature of episodic memory is that the event being
remembered may be one in which subjects actually partici-
pated themselves (e.g., playing in a game of football) or one
that was merely observed (e.g., seeing a dangerous animal).
Tulving (1983) revised his earlier definition of episodic mem-
ory by adding that the observer needed to have a particular
sense of self-identity. He called it “autonoetic consciousness.”
This form of self-consciousness is necessary for memories of
the form “I remember scoring a goal” because such a memory
entails a sense of self-identity. Opinion is divided, however, on
the need for so sophisticated a form of awareness for all types
of event memory. It is far from clear that one needs a sense of
self-identity to recall having seen a dangerous animal, yet one
can readily imagine evolutionary selection pressures that led
to a brain system, present in nonhuman animals, for the effec-
tive memory of having seen a predator at a particular place
and a specific time. Animals that stay that much safer by hav-
ing such a memory system may or may not have a sense of
self-identity. Tulving nonetheless insisted on his autonoetic
consciousness criterion for “true” episodic memory, but this is
impossible to apply to animals because we are unlikely ever to
know whether animals have a sense of self-identity (Clayton et
al., 2003b). Episodic memory may or may not be uniquely
human and so may or may not reflect distinctive features of
the human brain.
Given this somewhat metaphysical state of affairs, Griffiths
et al. (1999) introduced the term “episodic-like” memory. This
is something that can be defined behaviorally in terms of
Tulving’s older 1972 definition as memory for “what, where,
and when” with the autonoetic aspect finessed for the time
being. Meeting even this ostensibly more modest requirement
for demonstrating episodic memory remains far from easy.
Considerable ingenuity is required of behavioral scientists to
invent novel memory tasks with the appropriate attributes to
do it successfully (Morris, 2001). Much of what is available so
far, including the many hippocampus-dependent tasks dis-
cussed in this chapter are not up to the task (Hampton and
Schwartz, 2004).
However, suppose that an animal model of episodic-like
memory were possible. Its value would be that we would then
be able to get a handle on the neurobiology of rapid, context-
specific memory formation with access to the usual range of
anatomical, physiological, and biochemical data necessary for
making causal inferences. It could then be classified as the class
of vertebrate memory that refers to unique events and for
which the representational content varies in a species-specific
manner. There is nothing special about such a comparative
perspective. Procedural learning is just the same—the motor
skills that can be learned by a bird, bee, or human vary con-
siderably—yet Tulving would not deny that all three species
have a nonpropositional form of learning that can be classified
as “sensorimotor procedural skills.” By analogy, perhaps
episodic-like memory may vary as a function of phylogeny
with respect to both content and degree of awareness. It may
have evolved in vertebrates as a “one-shot” memory because it
was evolutionary adaptive to be able to encode and then later
recall unique events, such as seeing an unexpected predator.
This eclectic approach opens the prospect of a neurobio-
logical analysis provided the distinctive features of episodic-
like memory can be tied down operationally. Various research
groups are feeling their way toward a range of tasks that may
be suitable assays. An unfortunate problem is that several such

tasks are asserted to be “episodic” in character when the clas-
sification is barely deserved. In the context of the theories that
have been considered in this chapter, episodic-like memory is
over and beyond being merely “declarative” and something
more than a “spatial memory” because, in both cases, the
memory must include information about what and when
something happened as well as information about a specific
location. Episodic-like memory is also more than one-trial
memory. Certain emotional dispositions, such as learned
fear associated with a specific stimulus, can be acquired dur-
ing one trial. There may be configural or relational compo-
nents of episodic memory as well, yet neither of these two
categories quite captures the distinctive features of the full
“episodic” concept or the possible role of the hippocampus in
mediating it.
13.6.2 Scene Memory as a Basis for
Episodic Memory and Top-down
Control by the Prefrontal Cortex
One idea about the neurobiological basis of episodic memory
was Gaffan’s proposal that it was a form of scene-specific
encoding (Gaffan, 1991, 1994). His hypothesis included state-
ments about the role of hippocampal formation and related
structures, but the framework is anatomically more wide-
ranging (Box 13–10).
These and related ideas were developed gradually through
an extensive series of experiments on Old World primates,
beginning with Gaffan’s seminal study of recognition memory
published in 1974 (see Section 13.3).
Recognition Memory, Associative Memory,
and Scene Memory
The distinction between recognition and association (Gaffan,
1974) was, in certain respects, a forerunner of Mishkin’s dis-
tinction between memory and habit (Mishkin, 1982).
However, later work by Gaffan and his colleagues overturned
Box 13–10
“Scene Memory” Basis of Episodic Memory
1. Part of a circuit: The hippocampus is part of a circuit that
includes the fornix, mammillary bodies, and anteror thal-
amus (the Delay-Brion circuit), which collectively encodes
events with respect to the spatially organized scenes in
which they occur.
2. Spatial basis: There is a spatial component to episodic
memory (“scene-encoding”) that is neuropsychologically
dissociable from place memory per se and from spatial
navigation.
3. Anatomical dissociation of function: The functions of the
Delay-Brion circuit are also dissociable from the roles in
memory played by the frontal lobe and the medial tempo-
ral lobe (which it connects), including the amygdala and
perirhinal cortex.
Theories of Hippocampal Function 679
the idea that fornix lesions leave associative memory unaf-
fected following a series of ingenious experiments reported
in somewhat impenetrable papers published during the early
1980s. The insight buried within this work was a new set of
ideas about what it means to say that an animal has learned
an “association.” The traditional view was that associations
were, to all intents and purposes, conditional reflexes or
autonomous “habits,” but Gaffan’s studies and modern animal
learning theory have considerably enriched our view of asso-
ciative learning.
Inspired in part by the work of the animal learning theorist
(Capaldi, 1971), Gaffan explored tasks in which an object
might lead to food the first time it was seen but not the next
time it was presented (Gaffan et al., 1984). That is, reward was
available when the object was novel but not when it was famil-
iar. Normal monkeys could learn this type of task, correctly
reaching for the “other” object of a pair in two-alternative
choice tests as quickly as in a simple association task. This
novelty association task is sensitive to fornix lesions. The
implication was that certain types of rapidly acquired “associ-
ations” between an object and reward are not autonomous
habits at all; they are bits of stored “knowledge.” Gaffan went
on to suggest that the type of actions a monkey is required to
perform during an association task may be critical to whether
a fornix lesion that partially disconnects the hippocampus
from anterior parts of the brain has a deleterious effect
(Gaffan, 1983). However, this hypothesis, although supported
by studies suggesting that spatially directed actions are sensi-
tive to fornix transection (Rupniak and Gaffan, 1987), failed
to attract general support.
A turning point that led to a key insight about episodic-like
memory was the outcome of a study on conditional learning
(Gaffan and Harrison, 1989). Monkeys were trained to reach
for one of two objects (A or B), such that A was correct when
the animals were facing one way in the laboratory room, but
B was correct when they were facing in the other direction.
This task was relatively easy for control animals to learn, but a
severe deficit occurred after fornix lesioning. Interestingly, the
deficit was not a general problem of conditional learning
because there was no deficit if the objects were presented on
trays of different colors with object A correct on an orange
tray but B correct on a white tray. From a strictly logical per-
spective, the two tasks are identical, but they were differen-
tially sensitive to fornix lesions. Gaffan suspected that it might
have something to do with the spatial organization of the
“scene” before the animal. When the monkey had to face in
different directions, the scene before him changed even
though many elements of it were still there (e.g., the window).
When the monkey looked at objects on trays of different col-
ors, however, the global scene was unchanged despite the
minor change of a specific element in it. A third experiment
contrasted “scenes” with “places.” To compare them, a new set
of monkeys was trained to reach for one of five objects as a
function of their place in the room. The places were chosen to
be as visually different from one another as possible so loca-
tion rather than the spatial arrangement of the scene would
differentiate the places (e.g., in front of bookshelves, by the
680 The Hippocampus Book
door, by the window). Surprisingly, no fornix lesion deficit
was seen. Accordingly, Gaffan argued for a distinction between
places and scenes, arguing that a fornix lesion disrupts some
aspect of scene encoding or the capacity of scene information
to set the occasion for other stimuli to be rewarded or nonre-
warded. This analysis was a forerunner of Aggleton and
Brown’s later description of hippocampal memory as involv-
ing the spatial organization of familiar objects (Aggleton and
Brown, 1999). The elegance of Gaffan and Harrison’s (1989)
classic paper is marred only by a minor confounding between
the selectivity of the deficit and the change from two scenes to
five places, with a fornix lesion impairment in the two scenes
experiment but no such effect in the five places experiment.
However, subsequent work went on to establish the sensitivity
of scene memory to fornix transection in monkeys that were
not moved from place to place.
The emphasis on scenes rather than places is a subtle but
important way in which Gaffan’s hypothesis differs from the
cognitive map theory. What is important is not where the
monkey is itself located but what he is looking at. Later unit-
recording work by Rolls revealing the existence of “view cells”
in the hippocampus (Robertson et al., 1998) raised the possi-
bility that fornix transection disrupts a view-specific physio-
logical process that is implemented in or, at least projected to,
the hippocampus. Gaffan, being apparently uninterested in
localization of function as a scientific goal, was noncommittal
about what aspect of cognitive function is mediated within
the hippocampus itself—perhaps uncertain that it would map
onto any psychologically defined category. His interest has
generally been restricted to whether damage to area X disrupts
some psychological process or, in disconnection studies, that
conjoint unilateral damage to area X and contralateral damage
to area Y does likewise (e.g., Easton et al., 2001). Thus, despite
the extensive use of lesions as a tool to dissociate putative
memory systems and a detailed specification of relevant pri-
mate neuroanatomy, the “scene-processing” theory does not
identify the episodic system as located in one or more struc-
tures. The closest Gaffan’s theory comes to anatomical local-
ization is with reference to the Delay-Brion circuit, a group of
structures that includes the hippocampal formation as well as
the fornix, mammillary bodies, and anterior thalamus. This
emphasis on the anterior connections of the hippocampal for-
mation, not least the connectivity with the prefrontal lobe
(Browning et al., 2005), is strikingly different from the per-
spective offered in the declarative memory theory, where the
emphasis is much more on posterior neocortical connections
(as in Chapter 3). Gaffan’s ideas can be said to be forward-
looking.
Subsequent experiments have elaborated and developed
the idea of scene-specific memory as the essential basis of
episodic-like memory. Before describing them, two points
should be noted. First, the hypothesis captures an important
phenomenological feature of episodic memory—our ability
to remember the context in which an event occurred. This is
relevant to many memory tasks given to humans and animals
where the description provided by the experimenter can all
too easily ignore features of the surrounding scene because
they do not seem relevant to the solution of the task.
Recognition memory is, once again, a case in point. Although
the surrounding features of the testing apparatus (the WGTA
or computer touchscreen) may not help the monkey remem-
ber whether object X or image Y was seen most recently,
an incidental and noncontingent memory of the context
may still be important. Normal animals automatically form
object–context associations, and the context may then act as a
retrieval cue or as an element of the remembered visual scene.
This was, it should be remembered, the basis of Nadel’s con-
cern about removing monkeys from a WGTA during long
memory delays in some of the primate studies of recognition
memory (Nadel, 1995).
The second point to note is methodological. The first
scene-encoding study (Gaffan and Harrison, 1989) and other
studies conducted around the same time involved moving the
transport cage containing the monkey to change the scene
around the animal. This procedural burden, coupled with the
laborious nature of training monkeys by hand in WGTAs,
encouraged the Oxford laboratory to be pioneers in the devel-
opment of automated stimulus presentation systems.
Automated testing was introduced by the National Institutes
of Health (NIH) group around the same time (Murray et al.,
1993), following innovative developments of T. Aigner, and
later by a San Diego group to test monkeys at very short mem-
ory delays in DNMTS (Alvarez et al., 1994). Horel’s group also
developed systems that used slide projectors under computer
control. The development of touchscreen technology was
partly a matter of convenience, but it was also to have an
impact on Gaffan’s developing theoretical ideas. He began by
using laser disks to show a large number of scenes to monkeys
(e.g., scenes from a film) and later digital graphics software to
generate a nearly infinite variety of abstract designs as scenes
(typically of colored circles, ellipses, and typographical char-
acters of different sizes).
Scene Memory, Place Memory, and Object-in-Place
Memory Are Sensitive to Fornix Lesions
In one study, monkeys were shown scene after scene from
George Lucas’s film “Raiders of the Lost Ark” (Gaffan, 1992).
Although we may wonder what they made of it, their task was
to remember which scenes were associated with reward and
which were not.* A highly significant deficit in the rate of
learning was observed in fornix-lesioned monkeys. This result
is important because it suggests that scene learning and object
discrimination learning may be quite different. From a strictly
*An unpublished study by Nicholas Humphrey should be mentioned in
passing. He showed cartoon movies to monkeys on a television screen. To
test whether there was any understanding of what was happening, he
allowed the animals to choose between seeing the film run normally or in
the reverse direction. Although the scenes the animals saw were visually
similar, only the normal movie “made sense.” The monkeys preferred the
regular film.

logical point of view, learning the reward significance of a set
of scenes is identical to concurrent object discrimination
learning in a WGTA. We saw earlier that concurrent discrimi-
nation tasks are sensitive to MTL lesions when multiple trials
of each pair of objects are given each day but not when only
one trial of each object is given per day (Malamut et al., 1984).
However, somewhat paradoxically, when several hundred
scenes are presented each day, each for no more than one or a
few trials, a clear deficit is observed with a fornix lesion that
typically produces only mild amnesia in humans and little or
no deficit in many classic primate declarative memory tasks.
Gaffan’s finding of a positive effect of fornix lesions therefore
supports his case that there may be something special about
scene processing in the Delay-Brion circuit compared to
merely discriminating objects.
Computer-generated abstract scenes provided another way
to compare three conditions: place memory, object memory
and object-in-place memory (Gaffan, 1994a). In the object-
in-place task, for example, the correct response for the mon-
key is to reach out and touch the one object of a pair that
always occupies a particular position in a unique background
of abstract colors and shapes (Fig. 13–44A). The use of the
term “object” is slightly misleading because, in practice, the
monkeys are often required to reach toward nothing more
than alpha-numeric characters on the screen. However, work
by Buckley and Gaffan (1998b) showed that monkeys are very
good at equating objects seen on a computer screen with real
three-dimensional objects they can handle and pick up. In a
series of experiments using such abstract scenes, a central
finding has been that hippocampal, fornix, mammillary body,
or anterior thalamic lesions each cause a deficit in recalling
scenes learned preoperatively and in learning new scenes (Fig.
13–44B). The reasons for arguing that the Delay-Brion struc-
tures form a circuit responsible for this type of memory are
(1) lesions to each structure individually cause a learning
Figure 13–44. Scene-encoding as a component of episodic mem-
ory. A. The monkey faces a computer screen that typically consists
of a number of randomly placed shapes and alpha-numeric charac-
ters. The task is to reach for one object (e.g., the number 7 in the
bottom-left corner) against one background and for another object
A. Greyscaled display of ‘object-in-place’ task
7
X u
Theories of Hippocampal Function 681
impairment of comparable magnitude; and (2) adding a
fornix lesion to a preexisting mammillary body lesion does
not exacerbate the deficit. It therefore seems likely that these
structures make their contribution to processing in a serial
manner. Later studies have gone on to implicate the entorhi-
nal and perirhinal cortex in object-in-place memory as well
(Gaffan and Parker, 1996; Murray et al., 1998; Easton and
Gaffan, 2000; Charles et al., 2004b).
A possible weakness is that although described as an “ani-
mal model of episodic memory” there is nothing very “event-
like” in remembering which of two alpha-numeric characters
is rewarded with food within particular scenes. The task could
be learned semantically and stored as a “fact” rather than as an
episode. There is also no “recall” element to the task. The cues
are always provided on the computer screen, and the animal
can recognize them rather than bring them to mind. To be
sure, learning is remarkably rapid such that, in experienced
monkeys, it takes only two to five trials to learn each object-in-
place task. At that speed of learning—clearly much faster than
traditional visual discrimination tasks—the animal is remem-
bering correctly what to do on the basis of only one or at most
a few trials. However, the episodic character of the task, like
paired-associate learning in humans, is being inferred more
from the sheer speed of learning than its formal logic. It does
not absolutely require memory of “what, where, and when.”
Memory Functions of the Temporal Lobe,
Delay-Brion Circuit, and Prefrontal Cortex
The proposal that the hippocampus is part of the Delay-Brion
circuit and that this circuit has a highly specialized role in
memory is only one aspect of Gaffan’s wider ideas about
multiple types of memory mediated by circuits interconnect-
ing the temporal and frontal lobes. His scene-specific theory
of episodic memory differs from the declarative memory
(e.g., the letter u in the right-hand side) when the presented against
a different background scene (not shown). B. Monkeys soon learn
these types of problem rapidly, whereas fornix-lesioned animals
show a clear deficit. (Source: After Gaffan, 1994.)
B. Mean performance during
learning
3
Errors per scene (early trials)
2 
1
0




NC Fx
682 The Hippocampus Book
theory in two major respects: First, it regards hippocampus-
dependent scene memory as one of several qualitatively dis-
tinct propositional memory systems. Like the position taken
by Horel a generation ago (Horel, 1978b, 1994), Gaffan there-
fore takes exception to the concept of a monolithic MTL
memory system (Gaffan, 2002). Second, unlike both the
declarative memory and relational processing theories, there
is apparently no explicit consolidation mechanism. Gaffan
believed that the Delay-Brion circuit is important for memory
encoding and memory retrieval, even if retrieval is not
required until months after original training, but it is non-
committal on sites of storage. The perspective on the role of
the MTL in memory is difficult to pin down. This is because
the dissociations that have emerged from his experiments
(using a variety of tasks and both bilateral and crossed unilat-
eral lesion techniques) are generally described only in terms of
the “necessity of a given area/pathway for a given task.” No
statement of what cognitive operations are carried out by each
structure in the temporal lobe has been presented, nor state-
ments about the nature of different memory systems, their
rules of operation, or the type(s) of information on which
they operate. Even for the best studied part of the system, the
Delay-Brion circuit, there is no statement of the presumably
distinct contributions to scene-specific episodic encoding and
recall made by the four distinct structures of the circuit. This
is puzzling because the local circuitry, numbers of cells, and
extrinsic connections of the mammillary body differ radically
from those of the hippocampus. Even if Gaffan is correct
that they work together serially to enable a common func-
tion, it would be helpful to unpack this statement by outlin-
ing their distinct contributions to that “common function.”
Electrophysiological studies may be helpful here.
These points may seem unduly harsh criticisms. However,
the picture that is emerging from Gaffan’s seminal work is that
he envisages a widespread but loose hierarchy of cortical areas
involved in memory (Murray et al., 2005). The middle and
inferior temporal gyri have the job of analyzing visual object
features, such as shape and color. The most anterior neocorti-
cal region, the perirhinal cortex, associates these object fea-
tures with nonvisual features of objects (e.g., reward
significance) to form perceptual representations of unique
individual objects (Buckley and Gaffan, 1998a). Information
about unique objects processed in the perirhinal cortex is put
together, in circuits involving the hippocampus, with spatial
information to form representations of unique complex
scenes (Murray et al., 2005). Finally, the entire system has
important functional connections with the prefrontal lobe.
By way of a coda, it should be noted that Gaffan’s empha-
sis on networks for memory processing has points of similar-
ity to Miyashita’s distinction between “automatic” and “active”
retrieval of information (Miyashita, 2004). Although
Miyashita accepted the idea of an MTL memory system, he
also identified a “semantic” associational system in area 36 of
the limbic cortex and the important role of “top-down” con-
nections from the prefrontal cortex in guiding the intentional
or active retrieval of information from these systems.
Miyashita’s analysis extended beyond the domain of episodic
memory to paired-associate learning following his identifica-
tion of neocortical “pair-coding” neurons that fire differen-
tially as a function of the association that binds two stimuli
together (Sakai and Miyashita, 1991). His suggestion is that
automatic retrieval of episodic or semantic information
requires no more than activation in the MTL or area 36,
respectively, but effortful retrieval also involves a top-down
signal from the prefrontal lobe. Two main lines of evidence
from animal studies support this idea. First, the latency at
which unit-recording signals are observed during automatic
retrieval is shortest in the most medial structures and spreads
backward from the hippocampus (for episodic information)
and from area 36 (for semantic information) to area TE where
neurons are gradually recruited that provide a fuller represen-
tation of remembered information, be it an event or a fact.
Second, during active retrieval, neurons in the inferotemporal
cortex received categorical rather than stimulus-specific
information from the frontal cortex to help guide the effort-
ful process of memory searching (Tomita et al., 1999).
Miyashita saw the challenge ahead as being “to understand
the hierarchical interactions between multiple cortical areas”
(Miyashita, 2004, p. 435) in the expression of memory.
Notwithstanding differences of approach and detail, Gaffan’s
perspective was similar.
13.6.3 What, Where, and When: Studies
of Food-caching and Sequence Learning
Food Caching by Corvids as a Model
of “What, Where, and When”
A different approach to thinking about the recollection of past
experience was taken by Clayton and her colleagues, who con-
ducted a series of studies examining food caching and recov-
ery by a corvid species, the Californian scrub-jay. Clayton and
Dickinson (1998) allowed adult hand-raised scrub jays to
cache food items in sand-filled ice-cube trays surrounded by
trial-unique arrays of distinctive visual objects (Lego bricks).
The two food items were (1) preferred but perishable wax
moth larvae and (2) less-preferred but nonperishable peanuts,
each of which were in plentiful supply in bowls at the time of
caching. During two successive caching periods separated by
an interval of 120 hours, the birds transported and then
cached wax moths on one side of the caching tray and peanuts
on the other, with each side being covered by a Perspex strip
during the caching of the other food item (Fig. 13–45A,B).
These caching periods constituted the opportunity for the
animals to encode into memory what food item was being
cached where and when. Recovery was permitted either 4 or
124 hours later with both sides of the ice-cube tray uncovered
and available. Over a series of pretraining trials extending over
several weeks, with the visual arrangement of Lego bricks and
the assignment of tray sides for the two foods randomly
changed on each trial, birds in the “degrade” group had the
opportunity to learn that wax moths recovered after 4 hours

A Experimental Design for Training Trials
Caching experience 1
Worms
120 hours later
or on other trials....
120 hours later
Peanuts
B Scrub-jay retrieving food caches
Percentage of Birds
Figure 13–45. Memory for what, where, and when in California
scrub-jays. A. Experimental design for the training trials in which
the birds in the “degrade” condition (shown) had the opportunity
to learn that wax-moths become unpalatable after a period of 4
days. B. California scrub-jay performing the task. C. Results from
are tasty and worth eating, whereas those retrieved after 124
hours are not. Birds assigned to a separate “replenish” control
group always found fresh wax worms at recovery whether it
occurred at the short or the long recovery interval. The main
finding of this ingenious study was that in nonrewarded probe
tests in which neither food was actually available (thereby pre-
venting any olfactory or visual cues to guide search location)
the scrub-jays in the degrade group searched preferentially for
wax moths when recovery occurred 4 hours after caching but
switched their search to peanuts when recovery did not occur
until 124 hours had elapsed (Fig. 13–45C). This switchover
was not shown by the birds in the “replenish” condition but
partially shown by birds in a third, “pilfer,” group in which the
preferred wax worms were missing after a 124-hour delay.
This pattern of results implies that, at recovery and using free
recall, the birds can recollect what food item was stored where
and, at least in some sense, how long ago.
In addition to careful controls, such as nonrewarded probe
tests, there are several innovations in this experimental
approach. The use of food caching is helpful as it ensures the
animal is attentive and engaged in the task at the time of
memory encoding; the act of caching is not, however, a
requirement for the formation of episodic-like memory. The
use of very long memory intervals over hours and days
ensures that memory retrieval is outside the range over which
interval timing mechanisms, mediated by short-term mem-
ory, can operate (Hampton and Schwartz, 2004). The use of
Theories of Hippocampal Function 683
Caching experience 2 Recovery
Degraded
Worms
4 hours
Peanuts
Peanuts
Worms
4 hours Worms
Peanuts
C First Search in non-rewarded Probe
100
Peanuts
Worms
50
0
Retrieval after... 4 hr 124 hr
the “degrade” group showing proportions of birds making their first
pecks to the wax-moth or peanut side of the caching tray. (Source:
After Clayton and Dickinson, 1998. Photograph courtesy of N.
Clayton.)
free recall at cache recovery obviates the use of a familiarity
strategy that has been so problematic in studies of recognition
memory. Clayton and Dickinson’s claim that their initial study
provides “the first conclusive behavioral evidence of ‘episodic-
like’ memory in animals other than humans” (Clayton and
Dickinson, 1998, p. 274) was vindicated by later studies in
which a number of detailed issues were taken further (Clayton
et al., 2001). In one of the studies, scrub-jays were allowed
to cache preferred crickets (rather than wax moths) and less-
preferred peanuts (Clayton et al., 2003a). After caching but
before recovery, the birds were taught that the crickets, which
used to remain edible for at least 3 days, now degraded rapidly
and became inedible. During cache recovery, the birds now
avoided the crickets and sought out the less preferred peanuts.
Thus, the birds seem to be able to apply information retro-
spectively at the time of retrieval to guide their choice of food
item. Other studies have investigated prospective aspects of
food caching with the scrub-jays (Emery and Clayton, 2001)
and refined the criteria for identifying mental time travel in
animals (Clayton et al., 2003b).
Sequence-dependent and Context-
dependent Object–Place Memory in Rats
The scrub-jays are not suitable subjects for neurobiological
studies, particularly because these studies use hand-raised
birds to ensure control of the caching experience. Unfortu-
684 The Hippocampus Book
nately, efforts to demonstrate “what, where, and when” in
other species, particularly laboratory animals such as rats,
have not been successful so far. This has led some to question
whether animals can undertake mental time travel (Roberts,
2000). Others have raised the possibility that the “when” com-
ponent should be thought about in other ways, such as with
reference to the context in which they occur or their sequence
of occurrence. Both sequence and context sometimes act as
mental surrogates for the experience of time.
With respect to context, which we have already discussed
extensively (see Section 13.5), Eacott made the interesting
argument that when asked about “when” an event happened
people often work out where they were at the time and then
use the retrieved contextual information to remember events.
In this mental exercise, there is not necessarily any mental
time travel as such (Eacott et al., 2005). Based on this thought,
she has developed a modification of Ennaceur’s novel object
recognition (NOR) task for rats in which context, place, and
object can be a triad of associations. That is, the rat has to
remember whether a specific object was in a specific location
in a particular context. Using this task, she reported that
fornix lesions disrupt recollection of the what–where–where
triad while, somewhat surprisingly, appearing not to affect a
simpler what–where task (Eacott and Gaffan, 2005). This is a
challenging result being, on the face of it, very different from
the extensive results reported by Gaffan’s group using object-
in-place memory. However, given the importance of testing
numerous memory delays in NOR tasks before claiming selec-
tivity (Clark and Martin, 2005), it would be valuable for vari-
ous memory delays to be investigated in her task. On a
conceptual level also, that people sometimes cannot remem-
ber time and use the surrogate of context does not imply that
they always do so, nor does it undermine the central impor-
tance of mental time travel to “true” episodic memory.
With respect to memory for the sequence of events, Honey
et al. (1998) showed that rats with hippocampal lesions were
less able than sham-lesioned animals to notice any change in
the order of two sequentially presented stimuli during an ori-
enting task. In another study, Fortin et al. (2002) showed that
rats could remember some information about the sequence in
which a series of distinctive odors was presented. Cups of sand
through which rats could dig to find food (like the paired-
associate and transitive inference experiments of Section 13.5)
were presented at 2.5-minute intervals. Common household
odors were mixed into the sand to determine its smell, with
different novel sequences of odors given during each series of
trials. For a given sequence, call it A-B-C-D-E, the animals
were later tested for whether they could remember that sand
odor A came before odor C, B before D, and so on; they were
also tested for whether odors A–E were familiar compared to
other novel odors. The results showed that whereas the sham-
lesioned animals could perform above chance on both the
sequence and recognition components of these tests, the rats
with radiofrequency-induced hippocampal lesions could
remember the individual odors but not the sequential infor-
mation. A follow-up study by Agster et al. (2002) indicated
that rats with hippocampal lesions could eventually acquire
some sequential information about repeated sequences with
extended training but never well enough to disambiguate
overlapping sequences. That the hippocampus could be
involved in rapidly learning sequences has also been the sub-
ject of modeling studies focusing on the recurrent associative
network organization of area CA3 (Jensen and Lisman, 1996;
Levy, 1996). Nakazawa’s observation that mice with a knock-
out of the NMDA receptor specific to area CA3 are poor at
“one-shot” learning but successfully master a spatial reference
memory task (Nakazawa et al., 2003) represents partial sup-
port for this view, but it would be interesting if such mice
could also be shown to be poor at learning explicit sequences
of information. Given these findings, Eichenbaum (2004)
extended his relational processing theory into the episodic
domain by arguing that the hippocampus is also the neural
substrate for mediating the sequential organization of mem-
ory. In his view, the mediation of memory for time can often
be achieved by reference to memory of the order in which
items occurred rather than some absolute memory of the time
at which they were initially presented.
In a complementary approach to thinking about sequential
aspects of events, single-unit recording studies in rats have
indicated that hippocampal pyramidal cells can sometimes
mediate more than “place” information. Chapter 11 docu-
mented the extensive evidence for the spatial sensitivity of
single-units in rodents, including place cells, theta units, head-
direction cells, and grid cells. However, we have already seen
that a growing body of evidence indicates that CA1 and CA3
pyramidal cells may have cognitive and behavioral correlates
that extend beyond the domain of space (Eichenbaum et al.,
1999; Wood et al., 1999; Jeffery, 2003).
Wood et al. (2000) recorded hippocampal complex-spike
neurons in rats running a figure-of-eight maze in which, after
moving through the common stem of the maze, the animals
had to alternate between turning left or right to secure reward
(Fig. 13–46). A high proportion of the cells (up to 67%) with
place fields on the stem fired differentially as the rat traversed
the common stem on left-turn and right-turn trials, even
when potentially confounding variations in running speed
and exact heading direction were taken into account. Other
cells fired similarly on both trial types (i.e., were place cells),
and other factors contributed to the firing determinants of the
remaining cells. They argued that hippocampal representa-
tions, as represented by cell firing, encoded information about
other nonspatial features of specific memory episodes. An ear-
lier study had also found evidence of patterns of nonspatial
firing correlates in an olfactory nonmatching task (Wood et
al., 1999). CA1 cell firing was related consistently to percep-
tual, behavioral, or cognitive events, irrespective of the loca-
tion where these events occurred.
Ferbinteanu and Shapiro (2003) went on to disentangle
factors that could be responsible for journey sensitivity, show-
ing that CA1 complex-spike cells certainly display “retrospec-
tive coding” and possibly also “prospective coding. They used
a hippocampus-dependent plus maze task (Packard and

A Figure-of-Eight Maze
Choice Point
R R
Percent correct choices
Base
C Differential CA1 unit firing as a function of journey
Figure 13–46. Task-dependent modulation of CA1 pyramidal cell
firing. A. Figure-of-eight maze as used by Wood et al. (2000) to
establish additional properties of complex-spike cells than just place
and direction sensitivity. The dotted and continuous lines refer to
alternating traverses through the central stem as the animal runs the
maze, receiving a reward (R) soon after each alternating choice. B.
Choice performance of normal and hippocampus-lesioned animals
McGaugh, 1996) in which the start arms of the maze were
switched between north and south within a block of trials, and
the goal arm switched from east to west between blocks of tri-
als. In this way, the animals could make one of four kinds of
journey: from north to either east or west and from south to
either east or west. Thus, in traversing the east arm up to the
goal, the rats could have come either from north or from
south. During recording studies, many “pure” place cells were
observed, as described by Wood et al. (2000), but differential
cell firing in the east and west arms was also observed,
depending on whether entry was from the north or south
arms, reflecting journey-sensitive “retrospective coding.”
“Prospective coding” is also claimed in the paper but accept-
ing that it is slightly problematic because the recordings were
then compared across blocks of trials in which the reward
location had been switched. Although unlikely, this goal rever-
sal might itself alter the firing properties of the cells, a com-
plication that does not arise with retrospective coding in
which reward location is fixed and recordings are taken from
within a single block of trials in which some journeys began in
the north and others in the south. One way in which this con-
founding might be addressed is through the use of two
Theories of Hippocampal Function 685
B Sensitivity to Memory Delay
100
NC
90 HPC
80
70
60
50 Chance
0 2 10 (sec)
as a function of an imposed delay before entering the stem of the
maze. C. Differential firing as a function of whether the rat has
come from the left side of the maze and is about to turn right (filled
symbols) or has come from the right side and is about to turn left
(open symbols). (Source: After Wood et al., 2000 and courtesy of
James Ainge.)
rewards (e.g., different flavors of food) at the east and west
arms of the plus maze, respectively. Then, through cueing of
the animals with a “pre-taste” of one or other food reward in
the start box, it might be possible to secure recordings within
a block of trials in which the animals turned toward the east
when cued with flavor 1 (from either north or south) and
turned to the west when cued with flavor 2. Evidence for
prospective coding uncontaminated by the use of a reversal
schedule might then be obtained.
What do these unit-recording findings imply for hip-
pocampal involvement in episodic-like memory? With respect
to neurobiological mechanisms mediating the “what-where-
when” trilogy, it is clearly necessary for place representations
to be supplemented by temporal or sequence information
(time-tagging or relative order) and by representations of
objects that are to be found at each place. These new studies of
the correlates of hippocampal cells raise the possibility that at
least “when” and “ where” might both be represented in pop-
ulation vectors of such cell firing. To date, the data are limited
to short time periods (seconds) and to the simplest descrip-
tors of sequence (e.g., south-before-east is encoded by cells
different from those representing north-before-east).
686 The Hippocampus Book
13.6.4 Problem of Awareness
Experiments with animals that lay claim to revealing a “recol-
lective” component of memory should somehow discriminate
explicit and implicit processing. We should try to face up to
the formidable problem of understanding the neurobiology of
“awareness.” Following Gaffan’s work, “asking” a monkey if it
saw a specific image on a computer screen 40 minutes ago is
now routine. Another example is the seminal work on “blind-
sight” in monkeys (Cowey and Stoerig, 1995) that was dis-
cussed earlier (Fig. 13–13). The discovery of “mirror neurons”
in primates (Rizzolatti and Craighero, 2004) further extends
our appreciation of a possible neurobiology of animal aware-
ness, as do Iriki’s studies of tool use (Maravita and Iriki, 2004).
These are telling examples because they are precisely the kinds
of phenomenon that an honorable and scholarly skeptic who
asserts the importance of language to awareness, such as
Macphail (1998), might have predicted to be impossible to
demonstrate in animals.
To be sure, cogent arguments have been put forward to
suggest that we may never know whether any animal can ever
be said to possess “autonoetic” consciousness or engage in
mental time travel and that seeking true episodic memory in
animals is therefore a forlorn exercise (Suddendorf and Busby,
2003). However, there is value in taking a positive view of the
possibility of identifying aspects of memory awareness in ani-
mals using behavioral criteria (Clayton et al., 2003b).
Language may not be central to its expression, a sense of the
self may not be required, and new behavioral protocols will
hopefully continue to reveal more about animal awareness
and the role, if any, of the hippocampus in it. In this way, the
problem of awareness is being broken down into bite-sized
chunks at which experimentalists can nibble away. We have
already seen (Fig. 13–14) that monkeys can reveal an aware-
ness of whether they can remember (Hampton, 2001). Can
animals also display quantitative characteristics of memory
that parallel those shown by humans when they are using rec-
ollection or familiarity to solve a recognition task? Can they
do one-trial cued recall? For each of these, we can go on to ask
whether performance is affected by hippocampal dysfunction.
Recollection, Familiarity,
and Region of Interest Curves
Studies examining the hippocampal contribution to the phe-
nomenological experience of memory have also been inspired
by developments in cognitive psychology. Fortin et al. (2004)
exploited a quantitative approach for distinguishing the two
components of dual-process theories of recognition mem-
ory—recollection and familiarity—that was developed during
the 1990s by Yonelinas (2001). His approach, discussed in
Chapter 12, involved plotting graphs called receiver operating
characteristic (ROC) curves, which are an important part of
the signal-detection theory (Tanner and Swets, 1954). Two
sets of probability functions, usually Gaussian, are plotted that
vary as a function of cost and benefit: “hits” and “false alarms.”
Hits are recorded when a subject correctly identifies a previ-
ously presented stimulus (i.e., a signal); false alarms are
recorded when the subject incorrectly identifies a stimulus as
seen before when it had not in fact been presented earlier (i.e.,
noise). The range of conditions involves systematic variation
of relative value, that is, the costs and benefits of correctly
reporting hits and mistakenly making false alarms. If the cost
of missing a correct stimulus is high and that of making a false
alarm is low, subjects adjust their reporting criterion in the
direction of reporting hits (i.e., to the right of the graphs in
Figure 13–47A.) Conversely, if the risks associated with incor-
rect memory are high (as in eye-witness testimony), then sub-
jects shift their response criterion downward (i.e., to the left in
the graphs).
How can this seemingly overcomplicated way of thinking
about memory performance be helpful? Yonelinas’s important
contribution was to recognize that the divergence from the
diagonal can take two forms. It may be symmetrical or asym-
metrical. In the former case, p (hit) initially rises faster than p
(false alarm) until both reach 0.5, and then the increase in p
(false alarm) catches up. This creates a symmetrical ROC
curve and reflects the familiarity component of recognition
(curved line in Fig. 13–47B). In the latter case, p (hit) rises
rapidly at low values of p (false alarm), an asymmetry thought
to be due to the second component of recognition memory—
recollection (dotted line in Fig. 13–47B). This component
operates as a threshold process in which items and their spa-
tiotemporal context are either recollected (above threshold)
or not recollected (below threshold). There is no gradation of
subjective familiarity as occurs in other components of recog-
nition memory. Like pregnancy, there are no half-measures.
Fortin et al. (2004) have now successfully established that
ROC curves for recognition memory can also be plotted for
rats. This was done taking advantage of their spectacular
memory for odors, using a continuous recognition task in
which the rats had to dig in sand wells for reward when the
odor of the adulterated sand was novel and refrain from doing
so when it was familiar (and then search elsewhere for
reward). Wood et al. (1999) had previously shown that this is
a task that displays a number of hippocampal unit correlates.
Costs and benefits were manipulated by altering the height of
the sand wells (making digging easier or more difficult) and
the size of the reward the animals got when they dug success-
fully (from one-fourth of a pellet to three pellets of food
reward). Their findings indicate that when normal rats were
tested 30 minutes after exposure to a “list” of odors (analogous
to a list of words in a human recognition experiment), the
ROC curve was asymmetrical (Fig. 13–47C). When half the
rats were then subjected to hippocampal lesioning, the ROC
curve became symmetrical (Figure 13.47D). Together with
other data in this study, it seems that recognition memory in
rats complies with dual-process models of recognition mem-
ory in humans and that hippocampal lesions may remove
the “recollection” component of this memory. The task of
realizing a neurobiological understanding of episodic and
episodic-like memory and the nature of hippocampal involve-
 Theories of Hippocampal Function 687

A Gaussian ‘noise’ and ‘signal’ B Two components of ROC curves

1.0
ty
iari
mil

Probability of hits
0.8 Fa
noise signal
0.6
tion
collec
0.4 Re

0.2

0.0

Θ
C Preoperative odor recognition D Postoperative odor recognition
1.0 1.0 Controls
Hippocampus
Probability of hits

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Probability of false alarms Probability of false alarms

Figure 13–47. Receiver operating characteristic (ROC) analysis of
recognition memory in the rat. A. Hypothetical Gaussian functions
plotting the probability of “signal” and “noise” during memory of
a prior event. The receiver operates at various points along the x-
axis and can so make estimates of what constitutes a “hit” (i.e., a
true signal) and what constitutes a “false-alarm” (i.e., from the noise
distribution but claimed to be a memory). The threshold  can
be adjusted up or down the x-axis to enable a more conservative (to
the right) or more risky (to the left) detection policy, as described
in the text. B. Quantitative analysis of ROC functions reveals a lin-
ear recollection component and a symmetrical familiarity compo-
nent of memory. The sum of the two components is a curved but
asymmetrical function. C. Normal rats show asymmetrical function
in odor recognition memory. D. After hippocampal lesions, rats
show a symmetrical ROC function. (Source: After Fortin et al.,
2004.)

ment in it seems no longer quite as forlorn an enterprise as
once it did. It is to one developing theory of this type that we
now turn.
13.6.5 Elements of a Neurobiological
Theory of the Role of the Hippocampus
in Episodic-like Memory
It is premature but not inappropriate to end this chapter with
an attempt to tie several strands together. The focus in this
chapter has been on the major historical contribution that
animal lesioning studies over the past 30 years have made to
our understanding of hippocampal function. Modern work
has greatly expanded the levels of analysis at which neurosci-
entists think about function and has been converging on the
idea of a role in the automatic aspects of episodic-like mem-
ory. There has been little discussion so far of the specific role
hippocampal synaptic plasticity might play in this process or
how the cell biological determinants of that plasticity fit with
neuropsychological ideas about memory consolidation. Given
the shift toward more interdisciplinary approaches in neuro-
science, it is fitting to end on this note.
Automatic Recording and Retrieval
of Attended Experience
One neurobiological theory of hippocampal function does
attempt to specify a role for synaptic plasticity in memory for-
mation (Morris et al., 2003). This theory supposes that hip-
pocampal memory can, like other forms of memory, be
divided into four processes: encoding, storage, consolidation,
and retrieval. A key aspect of the neurobiological approach is
the assertion that activity-dependent synaptic plasticity (e.g.,
LTP) is critical for encoding and the intermediate storage of
memory traces in the hippocampus that correspond to the
“indices” or “pointers” to cortical regions where detailed per-
ceptual information is temporarily stored (as discussed in
Section 13.3). These hippocampal indices are cartoons that
bind event information to the context in which they occur. A
novel feature of the theory is that these index memory traces
decay rapidly but are sometimes rendered more persistent by
the process of cellular consolidation. That is, LTP is actually
something of a misnomer as many, perhaps most, memories
of events captured online are soon lost. Protein synthesis-
independent E-LTP is more of a system for creating the
688 The Hippocampus Book
“potential” for memory with synaptic tagging at the time of
E-LTP induction (see Chapter 10) playing a part in determin-
ing what traces are rendered more persistent. The rapid “snap-
shot” creation of a temporary event-context index involves
synaptic potentiation at a large number of synapses across
parts of the hippocampal network (probably in CA1 and
CA3). Tags set automatically at the time of an event, irrespec-
tive of how trivial that event may be, can capture the products
of the somatic or dendritic synthesis of plasticity proteins set
in train by events before, during, or even after an event to be
remembered. This capturing process renders the temporary
indices in the hippocampus more permanent and thus allows
sufficient time for systems-level consolidation to occur. This
two-stage process of consolidation serves as a “protective fil-
ter” in the sense that rapidly decaying index traces do not last
long enough to guide neocortical consolidation; and thus
events that are not subject to cellular consolidation in the hip-
pocampus do not become represented in a persistent manner
in the cortex. Although hippocampal synaptic plasticity is
critical for encoding and its expression subject to intrahip-
pocampal cellular consolidation, the theory also asserts that it
is not involved in memory retrieval. It is also unlikely to be
involved in the widely discussed systems-level consolidation
process that depends on hippocampus–neocortex interac-
tions, although it does not preclude the possibility that neo-
cortical synaptic plasticity is critical. More formally, the
propositions of this developing theory are presented in Box
13–11.
Box 13–11
Neurobiological Theory of Hippocampal Function
1. Some animals have episodic-like memory, and the hip-
pocampal formation is one group of brain structures that
can encode and retrieve such memories.
2. Activity-dependent, associative, NMDA receptor-depend-
ent synaptic plasticity in the hippocampus is the primary
neural mechanism responsible for inducing temporary
encoding and storage of hippocampal “indices” of rapidly
acquired event-context memories. The detailed perceptual
information associated with events is represented in the
cortex.
3. An essential feature of cellular consolidation processes to
which these index traces are subject is the interaction
between local “synaptic tags”(set by glutamatergic activa-
tion), and diffusely targeted “plasticity proteins” (which
can be triggered by heterosynaptic activation of neuro-
modulatory and glutamatergic inputs).
4. Systems-level consolidation requires both hippocampus
and neocortical neural activity and is therefore not solely
time-dependent although certainly a process that takes
time. It does not require hippocampal plasticity but may
engage mechanisms of neocortical plasticity.
5. The hippocampal formation is one of a number of brain
regions that detects novelty, and its neural activity alters
in ways that enable a representation of novel events to be
encoded rapidly online.
This framework builds unashamedly on the functional/
neuropsychological theories discussed earlier in the chapter,
but it also makes a number of novel predictions, including
some counterintuitive predictions, at other levels of analysis.
Some examples of the shoulders on which it stands and these
predictions are as follows. From a neuropsychological per-
spective, it recognizes that events happen in particular places
at particular times, and their later recall generally includes the
memory of where and when an event occurred (Gaffan,
1994b; Clayton et al., 2001). Thus, event encoding is necessar-
ily associative and “contextual” in character, a point empha-
sized in several functional theories (e.g., cognitive mapping,
configural associative, relational). Many events cannot be
anticipated, occur only once, and may contain distinct fea-
tures that, in sequence, form short episodes—a point about
episodic memory emphasized by Eichenbaum (2004). It is
vital that traces representing information about such episodes
are encoded and stored in real time (as they happen), a
process that Frey and Morris characterized as the “automatic
recording of attended experience” (Morris and Frey, 1997).
Not all events are remembered for any length of time, and to
do so is not only unnecessary but might also saturate the stor-
age capacity of relevant areas of the brain. Moser et al. (1998)
exploited this feature of hippocampal memory in their contri-
bution to studies examining the functional significance of
hippocampal LTP (see Chapter 10). With respect to the level
of representational detail, it is also unclear whether the hip-
pocampus need receive, via its extrinsic afferents, detailed sen-
sory/perceptual information pertaining to individual objects
or events. The networks and local circuits for detailed percep-
tual processing are in the cortex. Rather, all it needs to remem-
ber events and the sequence in which they occur is binding
information about the locations in the neocortex where this
detail is processed and temporarily encoded—a point about
hippocampus–neocortex interactions discussed in Chapter 3.
Later processing, guided by these indices, determines whether
the temporary neocortical encoding becomes permanent.
This view of hippocampal memory processing capitalizes on
the advances made in other theories; and, like them, it implies
that the hippocampus is far from being an island. Realizing its
functions absolutely requires the synergistic activity of other
brain areas both upstream and downstream. This is therefore
a framework that, unlike some others (McClelland et al.,
1995), implies that the cortex must be able to encode infor-
mation rapidly online. The view here is not only that the cor-
tex can do this but that it loses information rapidly unless new
connections are stabilized by systems-level consolidation. In
this sense, consolidation is permissive rather than instructive.
The next step of this framework takes us farther into the
neurobiological domain. If events are to be encoded, there
must exist physiological mechanisms for capturing events as
they happen and encoding traces that could later enable
retrieval of event or event-associated information. A promi-
nent candidate for the neural substrate of event memory is
hippocampal NMDA receptor-dependent synaptic plasticity,
so-called Hebbian plasticity. Such plasticity, as revealed by the

physiological phenomenon of LTP, exhibits many properties
that are suitable for a role in memory formation (Bliss and
Collingridge, 1993; Martin et al., 2000), and a growing body of
evidence offers support for this view (see Chapter 10). Index
memory traces are achieved by an alteration in the connectiv-
ity between neurons such that retrieval cues, expressed as pat-
terns of neural activity, are able to induce new patterns of
neural activity that are one part of the substrate of “memo-
ries” of prior events.
Another strand relates to the persistence of such traces. A
key supposition is that, notwithstanding the fantastic storage
capacity of the brain, there is no need to store everything per-
manently, and the system is absolutely not evolved to do so.
Indeed, most memories fade and are lost. Of those that persist,
retrieval processes contribute to their accessibility such that
“forgetting” is a dual function of trace decay and retrieval fail-
ure. If most automatically encoded event memories are tem-
porary, there must be psychological processes and neural
mechanisms for selecting the subset of traces that are to be
rendered longer lasting or even permanent, such as the emo-
tional significance of the event to be remembered or, more
interstingly, the significance of other events happening close
together in time or space. Experiments on “behavioral tag-
ging” are needed to explore the predictions of the synaptic
tagging framework in the context of real memory (see
Chapter 10, Section 10.10). There is also evidence that the per-
sistence of memory can also be determined by the relevance of
ongoing events to the existing knowledge structures and
interests of the person witnessing them (Bartlett, 1932;
Bransford and Johnson, 1972; Bransford, 1979), an idea long
known in educational circles but that has had curiously little
impact on neuroscience until recently (Maguire et al., 1999).
Mediating these psychological processes of persistence are two
neuronal mechanisms of memory consolidation: (1) cellular
consolidation mechanisms, which include the synthesis and
synaptic capture of plasticity proteins that stabilize memory
traces at the level of the individual synapse in neurons (Goelet
et al., 1986; Frey and Morris, 1998; Sweatt, 2003); and (2) sys-
tems-level consolidation mechanisms, as discussed in previ-
ous sections of this chapter, which reflect a dynamic
interaction between populations of neurons in the hippocam-
pus and neocortex (Dudai and Morris, 2001). These forms of
consolidation are distinct but also interdependent. This inter-
dependence arises because cellular consolidation enables
memory indices in hippocampus to last long enough for the
slower systems-level consolidation process to work and thus
for their time courses to dovetail.
The defining functional characteristics of associative net-
works such as the hippocampus are believed by several theo-
rists to include distributed representations, interleaved storage
across multiple synapses, and associative retrieval. These
processes enable patterns of activity to be stored as a matrix of
synaptic changes and retrieval to be “completed” from partial
fragments of the original input (Marr, 1971; McNaughton and
Morris, 1987; Paulsen and Moser, 1998; Nakazawa et al.,
2002). Several factors determine the operating characteristics
Theories of Hippocampal Function 689
and storage capacity of such networks. One of them, connec-
tivity density (i.e., the number of connections per cell), pro-
vides an anatomical basis for understanding an important
feature of the relation between hippocampus and neocortex
(McNaughton et al., 2003). Specifically, the average connectiv-
ity in the cortex is, in general, too low to support the encoding
of arbitrary associations (Rolls and Treves, 1998). The cortical
mantle contains on the order of 1010 neurons, but each corti-
cal principal neuron receives only about 104 connections.
Thus, the average connection probability in cortex is only
1:106. To overcome this apparent biological limitation, mam-
mals seem to have evolved an arrangement whereby distrib-
uted associative memory between items represented in
different sensory modalities can be accomplished through
indirect associations mediated by a hierarchical organization
(McNaughton et al., 2003). In such a scheme, neocortical
modules at the base of the hierarchy are reciprocally con-
nected via modifiable synapses with one or more hippocam-
pal modules at the apex. The hippocampal modules include
CA3 as well as the dentate hilus, both characterized by high
internal connectivity as well as modifiable synapses. In CA3,
each pyramidal cell is contacted by ~ 4% of the pyramidal
cells of the same subfield (Amaral, 1990), implying that most
CA3 pyramidal cells are connected via two or three synaptic
steps (Rolls and Treves, 1998). This high degree of internal
recurrent connectivity is probably sufficient to allow autoas-
sociation, or association among individual elements of a pat-
terned input (Marr, 1971). Activity patterns reflecting sensory
detail in neocortical modules may generate a unique identify-
ing index in such a network (Teyler and DiScenna, 1987). This
high-level hippocampal trace index is no longer “sensory” in
any strict sense but is stored associatively with other indices
and the output fed back to neocortical modules via modifiable
synapses. Activation of a cortical pattern (e.g., a specific flavor
of food) could then result in activation of its index in the hip-
pocampus. In turn, this enables retrieval of associated indices
and thence the complementary pattern in the other cortical
modules (e.g., where the food is found). Indirect associations
enable memory retrieval between cortical modules that are
too sparsely connected to do this directly.
This principle of indirect association in memory places
high demands on the synaptic storage capacity of the hip-
pocampus that, as noted briefly above, is otherwise in danger
of becoming saturated. Once saturated, learning can no longer
proceed effectively (McNaughton and Barnes, 1986; Moser et
al., 1998). There are several ways in which this burden may be
limited. One, as already noted, is via rapid decay of a high pro-
portion of the traces that are automatically encoded online (a
property of E-LTP). Heterosynaptic depression may also serve
a normalizing function and so increase effective storage
capacity (Willshaw and Dayan, 1990). A third way to limit the
rate of use of storage space, also an element of the present the-
oretical ideas, would be to ensure that what is stored in the
hippocampus is only a cartoon, with the full sensory/percep-
tual details encoded in the cortex. The fourth way—the
process of systems-level memory consolidation itself—would
690 The Hippocampus Book
be to enable hippocampal associations that are strongly
and/or repeatedly recalled, such as those representing envi-
ronmental regularities, to trigger the development and stabi-
lization of low-level intermodular connections in the
neocortex. It is these connections that could later enable cor-
tical retrieval in the absence of neural activity in the hip-
pocampus. To work, it is vital that the intermodular
connections that develop are appropriate to the associations
represented. This requires the gradual interleaving of appro-
priate connections (McClelland et al., 1995), perhaps during
sleep (McNaughton et al., 2003). Interestingly, the rate of con-
solidation may not be strictly time-dependent. Rather, the
process may be “cladistic,” with the rate of consolidation
affected by the frequency with which hippocampal indices are
reactivated. It may also depend on an activated neocortical
schema to receive the new information.
The last step to be considered relates to the detection of
novelty and acting on it. If the hippocampus is critical for
building context representations and encoding events in rela-
tion to them, it is likely that its neural activity reflects changes
to the environment. This is not a new idea, as Gray (1982),
Vinogradova (1995), and more recently Gray (2000) have
extensively discussed possible “comparator” functions of the
hippocampus. Novelty detection is by no means unique to the
hippocampus, it being a process required in several cognitive
systems; but evidence in favor of a hippocampal contribution
has come from many experiments on different forms of nov-
elty detection, including the studies of exploratory behavior in
which rats examine objects and landmarks in simple arenas
(see Section 13.4). The misplace cells of O’Keefe (1976) and
the more recent observations of neural firing in response to
the movement of a hidden escape platform in a watermaze
(Fyhn et al., 2002) also suggest that the hippocampus, perhaps
acting in concert with neuromodulatory systems, detects and
represent changes in the conjunction of objects and their loca-
tions. Several cells fired vigorously at the new platform loca-
tion, despite previously having been silent. Others that fired in
different locations around the maze continued to do so after
platform relocation, arguing against spatial remapping. The
new activity was paralleled by reduced discharge in a subset of
simultaneously recorded interneurons. The pattern of activity
largely returned toward its original configuration as the rat
learned the new location. However, a few of the newly
recruited neurons remained active. This persistent firing may
reflect facilitated synaptic plasticity during the temporary
reduction in inhibition (Wigstrom and Gustafsson, 1983;
Paulsen and Moser, 1998). NMDA receptor-dependent LTP
may be necessary for these permanent modifications in firing
patterns when novel events occur in a familiar environment.
Hippocampal Synaptic Plasticity, Neural
Activity, and Rapid Paired-associate Learning
Day et al. (2003) developed a one-trial paired-associate task in
which rats were trained to recall in which of two locations a
particular flavor of rat chow was to be found in a large “event
arena” (Fig. 13–48). In each of two sample trials on each day a
few minutes apart, rats left the start box to find a single open
sand well where they could dig for a flavored food. No choices
were required on this sample trial, and there was nothing in
the procedure that would have required the animals to engage
a contingent learning mechanism. The animals merely ran out
into the arena, found a sand well, dug in it, and found food.
This constituted one “event.” The two sample trials did, how-
ever, use two locations and two foods. This gave the animals
the opportunity of automatically encoding—in one trial for
each sandell—that one flavor of chow was at one location and
another flavor at another (Fig. 13–48A). They therefore had
two “events” prior to the daily choice trial, which was started
5 to 20 minutes later. It began by giving the rats one of the two
flavors to eat in the start box (their recall cue). A short while
later, they were allowed to enter the arena where they now
found two sand wells (Fig. 13–48B). On choice trials during
training, going to the location associated with the flavor cue
was rewarded with more of that same flavor. Nonrewarded
probe trials were also run. Up to 30 locations and flavors were
paired in novel combinations at the rate of two pairs per day.
The results obtained with this “what–where” paradigm pro-
vided evidence that rats are capable of a limited form of
episodic-like memory—and in a species more amenable to
neurobiological study than the scrub-jays used by Clayton.
The protocol lacks the dimension of time (the “when” ele-
ment), which is a concern; but it shares with the avian food-
caching paradigms that recall is required rather than
recognition, and this represents a step forward from the
object-in-place tasks used by Gaffan.
Day et al. (2003) also established that the one-trial encod-
ing of such paired associates lasted no longer than 60 minutes
before recall performance dropped to chance (Fig. 13–48C).
Nonrewarded probe trials established that there was a prefer-
ence for digging at the cued location despite the limited nature
of the “event” that had uniquely taken place there that day.
Encoding was sensitive to the acute intrahippocampal infu-
sion of an NMDA receptor antagonist (D-AP5) without any
effect on retrieval (Fig. 13–48D,E), whereas retrieval was
insensitive to the AMPA receptor antagonist CNQX (Fig.
13–48E). These pharmacological observations were made by
timing the infusions to be either just before or just after the
encoding trial. That a difference between encoding and
retrieval was obtained when the dorsal hippocampus was
infused with D-AP5 indicates that the mere presence of the
drug has no effect on memory retrieval, but it definitively
blocks encoding. It seems more likely, and is in keeping with
physiological studies revealing its importance for associative
LTP, that an NMDA receptor-dependent mechanism in the
hippocampus is involved in encoding the paired associate of
an event and its location into memory in such a manner that
the reactivation of one member of the pair can retrieve the
memory of the other.
The “event arena” has the potential to serve as a flexible test
bed for examining the neurobiology of rapid associative
memory of the kind that may underlie human episodic mem-

A Sample trial 1 (encoding) Sample trial 2 (encoding)
Start-boxes
F1
sand wells
door
C Within-day forgetting D Drug infusion prior to encoding
70
100
Dig times at correct
Sand-well digging (%)
80
sand-well (%)
60
60
40
50
Chance 20
40 0
5 min 90 min 540 min aCSF
Figure 13–48. Cued recall of the location of a single event. A. The
1.6 m2 “event arena” made of plexiglass consisted of 49 sand wells, 2
intramaze landmarks, and 4 start boxes. On sample trial 1, the door
to a start box is drawn back, and the rat runs out into the arena
(dotted line) where it displays occasional lateral head movements
to find food 1 at the single open well. Sample trial 2 is to a different
food (F2) at a different location. B. The cued-recall choice trial
begins with presentation of either of the two sample trial foods in
the start box (food F1 is shown) followed by the rat being rewarded
ory (what, where, and when). However, there are several fea-
tures of behavioral performance in this task that need to be
clarified and improved. First, the temporal component has yet
been addressed, though it may be possible to use the surrogate
of contextual specificity. Second, no hippocampal lesion study
has yet been reported, and there is the difficulty that the drugs
used may be directly affecting only one component of the
what-where-when triad, such as the spatial component. That
is, the drugs’ apparent impact on associative encoding or
retrieval could be misleading. Studies of one-trial spatial
memory in the event arena have helped clarify this point (Bast
et al., 2005). Third, alterations in the protocol to look at the
concurrent training of multiple places and flavors, analogous
to learning a list of object–place pairs, are also underway.
Conducted in the spatial domain, such a study would cast
light on how multiple spatial associates could become inte-
grated into “spatial schema” that might be constructed and
stored in the neocortex. Preliminary data from such a study by
Tse, Kakeyama and Langston (personal communication) indi-
cates that such spatial schema can be gradually acquired over
Theories of Hippocampal Function 691
B Choice trial (retrieval)
F1+
F1
F2 No F2
E Drug infusion prior to retrieval
p < 0.025 NS
p < 0.01
NS
Chance
AP5 CNQX aCSF AP5 CNQX
selectively for digging at the sand well containing this same food.
C. Rapid decay of event-place memory over the course of the day.
D, E. Nonrewarded probe trials show that the animals preferentially
digs in the sand well appropriate to the cue flavor given in the start
box. Memory encoding in this task is sensitive to blockade of
NMDA receptors (D), whereas memory expression requires only
fast synaptic transmission via AMPA receptors (E). (Source: After
Day et al., 2003.)
several weeks; and once fully encoded and stored, new infor-
mation can be added to the schema rapidly. If the schema can
be shown to be neocortical, which preliminary evidence indi-
cates they are, such a study may offer some handle on whether
neocortical storage of information in a hippocampus-
dependent learning task can be achieved during a single trial.
This would provide vindication for the unusual prediction of
this neurobiological theory that cortical learning is sometimes
just as fast as hippocampal learning. Fourth, it would also be
valuable to complement lesion and inactivation studies with
physiological studies investigating whether there is differential
c-fos or Arc activation in the hippocampus during this task,
and if it is possible to identify neural representations of cued
recall using ensemble single-unit recording. The latter would
be a formidable undertaking but a valuable one, as it might
reveal anticipatory components of place-specific unit firing
driven by the task demands in which a flavor cue evokes a
memory of a place. This would be a logical extension of the
concepts of prospective and retrospective coding introduced
by Ferbinteanu and Shapiro (2003).
692 The Hippocampus Book
Storage of Spatial/Contextual Information
Outside the Hippocampus?
An important milestone in the understanding of spatial cog-
nition was the discovery that most pyramidal cells in the hip-
pocampus exhibit location-specific activity and that the
activity of such place cells is influenced by the training history
of the animal (see Chapter 11). The predominantly spatial
nature of hippocampal neuronal activity led to the proposal
that place cells form a distributed map-like representation of
the spatial environment that an animal uses for efficient nav-
igation (O’Keefe and Nadel, 1978). When a rat is exposed to a
new environment, pyramidal cells develop distinct firing fields
within a few minutes (Hill, 1978; Wilson and McNaughton,
1993b). The place fields then remain stable for weeks or more
if the environment is constant (Thompson and Best, 1990;
Lever et al., 2002), as predicted if these cells contribute to a
particular spatial memory.
It is commonly believed that the development of hip-
pocampal firing patterns, like hippocampal memory, might
depend on LTP at hippocampal excitatory synapses. Several
studies have investigated the contribution of LTP to spatial fir-
ing in hippocampal pyramidal cells. Surprisingly, interven-
tions that abolish hippocampal LTP have only subtle effects on
place fields. For example, place-specific firing fields develop
normally in rats treated systemically with an NMDA receptor
antagonist at a dose that prevents new LTP in the hippocam-
pus (Kentros et al., 1998). Despite blockade of the NMDA
receptor, place fields can be maintained between consecutive
test sessions in the same environment for at least 1.5 hours.
They do, however, show instability across days. Place fields
recorded in CA1 in mice with mutations of NMDA receptors
in either CA3 or CA1 are somewhat less distinct than in
control mice under certain conditions; but in contrast to
the pharmacological findings, firing fields remain stable across
repeated tests on the same day once they have been established
(McHugh et al., 1996; Nakazawa et al., 2002). Similar results
were obtained when LTP was blocked by interference with
Ca2/calmodulin-dependent protein kinase II (Rotenberg et
al., 1996) or protein kinase A (Rotenberg et al., 1996).
However, these interventions did decrease the long-term
stability of the place fields, as measured 24 hours after the ini-
tial exposure to the environment. Together, these studies sug-
gest that NMDA receptor activation may not be necessary for
the development or initial maintenance of place-related activ-
ity in hippocampal neurons, although it may contribute to
the fine-tuning of place fields and aspects of the long-term sta-
bilization of spatial representations (see Chapter 11). The lat-
ter function could involve the neocortex as well as the
hippocampus.
The development of place fields during NMDA receptor
blockade or other forms of disruption of LTP is consistent
with several lines of work suggesting that spatial information
can be generated and stored upstream of the hippocampus.
First, location-specific firing is already expressed in the super-
ficial layers of the entorhinal cortex (Quirk et al., 1992; Frank
et al., 2000; Fyhn et al., 2004; Hafting et al., 2005). The fact
that most principal cells in entorhinal cortex exhibit view-
independent spatial modulation suggests that, by this stage, a
fundamental spatial computation may already have been
made. It is also possible, however, that spatial firing in super-
ficial entorhinal neurons depends on spatial input from cells
in deep layers, which in turn may rely on associative compu-
tations in afferent hippocampal structures. The manner in
which entorhinal cells code space remains unclear, but that
they do helps to understand why pyramidal cells in CA1
exhibit spatial firing both after selective lesions of the dentate
gyrus (McNaughton et al., 1989) and after disconnection of
CA1 from CA3 (Brun et al., 2002). In CA3-lesioned animals,
CA1 pyramidal neurons receive cortical input only via direct
connections from the entorhinal cortex. The presence of place
fields in these preparations suggests that direct entorhinal–
hippocampal circuitry has significant capacity for transform-
ing weak location-modulated signals from superficial layers of
the entorhinal cortex into accurate spatial firing in CA1.
Several simple filter mechanisms could accomplish such a
transformation. For example, firing rates of perforant path
fibers to CA1 could be thresholded by feedforward inhibition,
such that only the highest afferent firing rates (i.e., those in the
center of the entorhinal place field, are able to drive postsy-
naptic neurons in the hippocampus. Alternatively, single exci-
tatory postsynaptic potentials (EPSPs) in distal pyramidal cell
dendrites of CA1 may often not be sufficient to trigger
somatic action potentials in these cells; reliable discharge may
require summation of EPSPs (i.e., high afferent firing rates
(Golding and Spruston, 1998; Golding et al., 2002).
It is important to note that the computation and storage of
positional information outside the hippocampus does not
preclude long-term storage of some spatial information (or
the indices of such information) in the hippocampus. Indeed,
the internal recurrent connectivity of hippocampal area CA3
makes the region highly suitable for storage of just these types
of patterned “cartoons,” at least for an intermediate period of
time and perhaps longer in some cases. Other results suggest
that plasticity in associational synapses of CA3 is necessary for
successful “snapshot” encoding of spatial information in a
manner that later allows recall with partial cues (Nakazawa et
al., 2002). Longitudinal axon collaterals in CA3 may be
important for successful retrieval of such information, as
memory retention may be impaired by a single transversely
oriented cut through the dorsal CA3 region of each hip-
pocampus (Steffenach et al., 2002).
A Last Word
It remains to be seen how this and other neurobiological the-
ories of episodic-like memory (Hasselmo et al., 2002;
McNaughton et al., 2003; Nakazawa et al., 2004; Buzsaki, 2005;
Hasselmo and Howard, 2005) will handle the growing body of
data that point to a specific role for the hippocampus in at
least aspects of “episodic-like” memory. The approach just
described builds on the neuropsychological framework devel-
oped in this chapter. It accepts that understanding the neu-
roanatomical basis of recognizing a previously seen stimulus
 Theories of Hippocampal Function 693

is an important foundation stone of declarative memory (see
Section 13.3) but asserts that there is more to memory than
this (as, indeed, declarative memory theory has long recog-
nized). It also accepts that hippocampal cells agree that space
is special but argues that they can do more than encode spa-
tial information and calls into question the supposition that
allocentric spatial memory is fundamentally different from
other forms of associative memory (see Section 13.4). It tries
to take on board the somewhat impenetrable developments in
contemporary animal learning theory and the progress that
has been made in understanding predictable ambiguity (see
Section 13.5). Perhaps most difficult of all, it is part of an
international effort seeking to integrate functional and mech-
anistic ideas about rapid, one-trial, context-specific learning.
This task requires an exquisite understanding of hippocampal
neuroanatomy, including that of the myriad inhibitory
interneurons and the local circuits of which they are a part
that surely make a major contribution to the information pro-
cessing “algorithms” computed in the hippocampus. Cell bio-
logical approaches will also prove increasingly valuable when
thinking about the problems of trace stabilization and mem-
ory consolidation. A major obstacle ahead is understanding
better how information is “represented” as patterns of neural
activity, and progress in unraveling the physiology of hip-
pocampal neurons will aid in that formidable task (see
Chapters 5 and 6). In this research program, a wide range of
behavioral tasks with animals will continue to play a key part
in unraveling the functions and neural mechanisms of the
hippocampal formation. This chapter has drawn on data
derived from primate experiments investigating recognition
memory and object-in-place memory and from rats running
in radial mazes, digging in odorized sand wells for food, or
swimming in watermazes; and it has touched on new behav-
ioral techniques to investigate episodic-like memory. It is fit-
ting to acknowledge and celebrate the contribution that
laboratory animals have made to this effort (Fig. 13–49, see
color insert). Progress is being made in tackling the mysteries
of the hippocampal “memory machine.” The challenges ahead
are as formidable as have been the achievements in character-
izing its role in memory over the past three decades.

Figure 13–49. Examples of common behavioral tasks used to study hippocampal function in
animals. A. Classic Wisconsin General Testing Apparatus for macaque monkeys. B. Screen shot
of modern “touchscreen” display. C. Radial maze foraging task for rats, shown with doors. D.
Rat making a choice in a two-pot olfactory discrimination task. Only one odor is familiar or
may be cued by another odor in a paired-associate manner.
694 The Hippocampus Book
Figure 13–49. (Continued) E. Foraging monkey exploring objects in
a laboratory. F. Watermaze. Rat swimming in a large 2-m pool try-
ing to find a hidden escape platform under the water surface (just
visible on the right). Inset (below). The rat is on the platform, rear-
ing and working out his location. G. Scrub-jay in an episodic-like
Á ACKNOWLEDGMENTS
I am grateful to many people for their help in preparing this chapter.
In addition to my co-editors of this book, Mark Baxter (Oxford
University) read a draft of the entire manuscript and offered numer-
ous comments and criticisms. Tobias Bast, Stephen Martin
(University of Edinburgh and other members of the Laboratory for
Cognitive Neuroscience), Edvard Moser and May-Britt Moser
(Center for the Biology of Memory, Trondheim), and Craig Stark
(Johns Hopkins) also read large parts and directed me to additional
literature. I am grateful to Helen Knight, Marie Pezze, and Carolyn
Smith for collecting material for and assisting with many of the fig-
ures. I am also grateful to many neuroscience colleagues around the
world who provided material for the figures or advice of other kinds
(including Jamie Ainge, Verner Bingman, Norbert Fortin, John
“what-where-when” memory task in which it caches various foods
in an ice-cube tray and then retrieves them selectively later. H.
Paired-associate event arena task for rats. Different foods are found
at specific locations (e.g., one shown) in the apparatus.
Guzowski, Lucia Jacobs, Len Jarrard, Kazu Nakazawa, and Simon
Rempel). Patrick Spoooner helped with many aspects of the comput-
ing. I apologize to yet others who helped but whom I have not men-
tioned. Finally, one institution should not be forgotten. I am eternally
grateful to the UK Medical Research Council without whose support
throughout my career this work would not have been possible.
Á REFERENCES
Abel T, Lattal KM (2001) Molecular mechanisms of memory acqui-
sition, consolidation and retrieval. Curr Opin Neurobiol
11:180–187.
Adams CD, Dickinson A (1981) Instrumental responding following
reinforcer devaluation. Q J Exp Psychol 33B:109–122.

Adolphs R (2003) Cognitive neuroscience of human social behav-
iour. Nat Rev Neurosci 4:165–178.
Aggleton JP (1985) One-trial object recognition by rats. Q J Exp
Psychol 37B:279–294.
Aggleton JP (1999) Mapping recognition memory in the primate
brain: why it’s sometimes right to be wrong. Brain Res Bull
50:447–448.
Aggleton JP, Brown MW (1999) Episodic memory, amnesia, and the
hippocampal-anterior thalamic axis. Behav Brain Sci 22:
425–489.
Aggleton JP, Pearce JM (2001) Neural systems underlying episodic
memory: insights from animal research. Philos Trans R Soc Lond
B Biol Sci 356:1467–1482.
Aggleton JP, Shaw C (1996) Amnesia and recognition memory: a re-
analysis of psychometric data. Neuropsychologia 34:51–62.
Aggleton JP, Vann SD (2004) Testing the importance of the retrosple-
nial navigation system: lesion size but not strain matters: a reply
to Harker and Whishaw. Neurosci Biobehav Rev 28:525-531.
Aggleton JP, Hunt PR, Rawlins JN (1986) The effects of hippocampal
lesions upon spatial and non-spatial tests of working memory.
Behav Brain Res 19:133–146.
Agster KL, Fortin NJ, Eichenbaum H (2002) The hippocampus and
disambiguation of overlapping sequences. J Neurosci 22:
5760–5768.
Aiba A, Chen C, Herrup K, Rosenmund C, Stevens CF, Tonegawa S
(1994) Reduced hippocampal long-term potentiation and con-
text-specific deficit in associative learning in mGluR1 mutant
mice. Cell 79:365–375.
Alvarado MC, Rudy JW (1995a) A comparison of kainic acid plus
colchicine and ibotenic acid-induced hippocampal formation
damage on four configural tasks in rats. Behav Neurosci
109:1052–1062.
Alvarado MC, Rudy JW (1995b) Rats with damage to the hippocam-
pal-formation are impaired on the tranverse-patterning prob-
lem but not on elemental discriminations. Behav Neurosci
109:204–211.
Alvarez-Royo P, Zola-Morgan S, Squire LR (1992) Impairment of
long-term memory and sparing of short-term memory in mon-
keys with medial temporal lobe lesions: a response to Ringo.
Behav Brain Res 52:1–5.
Alvarez P, Zola-Morgan S, Squire LR (1994) The animal model of
human amnesia: long-term memory impaired and short-term
memory intact. Proc Natl Acad Sci USA 91:5637–5641.
Alvarez P, Zola-Morgan S, Squire LR (1995) Damage limited to the
hippocampal region produces long-lasting memory impair-
ment in monkeys. J Neurosci 15:3796–3807.
Alyan S, McNaughton BL (1999) Hippocampectomized rats are capa-
ble of homing by path integration. Behav Neurosci 113:19-31.
Amaral DG (1990) Neurons, numbers and the hippocampal net-
work. Prog Brain Res 83:1–11.
Amaral DG, Witter MP (1989) The three-dimensional organization
of the hippocampal formation: a review of anatomical data.
Neuroscience 31:571–591.
Amaral DG, Insausti R, Cowan WM (1987) The entorhinal cortex of
the monkey. I. Cytoarchitectonic organization. J Comp Neurol
264:326–355.
Anagnostaras SG, Maren S, Fanselow MS (1999) Temporally graded
retrograde amnesia of contextual fear after hippocampal damage
in rats: within-subjects examination. J Neurosci 19:1106–1114.
Anagnostaras SG, Gale GD, Fanselow MS (2001) Hippocampus and
contextual fear conditioning: recent controversies and advances.
Hippocampus 11:8–17.
Theories of Hippocampal Function 695
Anagnostaras SG, Gale GD, Fanselow MS (2002) The hippocam-
pus and pavlovian fear conditioning: reply to Bast et al.
Hippocampus 12:561–565.
Andersen P, Bliss TVP, Skrede KK (1971) Lamellar organization of
hippocampal excitatory pathways. Exp Brain Res 13:222–238.
Andersen P, Soleng AF, Raastad M (2000) The hippocampal lamella
hypothesis revisited. Brain Res 886:165–171.
Andersson M, Krebs JR (1978) On the evolution of hoarding behav-
iour. Anim Behav 26:707–711.
Angeli SJ, Murray EA, Mishkin M (1993) Hippocampetized monkeys
can remember one place but not two. Neuropsychologia
31:1021–1030.
Archer J, Birke L (1983) Exploration in animals and humans. London:
Van Nostrand Reinhold.
Attwell PJ, Rahman S, Yeo CH (2001) Acquisition of eyeblink condi-
tioning is critically dependent on normal function in cerebellar
cortical lobule HVI. J Neurosci 21:5715–5722.
Bachevalier J, Brickson M, Hagger C (1993) Limbic-dependent
recognition memory in monkeys develops early in infancy.
Neuroreport 4:77–80.
Bachevalier J, Nemanic S, Alvarado MC (2002) The medial temporal
lobe structures and object recognition memory in nonhuman
primates. In: Neuropsychology of memory, 3rd ed. (Squire LR,
Schacter DL, eds), pp 326–339. New York: Guildford Press.
Bachevalier J, Vargha-Khadem F (2005) The primate hippocampus:
ontogeny, early insult and memory. Curr Opin Neurobiol
15:168–174.
Baddeley A (2001) The concept of episodic memory. Philos Trans R
Soc Lond B Biol Sci 356:1345–1350.
Baddeley A, Conway M, Aggleton JP (2002) Episodic memory: new
directions in research. Oxford, UK: Oxford University Press.
Balda RP, Kamil AC (1989) A comparative study of cache recovery by
three corvid species. Anim Behav 38:486–495.
Balsam PD, Tomie A (1985) Context and learning. Hillsdale, NJ:
Erlbaum.
Bannerman DM, Yee BK, Good MA, Heupel MJ, Iversen SD, Rawlins
JN (1999) Double dissociation of function within the hip-
pocampus: a comparison of dorsal, ventral, and complete hip-
pocampal cytotoxic lesions. Behav Neurosci 113:1170–1188.
Bannerman DM, Yee BK, Lemaire M, Wilbrecht L, Jarrard L, Iversen
SD, Rawlins JN, Good MA (2001) The role of the entorhinal
cortex in two forms of spatial learning and memory. Exp Brain
Res 141:281–303.
Bannerman DM, Deacon RM, Offen S, Friswell J, Grubb M, Rawlins
JN (2002) Double dissociation of function within the hip-
pocampus: spatial memory and hyponeophagia. Behav Neurosci
116:884–901.
Bannerman DM, Rawlins JN, McHugh SB, Deacon RM, Yee BK, Bast
T, Zhang WN, Pothuizen HH, Feldon J (2004) Regional dissoci-
ations within the hippocampus—memory and anxiety. Neurosci
Biobehav Rev 28:273–283.
Barnea A, Nottebohm F (1994) Seasonal recruitment of hippocampal
neurons in adult free-ranging black-capped chickadees. Proc
Natl Acad Sci USA 91:11217–11221.
Barnes CA (1979) Memory deficits associated with senescence: a neu-
rophysiological and behavioral study in the rat. J Comp Physiol
Psychol 93:74–104.
Barnes CA (1988) Spatial learning and memory processes: the search
for their neurobiological mechanisms in the rat. Trends Neurosci
11:163–169.
Bartlett FC (1932) Remembering. Cambridge, UK: Cambridge
University Press.
696 The Hippocampus Book
Barton RA, Harvey PH (2000) Mosaic evolution of brain structure in
mammals. Nature 405:1055–1058.
Bast T, Zhang WN, Feldon J (2001) Hippocampus and classical fear
conditioning. Hippocampus 11:828–831.
Bast T, da Silva BM, Morris RGM (2005) Distinct contributions of
hippocampal NMDA and AMPA receptors to encoding and
retrieval of one-trial place memory. J Neurosci 25:5845–5856.
Baxter MG, Murray EA (2001a) Effects of hippocampal lesions on
delayed nonmatching-to-sample in monkeys: a reply to Zola
and Squire (2001). Hippocampus 11:201–203.
Baxter MG, Murray EA (2001b) Opposite relationship of hip-
pocampal and rhinal cortex damage to delayed nonmatching-
to-sample deficits in monkeys. Hippocampus 11:61–71.
Bayley PJ, Frascino JC, Squire LR (2005a) Robust habit learning in
the absence of awareness and independent of the medial tem-
poral lobe. Nature 436:550–553.
Bayley PJ, Gold JJ, Hopkins RO, Squire LR (2005b) The neu-
roanatomy of remote memory. Neuron 46:799–810.
Bear MF, Connors BW, Paradiso MA (2001) Neuroscience: exploring
the brain. Philadelphia: Lippincott.
Beason-Held LL, Rosene DL, Killiany RJ, Moss MB (1999)
Hippocampal formation lesions produce memory impairment
in the rhesus monkey. Hippocampus 9:562–574.
Beggs J, Brown T, Byrne J, Crow T, LeDoux J, LeBar K, Thompson R
(1999) Learning and memory: basic mechanisms. In: Funda-
mental neuroscience, pp 1411–1454. London: Academic Press.
Behrmann M, Geng JJ, Shomstein S (2004) Parietal cortex and atten-
tion. Curr Opin Neurobiol 14:212–217.
Benhamou S (1997) On systems of reference involved in spatial
memory. Behav Processes 40:149–163.
Bennett AT (1996) Do animals have cognitive maps? J Exp Biol 199
(Pt 1):219–224.
Berger TW, Rinaldi PC, Weisz DJ, Thompson RF (1983) Single-unit
analysis of different hippocampal cell types during classical
conditioning of rabbit nictitating membrane response. J
Neurophysiol 50:1197–1219.
Berlyne DE (1950) Novelty and curiousity as determinants of
exploratory behavior. Br J Psychol 41:68–80.
Berlyne DE (1966) Curiosity and exploration. Science 153:25–33.
Bernasconi-Guastalla S, Wolfer DP, Lipp HP (1994) Hippocampal
mossy fibers and swimming navigation in mice: correlations
with size and left-right asymmetries. Hippocampus 4:53–63.
Biegler R (2000) Possible uses of path integration in animal naviga-
tion. Anim Learn Behav 28:257–277.
Biegler R (2003) Reading cognitive and other maps: how to avoid get-
ting buried in thought. In: The neurobiology of spatial behaviour
(Jeffery KJ, ed), pp 259–274. New York: Oxford University Press.
Biegler R, Morris RGM (1993) Landmark stability is a prerequisite
for spatial but not discrimination learning. Nature 361:631–633.
Biegler R, Morris RGM (1999) Blocking in the spatial domain with
arrays of discrete landmarks. J Exp Psychol Anim Behav Process
25:334–351.
Biegler R, McGregor A, Krebs JR, Healy SD (2001) A larger hip-
pocampus is associated with longer-lasting spatial memory. Proc
Natl Acad Sci USA 98:6941–6944.
Bingman VP, Jones TJ (1994) Sun compass-based spatial learning
impaired in homing pigeons with hippocampal lesions. J
Neurosci 14:6687–6694.
Bingman VP, Bagnoli P, Ioale P, Casini G (1984) Homing behav-
ior of pigeons after telencephalic ablations. Brain Behav Evol
24:94–108.
Bingman VP, Ioale P, Casini G, Bagnoli P (1987) Impaired retention
of preoperatively acquired spatial reference memory in homing
pigeons following hippocampal ablation. Behav Brain Res
24:147–156.
Bingman VP, Hough GE 2nd, Kahn MC, Siegel JJ (2003) The homing
pigeon hippocampus and space: in search of adaptive special-
ization. Brain Behav Evol 62:117–127.
Bliss TVP, Collingridge GL (1993) A synaptic model of memory:
long-term potentiation in the hippocampus. Nature 361:31–39.
Bliss TVP, Collingridge GL, Morris RGM (2004) Long term potentia-
tion: enhancing neuroscience for 30 years. Oxford, UK: Oxford
University Press.
Bolhuis JJ, Macphail EM (2001) A critique of the neuroecology of
learning and memory. Trends Cogn Sci 5:426–433.
Bolhuis JJ, Stewart CA, Forrest EM (1994) Retrograde amnesia and
memory reactivation in rats with ibotenate lesions to the hip-
pocampus or subiculum. Q J Exp Psychol B 47:129–150.
Bolles RC (1970) Species-specific defense reactions and avoidance
learning. Psychol Rev 77:32–48.
Bontempi B, Laurent-Demir C, Destrade C, Jaffard R (1999) Time-
dependent reorganization of brain circuitry underlying long-
term memory storage. Nature 400:671–675.
Bostock E, Muller RU, Kubie JL (1991) Experience-dependent
modifications of hippocampal place cell firing. Hippocampus
1:193–205.
Bouton ME (2004) Context and behavioral processes in extinction.
Learn Mem 11:485–494.
Bouton ME, Moody EW (2004) Memory processes in classical condi-
tioning. Neurosci Biobehav Rev 28:663–674.
Brandeis R, Brandys Y, Yehuda S (1989) The use of the Morris water-
maze in the study of memory and learning. Int J Neurosci
48:29–69.
Bransford JD (1979) Human cognition: learning, understanding and
remembering. Belmont, CA: Wadsworth.
Bransford JD, Johnson MK (1972) Contextual prerequisites for
understanding: some investigations of comprehension and
recall. J Verb Learn Verb Behav 11:717–726.
Brasted PJ, Bussey TJ, Murray EA, Wise SP (2002) Fornix transec-
tion impairs conditional visuomotor learning in tasks involv-
ing nonspatially differentiated responses. J Neurophysiol 87:
631–633.
Broadbent NJ, Gallagher S, Colombo M (1999) Hippocampal lesions
and negative patterning in pigeons. Psychobiology 27:51–56.
Broadbent NJ, Squire LR, Clark RE (2004) Spatial memory, recogni-
tion memory, and the hippocampus. Proc Natl Acad Sci USA
101:14515–14520.
Brown MF (1992) Does a cognitive map guide choices in the radial-
arm maze? J Exp Psychol Anim Behav Process 21:147–156.
Brown MF, Rish PA, Von Culin JE, Edberg JA (1993) Spatial guidance
of choice behavior in the radial arm maze. J Exp Psychol Anim
Behav Process 19:195–214.
Brown MW (1996) Neuronal responses and recognition memory.
Semin Neurosci 8:23–32.
Brown MW, Aggleton JP (2001) Recognition memory: what are the
roles of the perirhinal cortex and hippocampus? Nat Rev
Neurosci 2:51–61.
Brown MW, Xiang J-Z (1998) Recognition memory: neuronal sub-
strates of the judgement of prior occurrence. Prog Neurobiol
55:149–189.
Brown MW, Wilson FAW, Riches IP (1987) Neuronal evidence
that inferomedial temporal cortex is more important than

hippocampus in certain processes underlying recognition mem-
ory. Brain Res 409:158–162.
Browning PG, Easton A, Buckley MJ, Gaffan D (2005) The role of
prefrontal cortex in object-in-place learning in monkeys. Eur J
Neurosci 22:3281–3291.
Brun VH, Otnass MK, Molden S, Steffenach HA, Witter MP, Moser
MB, Moser EI (2002) Place cells and place recognition main-
tained by direct entorhinal-hippocampal circuitry. Science
296:2243–2246.
Buckley MJ, Gaffan D (1998a) Perirhinal cortex ablation impairs
visual object identification. J Neurosci 18:2268–2275.
Buckley MJ, Gaffan D (1998b) Learning and transfer of object-
reward associations and the role of the perirhinal cortex. Behav
Neurosci 112:15–23.
Buckmaster CA, Eichenbaum H, Amaral DG, Suzuki WA, Rapp PR
(2004) Entorhinal cortex lesions disrupt the relational organiza-
tion of memory in monkeys. J Neurosci 24:9811–9825.
Buffalo EA, Reber PJ, Squire LR (1998a) The human perirhinal cor-
tex and recognition memory. Hippocampus 8:330–339.
Buffalo EA, Stefanacci L, Squire LR, Zola SM (1998b) A reexamina-
tion of the concurrent discrimination learning task: the impor-
tance of anterior inferotemporal cortex, area TE. Behav Neurosci
112:3–14.
Bunsey M, Eichenbaum H (1995) Selective damage to the hip-
pocampal region blocks long-term retention of a natural
and nonspatial stimulus-stimulus association. Hippocampus
5:546–556.
Bunsey M, Eichenbaum H (1996) Conservation of hippocampal
memory function in rats and humans. Nature 379:255–257.
Bures J, Fenton AA, Kaminsky Y, Wesierska M, Zahalka A (1998)
Rodent navigation after dissociation of the allocentric and idio-
thetic representations of space. Neuropharmacology 37:689–699.
Burgess N, Maguire EA, O’Keefe J (2002) The human hippocampus
and spatial and episodic memory. Neuron 35:625–641.
Bussey TJ, Clea Warburton E, Aggleton JP, Muir JL (1998) Fornix
lesions can facilitate acquisition of the transverse patterning
task: a challenge for ′′configural′′ theories of hippocampal func-
tion. J Neurosci 18:1622–1631.
Bussey TJ, Muir JL, Aggleton JP (1999) Functionally dissociating
aspects of event memory: the effects of combined perirhinal
and postrhinal cortex lesions on object and place memory in the
rat. J Neurosci 19:495–502.
Bussey TJ, Dias R, Redhead ES, Pearse JM, Muir JL, Aggleton JP
(2000) Intact negative patterning in rats with fornix or com-
bined perirhinal and postrhinal cortex lesions. Exp Brain Res
134:506–519.
Bussey TJ, Saksida LM, Murray EA (2002) Perirhinal cortex resolves
feature ambiguity in complex visual discriminations. Eur J
Neurosci 15:365–374.
Bussey TJ, Saksida LM, Murray EA (2003) Impairments in visual dis-
crimination after perirhinal cortex lesions: testing ‘declarative’
vs. ‘perceptual-mnemonic’ views of perirhinal cortex function.
Eur J Neurosci 17:649–660.
Buzsaki G (2005) Theta rhythm of navigation: link between path
integration and landmark navigation, episodic and semantic
memory. Hippocampus 15:827–840.
Cahill L, McGaugh JL (1996) Modulation of memory storage. Curr
Opin Neurobiol 6:237–242.
Capaldi EJ (1971) Memory and learning: a sequential view. In:
Animal memory (Honig WK, James PHR, eds), pp 111–154. San
Diego: Academic Press.
Theories of Hippocampal Function 697
Cassaday HJ, Rawlins JN (1995) Fornix-fimbria section and working
memory deficits in rats: stimulus complexity and stimulus size.
Behav Neurosci 109:594–606.
Cassaday HJ, Rawlins JN (1997) The hippocampus, objects, and their
contexts. Behav Neurosci 111:1228–1244.
Cavaco S, Anderson SW, Allen JS, Castro-Caldas A, Damasio H
(2004) The scope of preserved procedural memory in amnesia.
Brain 127:1853–1867.
Chamizo VD (2002) Spatial learning: conditions and basic effects.
Psicologica 23:33–57.
Chamizo VD (2003) Acquisition of knowledge about spatial location:
assessing the generality of the mechanism of learning. Q J Exp
Psychol B 56:102–113.
Chamizo VD, Sterio D, Mackintosh NJ (1985) Blocking and over-
shadowing between intra-maze and extra-maze cues: a test of
the independence of locale and guidance learning. Q J Exp
Psychol 37B:235–263.
Chan KH, Morell JR, Jarrard LE, Davidson TL (2001) Reconsideration
of the role of the hippocampus in learned inhibition. Behav
Brain Res 119:111–130.
Charles DP, Gaffan D, Buckley MJ (2004a) Impaired recency judg-
ments and intact novelty judgments after fornix transection in
monkeys. J Neurosci 24:2037–2044.
Charles DP, Browning PG, Gaffan D (2004b) Entorhinal cortex
contributes to object-in-place scene memory. Eur J Neurosci
20:3157–3164.
Cheng K (1986) A purely geometric module in the rat’s spatial repre-
sentation. Cognition 23:149–178.
Cho YH, Beracochea D, Jaffard R (1993) Extended temporal
gradi-ent for the retrograde and anterograde amnesia produced
by ibotenate entorhinal cortex lesions in mice. J Neurosci
13:1759–1766.
Cho YH, Kesner RP, Brodale S (1995) Retrograde and anterograde
amnesia for spatial discrimination in rats: role of hippocam-
pus, entorhinal cortex, and parietal cortex. Psychobiology 23:
185–194.
Cipolotti L, Shallice T, Chan D, Fox N, Scahill R, Harrison G, Stevens
J, Rudge P (2001) Long-term retrograde amnesia...the crucial
role of the hippocampus. Neuropsychologia 39:151–172.
Cirillo RA, Horel JA, George PJ (1989) Lesions of the anterior tem-
poral stem and the performance of delayed match-to-sample
and visual discriminations in monkeys. Behav Brain Res
34:55–69.
Claparede E (1911) Recognition et moite. Arch Psychol Geneva
11:79–90.
Clark GA, McCormick DA, Lavond DG, Thompson RF (1984)
Effects of lesions of cerebellar nuclei on conditioned behav-
ioral and hippocampal neuronal responses. Brain Res
291:125–136.
Clark RE, Martin SJ (2005) Interrogating rodents regarding their
object and spatial memory. Curr Opin Neurobiol 15:593–598.
Clark RE, Squire LR (1998) Classical conditioning and brain systems:
the role of awareness. Science 280:77–81.
Clark RE, Zola SM, Squire LR (2000) Impaired recognition memory
in rats after damage to the hippocampus. J Neurosci 20:
8853–8860.
Clark RE, West AN, Zola SM, Squire LR (2001) Rats with lesions of
the hippocampus are impaired on the delayed nonmatching-to-
sample task. Hippocampus 11:176–186.
Clark RE, Broadbent NJ, Squire LR (2004) Hippocampus and remote
spatial memory in rats. Hippocampus 15:260–272.
698 The Hippocampus Book
Clark RE, Broadbent NJ, Squire LR (2005a) Impaired remote spatial
memory after hippocampal lesions despite extensive training
beginning early in life. Hippocampus 15:340–346.
Clark RE, Broadbent NJ, Squire LR (2005b) Hippocampus and
remote spatial memory in rats. Hippocampus 15:260–272.
Clayton NS (1995) Development of memory and the hippocampus:
comparison of food-storing and nonstoring birds on a one-trial
associative memory task. J Neurosci 15:2796–2807.
Clayton NS, Dickinson A (1998) Episodic-like memory during cache
recovery by scrub jays. Nature 395:272–274.
Clayton NS, Krebs JR (1995) Memory in food-storing birds: from
behaviour to brain. Curr Opin Neurobiol 5:149–154.
Clayton NS, Griffiths DP, Emery N, Dickinson A (2001) Elements of
episodic-like memory in animals. In: Episodic memory: new
directions in research (Baddeley A, Conway M, Aggleton JP, eds),
pp 232–248. Oxford, UK: Oxford University Press.
Clayton NS, Yu KS, Dickinson A (2003a) Interacting cache memories:
evidence for flexible memory use by Western scrub-jays
(Aphelocoma californica). J Exp Psychol Anim Behav Process
29:14–22.
Clayton NS, Bussey TJ, Emery NJ, Dickinson A (2003b) Prometheus
to Proust: the case for behavioural criteria for ‘mental time
travel.’ Trends Cogn Sci 7:436–437.
Cohen NJ, Eichenbaum HE (1993) Memory, amnesia and the hip-
pocampal system, 1st ed. Cambridge, MA: MIT Press.
Cohen NJ, Squire LR (1980) Preserved learning and retention of
pattern-analyzing skill in amnesia: dissociation of knowing how
and knowing that. Science 210:207–210.
Cohen NJ, Ryan J, Hunt C, Romine L, Wszalek T, Nash C (1999)
Hippocampal system and declarative (relational) memory:
summarizing the data from functional neuroimaging studies.
Hippocampus 9:83–98.
Collett TS (1992) Landmark learning and guidance in insects. Philos
Trans R Lond B 337:295–303.
Collett TS, Cartwright BA, Smith BA (1986) Landmark learning
and visuo-spatial memories in gerbils. J Comp Psychol A
158:835–851.
Collins AM, Quillian MR (1969) Retrieval time from semantic mem-
ory. J Verbal Learn Verbal Behav 8:240–247.
Columbo M, Cawley S, Broadbent NJ (1997) The effects of hip-
pocampal and area parahippocampalis lesions in pigeons. II.
Concurrent discrimination and spatial memory. Q J Exp Psychol
50B:172–189.
Conference B (1997) Mutant mice and neuroscience: recommenda-
tions concerning genetic background; Banbury conference on
genetic background in mice. Neuron 19:755–759.
Consortium RGSP (2004) Genome sequence of the brown
Norway rat yields insights into mammalian evolution. Nature
428:493–521.
Cooper BG, Mizumori SJ (2001) Temporary inactivation of the ret-
rosplenial cortex causes a transient reorganization of spatial
coding in the hippocampus. J Neurosci 21:3986–4001.
Corcoran KA, Maren S (2001) Hippocampal inactivation disrupts
contextual retrieval of fear memory after extinction. J Neurosci
21:1720–1726.
Corcoran KA, Maren S (2004) Factors regulating the effects of hip-
pocampal inactivation on renewal of conditional fear after
extinction. Learn Mem 11:598–603.
Corcoran KA, Desmond TJ, Frey KA, Maren S (2005) Hippocampal
inactivation disrupts the acquisition and contextual encoding of
fear extinction. J Neurosci 25:8978–8987.
Corkin S (1984) Lasting consequences of bilateral medial temporal
lobectomy: clinical course and experimental findings in H.M.
Semin Neurol 4:249–259.
Cowey A, Stoerig P (1995) Blindsight in monkeys. Nature
373:247–249.
Crabbe JC, Morris RGM (2004) Festina lente: late-night thoughts on
high-throughput screening of mouse behavior. Nat Neurosci
7:1175–1179.
Crusio WE, Schwegler H, Lipp HP (1987) Radial-maze performance
and structural variation of the hippocampus in mice: a correla-
tion with mossy fibre distribution. Brain Res 425:182–185.
Dale N, Roberts A (1985) Dual-component amino-acid-mediated
synaptic potentials: excitatory drive for swimming in Xenopus
embryos. J Physiol 363:35–59.
Davidson TL (1993) The nature and function of interoceptive signals
to feed: toward integration of physiological and learning per-
spectives. Psychol Rev 100:640–657.
Davidson TL, Jarrard LE (1993) A role for hippocampus in the uti-
lization of hunger signals. Behav Neural Biol 59:167–171.
Davidson TL, Jarrard LE (2004) The hippocampus and inhibitory
learning: a ‘gray’ area? Neurosci Biobehav Rev 28:261–271.
Davidson TL, McKernan MG, Jarrard LE (1993) Hippocampal
lesions do not impair negative patterning: a challenge to config-
ural association theory. Behav Neurosci 107:227-234.
Davis M, Falls WA, Campeau S, Kim M (1993) Fear-potentiated star-
tle: a neural and pharmacological analysis. Behav Brain Res
58:175-198.
Day M, Langston RF, Morris RGM (2003) Glutamate receptor
dependent encoding and retrieval of paired associate learning.
Nature 424:205-209.
Deacon RM, Bannerman DM, Rawlins NP (2001) Conditional dis-
criminations based on external and internal cues in rats with
cytotoxic hippocampal lesions. Behav Neurosci 115:43–57.
Dean P, Weiskrantz L (1974) Loss of preoperative habits in rhesus
monkeys with inferotemporal lesions: recognition failure or
relearning deficit? Neuropsychologia 12:299–311.
De Hoz L, Knox J, Morris RGM (2003) Longitudinal axis of the hip-
pocampus: both septal and temporal poles of the hippocampus
support watermaze spatial learning depending on the training
protocol. Hippocampus 13:587–603.
De Hoz L, Martin SJ, Morris RGM (2004) Forgetting, reminding, and
remembering: the retrieval of lost spatial memory. PLoS Biol
2:E225.
De Kloet ER, Oitzl MS, Joels M (1999) Stress and cognition: are cor-
ticosteroids good or bad guys? Trends Neurosci 22:422–426.
D’Hooge R, De Deyn PP (2001) Applications of the Morris water-
maze in the study of learning and memory. Brain Res Brain Res
Rev 36:60–90.
Dickinson A (1980) Contemporary animal learning theory.
Cambridge: Cambridge University Press.
Dickinson A, Mackintosh NJ (1978) Classical conditioning in ani-
mals. Annu Rev Psychol 29:587–612.
DiMattia BV, Kesner RP (1988a) Role of the posterior parietal associ-
ation cortex in the processing of spatial event information.
Behav Neurosci 102:397–403.
Dix SL, Aggleton JP (1999) Extending the spontaneous preference
test of recognition: evidence of object-location and object-
context recognition. Behav Brain Res 99:191–200.
Dore FY, Thornton JA, White NM, Murray EA (1998) Selective hip-
pocampal lesions yield nonspatial memory impairments in rhe-
sus monkeys. Hippocampus 8:323–329.

Douglas RJ, Martin KA (2004) Neuronal circuits of the neocortex.
Annu Rev Neurosci 27:419–451.
Dragunow M (1996) A role for immediate-early transcription factors
in learning and memory. Behav Genet 26:293–299.
Dudai Y, Morris RGM (2001) To consolidate or not to consolidate:
what are the questions? In: Brain, perception and memory:
advances in cognitive sciences (Bolhuis J, ed), pp 147–162.
Oxford, UK: Oxford University Press.
Dudchenko PA (2003) The head-direction system and navigation. In:
The neurobiology of spatial behaviour (Jeffery K, ed). Oxford,
UK: Oxford University Press.
Dudchenko PA, Taube JS (1997) Correlation between head direction
cell activity and spatial behavior on a radial arm maze. Behav
Neurosci 111:3–19.
Dudchenko PA, Wood ER, Eichenbaum H (2000) Neurotoxic hip-
pocampal lesions have no effect on odor span and little effect on
odor recognition memory but produce significant impairments
on spatial span, recognition, and alternation. J Neurosci
20:2964–2977.
Dusek JA, Eichenbaum H (1997) The hippocampus and mem-
ory for orderly stimulus relations. Proc Natl Acad Sci USA
94:7109–7114.
Dyer F (1998) Cognitive ecology of navigation. In: Cognitive ecology
(Dukas R, ed), pp 201–260. Chicago: University of Chicago
Press.
Eacott MJ, Gaffan EA (2005) The roles of perirhinal cortex, postrhi-
nal cortex, and the fornix in memory for objects, contexts, and
events in the rat. Q J Exp Psychol B 58:202–217.
Eacott MJ, Norman G (2004) Integrated memory for object, place,
and context in rats: a possible model of episodic-like memory?
J Neurosci 24:1948–1953.
Eacott MJ, Easton A, Zinkivskay A (2005) Recollection in an episodic-
like memory task in the rat. Learn Mem 12:221–223.
Easton A, Gaffan D (2000) Comparison of perirhinal cortex
ablation and crossed unilateral lesions of the medial forebrain
bundle from the inferior temporal cortex in the rhesus
monkey: effects on learning and retrieval. Behav Neurosci
114:1041–1057.
Easton A, Parker A, Gaffan D (2001) Crossed unilateral lesions of
medial forebrain bundle and either inferior temporal or frontal
cortex impair object recognition memory in rhesus monkeys.
Behav Brain Res 121:1–10.
Eichenbaum H (1996) Is the rodent hippocampus just for ′′place′′?
Curr Opin Neurobiol 6:187–195.
Eichenbaum H (2004) Hippocampus: cognitive processes and neural
representations that underlie declarative memory. Neuron
44:109–120.
Eichenbaum H, Cohen NJ (2001) From conditioning to conscious rec-
ollection. New York: Oxford University Press.
Eichenbaum H, Stewart C, Morris RGM (1990) Hippocampal repre-
sentation in place learning. J Neurosci 10:3531–3542.
Eichenbaum H, Dudchenko P, Wood E, Shapiro M, Tanila H (1999)
The hippocampus, memory, and place cells: is it spatial memory
or a memory space? Neuron 23:209–226.
Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman
EL, Fried I (2003) Cellular networks underlying human spatial
navigation. Nature 425:184–188.
Emery NJ, Clayton NS (2001) Effects of experience and social context
on prospective caching strategies by scrub jays. Nature
414:443–446.
Emery NJ, Clayton NS (2004) The mentality of crows: convergent
Theories of Hippocampal Function 699
evolution of intelligence in corvids and apes. Science
306:1903–1907.
Ennaceur A, Aggleton JP (1994) Spontaneous recognition of object
configurations in rats: effects of fornix lesions. Exp Brain Res
100:85–92.
Ennaceur A, Delacour J (1988) A new one-trial test for neurobiolog-
ical studies of memory in rats. 1. Behavioral data. Behav Brain
Res 31:47–59.
Etienne AS, Jeffery KJ (2004) Path integration in mammals.
Hippocampus 14:180–192.
Etienne AS, Maurer R, Berlie J, Reverdin B, Rowe T, Georgakopoulos
J, Seguinot V (1998) Navigation through vector addition. Nature
396:161–164.
Fanselow MS (1990) Factors governing one-trial contextual condi-
tioning. Anim Learn Behav 18:264–270.
Fanselow MS (2000) Contextual fear, gestalt memories and the hip-
pocampus. Behav Brain Res 110:73–81.
Fantz RL (1964) Visual experience in infants: decreased attention to
familiar patterns relative to novel ones. Science 146:668–670.
Farah MJ (2004) Visual agnosia. Cambridge, MA: MIT Press.
Ferbinteanu J, Shapiro ML (2003) Prospective and retrospective
memory coding in the hippocampus. Neuron 40:1227–1239.
Finger S, Koehler PJ, Jagella C (2004) The Monakow concept of
diaschisis: origins and perspectives. Arch Neurol 61:283–288.
Fletcher PC, Henson RN (2001) Frontal lobes and human memory:
insights from functional neuroimaging. Brain 124:849–881.
Fortin NJ, Agster KL, Eichenbaum HB (2002) Critical role of the hip-
pocampus in memory for sequences of events. Nat Neurosci
5:458–462.
Fortin NJ, Wright SP, Eichenbaum H (2004) Recollection-like mem-
ory retrieval in rats is dependent on the hippocampus. Nature
431:188–191.
Forwood SE, Winters BD, Bussey TJ (2005) Hippocampal lesions
that abolish spatial maze performance spare object recogni-
tion memory at delays of up to 48 hours. Hippocampus
15:347–355.
Foster TC, Castro CA, McNaughton BL (1989) Spatial selectivity of
rat hippocampal neurons: dependence on preparedness for
movement. Science 244:1580–1582.
Fox GD, Holland PC (1998) Neurotoxic hippocampal lesions fail to
impair reinstatement of an appetitively conditioned response.
Behav Neurosci 112:255–260.
Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the
hippocampus and entorhinal cortex. Neuron 27:169–178.
Frank MJ, Rudy JW, O’Reilly RC (2003) Transitivity, flexibility, con-
junctive representations and the hippocampus. II. A computa-
tional analysis. Hippocampus 13:299–312.
Frankland PW, Bontempi B (2005) The organization of recent and
remote memories. Nat Rev Neurosci 6:119–130.
Frankland PW, Cestari V, Filipkowski RK, McDonald RJ, Silva AJ
(1998) The dorsal hippocampus is essential for context discrim-
ination but not for contextual conditioning. Behav Neurosci
112:863–874.
Frankland PW, O’Brien C, Ohno M, Kirkwood A, Silva AJ (2001)
Alpha-CaMKII-dependent plasticity in the cortex is required
for permanent memory. Nature 411:309–313.
Frankland PW, Bontempi B, Talton LE, Kaczmarek L, Silva AJ (2004)
The involvement of the anterior cingulate cortex in remote con-
textual fear memory. Science 304:881–883.
Frey U, Morris RGM (1997) Synaptic tagging and long-term potenti-
ation. Nature 385:533–536.
700 The Hippocampus Book
Frey U, Morris RGM (1998) Synaptic tagging: implications for late
maintenance of hippocampal long-term potentiation. Trends
Neurosci 21:181–188.
Fried I, Wilson CL, MacDonald KA, Behnke EJ (1998) Electric cur-
rent stimulates laughter. Nature 391:650.
Friston KJ (1995) Functional and effective connectivity in neu-
roimaging: A synthesis. Hum Brain Mapp 2:56–78.
Friston KJ (2004) Models of brain function in neuroimaging. Annu
Rev Psychol 56:57–87.
Fritz J, Mishkin M, Saunders RC (2005) In search of an auditory
engram. Proc Natl Acad Sci USA 102:9359–9364.
Frohardt RJ, Guarraci FA, Bouton ME (2000) The effects of neuro-
toxic hippocampal lesions on two effects of context after fear
extinction. Behav Neurosci 114:227–240.
Fukuda M, Kobayashi T, Bures J, Ono T (1992) Rat exploratory
behavior controlled by intracranial self-stimulation improves
the study of place cell activity. J Neurosci Methods 44:121–131.
Fyhn M, Molden S, Hollup SA, Moser M-B, Moser EI (2002)
Hippocampal neurons responding to first-time dislocation of a
target object. Neuron 35:555–566.
Fyhn M, Molden S, Witter MP, Moser EI, Moser MB (2004)
Spatial representation in the entorhinal cortex. Science
305:1258–1264.
Gabrieli JDE, Fleischman DA, Deane MM, Reminger SL, Morrell F
(1995) Double dissociation between memory systems underly-
ing explicit and implicit memory in the human brain. Psychol
Sci 6:76–82.
Gaffan D (1974) Recognition impaired and association intact in the
memory of monkeys after transection of the fornix. J Comp
Physiol Psychol 86:1100–1109.
Gaffan D (1983) Animal amnesia: some disconnection syndromes?
In: Neurobiology of the hippocampus (Seifert W, ed), pp 513–528.
London: Academic Press.
Gaffan D (1991) Spatial organization of episodic memory. Hippocam-
pus 1:262–264.
Gaffan D (1992) Amnesia for complex naturalistic scenes and for
objects following fornix transection in the rhesus monkey. Eur J
Neurosci 4:381–388.
Gaffan D (1993) Additive effects of forgetting and fornix tran-
section in the temporal gradient of retrograde amnesia.
Neuropsychologia 31:1055–1066.
Gaffan D (1994) Scene-specific memory for objects: a model of
episodic memory impairment in monkeys with fornix transec-
tion. J Cogn Neurosci 6:305–320.
Gaffan D (2002) Against memory systems. Philos Trans R Soc Lond B
Biol Sci 357:1111–1121.
Gaffan EA, Eacott MJ (1995) A computer-controlled maze environ-
ment for testing visual memory in the rat. J Neurosci Methods
60:23–37.
Gaffan D, Harrison S (1989) Place memory and scene memory: effects
of fornix transection in the monkey. Exp Brain Res 74:202–212.
Gaffan D, Parker A (1996) Interaction of perirhinal cortex with the
fornix-fimbria: memory for objects and ′′object-in-place′′
memory. J Neurosci 16:5864–5869.
Gaffan D, Gaffan EA, Harrison S (1984) Effects of fornix transection
on spontaneous and trained non-matching by monkeys. Q J Exp
Psychol 36B:285–303.
Gagliardo A, Ioale P, Bingman VP (1999) Homing in pigeons: the role
of the hippocampal formation in the representation of land-
marks used for navigation. J Neurosci 19:311–315.
Gagliardo A, Ioale P, Odetti F, Kahn MC, Bingman VP (2004)
Hippocampal lesions do not disrupt navigational map retention
in homing pigeons under conditions when map acquisition is
hippocampal dependent. Behav Brain Res 153:35–42.
Galea LA, McEwen BS (1999) Sex and seasonal differences in the rate
of cell proliferation in the dentate gyrus of adult wild meadow
voles. Neuroscience 89:955–964.
Galef BG (1990) Necessary and sufficient conditions for communi-
cation of diet preferences by Norway rats. Anim Learn Behav
18:347–351.
Galef BG, Wigmore SW (1983) Transfer of information concerning
distant foods: a laboratory investigation of the information-
centre hypothesis. Anim Behav 31:748–758.
Gallagher M, Holland PC (1992) Preserved configural learning and
spatial learning impairment in rats with hippocampal damage.
Hippocampus 2:81–88.
Gallagher M, Rapp PR (1997) The use of animal models to study the
effects of aging on cognition. Annu Rev Psychol 48:339–370.
Gallistel C (1980) The organisation of action: a new synthesis.
Hillsdale, NJ: Erlbaum.
Garrud P, Rawlins JN, Mackintosh NJ, Goodall G, Cotton MM,
Feldon J (1984) Successful overshadowing and blocking in hip-
pocampectomized rats. Behav Brain Res 12:39–53.
Gattis M (2001) Spatial schemas and abstract thought. Cambridge,
MA: MIT Press.
Gaulin SJ, Fitzgerald RW (1989) Sexual selection for spatial-learning
ability. Anim Behav 37:322–331.
Gaulin SJ, FitzGerald RW, Wartell MS (1990) Sex differences in
spatial ability and activity in two vole species (Microtus
ochrogaster and M. pennsylvanicus). J Comp Psychol 104:88–93.
Gazzaniga MS (2002) Cognitive neuroscience: the biology of the mind.
New York: Norton.
Gerlai R (1996) Gene-targeting studies of mammalian behavior: is it
the mutation or the background genotype? Trends Neurosci
19:177–181.
Gewirtz JC, McNish KA, Davis M (2000) Is the hippocampus neces-
sary for contextual fear conditioning? Behav Brain Res
110:83–95.
Gilbert P, Kesner R, DeCoteau R (1998) Memory for spatial location:
role of the hippocampus in mediating spatial pattern separa-
tion. J Neurosci 18:804–810.
Gilbert PE, Kesner RP, Lee I (2001) Dissociating hippocampal subre-
gions: double dissociation between dentate gyrus and CA1.
Hippocampus 11:626–636.
Gisquet-Verrier P, Dutrieux G, Richer P, Doyere V (1999) Effects of
lesions to the hippocampus on contextual fear: evidence for a
disruption of freezing and avoidance behavior but not context-
conditioning. Behav Neurosci 113:507–522.
Gladwell M (2002) The tipping point: how little things can make a big
difference. Boston: Back Bay Books.
Goelet P, Castellucci VF, Schacher S, Kandel ER (1986) The long and
the short of long-term memory—a molecular framework.
Nature 322:419–422.
Golding NL, Spruston N (1998) Dendritic sodium spikes are variable
triggers of axonal action potentials in hippocampal CA1 pyram-
idal neurons. Neuron 21:1189–1200.
Golding NL, Staff NP, Spruston N (2002) Dendritic spikes as a
mechanism for cooperative long-term potentiation. Nature
418:326–331.
Golob EJ, Taube JS (2002) Differences between appetitive and aver-
sive reinforcement on reorientation in a spatial working mem-
ory task. Behav Brain Res 136:309–316.

Good M, Honey RC (1991) Conditioning and contextual retrieval in
hippocampal rats. Behav Neurosci 105:499–509.
Good M, de Hoz L, Morris RGM (1998) Contingent versus inciden-
tal context processing during conditioning: dissociation
after excitotoxic hippocampal plus dentate gyrus lesions.
Hippocampus 8:147–159.
Good MA, Morris RGM (1994) A step linking memory to under-
standing? Behav Brain Sci 17:477–479.
Gothard KM, Skaggs WE, Moore KM, McNaughton BL (1996)
Binding of hippocampal CA1 neural activity to multiple refer-
ence frames in a landmark-based navigation task. J Neurosci
16:823–835.
Gould SJ, Eldredge N (1977) Punctuated equilibria: the tempo and
mode of evolution reconsidered. Paleobiology 3:115–151.
Graf P, Schacter DL (1985) Implicit and explicit memory for new
associations in normal and amnesic subjects. J Exp Psychol Learn
Mem Cogn 11:501–518.
Graf P, Squire LR, Mandler G (1984) The information that amnesic
patients do not forget. J Exp Psychol 10:164–178.
Graham KS, Simons JS, Pratt KH, Patterson K, Hodges JR
(2000) Insights from semantic dementia on the relationship
between episodic and semantic memory. Neuropsychologia
38:313–324.
Grant SG, O’dell TJ, Karl KA, Stein PL, Soriano P, Kandel ER
(1992) Impaired long-term potentiation, spatial learning, and
hippocampal development in fyn mutant mice. Science
258:1903–1910.
Gray JA (1982) The neuropsychology of anxiety: an enquiry into the
functions of the septo-hippocampal system. Oxford, UK: Oxford
University Press.
Gray JA (2000) The neuropsychology of anxiety: an enquiry into the
functions of the septo-hippocampal system, 2nd ed. Oxford:
Oxford University Press.
Gregory RL (1961) The brain as an engineering problem. In: Current
problems in animal behaviour (Thorpe WH, Zangwill OL, eds).
Cambridge, UK: Cambridge University Press.
Griffiths D, Dickinson A, Clayton N (1999) Episodic memory: what
can animals remember about their past? Trends Cogn Sci
3:74–80.
Grillner S, Ekeberg, El Manira A, Lansner A, Parker D, Tegnér J,
Wallén P (1998) Intrinsic function of a neuronal network—a
vertebrate central pattern generator. Brain Res Brain Res Rev
26:184–197.
Guillot PV, Roubertoux PL, Crusio WE (1994) Hippocampal mossy
fiber distributions and intermale aggression in seven inbred
mouse strains. Brain Res 660:167–169.
Guzowski JF (2002) Insights into immediate-early gene function in
hippocampal memory consolidation using antisense oligonu-
cleotide and fluorescent imaging approaches. Hippocampus
12:86–104.
Guzowski JF, McNaughton BL, Barnes CA, Worley PF (1999)
Environment-specific expression of the immediate-early gene
Arc in hippocampal neuronal ensembles. Nat Neurosci
2:1120–1124.
Guzowski JF, Setlow B, Wagner EK, McGaugh JL (2001) Experience-
dependent gene expression in the rat hippocampus after spatial
learning: a comparison of the immediate-early genes Arc, c-fos,
and zif268. J Neurosci 21:5089–5098.
Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005)
Microstructure of a spatial map in the entorhinal cortex. Nature
436:801–806.
Theories of Hippocampal Function 701
Hagan JJ, Salamone JD, Simpson J, Iversen SD, Morris RG (1988)
Place navigation in rats is impaired by lesions of medial septum
and diagonal band but not nucleus basalis magnocellularis.
Behav Brain Res 27:9–20.
Hall G, Purves D, Bonardi C (1996) Contextual control of condi-
tioned responding in rats with dorsal hippocampal lesions.
Behav Neurosci 110:933–945.
Halliday MS (1968) Exploratory behaviour. In: Analysis of behav-
ioural change (Weiskrantz L, ed), pp 107–126. New York: Harper
& Row.
Hamann SB, Squire LR (1997) Intact perceptual memory in the
absence of conscious memory. Behav Neurosci 111:850–854.
Hampton RR (2001) Rhesus monkeys know when they remember.
Proc Natl Acad Sci USA 98:5359–5362.
Hampton RR, Schwartz BL (2004) Episodic memory in nonhumans:
what, and where, is when? Curr Opin Neurobiol 14:192–197.
Hampton RR, Shettleworth SJ (1996) Hippocampal lesions impair
memory for location but not color in passerine birds. Behav
Neurosci 110:831–835.
Hampton RR, Healy SD, Shettleworth SJ, Kamil AC (2002)
Neuroecologists’ are not made of straw. Trends Cogn Sci 6:
6–7.
Hardiman MJ, Ramnani N, Yeo CH (1996) Reversible inactivities of
the cerebellum with muscimol prevent the acquisition and
extinction of conditioned nictitating membrame responses in
the rabbit. Exp Brain Res 110:235–247.
Harker KT, Whishaw IQ (2004) A reaffirmation of the retrosplenial
contribution to rodent navigation: reviewing the influences of
lesion, strain, and task. Neurosci Biobehav Rev 28:485–496.
Hartley T, Burgess N, Lever C, Cacucci F, O’Keefe J (2000) Modeling
place fields in terms of the cortical inputs to the hippocampus.
Hippocampus 10:369–379.
Hasselmo ME, Howard E (2005) Hippocampal mechanisms for
the context-dependent retrieval of episodes. Neural Netw
18:1172–1190.
Hasselmo ME, Bodelon C, Wyble BP (2002) A proposed function for
hippocampal theta rhythm: separate phases of encoding and
retrieval enhance reversal of prior learning. Neural Comput
14:793–817.
Hauser MD (2003) Knowing about knowing: dissociations between
perception and action systems over evolution and during devel-
opment. Ann N Y Acad Sci 1001:79–103.
Healy SD (1998) Spatial representations in animals. New York: Oxford
University Press.
Healy SI, Braithwaite II (2000) Cognitive ecology: a field of sub-
stance? Trends Ecol Evol 15:22–26.
Hebb DO (1949) The organization of behaviour. New York: Wiley.
Hebben N, Corkin S, Eichenbaum H, Shedlack K (1985) Dimini-
shed ability to interpret and report internal states after bilat-eral
medial temporal resection: case H.M. Behav Neurosci
99:1031–1039.
Henke K, Buck A, Weber B, Wieser HG (1997) Human hippocampus
establishes associations in memory. Hippocampus 7:249–256.
Herrera DG, Robertson HA (1996) Activation of c-fos in the brain.
Prog Neurobiol 50:83–107.
Hess US, Lynch G, Gall CM (1995) Regional patterns of c-fos mRNA
expression in rat hippocampus following exploration of a novel
environment versus performance of a well-learned discrimina-
tion. J Neurosci 15:7796–7809.
Hill AJ (1978) First occurrence of hippocampal spatial firing in a new
environment. Exp Neurol 62:282–297.
702 The Hippocampus Book
Hirsh R (1974) The hippocampus and contextual retrieval of infor-
mation from memory: a theory. Behav Biol 12:421–444.
Hok V, Save E, Lenck-Santini PP, Poucet B (2005) Coding for spatial
goals in the prelimbic/infralimbic area of the rat frontal cortex.
Proc Natl Acad Sci USA 102:4602–4607.
Holland PC, Fox GD (2003) Effects of hippocampal lesions in
overshadowing and blocking procedures. Behav Neurosci
117:650–656.
Holland PC, Lamoroureux JA, Han JS, Gallagher M (1999)
Hippocampal lesions interfere with negative occasion-setting.
Hippocampus 9:143–157.
Hollup SA, Kjelstrup KG, Hoff J, Moser MB, Moser EI (2001)
Impaired recognition of the goal location during spatial naviga-
tion in rats with hippocampal lesions. J Neurosci 21:4505–4513.
Honey RC, Good M (1993) Selective hippocampal lesions abolish the
contextual specificity of latent inhibition and of conditioning.
Behav Neurosci 107:22–33.
Honey RC, Watt A, Good M (1998) Hippocampal lesions disrupt an
associative mismatch process. J Neurosci 18:2226–2230.
Honig WK (1978) Studies of working memory in the pigeon. In:
Cognitive aspects of animal behavior (Hulse SH, Fowler HF,
Honig WK, eds), pp 211–248. Hillside, NJ: Erlbaum.
Horel JA (1978) The neuroanatomy of amnesia: a critique of the hip-
pocampal memory hypothesis. Brain 101:403–445.
Horel JA (1994) Some comments on the special cognitive functions
claimed for the hippocampus. Cortex 30:269–280.
Insausti R, Amaral DG, Cowan WM (1987a) The entorhinal cortex
of the monkey. III. Subcortical afferents. J Comp Neurol
264:396–408.
Insausti R, Amaral DG, Cowan WM (1987b) The entorhinal cortex of
the monkey. II. Cortical afferents. J Comp Neurol 264:356–395.
Ioale P, Gagliardo A, Bingman VP (2000) Hippocampal participation
in navigational map learning in young homing pigeons is
dependent on training experience. Eur J Neurosci 12:742–750.
Iversen SD (1976) Do hippocampal lesions produce amnesia in ani-
mals? Int Rev Neurobiol 19:1–49.
Jacobs LF (1995) The ecology of spatial cognition. In: Behavioural
brain research in naturalistic and semi-naturalistic settings
(Alleva E, ed), pp 301–322. Boston: Kluwer Academic.
Jacobs LF (2003) The evolution of the cognitive map. Brain Behav
Evol 62:128–139.
Jacobs LF, Schenk F (2003) Unpacking the cognitive map: the
parallel map theory of hippocampal function. Psychol Rev
110:285–315.
Jacobs LF, Spencer WD (1994) Natural space-use patterns and
hippocampal size in kangaroo rats. Brain Behav Evol 44:
125–132.
Jacobs LF, Gaulin SJ, Sherry DF, Hoffman GE (1990) Evolution
of spatial cognition: sex-specific patterns of spatial behavior
predict hippocampal size. Proc Natl Acad Sci USA 87:
6349–6352.
James W (1890) The principles of psychology, 1918 ed. New York:
Dover.
Jarrard LE (1978) Selective hippocampal lesions: differential
effects on performance by rats of a spatial task with preopera-
tive versus postoperative training. J Comp Physiol Psychol
92:1119–1127.
Jarrard LE (1986) Selective hippocampal lesions and behavior: impli-
cations for current research and theorizing. In: The hippocam-
pus (Isaacson RL, Pribram KH, eds), pp 93–126. New York:
Plenum.
Jarrard LE (1989) On the use of ibotenic acid to lesion selectively dif-
ferent components of the hippocampal formation. J Neurosci
Methods 29:251–259.
Jarvis ED, Gunturkun O, Bruce L, Csillag A, Karten H, Kuenzel W,
Medina L, Paxinos G, Perkel DJ, Shimizu T, Striedter G, Wild
JM, Ball GF, Dugas-Ford J, Durand SE, Hough GE, Husband S,
Kubikova L, Lee DW, Mello CV, Powers A, Siang C, Smulders
TV, Wada K, White SA, Yamamoto K, Yu J, Reiner A, Butler AB
(2005) Avian brains and a new understanding of vertebrate
brain evolution. Nat Rev Neurosci 6:151–159.
Jeffery KJ (2003) The neurobiology of spatial behaviour. New York:
Oxford University Press.
Jeffery KJ, Anderson MI, Hayman R, Chakraborty S (2004) A pro-
posed architecture for the neural representation of spatial con-
text. Neurosci Biobehav Rev 28:201–218.
Jenkins TA, Amin E, Pearce JM, Brown MW, Aggleton JP (2004)
Novel spatial arrangements of familiar visual stimuli promote
activity in the rat hippocampal formation but not the parahip-
pocampal cortices: a c-fos expression study. Neuroscience
124:43–52.
Jensen O, Lisman JE (1996) Theta/gamma networks with slow
NMDA channels learn sequences and encode episodic memory:
role of NMDA channels in recall. Learn Mem 3:2264–278.
Jung MW, Wiener SI, McNaughton BL (1994) Comparison of spatial
firing characteristics of units in dorsal and ventral hippocampus
of the rat. J Neurosci 14:7347–7356.
Kamin LJ (1968) Attention-like processes in classical conditioning. In:
Miami symposium on the prediction of behavior: aversive stimula-
tion (Jones MR, ed), pp 9–33. Miami: University of Miami Press.
Kaminski J, Call J, Fischer J (2004) Word learning in a domestic dog:
evidence for ′′fast mapping.′′ Science 304:1682–1683.
Kandel ER, Schwartz JH, Jessell TM (2000) Principles of neural sci-
ence. New York: McGraw-Hill.
Kao YC, Davis ES, Gabrieli JD (2005) Neural correlates of actual and
predicted memory formation. Nat Neurosci 8:1776–1783.
Kapur N, Brooks DJ (1999) Temporally-specific retrograde
amnesia in two cases of discrete bilateral hippocampal pathol-
ogy. Hippocampus 9:247–254.
Kay RF, Ross C, Blythe A, Williams BA (1997) Anthropoid origins.
Science 275:797–804.
Kelsey BA, Landry JE (1988) Medial septal lestions disrupt spatial
mapping ability in rats. Behav Neurosci 102:289–293.
Kempermann G, Brandon EP, Gage FH (1998) Environmental stim-
ulation of 129/SvJ mice causes increased cell proliferation and
neurogenesis in the adult dentate gyrus. Curr Biol 8:939–942.
Kennedy PJ, Shapiro ML (2004) Retrieving memories via internal
context requires the hippocampus. J Neurosci 24:6979–6985.
Kentros C, Hargreaves E, Hawkins RD, Kandel ER, Shapiro M, Muller
RV (1998) Abolition of long-term stability of new hippocam-
pal place cell maps by NMDA receptor blockade. Science
280:2121–2126.
Kesner RP (1998) Neurobiological views of memory. In: Neurobiology
of learning and memory (Martinez JL, Kesner RP, eds), pp
361–416. San Diego: Academic Press.
Kesner RP (2000) Behavioral analysis of the contribution of the hip-
pocampus and parietal cortex to the processing of information:
interactions and dissociations. Hippocampus 10:483–490.
Kesner RP, Bolland BL, Dakis M (1993) Memory for spatial locations,
motor responses, and objects: triple dissociation among the hip-
pocampus, caudate nucleus, and extrastriate visual cortex. Exp
Brain Res 93:462–470.

Kiernan MJ, Westbrook RF (1993) Effects of exposure to a to-be-
shocked environment upon the rat’s freezing response: evidence
for facilitation, latent inhibition and perceptual learning. Q J
Exp Psychol 46B:271–288.
Kim JJ, Fanselow MS (1992) Modality-specific retrograde amnesia of
fear. Science 256:675–677.
Kimble DP (1968) Hippocampus and internal inhibition. Psychol
Bull 70:285–295.
Kjelstrup KG, Tuvnes FA, Steffenach HA, Murison R, Moser
EI, Moser MB (2002) Reduced fear expression after lesions
of the ventral hippocampus. Proc Natl Acad Sci USA
99:10825–10830.
Knowlton BJ, Squire LR (1993) The learning of categories: parallel
brain systems for item memory and category knowledge. Science
262:1747–1749.
Knuttinen MG, Power JM, Preston AR, Disterhoft JF (2001)
Awareness in classical differential eyeblink conditioning in
young and aging humans. Behav Neurosci 115:747–757.
Kowalska DM, Kusmierek P, Kosmal A, Mishkin M (2001) Neither
perirhinal/entorhinal nor hippocampal lesions impair
short-term auditory recognition memory in dogs. Neuroscience
104:965–978.
Krebs JR, Sherry DF, Healy SD, Perry VH, Vaccarino AL (1989)
Hippocampal specialization of food-storing birds. Proc Natl
Acad Sci USA 86:1388–1392.
Krupa DJ, Thompson JK, Thompson RF (1993) Localization of a
memory trace in the mammalian brain. Science 260:989–991.
Kumaran D, Maguire EA (2005) The human hippocampus: cognitive
maps or relational memory? J Neurosci 25:7254–7259.
LaBar KS, Disterhof JF (1998) Conditioning, awareness, and the hip-
pocampus. Hippocampus 8:620–626.
Lashley KS (1950) In search of the engram. In: Symposia for the
Society for Experimental Biology, pp 454–482. New York:
Cambridge University Press.
Lavenex P, Steele MA, Jacobs LF (2000) Sex differences, but no sea-
sonal variations in the hippocampus of food-caching squirrels:
a stereological study. J Comp Neurol 425:152–166.
LeDoux JE (2000) Emotion circuits in the brain. Annu Rev Neurosci
23:155–184.
Lee AC, Bussey TJ, Murray EA, Saksida LM, Epstein RA, Kapur N,
Hodges JR, Graham KS (2005) Perceptual deficits in amnesia:
challenging the medial temporal lobe ‘mnemonic’ view.
Neuropsychologia 43:1–11.
Lee DW, Miyasato LE, Clayton NS (1998) Neurobiological bases of
spatial learning in the natural environment: neurogenesis and
growth in the avian and mammalian hippocampus. Neuroreport
9:R15–27.
Lee DW, Smith GT, Tramontin AD, Soma KK, Brenowitz EA,
Clayton NS (2001) Hippocampal volume does not change sea-
sonally in a non food-storing songbird. Neuroreport 12:1925-
1928.
Lee I, Kesner RP (2004) Differential contributions of dorsal hip-
pocampal subregions to memory acquisition and retrieval in
contextual fear-conditioning. Hippocampus 14:301–310.
Lenck-Santini PP, Save E, Poucet B (2001) Evidence for a relationship
between place-cell spatial firing and spatial memory perform-
ance. Hippocampus 11:377–390.
Lenck-Santini PP, Muller RU, Save E, Poucet B (2002) Relationships
between place cell firing fields and navigational decisions by
rats. J Neurosci 22:9035–9047.
Leonard BW, Amaral DG, Squire LR, Zola-Morgan S (1995)
Theories of Hippocampal Function 703
Transient memory impairment in monkeys with bilateral
lesions of the entorhinal cortex. J Neurosci 15:5637–5659.
Leutgeb S, Leutgeb JK, Treves A, Moser MB, Moser EI (2004) Distinct
ensemble codes in hippocampal areas CA3 and CA1. Science
305:1295–1298.
Lever C, Wills T, Cacucci F, Burgess N, O’Keefe J (2002) Long-term
plasticity in hippocampal place-cell representation of environ-
mental geometry. Nature 416:90–94.
Levy DA, Stark CE, Squire LR (2004) Intact conceptual priming in
the absence of declarative memory. Psychol Sci 15:680–686.
Levy WB (1996) A sequence predicting CA3 is a flexible associator
that learns and uses context to solve hippocampal-like tasks.
Hippocampus 6:579–590.
Lindner MD, Plone MA, Schallert T, Emerich DF (1997) Blind rats
are not profoundly impaired in the reference memory Morris
watermaze and cannot be clearly discriminated from rats with
cognitive deficits in the cued platform task. Brain Res Cogn
Brain Res 5:329–333.
Link W, Konietzko U, Kauselmann G, Krug M, Schwanke B, Frey U,
Kuhl D (1995) Somatodendritic expression of an immediate
early gene is regulated by synaptic activity. Proc Natl Acad Sci
USA 92:5734–5738.
Lipp HP, Wolfer DP (1998) Genetically modified mice and cognition.
Curr Opin Neurobiol 8:272–280.
Lipp HP, Schwegler H, Heimrich B, Driscoll P (1988) Infrapyramidal
mossy fibers and two-way avoidance learning: developmental
modification of hippocampal circuitry and adult behavior of
rats and mice. J Neurosci 8:1905–1921.
Lipp HP, Collins RL, Hausheer-Zarmakupi Z, Leisinger-Trigona
MC, Crusio WE, Nosten-Bertrand M, Signore P, Schwegler H,
Wolfer DP (1996) Paw preference and intra/infrapyramidal
mossy fibers in the hippocampus of the mouse. Behav Genet
26:379–390.
Lipp HP, Amrein I, Slominanka L, Wolfer DP (1999) Natural genetic
variation of hippocampal structures and behavior. In: Cellular
and quantitative methods in neurogenetics: methods in life sci-
ences (Jones BC, Mormede P, eds), pp 217–235. Boca Raton, FL:
CRC Press.
Lipp HP, Vyssotski AL, Wolfer DP, Renaudineau S, Savini M, Troster
G, Dell’Omo G (2004) Pigeon homing along highways and exits.
Curr Biol 14:1239–1249.
Lisman JE, Zhabotinsky AM (2001) A model of synaptic memory:
a Ca MKII/PP1 switch that potentiates transmission by
organizing an AMPA receptor anchoring assembly. Neuron
31:191–201.
Lyford GL, Yamagata K, Kaufmann WE, Barnes CA, Sanders LK,
Copeland NG, Gilbert DJ, Jenkins NA, Lanahan AA, Worley PF
(1995) Arc, a growth factor and activity-regulated gene, encodes
a novel cytoskeleton-associated protein that is enriched in neu-
ronal dendrites. Neuron 14:433–445.
Lynch G, Deadwyler S, Cotman C (1973) Postlesion axonal
rowth produces permanent functional connections. Science
180:1364–1366.
Mackintosh NJ (1983) Conditioning and associative learning. Oxford,
UK: Clarendon Press.
Mackintosh NJ (2002) Do not ask whether they have a cognitive map,
but how they find their way about. Psicologica 23:165–185.
Macphail EM (1982) Brain and intelligence in vertebrates. Oxford,
UK: Clarendon Press.
Macphail EM (1998) The evolution of consciousness. Oxford, UK:
Oxford University Press.
704 The Hippocampus Book
Macphail EM (2002) The role of the avian hippocampus in spatial
memory. Psicologica 23:93–108.
Macphail EM, Bolhuis JJ (2001) The evolution of intelligence: adap-
tive specializations versus general process. Biol Rev Camb Philos
Soc 76:341–364.
Maguire EA (2001) Neuroimaging studies of autobiographical event
memory. Philos Trans R Soc Lond B Biol Sci 356:1441–1451.
Maguire EA, Burgess N, Donnett JG, Frackowiak RS, Frith CD,
O’Keefe J (1998) Knowing where and getting there: a human
navigation network. Science 280:921–924.
Maguire EA, Frith CD, Morris RGM (1999) The functional neu-
roanatomy of comprehension and memory: the importance of
prior knowledge. Brain 122(Pt 10):1839–1850.
Malamut BL, Saunders RC, Mishkin M (1984) Monkeys with com-
bined amygdalo-hippocampal lesions succeed in object dis-
crimination learning despite 24-hour intertrial intervals. Behav
Neurosci 98:759–769.
Malkova L, Mishkin M (2003) One-trial memory for object-place
associations after separate lesions of hippocampus and
posterior parahippocampal region in the monkey. J Neurosci
23:1956–1965.
Malkova L, Gaffan D, Murray EA (1997) Excitotoxic lesions of the
amygdala fail to produce impairment in visual learning for
auditory secondary reinforcement but interfere with reinforcer
devaluation effects in rhesus monkeys. J Neurosci 17:6011–6020.
Malkova L, Lex CK, Mishkin M, Saunders RC (2001) MRI-based
evaluation of locus and extent of neurotoxic lesions in monkeys.
Hippocampus 11:361–370.
Mandler G (1980) Recognizing: the judgement of previous occur-
rence. Psychol Rev 87:252–271.
Manns JR, Clark RE, Squire LR (2000a) Parallel acquisition of aware-
ness and trace eyeblink classical conditioning. Learn Mem
7:267–272.
Manns JR, Stark CE, Squire LR (2000b) The visual paired-
comparison task as a measure of declarative memory. Proc Natl
Acad Sci USA 97:12375–12379.
Manns JR, Hopkins RO, Reed JM, Kitchener EG, Squire LR (2003)
Recognition memory and the human hippocampus. Neuron
37:171–180.
Maravita A, Iriki A (2004) Tools for the body (schema). Trends Cogn
Sci 8:79–86.
Maren S (2001) Neurobiology of pavlovian fear conditioning. Annu
Rev Neurosci 24:897–931.
Maren S (2005) Building and burying fear memories in the brain.
Neuroscientist 11:89–99.
Maren S, Fanselow MS (1997) Electrolytic lesions of the fim-
bria/fornix, dorsal hippocampus, or entorhinal cortex produce
anterograde deficits in contextual fear conditioning in rats.
Neurobiol Learn Mem 67:142–149.
Maren S, Aharonov G, Fanselow MS (1997) Neurotoxic lesions of the
dorsal hippocampus and pavlovian fear conditioning in rats.
Behav Brain Res 88:261–274.
Margules J, Gallistel CR (1988) Heading in the rat: determination by
environmental shape. Anim Learn Behav 16:404–410.
Markman EM, Abelev M (2004) Word learning in dogs? Trends Cogn
Sci 8:479–481.
Marr D (1971) Simple memory: a theory for archicortex. Philos Trans
R Soc Lond B: 262:23–81.
Martin A, Wiggs C, Weisberg J (1997) Modulation of human medial
temporal lobe activity by form, meaning and experience.
Hippocampus 7:587–593.
Martin SJ, Grimwood PD, Morris RGM (2000) Synaptic plasticity
and memory: an evaluation of the hypothesis. Annu Rev
Neurosci 23:649–711.
Martin SJ, de Hoz L, Morris RGM (2005) Retrograde amnesia: nei-
ther partial nor complete hippocampal lesions in rats result in
preferential sparing of remote spatial memory, even after
reminding. Neuropsychologia 43:609–624.
Matsumura N, Nishijo H, Tamura R, Eifuku S, Endo S, Ono T (1999)
Spatial- and task-dependent neuronal responses during real and
virtual translocation in the monkey hippocampal formation. J
Neurosci 19:2381–2393.
Maviel T, Durkin TP, Menzaghi F, Bontempi B (2004) Sites of neo-
cortical reorganization critical for remote spatial memory.
Science 305:96–99.
Mayford M, Bach ME, Huang Y-Y, Wang L, Hawkins R, Kandel ER
(1996) Control of memory formation through regulated
expression of a CaMKII transgene. Science 274:1678–1683.
McCarthy RA, Warrington EA (1990) Cognitive neuropsychology. San
Diego: Academic Press.
McClelland JL, Goddard NH (1996) Considerations arising from a
complementary learning systems perspective on hippocampus
and neocortex. Hippocampus 6:654–665.
McClelland JL, McNaughton BL, O’Reilly RC (1995) Why there are
complementary learning systems in the hippocampus and neo-
cortex: insights from the successes and failures of connectionist
models of learning and memory. Psychol Rev 102:419–457.
McDonald RJ, White NM (1994) Parallel information processing in
the watermaze: evidence for independent memory systems
involving dorsal striatum and hippocampus. Behav Neural Biol
61:260–270.
McDonald RJ, White NM (1995) Hippocampal and nonhippocam-
pal contributions to place learning in rats. Behav Neurosci
109:579–593.
McDonald RJ, Murphy RA, Guarraci FA, Gortler JR, White NM,
Baker AG (1997) Systematic comparison of the effects of hip-
pocampal and fornix-fimbria lesions on acquisition of three
configural discriminations. Hippocampus 7:371–388.
McDonald RJ, Devan BD, Hong NS (2004) Multiple memory
systems: the power of interactions. Neurobiol Learn Mem
82:333–346.
McEwen B, Sapolsky R (1995) Stress and cognitive function. Curr
Opin Neurobiol 5:205–212.
McGaugh JL (2000) Memory-a century of consolidation. Science
287:258–251.
McGlinchey-Berroth R, Carrillo MC, Gabrieli JDE, Brawn CM,
Disterhof JF (1997) Impaired trace eyeblink conditioning
in bilateral, medial-temporal lobe amnesia. Behav Neurosci
111:873–882.
McGonigle BO, Chalmers M (1977) Are monkeys logical? Nature
267:694–696.
McGonigle BO, Chalmers M (2003) The growth of cognitive struc-
tures in monkeys and men. In: Animal cognition and sequential
behavior: behavioral, biological and computational perspectives
(Fountain SB, Bunsey MD, Danks JH, McBeath MK, eds).
Boston: Kluwer Academic.
McGregor A, Good MA, Pearce JM (2004a) Absence of an interaction
between navigational strategies based on local and distal land-
marks. J Exp Psychol Anim Behav Process 30:34–44.
McGregor A, Hayward AJ, Pearce JM, Good MA (2004b) Hippocam-
pal lesions disrupt navigation based on the shape of the envi-
ronment. Behav Neurosci 118:1011–1021.

McHugh TJ, Blum KI, Tsien JZ, Tonegawa S, Wilson MA (1996)
Impaired hippocampal representation of space in CA1-specific
NMDAR1 knockout mice. Cell 87:1339–1349.
McNaughton BL, Barnes CA (1986) Long-term enhancement of hip-
pocampal synaptic transmission and the acquisition of spatial
information. J Neurosci 6:563–571.
McNaughton BL, Morris RGM (1987) Hippocampal synaptic
enhancement and information storage within a distributed
memory system. TINS 10:408–415.
McNaughton BL, Barnes CA, Meltzer J, Sutherland RJ (1989)
Hippocampal granule cells are necessary for normal spatial
learning but not for spatially-selective pyramidal cell discharge.
Exp Brain Res 76:485–496.
McNaughton BL, Chen LL, Markus EJ (1991) ′′Dead reckoning,′′
landmark learning, and the sense of direction: a neuro-
physiological and computational hypothesis. J Cogn Neurosci
3:190–202.
McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW,
Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver
KL (1996) Deciphering the hippocampal polyglot: the hip-
pocampus as a path integration system. J Exp Biol 199:173–185.
McNaughton BL, Barnes CA, Battaglia FP, Bower MR, Cowen SL,
Ekstrom AD, Gerrard JL, Hoffman KL, Houston FP, Karten Y,
Lipa P, Pennartz CMA, Sutherland GR (2003) Off-line repro-
cessing of recent memory and its role in memory consolidation:
a progress report. In: Sleep and synaptic plasticity (Smith C,
Maquet P, eds), pp 225–246. New York: Oxford University Press.
Meunier M, Bachevalier J, Mishkin M, Murray EA (1993) Effects on
visual recognition of combined and separate ablations of the
entorhinal and perirhinal cortex in rhesus monkeys. J Neurosci
13:5418–5432.
Mishkin M (1978) Memory in monkeys severely impaired by com-
bined but not by separate removal of amygdala and hippocam-
pus. Nature 273:297–298.
Mishkin M (1982) A memory system in the monkey. Philos Trans R
Soc Lond B Biol Sci 298:83–95.
Mishkin M, Delacour J (1975) An analysis of short-term visual mem-
ory in the monkey. J Exp Psychol Anim Behav Process 1:326–334.
Mishkin M, Murray EA (1994) Stimulus recognition. Curr Opin
Neurobiol 4:200–206.
Mishkin M, Suzuki WA, Gadian DG, Vargha-Khadem F (1997)
Hierarchical organization of cognitive memory. Philos Trans
Royal Soc Lond B Biol Sci 352:1461–1467.
Mishkin M, Vargha-Khadem F, Gadian DG (1998) Amnesia and
the organization of the hippocampal system. Hippocampus
8:212–216.
Mittelstaedt H, Mittelstaedt M (1982) Homing by path integration.
In: Avian navigation (Papi F, Wallraff HG, eds), pp 290–297.
Heidelberg: Springer-Verlag.
Miyashita Y (2004) Cognitive memory: cellular and network
machineries and their top-down control. Science 306:435–440.
Mizumori SJ, Miya DY, Ward KE (1994) Reversible inactivation of the
lateral dorsal thalamus disrupts hippocampal place representa-
tion and impairs spatial learning. Brain Res 644:168–174.
Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT (2003)
Hippocampal place cells acquire location-specific responses to
the conditioned stimulus during auditory fear conditioning.
Neuron 37:485–497.
Morgan DG, Diamond DM, Gottschall PE, Ugen KE, Dickey C,
Hardy JD, Duff K, Jantzen P, DiCarlo G, Wilcock D, Connor K,
Hatcher J, Hope C, Gordon MN, Arendash GW (2000) A beta
Theories of Hippocampal Function 705
peptide vaccination prevents memory loss in an animal model
of Alzheimer’s disease. Nature 482:982–986.
Morgan JI, Cohen DR, Hempstead JL, Curran T (1987) Mapping pat-
terns of c-fos expression in the central nervous system after
seizure. Science 237:192–197.
Morris RG, Mayes AR (2004) Long-term spatial memory: introduc-
tion and guide to the special section. Neuropsychology
18:403–404.
Morris RGM (1981) Spatial localisation does not depend on the pres-
ence of local cues. Learn Motiv 12:239–260.
Morris RGM (1983) Neural subsystems of exploration in rats. In:
Exploration in animals and humans (Archer J, Birke L, eds), pp
117–146. London: Van Nostrand Reinhold.
Morris RGM (1984) Developments of a watermaze procedure for
studying spatial learning in the rat. J Neurosci Methods 11:47–60.
Morris RGM (1991) Distinctive computations and relevant associa-
tive processes: hippocampal role in processing, retrieval, but not
storage of allocentric spatial memory. Hippocampus 1:287–290.
Morris RGM (2001) Episodic-like memory in animals: psychological
criteria, neural mechanisms and the value of episodic-like tasks
to investigate animal models of neurodegenerative disease.
Philos Trans R Soc Lond B Biol Sci 356:1453–1465.
Morris RGM (2006) Elements of a neurobiological theory of the hip-
pocampus: the role of activity-dependent synaptic plasticity in
episodic-like memory. Eur J Neurosci 23:2829–2846.
Morris RGM, Frey U (1997) Hippocampal synaptic plasticity: role in
spatial learning or the automatic recording of attended experi-
ence? Philos Trans R Soc Lond B Biol Sci 352:1489–1503.
Morris RGM, Kennedy MB (1992) The pierian spring. Curr Biol
2:511–514.
Morris RGM, Garrud P, Rawlins JN, O’Keefe J (1982) Place naviga-
tion impaired in rats with hippocampal lesions. Nature
297:681–683.
Morris RGM, Hagan JJ, Rawlins JN (1986a) Allocentric spatial learn-
ing by hippocampectomised rats: a further test of the ′′spatial
mapping′′ and ′′working memory′′ theories of hippocampal
function. Q J Exp Psychol B 38:365–395.
Morris RGM, Anderson E, Lynch GS, Baudry M (1986b) Selective
impairment of learning and blockade of long-term potentiation
by an N-methyl-D-aspartate receptor antagonist, AP5. Nature
319:774–776.
Morris RGM, Schenk F, Tweedie F, Jarrard LE (1990) Ibotenate
lesions of hippocampus and/or subiculum: dissociating compo-
nents of allocentric spatial learning. Eur J Neurosci 2:1016–1028.
Morris RGM, Martin SJ, Moser EI, Riedel G, Sandin J, Day M,
O’Carroll C (2003) Elements of a neurobiological theory of the
hippocampus: the role of activity-dependent synaptic plasticity
in memory. Philos Trans Roy Soc Lond B 358:773–786.
Moscovitch M (1995) Recovered consciousness: a hypothesis
concerning modularity and episodic memory. J Clin Exp
Neuropsychol 17:276–290.
Moscovitch M, Nadel L (1998) Consolidation and the hippocampal
complex revisited: in defense of the multiple-trace model. Curr
Opin Neurobiol 8:297–300.
Moser E, Moser MB, Anderson P (1993) Spatial learning impair-
ment parallels the magnitude of dorsal hippocampal lesions,
but is hardly present following ventral lesions. J Neurosci
13:3916–3925.
Moser EI, Krobert KA, Moser MB, Morris RGM (1998) Impaired spa-
tial learning after saturation of long-term potentiation. Science
281:2038–2042.
706 The Hippocampus Book
Moser MB, Moser EI (1998a) Distributed encoding and retrieval of
spatial memory in the hippocampus. J Neurosci 18:7535–7542.
Moser MB, Moser EI (1998b) Functional differentiation in the hip-
pocampus. Hippocampus 8:608–619.
Moser MB, Moser EI, Forrest E, Andersen P, Morris RGM (1995)
Spatial learning with a minislab in the dorsal hippocampus. Proc
Natl Acad Sci USA 92:9697–9701.
Moses SN, Ryan JD (2006) A comparison and evaluation of the pre-
dictions of relational and conjunctive accounts of hippocampal
function. Hippocampus 16:43–65.
Moss M, Mahut H, Zola-Morgan S (1981) Concurrent discrimina-
tion learning of monkeys after hippocampal, entorhinal, or
fornix lesions. J Neurosci 1:227–240.
Moyer JR, Thompson LT, Disterhoft JF (1996) Trace eyeblink condi-
tioning increases CA1 excitability in a transient and learning-
specific manner. J Neurosci 16:5536–5546.
Moyer JR, Power JM, Thompson LT, Disterhoft JF (2000) Increased
excitability of aged rabbit CA1 neurons after trace eyeblink con-
ditioning. J Neurosci 20:5476–5482.
Muller R (1996) A quarter of a century of place cells. Neuron
17:813–822.
Mumby DG (2001) Perspectives on object-recognition memory fol-
lowing hippocampal damage: lessons from studies in rats. Behav
Brain Res 127:159–181.
Mumby DG, Pinel JPJ, Wood ER (1990) Nonrecurring-items delayed
nonmatching-to-sample in rats: a new paradigm for testing
nonspatial working memory. Psychobiology 18:321–326.
Mumby DG, Wood ER, Pinel JP (1992a) Object-recognition memory
is only mildly impaired in rats with lesions of the hippocampus
and amygdala. Psychobiology 20:18–27.
Mumby DG, Wood ER, Pinel JPJ (1992b) Object recognition mem-
ory in rats is only mildly impaired by lesions of the hippocam-
pus and amygdala. Psychobiology 20:18–27.
Mumby DG, Pinel JPJ, Kornecook TJ, Shen MJ, Redila VA (1995)
Memory deficits following lesions of hippocampus or amygdala
in rats: an assesment by an object-memory tets battery.
Psychobiology 23:26–36.
Mumby DG, Astur RS, Weisend MP, Sutherland RJ (1999) Retrograde
amnesia and selective damage to the hippocampal formation:
memory for places and object discriminations. Behav Brain Res
106:97–107.
Murray EA, Bussey TJ (1999) Perceptual-mnemonic functions of the
perirhinal cortex. Trends Cogn Sci 3:142–151.
Murray EA, Bussey TJ (2001) Consolidation and the medial tempo-
ral lobe revisited: methodological considerations. Hippocampus
11:1–7.
Murray EA, Mishkin M (1983) Severe tactual memory deficits in
monkeys after combined removal of the amygdala and hip-
pocampus. Brain Res 270:340–344.
Murray EA, Mishkin M (1985) Amygdalectomy impairs crossmodal
association in monkeys. Science 228:604–605.
Murray EA, Mishkin M (1998) Object recognition and location
memory in monkeys with excitotoxic lesions of the amygdala
and hippocampus. J Neurosci 18:6568–6582.
Murray EA, Wise SP (2004) What, if anything, is the medial tempo-
ral lobe, and how can the amygdala be part of it if there is no
such thing? Neurobiol Learn Mem 82:178–198.
Murray EA, Gaffan D, Mishkin M (1993) Neural substrates of visual
stimulus-stimulus association in rhesus monkeys. J Neurosci
13:4549–4561.
Murray EA, Baxter MG, Gaffan D (1998) Monkeys with rhinal cortex
damage or neurotoxic hippocampal lesions are impaired on
spatial scene learning and object reversals. Behav Neurosci
112:1291–1303.
Murray EA, Graham KS, Gaffan D (2005) Perirhinal cortex and its
neighbours in the medial temporal lobe: contributions to mem-
ory and perception. Q J Exp Psychol 58:378–396.
Nadel L (1991) The hippocampus and space revisited. Hippocampus
1:221–229.
Nadel L (1995) The role of the hippocampus in declarative memory:
a comment on Zola-Morgan, Squire, and Ramus (1994).
Hippocampus 5:232–239.
Nadel L, Bohbot V (2001) Consolidation of memory. Hippocampus
11:56–60.
Nadel L, Hardt O (2004) The spatial brain. Neuropsychology
18:473–476.
Nadel L, Moscovitch M (1997) Memory consolidation, retrograde
amnesia and the hippocampal complex. Curr Opin Neurobiol
7:217–227.
Nadel L, Moscovitch M (1998) Hippocampal contributions to corti-
cal plasticity. Neuropharmacology 37:431–439.
Nadel L, Willner J (1980) Context and conditioning: a place for space.
Physiol Psychol 8:218–228.
Nadel L, O’Keefe J, Black AH (1975) Slam on the brakes: a critique of
Altman, Brunner and Bayer’s response-inhibition model of hip-
pocampal function. Behav Biol 14:151–162.
Nadel L, Samsonovich A, Ryan L, Moscovitch M (2000) Multiple
trace theory of human memory: computational, neuroimaging,
and neuropsychological results. Hippocampus 10:352–368.
Nader K (2003) Memory traces unbound. Trends Neurosci 26:65–72.
Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun
LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S
(2002) Requirement for hippocampal CA3 NMDA receptors in
associative memory recall. Science 297:211–218.
Nakazawa K, Sun LD, Quirk MC, Rondi-Reig L, Wilson MA, Tone-
gawa S (2003) Hippocampal CA3 NMDA receptors are crucial
for memory acquisition of one-time experience. Neuron
38:305–315.
Nakazawa K, McHugh TJ, Wilson MA, Tonegawa S (2004) NMDA
receptors, place cells and hippocampal spatial memory. Nat Rev
Neurosci 5:361–372.
Nelson W (1988) Food restriction, circadian disorder and longevity
of rats and mice. J Nutr 118:286–289.
Nemanic S, Alvarado MC, Bachevalier J (2004) The hippocampal/
parahippocampal regions and recognition memory: insights
from visual paired comparison versus object-delayed non-
matching in monkeys. J Neurosci 24:2013–2026.
Nunn JA, LePeillet E, Netto CA, Hodges H, Gray JA, Meldrum BS
(1994) Global ischaemia: hippocampal pathology and spatial
deficits in the watermaze. Behav Brain Res 62:41–54.
Ogden JA, Corkin S (1991) Memories of H.M. In: Memory mecha-
nisms: a tribute to G.V. Goddard (Abraham WC, Corballis M,
White KG, eds), pp 195–215. Hillsdale, NJ: Erlbaum,
Ohno M, Frankland PW, Chen AP, Costa RM, Silva AJ (2001)
Inducible, pharmacogenetic approaches to the study of learning
and memory. Nat Neurosci 4:1238–1243.
O’Keefe J (1976) Place units in the hippocampus of the freely mov-
ing rat. Exp Neurol 51:78–109.
O’Keefe J (1996) The spatial prepositions in English, vector grammar
and the cognitive map theory. In: Language and space (Bloom P,
Peterson MA, Nadel L, Garrett MF, eds), pp 277–316.
Cambridge, MA: MIT Press.
O’Keefe J (2005) Vector grammar, places, and the functional role of
the spatial prepositions in English. In: Representing direction in

language and space (van de Zee E, Slack J, eds), pp 69–85.
Oxford, UK: Oxford University Press.
O’Keefe J, Burgess N (1996) Geometric determinants of the place
fields of hippocampal neurons. Nature 381:425–428.
O’Keefe J, Conway DH (1978) Hippocampal place units in the freely
moving rat: why they fire where they fire. Exp Brain Res
31:573–590.
O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map:
preliminary evidence from unit activity in the freely-moving
rat. Brain Res 34:171–175.
O’Keefe J, Nadel L (1978) The hippocampus as a cognitive map.
Oxford, UK: Clarendon Press.
O’Keefe J, Nadel L (1979) Precis of O’Keefe & Nadel’s The
Hippocampus as a cognitive map. Behav Brain Sci 2:487–533.
O’Keefe J, Speakman A (1987) Single unit activity in the rat hip-
pocampus during a spatial memory task. Exp Brain Res 68:1–27.
O’Keefe J, Nadel L, Keightley S, Kill D (1975) Fornix lesions selec-
tively abolish place learning in the rat. Exp Neurol 48:152–166.
Olton DS (1977) The function of septo-hippocampal connections in
spatially organized behaviour. Ciba Found Symp 58:327–349.
Olton DS, Papas BC (1979) Spatial memory and hippocampal func-
tion. Neuropsychologia 17:669–682.
Olton DS, Samuelson RJ (1976) Remembrance of places passed: spa-
tial memory in rats. J Exp Psychol Anim Behav Process 2:97–116.
Olton DS, Becker JT, Handelmann GE (1979) Hippocampus, space,
and memory. Brain Behav Sci 2:313–365.
O’Reilly RC (1998) Six principles for biologically-based computa-
tional models of cortical cognition. Trends Cogn Sci 2:455–462.
O’Reilly RC, Norman KA (2002) Hippocampal and neocortical con-
tributions to memory: advances in the complementary learning
systems framework. Trends Cogn Sci 6:505–510.
O’Reilly RC, Rudy JW (2001) Conjunctive representations in learning
and memory: principles of cortical and hippocampal function.
Psychol Rev 108:311–345.
Otto T, Eichenbaum H (1992) Complementary roles of the orbital
prefrontal cortex and the perirhinal-entorhinal cortices in an
odor-guided delayed-nonmatching-to-sample task. Behav
Neurosci 106:762–775.
Overman WH, Ormsby G, Mishkin M (1990) Picture recognition vs.
picture discrimination learning in monkeys with medial tempo-
ral removals. Exp Brain Res 79:18–24.
Overton DA (1964) State-dependent or ‘dissociation’ learning pro-
duced with pentobarbital. J Comp Physiol Psychol 57:3–12.
Owen MJ, Butler SR (1981) Amnesia after transection of the fornix in
monkeys: long-term memory impaired, short-term memory
intact. Behav Brain Res 3:115–123.
Packard MG, McGaugh JL (1996) Inactivation of hippocampus or
caudate nucleus with lidocaine differentially affects expression
of place and response learning. Neurobiol Learn Mem 65:65–72.
Papadimitriou A, Wynne C (1999) Preserved negative patterning and
impaired spatial learning in pigeons (Columba livia) with
lesions of the hippocampus. Behav Neurosci 113:683–690.
Papi F (1990) Olfactory navigation in birds. Experientia 46:352–363.
Papini MR (2002) Pattern and process in the evolution of learning.
Psychol Rev 109:186–201.
Parkinson JK, Murray EA, Mishkin M (1988) A selective mnemonic
role for the hippocampus in monkeys: memory for the location
of objects. J Neurosci 8:4159–4167.
Parron C, Save E (2004) Comparison of the effects of entorhinal and
retrosplenial cortical lesions on habituation, reaction to spatial
and non-spatial changes during object exploration in the rat.
Neurobiol Learn Mem 82:1–11.
Theories of Hippocampal Function 707
Parslow DM, Rose D, Brooks B, Fleminger S, Gray JA, Giampietro V,
Brammer MJ, Williams S, Gasston D, Andrew C, Vythelingum
GN, Loannou G, Simmons A, Morris RG (2004) Allocentric
spatial memory activation of the hippocampal formation meas-
ured with fMRI. Neuropsychology 18:450–461.
Patel SN, Clayton NS, Krebs JR (1997) Spatial learning induces neu-
rogenesis in the avian brain. Behav Brain Res 89:115128.
Paulsen O, Moser EI (1998) A model of hippocampal memory
encoding and retrieval: GABAergic control of synaptic plastic-
ity. Trends Neurosci 21:273–278.
Pearce JM, Bouton ME (2001) Theories of associative learning in ani-
mals. Annu Rev Psychol 52:111–139.
Pearce JM, Hall G (1980) A model for pavlovian learning: variations
in the effectiveness of conditioned but not of unconditioned
stimuli. Psychol Rev 87:532–552.
Pearce JM, Roberts AD, Good M (1998) Hippocampal lesions disrupt
navigation based on cognitive maps but not heading vectors.
Nature 396:75–77.
Pearce JM, Good MA, Jones PM, McGregor A (2004) Transfer of
spatial behavior between different environments: implications
for theories of spatial learning and for the role of the hip-
pocampus in spatial learning. J Exp Psychol Anim Behav Process
30:135–147.
Phillips RG, LeDoux JE (1992) Differential contribution of amygdala
and hippocampus to cued and contextual for conditioning.
Behav Neurosci 106:274–285.
Phillips RG, LeDoux JE (1994) Lesions of the dorsal hippocampal
formation interfere with background but not foreground con-
textual fear conditioning. Learn Mem 1:34–44.
Pickens CL, Holland PC (2004) Conditioning and cognition.
Neurosci Biobehav Rev 28:651–661.
Pinker S (1994) The language instinct. New York: William Morrow.
Pitkanen A, Pikkarainen M, Nurminen N, Ylinen A (2000) Reciprocal
connections between the amygdala and the hippocampal for-
mation, perirhinal cortex, and postrhinal cortex in rat: a review.
Ann N Y Acad Sci 911:369–391.
Poldrack RA, Clark J, Pare-Blagoev EJ, Shohamy D, Creso Moyano J,
Myers C, Gluck MA (2001) Interactive memory systems in the
human brain. Nature 414:546–550.
Poucet B (1993) Spatial cognitive maps in animals: new hypotheses
on their structure and neural mechanisms. Psychol Rev
100:163–182.
Poucet B, Benhamou S (1997) The neuropsychology of spatial cogni-
tion in the rat. Crit Rev Neurobiol 11:101–120.
Poucet B, Lenck-Santini PP, Paz-Villagran V, Save E (2003) Place cells,
neocortex and spatial navigation: a short review. J Physiol Paris
97:537546.
Pravosudov VV, Clayton NS (2002) A test of the adaptive specializa-
tion hypothesis: population differences in caching, memory,
and the hippocampus in black-capped chickadees (Poecile atri-
capilla). Behav Neurosci 116:515–522.
Prusky GT, West PW, Douglas RM (2000a) Behavioral assessment of
visual acuity in mice and rats. Vision Res 40:2201–2209.
Prusky GT, West PW, Douglas RM (2000b) Reduced visual acuity
impairs place but not cued learning in the Morris water task.
Behav Brain Res 116:135–140.
Quirk GJ, Muller RU, Kubie JL, Ranck JB (1992) The positional firing
properties of medial entorhinal neurons: description and com-
parison with hippocampal place cells. J Neurosci 12:1945–1963.
Ramos JM (1998) Retrograde amnesia for spatial information: a dis-
sociation between intra- and extramaze cues following hip-
pocampus lesions in rats. Eur J Neurosci 10:3295–3301.
708 The Hippocampus Book
Rawlins JN, Lyford GL, Seferiades A, Deacon RM, Cassaday HJ
(1993) Critical determinants of nonspatial working memory
deficits in rats with conventional lesions of the hippocampus or
fornix. Behav Neurosci 107:420–433.
Rawlins JNP (1985) Associations across time: the hippocampus as a
temporary memory store. Behav Brain Sci 8:479497.
Redish AD (1999) Beyond the cognitive map: from place cells to
episodic memory. Cambridge, MA: MIT Press.
Reisel D, Bannerman DM, Schmitt WB, Deacon RM, Flint J,
Borchardt T, Seeburg PH, Rawlins JN (2002) Spatial memory
dissociations in mice lacking GluR1. Nat Neurosci 5:868–873.
Remondes M, Schuman EM (2004) Role for a cortical input to hip-
pocampal area CA1 in the consolidation of a long-term mem-
ory. Nature 431:699–703.
Rempel-Clower NL, Zola SM, Squire LR, Amaral DG (1996) Three
cases of enduring memory impairment after bilateral damage
limited to the hippocampal formation. J Neurosci 16:5233–5255.
Renner MJ (1990) Neglected aspects of exploratory and investigatory
behavior. Psychobiology 18:16–22.
Rescorla RA (1973) Effect of US habituation following conditioning.
J Comp Physiol Psychol 82:137–143.
Rescorla RA (1988) Pavlovian conditioning: it’s not what you think it
is. Am Psychol 43:151–160.
Rescorla RA, Wagner AR (1972) A theory of pavlovian conditioning:
the effectiveness of of reinforcement and nonreinforcement. In:
Classical conditioning. II. Current research and theory (Black AH,
Prokasy WF, eds), pp 64–99. New York: Appleton-Century-
Crofts.
Ressler KJ, Paschall G, Zhou X-L, Davis M (2002) Regulation of
synaptic plasticity genes during consolidation of fear condition-
ing. J Neurosci 22:7892–7902.
Richmond MA, Yee BK, Pouzet B, Veenman L, Rawlins JNP, Feldon J,
Bannerman DM (1999) Dissociating context and space within
the hippocampus: effects of complete, dorsal, and ventral exci-
totoxic hippocampal leesions on conditioned freezing and spa-
tial learning. Behav Neurosci 113:1189–1203.
Riedel G, Micheau J, Lam AG, Roloff E, Martin SJ, Bridge H, Hoz L,
Poeschel B, McCulloch J, Morris RGM (1999) Reversible neural
inactivation reveals hippocampal participation in several mem-
ory processes. Nat Neurosci 2:898–905.
Ringo JL (1991) Memory decays at the same rate in macaques with
and without brain lesions when expressed in d’ or arcsine terms.
Behav Brain Res 42:123–134.
Ringo JL (1993) Spared short-term memory in monkeys following
medial temporal lobe lesions is not yet established: a reply to
Alvarez-Royo, Zola-Morgan and Squire. Behav Brain Res
59:65–72.
Rizzolatti G, Craighero L (2004) The mirror-neuron system. Annu
Rev Neurosci 27:169–192.
Robbins TW, James M, Owen AM, Sahakian BJ, McInnes L, Rabbitt P
(1994) Cambridge Neuropsychological Test Automated Battery
(CANTAB): a factor analytic study of a large sample of normal
elderly volunteers. Dementia 5:266–281.
Roberts WA (2000) Are animals stuck in time? Psychol Bull
128:473–489.
Robertson RG, Rolls ET, Georges-Francois P (1998) Spatial view cells
in the primate hipocampus: effects of removal of view details. J
Neurophysiol 79:1145–1156.
Robinson L, Bridge H, Riedel G (2001) Visual discrimination learn-
ing in the watermaze: a novel test for visual acuity. Behav Brain
Res 119:77–84.
Rodrigo T (2002) Navigational strategies and models. Psicologica
23:2–32.
Rodrigo T, Chamizo VD, McLaren IPL, Mackintosh NJ (1997)
Blocking in the spatial domain. J Exp Psychol Anim Behav Process
23:110–118.
Rolls ET, Treves A (1998) Neural networks and brain function. Oxford,
UK: Oxford University Press.
Rosenbaum RS, Winocur G, Moscovitch M (2001) New views on old
memories: re-evaluating the role of the hippocampal complex.
Behav Brain Res 127:183–197.
Rosenzweig ES, Redish AD, McNaughton BL, Barnes CA (2003)
Hippocampal map realignment and spatial learning. Nat
Neurosci 6:609–615.
Ross RT, Orr WB, Holland PC, Berger TW (1984) Hippocampectomy
disrupts acquisition and retention of learned conditional
responding. Behav Neurosci 98:211–225.
Rossier J, Kaminsky Y, Schenk F, Bures J (2000) The place
preference task: a new tool for studying the relation between
behavior and place cell activity in rats. Behav Neurosci
114:273–284.
Rotenberg A, Mayford M, Hawkins RD, Kandel ER, Muller RU (1996)
Mice expressing activated CaMKII lack low frequency LTP and
do not form stable place cells in the CA1 region of the hip-
pocampus. Cell 87:1351–1361.
Rotenberg A, Abel T, Hawkins RD, Kandel ER, Muller RU (2000)
Parallel instabilities of long-term potentiation, place cells, and
learning caused by decreased protein kinase A activity. J
Neurosci 20:8096–8102.
Rothblat LA, Hayes LL (1987) Short-term object recognition mem-
ory in the rat: nonmatching with trial-unique junk stimuli.
Behav Neurosci 101:587–590.
Roullet P, Lassalle JM (1992) Behavioural strategies, sensorial
processes and hippocampal mossy fibre distribution in radial
maze performance in mice. Behav Brain Res 48:77–85.
Rudy JW, Sutherland RJ (1995) Configural association theory and the
hippocampal formation: an appraisal and reconfiguration.
Hippocampus 5:375–389.
Rudy JW, O’Reilly RC (1999) Contextual fear conditioning, conjunc-
tive representations, pattern completion, and the hippocampus.
Behav Neurosci 113:867–880.
Rudy JW, Barrientos RM, O’Reilly RC (2002) Hippocampal forma-
tion supports conditioning to memory of a context. Behav
Neurosci 116:530–538.
Rumelhart DE, McClelland JL (1986) Parallel distributed processing:
explorations in the microstructure of cognition. Cambridge, MA:
Bradford Books.
Rupniak NMJ, Gaffan D (1987) Monkey hippocampus and
learning about spatially directed movements. J Neurosci
7:2331–2337.
Ryle G (1949) The concept of mind. London: Hutchinson.
Sahgal A (1993) Behavioural neuroscience: a practical approach,.
Oxford: IRL Press.
Sakai K, Miyashita Y (1991) Neural organization for the long-term
memory of paired associates. Nature 354:152–155.
Salmon DP, Zola-Morgan S, Squire LR (1987) Retrograde amnesia
following combined hippocampus-amygdala lesions in mon-
keys. Psychobiology 15:37–47.
Sandi C, Loscertales M, Guaza C (1997) Experience-dependent facil-
itating effect of corticosterone on spatial memory formation in
the watermaze. Eur J Neurosci 9:637–642.
Sapolsky RM (1985) A mechanism for glucocorticoid toxicity in the

hippocampus: increased neuronal vulnerability to metabolic
insults. J Neurosci 5:1228–1232.
Saunders RC, Weiskrantz L (1989) The effects of fornix transection
and combined fornix transection, mammillary body lesions and
hippocampal ablations on object-pair association memory in
the rhesus monkey. Behav Brain Res 35:85–94.
Save E, Poucet B, Foreman N, Buhot MC (1992a) Object exploration
and reactions to spatial and nonspatial changes in hooded rats
following damage to parietal cortex or hippocampal formation.
Behav Neurosci 106:447–456.
Save E, Buhot MC, Foreman N, Thinus-Blanc C (1992b) Exploratory
activity and response to a spatial change in rats with hippocam-
pal or posterior parietal cortical lesions. Behav Brain Res
47:113–127.
Schacter DE, Tulving E (1994) Memory systems, pp 269–310.
Cambridge, MA: MIT Press.
Schenk F, Morris RGM (1985) Dissociation between components of
spatial memory in rats after recovery from the effects of retro-
hippocampal lesions. Exp Brain Res 58:11–28.
Schmajuk NA (1984) Psychological theories of hippocampal func-
tion. Physiol Psychol 12:166–183.
Schmitt WB, Sprengel R, Mack V, Draft RW, Seeburg PH, Deacon
RM, Rawlins JN, Bannerman DM (2005) Restoration of spatial
working memory by genetic rescue of GluR-A-deficient mice.
Nat Neurosci 8:270–272.
Schwegler H, Lipp HP (1981) Is there a correlation between hip-
pocampal mossy fiber distribution and two-way avoidance per-
formance in mice and rats? Neurosci Lett 23:25–30.
Schwegler H, Lipp HP (1983) Hereditary covariations of neu-
ronal circuitry and behavior: correlations between the
proportions of hippocampal synaptic fields in the regio infe-
rior and two-way avoidance in mice and rats. Behav Brain Res
7:138.
Schwegler H, Crusio WE, Brust I (1990) Hippocampal mossy fibers
and radial-maze learning in the mouse: a correlation with spa-
tial working memory but not with non-spatial reference mem-
ory. Neuroscience 34:293–298.
Schwegler H, Crusio WE, Lipp HP, Brust I, Mueller GG (1991) Early
postnatal hyperthyroidism alters hippocampal circuitry and
improves radial-maze learning in adult mice. J Neurosci
11:2102–2106.
Selden NRW, Everitt BJ, Jarrard LE, Robbins TW (1991)
Complementary roles for the amygdala and hippocampus in
aversive conditioning to explicit and contextual cues.
Neuroscience 42:335–350.
Seyfarth RM, Cheney DL, Marler P (1980) Monkey responses to three
different alarm calls: evidence of predator classification and
semantic communication. Science 210:801–803.
Shallice T (1988) From neuropsychology to mental structure. New
York: Cambridge University Press.
Sherry DF, Schacter DL (1987) The evolution of multiple memory
systems. Psychol Rev 94:439–454.
Sherry DF, Vaccarino AL (1989) Hippocampus and memory for food
caches in black-capped chickadees. Behav Neurosci
103:308–318.
Sherry DF, Vaccarino AL, Buckenham K, Herz RS (1989) The hip-
pocampal complex of food-storing birds. Brain Behav Evol
34:308–317.
Sherry DF, Jacobs LF, Gaulin SJ (1992) Spatial memory and
adaptive specialization of the hippocampus. Trends Neurosci
15:298–303.
Theories of Hippocampal Function 709
Shettleworth SJ (1998) Cognition, evolution, and behavior. New York:
Oxford University Press.
Shettleworth SJ (2001) Animal cognition and animal behaviour.
Anim Behav 61:277–286.
Shettleworth SJ (2003) Memory and hippocampal specialization in
food-storing birds: challenges for research on comparative cog-
nition. Brain Behav Evol 62:108–116.
Shettleworth SJ, Krebs JR (1982) How marsh tits find their hoards:
the roles of site preference and spatial memory. J Exp Psychol
Anim Behav Process 8:354–375.
Shiflett MW, Smulders TV, Benedict L, DeVoogd TJ (2003) Reversible
inactivation of the hippocampal formation in food-storing
black-capped chickadees (Poecile atricapillus). Hippocampus
13:437–444.
Shimamura AP, Squire LR (1987) A neuropsychological study of fact
memory and source amnesia. J Exp Psychol 13:464–473.
Siapas AG, Wilson MA (1998) Coordinated interactions between hip-
pocampal ripples and cortical spindals during slow-wave sleep.
Neuron 21:1123–1128.
Silva AJ, Paylor R, Wehner JM, Tonegawa S (1992) Impaired spatial
learning in alpha-calcium-calmodulin kinase II mutant mice.
Science 257:206–211.
Sirota A, Csicsvari J, Buhl D, Buzsaki G (2003) Communication
between neocortex and hippocampus during sleep in rodents.
Proc Natl Acad Sci USA 100:2065–2069.
Skelton RW, McNamara RK (1992) Bilateral knife cuts to the per-
forant path disrupt spatial learning in the Morris watermaze.
Hippocampus 2:73–80.
Smith ML, Milner B (1981) The role of the right hippocampus in the
recall of spatial location. Neuropsychologia 19:781–793.
Smulders TV, Sasson AD, DeVoogd TJ (1995) Seasonal variation in
hippocampal volume in a food-storing bird, the black-capped
chickadee. J Neurobiol 27:15–25.
Smulders TV, Shiflett MW, Sperling AJ, DeVoogd TJ (2000) Seasonal
changes in neuron numbers in the hippocampal formation of a
food-hoarding bird: the black-capped chickadee. J Neurobiol
44:414–422.
Sole LM, Shettleworth SJ, Bennett PJ (2003) Uncertainty in pigeons.
Psychon Bull Rev 10:738–745.
Spooner RIW, Thomson A, Hall J, Morris RGM, Salter SH (1994) The
Atlantis Platform: A new design and further developments of
Buresova’s on-demand platform for the watermaze. Learn Mem
1:203–211.
Squire LR (1992) Memory and the hippocampus: a synthesis from
findings with rats, monkeys, and humans. Psychol Rev
99:195–231.
Squire LR (1993) The hippocampus and spatial memory. TINS
16:56–57.
Squire LR (2004) Memory systems of the brain: a brief history and
current perspective. Neurobiol Learn Mem 82:171–177.
Squire LR, Alvarez P (1995) Retrograde amnesia and memory con-
solidation: a neurobiological perspective. Curr Opin Neurobiol
5:169–177.
Squire LR, Cohen NJ (1984) Human memory and amnesia.
In: Neurobiology of learning and memory (Lynch G, McGaugh
JL, Weinberger NM, eds), pp 3–64. New York: Gilford
Press.
Squire LR, Zola SM (1998) Episodic memory, semantic memory and
amnesia. Hippocampus 8:205–211.
Squire LR, Zola-Morgan S (1983) The neurology of memory: the case
for correspondence between the findings for human and non-
710 The Hippocampus Book
human primate. In: The physiological basis of memory, 2nd ed.
(Deutsch JA, ed). San Diego: Academic Press.
Squire LR, Zola-Morgan S (1991) The medial temporal lobe memory
system. Science 253:1380–1386.
Squire LR, Knowlton B, Musen G (1993) The structure and organi-
zation of memory. Annu Rev Psychol 44:453–495.
Squire LR, Clark RE, Knowlton BJ (2001) Retrograde amnesia.
Hippocampus 11:50–55.
Squire LR, Stark CE, Clark RE (2004) The medial temporal lobe.
Annu Rev Neurosci 27:279–306.
Staubli U, Ivy G, Lynch G (1984) Hippocampal denervation causes
rapid forgetting of olfactory information in rats. Proc Natl Acad
Sci USA 81:5885–5887.
Steele K, Rawlins JN (1993) The effects of hippocampectomy on per-
formance by rats of a running recognition task using long lists
of non-spatial items. Behav Brain Res 54:1–10.
Steele RJ, Morris RGM (1999) Delay-dependent impairment of a
matching-to-place task with chronic and intrahippocampal
infusion of the NMDA-antagonist D-AP5. Hippocampus
9:118–136.
Stefanacci L, Buffalo EA, Schmolck H, Squire LR (2000) Profound
amnesia after damage to the medial temporal lobe: a neu-
roanatomical and neuropsychological profile of patient E.P. J
Neurosci 20:7024–7036.
Steffenach HA, Sloviter RS, Moser EI, Moser MB (2002) Impaired
retention of spatial memory after transection of longitudinally
oriented axons of hippocampal CA3 pyramidal cells. Proc Natl
Acad Sci USA 99:3194–3198.
Stein DG, Finger S, Hart T (1983) Brain damage and recovery: prob-
lems and perspectives. Behav Neural Biol 37:185–222.
Steward O, Cotman CW, Lynch GS (1974) Growth of a new fiber pro-
jection in the brain of adult rats: re-innervation of the dentate
gyrus by the contralateral entorhinal cortex following ipsilateral
entorhinal lesions. Exp Brain Res 20:45–66.
Strasser R, Ehrlinger JM, Bingman VP (2004) Transitive behavior in
hippocampal-lesioned pigeons. Brain Behav Evol 63:181–188.
Suddendorf T, Busby J (2003) Mental time travel in animals? Trends
Cogn Sci 7:391–396.
Suddendorf T, Corballis MC (1997) Mental time travel and the evo-
lution of the human mind. Genet Soc Gen Psychol Monogr
123:133–167.
Sutherland RJ, Dyck RH (1984) Place navigation by rats in a swim-
ming pool. Can J Psychol 38:322–247.
Sutherland RJ, Rudy JW (1989) Confugural association theory: the
role of the hippocampal formation in learning, memory, and
amnesia. Psychobiology 17:129–144.
Sutherland RJ, Hamilton DA (2004) Rodent spatial navigation: at the
crossroads of cognition and movement. Neurosci Biobehav Rev
28:687–697.
Sutherland RJ, Whishaw IQ, Kolb B (1983) A behavioural analysis of
spatial localization following electrolytic, kainate- or colchicine-
induced damage to the hippocampal formation in the rat. Behav
Brain Res 7:133–153.
Sutherland RJ, Whishaw IQ, Kolb B (1988) Contributions of cingu-
late cortex to two forms of spatial learning and memory. J
Neurosci 8:1863–1872.
Sutherland RJ, Weisend MP, Mumby D, Astur RS, Hanlon FM,
Koerner A, Thomas MJ, Wu Y, Moses SN, Cole C, Hamilton DA,
Hoesing JM (2001) Retrograde amnesia after hippocampal
damage: recent vs. remote memories in two tasks. Hippocampus
11:27–42.
Suzuki S, Augerinos G, Black AH (1980) Stimulus control of spatial
behavior on the eight-arm maze in rats. Learn Motiv 1–8.
Suzuki WA, Amaral DG (1994a) Perirhinal and parahippocampal
cortices of the macaque monkey: cortical afferents. J Comp
Neurol 350:497–533.
Suzuki WA, Amaral DG (1994b) Topographic organization of the
reciprocal connections between the monkey entorhinal cortex
and the perirhinal and parahippocampal cortices. J Neurosci
14:1856–1877.
Suzuki WA, Clayton NS (2000) The hippocampus and memory: a
comparative and ethological perspective. Curr Opin Neurobiol
10:768–773.
Suzuki WA, Eichenbaum H (2000) The neurophysiology of memory.
Ann NY Acad Sci 911:175–191.
Suzuki WA, Zola-Morgan S, Squire LR, Amaral DG (1993) Lesions of
the perirhinal and parahippocampal cortices in the monkey
produce long-lasting memory impairment in the visual and tac-
tual modalities. J Neurosci 13:2430–2451.
Swanson LW (2000) Cerebral hemisphere regulation of motivated
behavior. Brain Res 886:113–164.
Swanson LW (2004) Brain maps: structure of the rat brain, 3rd ed.
Amsterdam: Elsevier.
Sweatt JD (2003) Mechanisms of memory. London: Academic
Press.
Tanner WP Jr, Swets JA (1954) A decision-making theory of visual
detection. Psychol Rev 61:401–409.
Taube JS (1998) Head direction cells and the neurophysiological basis
for a sense of direction. Prog Neurobiol 55:225–256.
Taube JS (1999) Some thoughts on place cells and the hippocampus.
Hippocampus 9:452–457.
Taube JS, Schwartzkroin PA (1987) Intracellular-recording from hip-
pocampal ca1 interneurons before and after development of
long-term potentiation. Brain Res 419:32–38.
Taube JS, Muller RU, Ranck JB Jr (1990a) Head-direction cells
recorded from the postsubiculum in freely moving rats. I.
Description and quantitative analysis. J Neurosci 10:420–435.
Taube JS, Muller RU, Ranck JB Jr (1990b) Head-direction cells
recorded from the postsubiculum in freely moving rats. II.
Effects of environmental manipulations. J Neurosci 10:438–447.
Taube JS, Kesslak JP, Cotman CW (1992) Lesions of the rat post-
subiculum impair performance on spatial tasks. Behav Neural
Biol 57:131–143.
Teng E, Squire LR (1999) Memory for places learned long ago is
intact after hippocampal damage. Nature 400:675–677.
Teng E, Stefanacci L, Squire LR, Zola SM (2000) Contrasting effects
on discrimination learning after hippocampal lesions and con-
joint hippocampal-caudate lesions in monkeys. J Neurosci
20:3853–3863.
Teuber HL (1955) Physiological psychology. Annu Rev Psychol
6:267–296.
Teyler TJ, DiScenna P (1986) The hippocampal memory indexing
theory. Behav Neurosci 100:147–154.
Teyler TJ, DiScenna P (1987) Long-term potentiation. Annu Rev
Neurosci 10:131–161.
Thinus-Blanc C, Bouzouba L, Chaix K, Chapuis N, Durup M, Poucet
B (1987) A study of spatial parameters encoded during explo-
ration in hamsters. J Exp Psychol Anim Behav Process
13:418–427.
Thomas KG, Hsu M, Laurance HE, Nadel L, Jacobs WJ (2001) Place
learning in virtual space. III. Investigation of spatial naviga-
tion training procedures and their application to fMRI and clin-

ical neuropsychology. Behav Res Methods Instrum Comput
33:21–37.
Thompson LT, Best PJ (1990) Long-term stability of the place-field
activity of single units recorded from the dorsal hippocampus of
freely behaving rats. Brain Res 509:299–308.
Thornton JA, Malkova L, Murray EA (1998) Rhinal cortex ablations
fail to disrupt reinforcer devaluation effects in rhesus monkeys
(Macaca mulatta). Behav Neurosci 112:1020–1025.
Tischmeyer W, Grimm R (1999) Activation of immediate early genes
and memory formation. Cell Mol Life Sci 55:564–574.
Tolman EC (1948) Cognitive maps in rats and men. Psychol Rev
55:189–208.
Tomita H, Ohbayashi M, Nakahara K, Hasegawa I, Miyashita Y
(1999) Top-down signal from prefrontal cortex in executive
control of memory retrieval. Nature 401:699–703.
Tommasi L, Thinus-Blanc C (2004) Generalization in place learning
and geometry knowledge in rats. Learn Mem 11:153–161.
Tonegawa S, Li Y, Erzurumlu RS, Jhaveri S, Chen C, Goda Y, Paylor R,
Silva AJ, Kim JJ, Wehner JM (1995) The gene knockout technol-
ogy for the analysis of learning and memory, and neural devel-
opment. Prog Brain Res 105:3–14.
Tooby J, Cosmides L (2000) Toward mapping the evolved functional
organisation of mind and brain. In: The new cognitive neuro-
sciences (Gazzaniga M, ed). Cambridge, MA: MIT Press.
Treves A, Rolls ET (1994) Computational analysis of the role of the
hippocampus in memory. Hippocampus 4:1–18.
Trullier O, Wiener SI, Berthoz A, Meyer JA (1997) Biologically based
artificial navigation systems: review and prospects. Prog
Neurobiol 51:483–544.
Trullier O, Shibata R, Mulder AB, Wiener SI (1999) Hippocampal
neuronal position selectivity remains fixed to room cues only in
rats alternating between place navigation and beacon approach
tasks. Eur J Neurosci 11:4381–4388.
Tsien JZ, Chen DF, Gerber D, Tom C, Mercer EH, Anderson DJ,
Mayford M, Kandel ER, Tonegawa S (1996) Subregion- and cell
type-restricted gene knockout in mouse brain. Cell
87:1317–1326.
Tulving E (1972) Episodic and semantic memory. In: Organisation of
memory (Tulving E, Donaldson W, eds), pp 381–403. San Diego:
Academic Press.
Tulving E (1983) Elements of episodic memory. New York: Oxford
University Press.
Tulving E (2004) Episodic memory and autonoesis: uniquely human?
In: The missing link in cognition: evolution of self-knowing con-
sciousness (Terrace H, Metcalfe J, eds). New York: Oxford
University Press.
Tulving E, Markowitsch HJ (1998) Episodic and declarative memory:
role of hippocampus. Hippocampus 8:198–204.
Ungerleider LG, Mishkin M (1982) Two cortical visual systems. In:
Analysis of visual behavior (Ingle DJ, Goodale MA, Mansfield
RJW, eds), pp 549–586. Cambridge, MA: MIT Press.
Vanderwolf CH, Cain DP (1994) The behavioral neurobiology of
learning and memory: a conceptual reorientation. Brain Res
Brain Res Rev 19:264–297.
Vanderwolf CH, Bland BH, Whishaw IQ (1973) Diencephalic hip-
pocampal and neocortical mechanisms in voluntary movement.
In: Efferent organisation and the integration of behavior (Masser
JD, ed), pp 229–263. San Diego: Academic Press.
Van Elzakker M, O’Reilly RC, Rudy JW (2003) Transitivity, flexibility,
conjunctive representations, and the hippocampus. I. An empir-
ical analysis. Hippocampus 13:334–340.
Theories of Hippocampal Function 711
Van Essen DC, Drury HA (1997) Structural and functional analyses
of human cerebral cortex using a surface based atlas. J Neurosci
17:7079–7102.
Vann SD, Aggleton JP (2004) Testing the importance of the retros-
plenial guidance system: effects of different sized retrosplenial
cortex lesions on heading direction and spatial working mem-
ory. Behav Brain Res 155:97–108.
Vann SD, Brown MW, Erichsen JT, Aggleton JP (2000) Fos imaging
reveals differential patterns of hippocampal and parahippocam-
pal subfield activation in rats in response to different spatial
memory tests. J Neurosci 20:2711–2718.
Van Praag H, Christie BR, Sejnowski TJ, Gage FH (1999) Running
enhances neurogenesis, learning, and long-term potentiation in
mice. Proc Natl Acad Sci USA 96:13427–13431.
Vargha-Khadem F, Gadian DG, Watkins KE, Connelly A, Van
Paesschen W, Mishkin M (1997) Differential effects of early hip-
pocampal pathology on episodic and semantic memory. Science
277:376–380.
Vazdarjanova A, Guzowski JF (2004) Differences in hippocampal
neuronal population responses to modifications of an environ-
mental context: evidence for distinct, yet complementary, func-
tions of CA3 and CA1 ensembles. J Neurosci 24:6489–6496.
Vazdarjanova A, McGaugh JL (1999) Basolateral amygdala is involved
in modulating consolidation of memory for classical fear condi-
tioning. J Neurosci 19:6615–6622.
Vinogradova OS (1995) Expression, control, and probable functional
significance of the neuronal theta rhythm. Prog Neurobiol
45:523–583.
Viskontas IV, McAndrews MP, Moscovitch M (2000) Remote
episodic memory deficits in patients with unilateral temporal
lobe epilepsy and excisions. J Neurosci 20:5853–5857.
Voermans NC, Petersson KM, Daudey L, Weber B, Van Spaendonck
KP, Kremer HP, Fernandez G (2004) Interaction between the
human hippocampus and the caudate nucleus during route
recognition. Neuron 43:427–435.
Wan H, Aggleton JP, Brown MW (1999) Different contributions of
the hippocampus and perirhinal cortex to recognition memory.
J Neurosci 19:1142–1148.
Wang R, Spelke E (2002) Human spatial representation: insights from
animals. Trends Cogn Sci 6:376.
Warburton EC, Baird A, Morgan A, Muir JL, Aggleton JP (2001) The
conjoint importance of the hippocampus and anterior thalamic
nuclei for allocentric spatial learning: evidence from a discon-
nection study in the rat. J Neurosci 21:7323–7330.
Warburton EC, Koder T, Cho K, Massey PV, Duguid G, Barker GR,
Aggleton JP, Bashir ZI, Brown MW (2003) Cholinergic neuro-
transmission is essential for perirhinal cortical plasticity and
recognition memory. Neuron 38:987–996.
Warrington EK, Weizkrantz L (1968) A study of learning and reten-
tion in amnesic patients. Neuropsychologia 6:283–291.
Wehner R, Lehrer M, Harvey WC (1996) Navigation: migration and
homing. J Exp Biol 199:1–260.
Weiskrantz L (1982) Comparative aspects of studies of amnesia.
Philos Trans R Soc Lond B Biol Sci 298:97–109.
Weiskrantz L (1997) Consciousness lost and found. Oxford, UK:
Oxford University Press.
Whishaw IQ (1998) Place learning in hippocampal rats and the path
integration hypothesis. Neurosci Biobehav Rev 22:209–220.
Whishaw IQ (2004a) Posterior neocortical (visual cortex) lesions in
the rat impair matching-to-place navigation in a swimming
pool: a reevaluation of cortical contributions to spatial behavior
712 The Hippocampus Book
using a new assessment of spatial versus non-spatial behavior.
Behav Brain Res 155:177–184.
Whishaw IQ (2004b) The behavior of the laboratory rat: a handbook
with tests. New York: Oxford University Press.
Whishaw IQ, Jarrard LE (1996) Evidence for extrahippocampal
involvement in place learning and hippocampal involvement in
path integration. Hippocampus 6:513–524.
Whishaw IQ, Tomie JA (1991) Acquisition and retention by hip-
pocampal rats of simple, conditional, and configural tasks using
tactile and olfactory cues: implications for hippocampal func-
tion. Behav Neurosci 105:787–797.
Whishaw IQ, Tomie JA (1997) Perseveration on place reversals in
spatial swimming pool tasks: further evidence for place learning
in hippocampal rats. Hippocampus 7:361–370.
Whishaw IQ, Kolb B, Sutherland RJ (1983) The analysis of behavior
in the laboratory rat. In: Behavioral approaches to brain research
(Robinson TE, ed), pp 141–211. New York: Oxford University
Press.
Whishaw IQ, Cassel JC, Jarrard LE (1995) Rats with fimbria-fornix
lesions display a place response in a swimming pool: a dissocia-
tion between getting there and knowing where. J Neurosci
15:5779–5788.
Whishaw IQ, Hines DJ, Wallace DG (2001) Dead reckoning (path
integration) requires the hippocampal formation: evidence from
spontaneous exploration and spatial learning tasks in light (allo-
thetic) and dark (idiothetic) tests. Behav Brain Res 127:49–69.
White NM, McDonald RJ (2002) Multiple parallel memory systems
in the brain of the rat. Neurobiol Learn Mem 77:125–184.
Wiener SI (1993) Spatial and behavioral correlates of striatal neurons
in rats performing a self-initiated navigation task. J Neurosci
13:3802–3817.
Wigstrom H, Gustafsson B (1983) Facilitated induction of hip-
pocampal long-lasting potentiation during blockade of inhibi-
tion. Nature 301:603–604.
Wiig KA, Bilkey DK (1995) Lesions of rat perirhinal cortex exacer-
bate the memory deficit observed following damage to the
fimbria-fornix. Behav Neurosci 109:620–630.
Willshaw D, Dayan P (1990) Optimal plasticity from matrix memo-
ries: what goes up must come down. Neural Commun 85–93.
Willshaw DJ, Buckingham JT (1990) An assessment of Marr’s theory
of the hippocampus as a temporary memory store. Philos Trans
R Soc Lond B Biol Sci 329:205–215.
Wilson A, Brooks DC, Bouton ME (1995) The role of the rat hip-
pocampal system in several effects of context in extinction.
Behav Neurosci 109:828–836.
Wilson BA (1999) Case studies in neuropsychological rehabilitation.
New York: Oxford University Press.
Wilson MA, McNaughton BL (1993) Dynamics of the hippocampal
ensemble code for space. Science 261:1055–1058.
Wilson MA, Tonegawa S (1997) Synaptic plasticity, place cells and
spatial memory: study with second generation knockouts.
Trends Neurosci 20:102–106.
Winocur G (1990) Anterograde and retrograde amnesia in rats with
dorsal hippocampal or dorsomedial thalamic lesions. Behav
Brain Res 38:145–154.
Winocur G, Rawlins JNP, Gray JA (1987) The hippocampus and con-
ditioning to contextual cues. Behav Neurosci 101:617–625.
Winocur G, Moscovitch M, Fogel S, Rosenbaum RS, Sekeres M
(2005) Preserved spatial memory after hippocampal lesions:
effects of extensive experience in a complex environment. Nat
Neurosci 8:273–275.
Winograd T (1975) Frame representations and the proce-
dural/declarative controversy. In: Representation and under-
standing: studies in cognitive science (Bobrow DG, Collins A,
eds), pp 185–210. San Diego: Academic Press.
Winters BD, Bussey TJ (2005) Glutamate receptors in perirhinal cor-
tex mediate encoding, retrieval, and consolidation of object
recognition memory. J Neurosci 25:4243–4251.
Winters BD, Forwood SE, Cowell RA, Saksida LM, Bussey TJ (2004)
Double dissociation between the effects of peri-postrhinal cor-
tex and hippocampal lesions on tests of object recognition and
spatial memory: heterogeneity of function within the temporal
lobe. J Neurosci 24:5901–5908.
Wirth S, Yanike M, Frank LM, Smith AC, Brown EN, Suzuki WA
(2003) Single neurons in the monkey hippocampus and learn-
ing of new associations. Science 300:1578–1581.
Witter MP, Naber PA, van Haeften T, Machielsen WC, Rombouts SA,
Barkhof F, Scheltens P, Lopes da Silva FH (2000) Cortico-
hippocampal communication by way of parallel parahippocam-
pal-subicular pathways. Hippocampus 10:398–410.
Wixted JT, Squire LR (2004) Recall and recognition are equally
impaired in patients with selective hippocampal damage. Cogn
Affect Behav Neurosci 4:58–66.
Wood ER, Dudchenko PA, Eichenbaum H (1999) The global
record of memory in hippocampal neuronal activity. Nature
397:613–616.
Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H (2000)
Hippocampal neurons encode information about different
types of memory episodes occurring in the same location.
Neuron 27:623–633.
Worden RP (1992) Navigation by fragment fitting: a theory of hip-
pocampal function. Hippocampus 2:165–187.
Wulff P, Wisden W (2005) Dissecting neural circuitry by combining
genetics and pharmacology. Trends Neurosci 28:44–50.
Yanike M, Wirth S, Suzuki WA (2004) Representation of well-
learned information in the monkey hippocampus. Neuron
42:477–487.
Yonelinas AP (2001) Components of episodic memory: the contri-
bution of recollection and familiarity. Philos Trans R Soc Lond B
Biol Sci 356:1363–1374.
Yonelinas AP, Kroll NE, Quamme JR, Lazzara MM, Sauve MJ,
Widaman KF, Knight RT (2002) Effects of extensive temporal
lobe damage or mild hypoxia on recollection and familiarity.
Nat Neurosci 5:1236–1241.
Zamanillo D, Sprengel R, Hvalby O, Jensen V, Burnashev N, Rozov
A, Kaiser KM, Koster HJ, Borchardt T, Worley P, Lubke J,
Frotscher M, Kelly PH, Sommer B, Andersen P, Seeburg PH,
Sakmann B (1999) Importance of AMPA receptors for hip-
pocampal synaptic plasticity but not for spatial learning. Science
284:1805–1811.
Zeman A (2002) Consciousness: a user’s guide. New Haven, CT: Yale
University Press.
Zhang WN, Bast T, Feldon J (2001) The ventral hippocampus and
fear conditioning in rats: different anterograde amnesias of fear
after infusion of N-methyl-D-aspartate or its noncompetitive
antagonist MK-801 into the ventral hippocampus. Behav Brain
Res 126:159–174.
Zhu XO, Brown MW, McCabe BJ, Aggleton JP (1995) Effects of the
novelty or familiarity of visual stimuli on the expression of the
immediate early gene c-fos in rat brain. Neuroscience
69:821–829.
Zinyuk L, Kubik S, Kaminsky Y, Fenton AA, Bures J (2000)

Understanding hippocampal activity by using purposeful
behavior: place navigation induces place cell discharge in both
task-relevant and task-irrelevant spatial reference frames. Proc
Natl Acad Sci USA 97:3771–3776.
Zola SM, Squire LR (2001) Relationship between magnitude of dam-
age to the hippocampus and impaired recognition memory in
monkeys. Hippocampus 11:92–98.
Zola SM, Squire LR, Teng E, Stefanacci L, Buffalo EA, Clark RE (2000)
Impaired recognition memory in monkeys after damage limited
to the hippocampal region. J Neurosci 20:451–463.
Zola-Morgan S, Squire LR (1984) Preserved learning in monkeys
with medial temporal lesions: sparing of motor and cognitive
skills. J Neurosci 4:107–1085.
Zola-Morgan S, Squire LR (1985) Medial temporal lesions in mon-
keys impair memory on a variety of tasks sensitive to human
amnesia. Behav Neurosci 99:22–34.
Zola-Morgan S, Squire LR (1990) The primate hippocampal forma-
Theories of Hippocampal Function 713
tion: evidence for a time-limited role in memory storage.
Science 250:288–290.
Zola-Morgan S, Squire LR, Mishkin M (1982) The neuroanatomy of
amnesia: amygdala-hippocampus versus temporal stem. Science
218:1337–1339.
Zola-Morgan S, Squire LR, Amaral DG (1986) Human amnesia and
the medial temporal region: enduring memory impairment fol-
lowing a bilateral lesion limited to field CA1 of the hippocam-
pus. J Neurosci 6:2950–2967.
Zola-Morgan S, Squire LR, Amaral DG (1989) Lesions of the amyg-
dala that spare adjacent cortical regions do not impair memory
or exacerbate the impairment following lesions of the hip-
pocampal formation. J Neurosci 9:1922–1936.
Zola-Morgan S, Squire LR, Ramus SJ (1994) Severity of memory
impairment in monkeys as a function of locus and extent of
damage within the medial temporal lobe memory system.
Hippocampus 4:483–495.
This page intentionally left blank
14 Á Neil Burgess
Computational Models of the Spatial
and Mnemonic Functions of the Hippocampus
Á data a vast body of knowledge has been collected regarding
14.1 Overview
Some of the most striking data relating cognitive behavior to
neuronal firing, damage, or metabolic activity in the brain
concerns the hippocampus. The use of computational models
has been invaluable for exploring the link between neurons
and behavior, enabling hypothetical mechanisms to be defined
precisely and examined quantitatively. Here I review many of
these models, including models of spatial functions, models
of more general associative mnemonic functions, models that
stress feedforward processing through the hippocampal sys-
tem, and those stressing recurrent processing within it. I
review the spatial models first, as they are most firmly rooted
in the known electrophysiology of the region. These models
cover both the representation of the animal’s spatial location
and orientation and the use of this information in spatial nav-
igation. The models of mnemonic function, specifically asso-
ciative or episodic memory, follow from Marr’s seminal 1971
model. I use this model as a generic framework in which to
consider the various subsequent developments to it. Finally, I
review those models attempting to bring together the spatial
and mnemonic functions of the hippocampus.
Á interpretation of a model, the way it works, and the values of
14.2 Introduction
There have been many attempts to understand and quantify
the contribution of the hippocampus to cognitive behavior. In
this chapter I focus on models of the link between the cogni-
tive ability of the animal and the action of individual cells
and synapses. As reviewed in Chapter 13, lesion studies in a
variety of mammals (including humans) have implicated the
hippocampus in spatial navigation, whereas human neu-
ropsychology has most notably implicated it in episodic or
declarative memory (see Chapter 12). In addition to these
the neural representation of the spatial location and orienta-
tion of freely moving rats (see Chapter 11). Hypotheses
regarding the function of the hippocampus have traditionally
been expressed in words. However, it is often possible to inter-
pret verbal descriptions in more than one way, or to change
their interpretation retrospectively to suit the facts. In addi-
tion, it can be difficult to tell whether the proposed explana-
tion would actually work as described and, if so, difficult to
make unambiguous quantitative predictions that can be used
to test it. These problems become even more acute when
hypotheses address the question of how a putative function
arises from the cooperative behavior of large numbers of neu-
rons and synapses. One way around this is to express such a
hypothesis in terms of equations or computer simulations,
referred to as a computational model. An advantage of this
approach is that all of the parameter values and assumptions
necessary to generate the behavior concerned need to be made
explicit; another is that the operation of the model is unam-
biguously specified. These advantages mean that computa-
tional modeling has an important role to play in the progress
of scientific understanding, most importantly in its interac-
tion with experimental investigation: by predicting critical
experiments, undergoing revision to reflect their results, and
predicting new ones. They are not a panacea for all ills; and
its parameters can still be changed or disputed. At the most
basic level a computational model can serve as an existence of
proof of the behavior that can result from a proposed mecha-
nism, but more generally it can serve to define a theoretical
understanding and provide a powerful framework within
which the nature of a theory relating brain to behavior can be
understood.
Examples of potential insights into the function of the
nervous system that would not have been fully understood,
and in some cases even suspected, without computational
modeling include the relation of voltage-dependent ion chan-
715
716 The Hippocampus Book
nels to the propagation of action potentials (Hodgkin and
Huxley, 1952); the possible relation of the pattern of firing of
dopamine neurons during conditioning experiments to the
process of learning from trial and error (interpretable via the
Rescorla-Wagner law and subsequent computational models
of “reinforcement learning”) (e.g., Shultz et al., 1997); the
relation of partially shifting gain field responses in parietal
cortex and optimal integration of multiple cues (made clear
by computational understanding of the way attractor net-
works can perform this function) (e.g., Deneve et al., 2001);
and the effect of genetic knockout of N-methyl-D-aspartate
(NMDA) receptors in hippocampal region CA3 on the
robustness of both the mouse’s spatial memory and the firing
of its place cells when a subset of environmental landmarks
are removed (interpretable as “pattern completion” in a model
of CA3 acting as an attractor network) (Marr, 1971; Nakazawa
et al., 2002) (see Section 14.3.3 and 14.5.3).
Computational modeling of the hippocampus has fol-
lowed two largely independent streams over the years: one
seeking to explain a general role in associative memory and
the other focusing on its role in spatial navigation. Models
usually start from as detailed a biophysical level as is useful for
the level of the hypothesis they seek to investigate. Although
many models reviewed here involve detailed simulation of cel-
lular and synaptic electrophysiology, the aim of this chapter is
to explain the neural bases of spatial and mnemonic behav-
ior—something that is made easier by focusing on the sim-
plest level of description capturing the likely functional
consequences of cellular and synaptic events (e.g., whether an
action potential was fired). Thus, the activity (firing rate) of a
neuron is simply viewed as a monotonic function of the
amount by which the net input to it exceeds some threshold
value. The net input to a neuron is the sum of the activity of
each neuron connected to it weighted by the strength of the
connection (occasionally inhibitory inputs are modeled as a
divisive term in the net input rather than a subtractive term,
see Figure 14–10, later). “Learning” corresponds to modifica-
tion of the connection strengths. Most commonly learning is
of a “Hebbian” nature (Hebb, 1949) such that simultaneous
pre- and post-connection activity leads to increased connec-
tion strength and is often used in explicit analogy to synaptic
processes such as long-term potentiation (LTP) (see Chapter
10) (see Box 14–1). Other concepts are explained as and where
necessary. Readers interested in neural computation more
generally should see the relevant literature (e.g., McClelland
and Rumelhart, 1986; Rumelhart and McClelland, 1986; Hertz
et al., 1990; Dayan and Abbott, 2002).
Because the anatomy of the region is similar in rats, pri-
mates, and humans, it seems sensible to start with the compu-
tational functions of the hippocampal region in the rat and
then consider how these functions might have been adapted
during evolution. This has the advantage of applying the most
detailed electrophysiological constraints at the outset.
Accordingly, in this chapter I concentrate initially on neural
models starting from the reliable and well understood body of
data regarding place cells and the neural representation of
space, drawing mostly on experimental data collected in rats.
I next consider models of the use of spatial representations in
guiding behavior. Together, models of the representation of
location and orientation from sensory input and models of
how these representations are used to guide behavior provide
one of the best examples of quantitative understanding of the
links between perception, cognition, and action and between
cells and systems and behavior. The second half of the chapter
concerns attempts to model the more general role of the
human hippocampus in memory for personal experience. The
chapter concludes with consideration of the models that
attempt to reconcile these two streams of research (spatial and
mnemonic). As we shall see, the role of the recurrent collater-
als in area CA3 of the hippocampus maintains a common
point of contact between these models: In both the spatial and
episodic memory frameworks they are assumed to perform an
associative memory function.
Á
14.3 Hippocampus and Spatial Representation
This section addresses the representation of spatial location
and orientation. Models of the representation of location
embodied by the firing of place cells are considered first. These
models take two often equivalent forms: those relying pre-
dominantly on feedforward connections to capture the data
and those relying predominantly on the recurrent connections
in CA3. It is not only the firing rates of place cells that encode
location but also the time of firing relative to the ongoing theta
rhythm on the electroencephalogram (EEG) (O’Keefe and
Recce, 1993). Both of these aspects of firing have been the sub-
jects of extensive computational modeling. The representation
of the animal’s orientation embodied by the head-direction
(HD) cells is equally striking; and the nature of this signal,
considered in the remainder of this section, has also been
investigated by computational modeling in more recent years.
Finally, the discovery of grid cells’ (Hafting et al., 2005) has
provided a new focus, reviewed by McNaughton et al. (2006).
14.3.1 Representing Spatial Location
and Orientation: Data
A rich set of experimental data has been gathered on the neu-
ral representation of spatial behavior found in and around the
hippocampus. Here I briefly summarize the results with the
greatest relevance to the models described below (see Chapter
11 for more details). The firing of place cells in the hip-
pocampi of freely moving rats encodes the location of the ani-
mal, each cell firing when the animal is within a particular
portion of its environment (the “place field”). Cells with sim-
ilar responses have also been observed in primates (Hori et al.,
2003; Ludvig et al., 2004) including humans (Ekstrom et al.,
2003). In open environments through which the rat can move
freely, firing rates are not influenced by the animal’s orienta-
tion, whereas in environments in which movement direction
 Computational Models of Spatial and Mnemonic Functions 717

Box 14–1
Learning via Synaptic Modification: Hebbian Learning Rules

In the simplest type of neural network model, the firing rate, or “activity,” of a neuron (a) is
simply a function (the “transfer function” f) of the net current coming into the neuron, which
in turn is simply a weighted sum of the firing rates (u i) of the neurons connecting to it. That is:
a f(Σi w i u i), often written as: a  f(w.u), where w is the vector of connection “weights”
modeling the strengths (e.g., net synaptic efficacy) of connections from the input neurons, and
the dot is the vector dot product. With the simplest, linear, transfer function, the activation is
given by:
a w.u (1)
In such networks, “learning” corresponds to modification of the connection weights w. Below I
discuss some of the Hebbian learning rules mentioned in the rest of the chapter, and their
effects, following the discussion in Dayan and Abbott (2002), where further details can be
found.
A learning rule directly implementing Hebb’s (1949) postulate of coincident firing leading to
increased coupling between neurons describes the change in connection weights in terms of the
product of pre- and post-synaptic firing rates:
dw dw
τ i  aui, or τ   au (2)
dt dt
where τ gives the rate of change of connection weights with time. When this rule is applied to a
“training set” of n example input patterns of activity u, each presented for an equal duration
over a total time T, we can integrate equation 2 to see the total change in w:
T
w → wΣau (3)
τ
where a  w.u from equation 1. If the connection weights are updated only after presenta-
tion of all of the input patterns, we can say:
T nT
w → wΣ(w.u)u  wQ.w (4)
τ τ
where Q is the correlation matrix of the input patterns (Q  u u , where  denotes the
average over input patterns, and u u is the outer product of u with itself). Thus, simple
Hebbian learning rules are also known as correlation-based learning rules. Inspection of equa-
tion 4 indicates that the weight vector w, if plotted in the same space as the input vectors u,
eventually follows the principal eigenvector of the correlation matrix; that is, it will lie along
the direction from the origin to the mean input pattern (u ) or, if u is at the origin,
along the first principal component of the set of input patterns. However, this learning rule is
not stable: Large weights produce large output activations, which produce large increases in
weights and so on. More formally, it can be seen from the dot product of w in equation 2 that
the length of the weight vector increases whenever the output neuron is active:
d ⏐w ⏐2 dw 2aw.u 2a2
  2w.    (5)
dt dt τ τ
One way to introduce balance into the learning rule is to allow for a connection weight to
increase or to decrease according to the levels of pre- and postconnection activity, by analogy
with long-term potentiation and depression (LTP and LTD) (see Chapter 10). In this way
equation 2 could become
dw dw
τ   (a θ)u, or τ   a(u-ϕ) (6)
dt dt
where either a postsynaptic threshold or a set of presynaptic thresholds are applied to deter-
mine the sense and size of weight changes (respectively: θ is the level postsynaptic activity must
surpass for the connection to increase rather than decrease; or ϕ, which is the vector of activity
levels each input neuron must surpass). The most obvious choice of threshold for the pre- or
postsynaptic neuron is its average activity over the training set. In this case, following a similar
derivation to equation 4, both versions produce the same learning rule:
nT
w → w C.w (7)
τ
(Continued)
718 The Hippocampus Book

Box 14–1
Learning via Synaptic Modification: Hebbian Learning Rules (Continued)

where C is the covariance matrix of the input patterns: C  (u- u )2   u u - u

2
 (u- u ) u . These learning rules are also known as covariance rules. Inspection of
equation 7 indicates that the weight vector will eventually follow the principal eigenvector of
the covariance matrix (i.e., it will lie along the direction of the first principal component of the
set of input patterns). It should be noted that these rules are also not stable; in this case d w 2/dt
is proportional to the variance (a2 a 2) of the output activity over the training set.
The BCM learning rule, derived from experimental investigation of visual cortical plasticity
(Bienenstock et al., 1982), proposes that:
dw
τ   au(a θ(a)) (8)
dt
This requires both pre- and postsynaptic activity for modification of a connection weight
(unlike the rules in equation 6), and also involves a sliding postsynaptic threshold (θ(a)), which
varies with postsynaptic activity. So long as the postsynaptic threshold increases as a power of
a 1 (typically following a time-averaged estimate of a2), it can ensure stability of the learning
rule: effectively increasing the threshold to an overactive output neuron so connection weights
to it tend to be reduced.
The other common way in which Hebbian learning rules are stabilized—e.g., in Rumelhart
and Zipser’s (1986) “competitive learning” algorithm—is to use divisive normalization: explic-
itly constraining the length of the weight vector to remain constant during learning by dividing
all weights by ⏐w ⏐. Although this is a nonlocal operation, because synaptic strengths must be
altered according to the state of other, distant synapses, a similar effect can be achieved by the
local learning rule of Oja (1982):
dw
τ   au a2w (9)
dt
An analysis similar to equation 5 shows that ⏐w⏐2 tends to a value 1/ under repeated applica-
tion of this rule.
As well as being stable, the BCM and normalized Hebbian learning rules involve competition
between connections: Increasing some of the connection weights onto a neuron leads to a de-
crease in the others. Under the BCM rule, this occurs because of increased activity leading to a
higher postsynaptic threshold and thus an increased incidence of LTD versus LTP. With normal-
ization it occurs directly owing to the increase in the length of the weight vector. These learning
rules tend to allow a neuron to become tuned to respond to specific patterns of input activation
(see text). There is at least some evidence for competitive interaction between synapses such
that increasing the strengths of one set of synapses leads to a decrease in the strengths of others
so as to normalize the total synaptic strength onto the neuron (Royer and Pare, 2003). Note
however, that recent evidence of the dependence of synaptic plasticity on the precise timing of
pre- and postsynaptic activity (see Chapter 10) changes the likely nature of learning rules based
on LTP and LTD, as in the effects of temporal asymmetry noted in Section 14.4.3.

is constrained (e.g., linear tracks, eight-arm mazes) firing is
strongly modulated by the rat’s direction of motion (see
Chapter 11). The orientation of the place cell representation is
controlled by “distal” cues at or beyond the edge of the envi-
ronment (O’Keefe and Conway, 1978; Muller and Kubie,
1987) but not by those within it (Cressant et al., 1997). A place
cell’s spatially localized firing appears to be robust to the
removal of subsets of cues and indeed removal of all of the
controlling visual cues while the rat remains in the environ-
ment (Muller and Kubie, 1987; O’Keefe and Speakman, 1987),
although remaining uncontrolled cues may be important in
these cases (Save et al., 2000).
The complementary representation of orientation inde-
pendent of location is found in “head direction cells” (HDCs)
in the mammillary bodies, anterior thalamic nuclei, and dor-
sal presubiculum (Taube et al., 1990). The HDCs code for
head direction within an environment, each firing whenever
the animal’s head points in a specific direction, independently
of the animal’s location. The orientation of the head direction
representation is controlled by distal visual cues in the same
way as the place cell representation. The overall orientation of
both place and HDC representations may be disrupted by dis-
orientation (rotating the rat in a covered container); and
when both have been recorded simultaneously, both represen-
tations have remained in register with each other (Knierim et
al., 1995). Interestingly, hints that an expanded representation
of neurons responding to specific combinations of place and
direction have been found in the presubiculum (Sharp, 1996;
Cacucci et al., 2004). A third type of representation, “grid
cells,” has now been found in dorsomedial entorhinal cortex
 Computational Models of Spatial and Mnemonic Functions
(Hafting et al., 2005). Each grid cell fires in a set of locations
which is laid out on a hexagonal grid. The intriguing proper-
ties of these cells are summarized at the end of Section 14.3.3.
However, at the time of this writing they have not yet been
fully incorporated into models of hippocampal function and
so are not a focus of this chapter.
Beyond the clear encoding of spatial location and orienta-
tion in the firing rates of these neurons, the picture becomes
slightly more complex. Initial experiments in which place cells
were recorded in environments of different shape (Muller and
Kubie, 1987) reported completely different patterns of firing in
the two environments (“remapping”). A place cell active in one
environment might fire in an unrelated location in the second
environment or might be silent. In other experiments, perhaps
involving less complete changes to the environment or less
extensive pretraining in the various environments, parametric
changes in the pattern of firing were observed (e.g., O’Keefe
and Burgess, 1996). Interestingly, although the firing rate of
place cells can be shown to encode the animal’s location within
a given environment (e.g., Wilson and McNaughton, 1993),
the times at which they fire relative to the theta rhythm
encodes additional information (see Chapter 11) (O’Keefe and
Recce, 1993; Skaggs et al., 1996; Jensen and Lisman, 2000).
14.3.2 Representing Spatial Location:
Feedforward Models
Computational modeling of place cell firing began with
Zipser’s (1985) model. In this model, sensory details of the
environment feedforward to landmark detectors and thence
to place cells. Landmark detectors are neurons specific to a
specific place cell and to a specific aspect of the sensory scene
(a “location parameter”). The output of these detectors is pro-
portional to the match between the stored state of a location
parameter and its currently perceived state. A place cell’s activ-
ity corresponds to a thresholded sum of the strengths of the
matches it receives from several landmark detectors. Interest-
ingly, the most obvious location parameter—distance from a
landmark—was rejected in favor of measures that scale with
environmental size, such as the retinal angle between two
landmarks, on the basis of a misinterpretation of Muller and
Kubie’s (1987) experiment. In this experiment, a significant
but small number of place fields were shown to expand fol-
lowing a doubling in the linear size of the environment.
However, this expansion corresponded to, at most, approxi-
mate doubling of the place field in area rather than the quad-
rupling predicted if everything scaled up proportionately.
Furthermore, in many cells the expansion often appeared to
be along only one environmental dimension rather than both.
The model captures some of the motility of place fields in the
presence of manipulations of environmental cues and some of
their robustness to the removal of subsets of cues and (incor-
rectly) produces a place field that scales up proportionately
with environmental expansion.
Sharp (1991) followed in the same vein of feedforward
modeling of the response of place cells to sensory input from
719
the environment but with the incorporation of an element of
“competitive learning” (Rumelhart and Zipser, 1986). Briefly,
this involves neurons arranged in groups dominated by lateral
inhibition such that only the neuron with the greatest input
can fire. Normalized Hebbian learning is then applied (i.e.,
increasing the strengths of connections between simultane-
ously active neurons while decreasing the others so the overall
strength of connections to a neuron does not change (see Box
14–1). This results in specific neurons coming to represent
specific patterns of input, with each neuron responding to a
particular pattern or patterns similar to it. Sharp’s model
envisaged two types of sensory input regarding each distal
cue: one representing its distance from the rat and the other
representing both its distance from the rat and its direction
relative to the rat’s heading. This sensory input passed forward
to a layer of entorhinal cortical cells and thence to a layer of
place cells (Fig. 14–1). Competitive learning at each layer
causes the entorhinal and place cells to learn to respond selec-
tively to the pattern of sensory input present in a particular
portion of the environment, and it produces reasonable
robustness to cue removal. The successive layers of competi-
tion produce sharper tuning to position and greater robust-
ness to cue removal in place cells than entorhinal cells. Due to
the use of distance-related inputs, expansion of the environ-
ment produces results that are qualitatively similar to those of
Muller and Kubie’s (1987) experiment. The most interesting
aspect of this model concerns the directional modulation of
place cell firing. In the model, place cell firing is initially direc-
tionally modulated owing to the partially directional sensory
inputs. During random exploration through an open environ-
ment, competitive learning allows a given place cell to learn to
respond to the sensory inputs occurring for different orienta-
tions at the place field, producing nondirectional firing. By
contrast, this does not occur during constrained motion (i.e.,
back and forth in a single direction). This provides a good,
simple model of the directionality of place cell firing,
although a more detailed look at directionality data indicates
that, if anything, place fields in open environments are ini-
tially nondirectional and become directional as a result of
experience. For alternative models see Blum and Abbott
(1996), Brunel and Trullier (1998) and Kali and Dayan (2000),
discussed below.
In an attempt to derive the form of the sensory input to the
place cells, O’Keefe and Burgess (1996) systematically varied
the shape and size of the rat’s environment while recording
from the same cells. The patterns of firing across environ-
ments included place fields that stretched or became bimodal
when the environment expanded. These patterns were not
consistent with those obtained in previous models of place
fields depending on the relative locations of discrete land-
marks from the rat (e.g., Zipser, 1985) but, rather, indicated
continuous dependence on environmental boundaries.
Specifically, place fields were viewed as a thresholded linear
sum of inputs tuned to respond to the presence of a boundary
at a given distance along a given allocentric direction (i.e.,
independent of the orientation of the rat and probably
720 The Hippocampus Book
Figure 14–1. Sharp’s (1991) model of place cell firing in response
to sensory input. A. Two groups of neocortical cells respond to the
distance or (egocentric) direction of specific cues as the rat moves
in a cylinder (shown in B, cues marked by letters). Competitive
learning in successive layers of entorhinal cells and place cells
leads, after exploration, to spatially tuned firing that is robust to
removal of subsets of cues. C. If exploration is unconstrained, as
determined relative to the head-direction system; see below)
(Fig. 14–2). These hypothetical inputs were termed “boundary
vector cells.” By fitting a place cell’s firing pattern across several
environmental shapes, the model can predict its firing pattern
in an environment of novel shape (Hartley et al., 2000). In an
experiment complementary to the variation of environmental
shape, Fenton et al. (2000) parametrically varied the position
of two cue cards around the edge of a cylindrical environment.
Although each card alone can control the overall orientation
of the recorded place fields, inconsistent movement of both
(i.e., moving them closer together or apart) produces inho-
mogeneous parametric movement of the place fields such that
their movement depends on their location. The added com-
plexity of the cue cards acting as both identifiable boundaries
and directional cues can be incorporated into the boundary
vector cell model by assuming that inconsistent cue-card
movement warps the representation of head direction, and
boundary vector cells become sensitive to variations in texture
after sufficient exposure (Burgess and Hartley, 2002).
in an open cylinder, place cell firing is only weakly modulated by
the heading direction of the animal (overall place field shown
in center, firing fields at each of eight heading direction shown
around the edge). D. If exploration is constrained to specific
directions of movement, as on a radial arm maze, place cell firing
is strongly modulated by the heading direction. (Source: Adapted
from Sharp, 1991.)
One phenomenon not addressed by the above models is
the “remapping” of place cell representations across different
environments. This remapping can be both partial and incre-
mental over time (e.g., Bostock et al., 1991; Skaggs and
McNaughton, 1998; Lever et al., 2002), with the eventual cre-
ation of stable but distinct patterns of firing in the two envi-
ronments. The factors influencing the speed and completeness
of remapping are not currently well understood, but one com-
mon change in an individual place cell is to continue to fire in
the environment in which it fires most strongly and to stop
firing in the other. This aspect of remapping was addressed by
Fuhs and Touretzky (2000) in a model of learning in the per-
forant path projection from entorhinal cortex to place cells in
CA3. They found that the usual learning rules relating synap-
tic modification to the product of pre- and postsynaptic activ-
ity (i.e.. Hebbian learning) or to its covariance were unable to
reproduce this behavior. In the case of Hebbian learning, a
place cell with strong firing in one environment and weak fir-
ing in the other strengthens its firing in both environments. In
 Computational Models of Spatial and Mnemonic Functions
b)
a) d)
Figure 14–2. Model of the geometrical influence on place fields
(Hartley et al., 2000), assuming a stable directional reference frame.
Place fields are composed from thresholded linear sums of the
firing rates of boundary vector cells (BVCs). a. Above: Each BVC
has a Gaussian tuned response to the presence of a boundary at
a given distance and bearing from the rat (independent of its
orientation). Below: The sharpness of tuning of a BVC decreases
as the distance to which it is tuned increases. The only free parame-
ters of a BVC are the distance and direction of the peak response.
the case of covariance learning, exposure to the second envi-
ronment tends to lead to loss of the place cell representation
of the first environment. By contrast, the BCM (Bienenstock-
Cooper-Munro) learning rule (Bienenstock et al., 1982) (see
Box 14–1), which explicitly makes the direction of synaptic
modification dependent on the strength of the postsynaptic
activity, did produce the desired result: strong firing remain-
ing stable and weak firing reducing with experience. This type
of learning can also capture the way place fields become more
coherent with time and the temporal dynamics of their
response to the introduction of a barrier into the environment
(Barry et al., 2006).
Evidence of experience-dependent change in place cell fir-
ing also comes from experiments by Mehta et al. (1997, 2000).
They found that, over the first few runs through a CA1 place
field on a linear track, the spatial distribution of firing changes
from roughly symmetrical to slightly asymmetrical, caused by
additional firing at low rates earlier on the track. They suggest
that this change results from the known temporal asymmetry
of LTP (which is greater when presynaptic activity precedes
postsynaptic activity than vice versa) acting on the CA3 to
CA1 pathway (see Chapter 10). Other models have implicated
the recurrent connections in CA3 as responsible for this effect.
Interestingly, the phase shift seen in place cell firing (see
Section 14.3.1 for data and Section 14.3.5 for models) (see
also Chapter 11) acts to increase this effect of temporal asym-
metry in LTP, causing place cells with fields early on the path
to fire before those with fields later along the path during each
theta cycle. The apparently unrelated effect of the remapping
721
c)
e)
b. Place fields recorded from the same cell in four environments of
different shape or orientation relative to distal cues. c. Simulation of
the place fields in b by the best fitting set of four BVCs constrained
to be in orthogonal directions (BVCs shown on the left, simulated
fields on the right). The simulated cell can now be used to predict
firing in novel situations. Real and predicted data from three novel
environments are shown in d and e, respectively, showing good
qualitative agreement. (Source: Adapted from Burgess and Hartley,
2002.)
of place cell representations across different environments has
also been ascribed to the effects of the recurrent connections
in CA3. Models stressing these connections are the subject of
the next section.
14.3.3 Representing Spatial Location
and Orientation: Feedback Models
The long-range recurrent connections between pyramidal
cells in area CA3 of the hippocampus have long been inter-
preted as enabling this region to work as an autoassociative
neural network (e.g., Marr, 1971; Hopfield, 1982; Amit, 1989).
This type of network is most often used to provide a content-
addressable memory, a subject explored in Section 14.5 (see
Box 14–2). In this section I consider the role played by recur-
rent collaterals in the spatial representations of place and HD
cells. Interestingly, direct experimental evidence for an asso-
ciative function for CA3 has emerged recently, with indica-
tions that the NMDA receptors in this region are involved in
making both the place fields and the rat’s spatial memory
robust to cue removal (Nakazawa et al., 2002). In parallel,
attractor dynamics have been found in the place cell represen-
tation of two environments of different shape after fast
remapping caused by exposure to the two environmental
shapes made of different materials (Wills et al., 2005). In these
representations, in contrast to those that have not fast-
remapped (Leutgeb et al., 2005), the two shapes act as attrac-
tors: all place cells in intermediate shaped environments
coherently returning to one or other representation.
722 The Hippocampus Book

Box 14–2
Attractors in Memory, Neural Coding, and Path Integration

POINT ATTRACTORS AND MEMORY

A network of recurrently connected neurons can be arranged so a finite number of discrete

patterns of activation across the neurons are stable states, or “attractors.” This means that any
pattern of activation similar enough to one of these attractors will evolve into the attractor
pattern under the dynamics of the network. These patterns of activation are “stored” in the
network in the sense that they are “retrieved” from any similar enough initial pattern. Such
networks are also referred to as “autoassociative” networks, and are an example of a “content-
addressable” memory in that a pattern of activity is retrieved by a pattern of similar content,
rather than, say, an unrelated index term or the address of a storage location.
In one of the simplest models (Hopfield, 1982), the activity of neuron i is modeled as ai 
 1. Connections between neurons i and j are symmetrical, with synaptic “weight” wij  wji
The dynamics of the network are given by: ai(t1)  sign(Σj wij aj(t)), such that the “energy”
or Lyapunov function of the network:
E∝ Σij wijaiaj
can only reduce. If connection weights undergo a form of Hebbian learning when the to-be-
stored patterns of activation (a , say) are present, such that wij ∝Σaiaj , these representa-
tions become attractors so long as the number of stored patterns is not too large (less than
around 0.14N, where N is the number of neurons, in this case). That is, a similar enough pat-
tern of activation converges onto the stored pattern under the dynamics of the network (Cohen
and Grossberg, 1983). This situation is often visualized by imagining how the “energy” of the
network varies as a function of the networks’ “state” au  (a1, a2 , .. aN)—the attractor states
being local minima of the energy surface to which nearby states evolve under the network’s
dynamics (Fig. 14–3A). Similar behavior is also shown by more biologically realistic models
(Amit, 1992; Treves and Rolls, 1992; McClelland et al., 1995) (Fig. 14–10).
LINE ATTRACTORS AND NEURAL CODING

The value of a continuous variable (or “stimulus”s) often seems to be represented in the firing
rates of a population of neurons, each of which is tuned to respond preferentially to a single
“preferred” value. For example, head-direction cells can be thought of in this way, with s repre-
senting the rat’s heading. The pattern of activation of the population is often visualized by
imagining the neurons arranged so their location reflects their preferred values: showing a
smooth bump of activity across the neurons peaked at the actual value of the stimulus.
However, if the firing rates are noisy it is difficult to estimate the precise value of the stimulus.
The presence of recurrent connections between neurons, arranged so that the weight of the
connection between each pair is simply a decreasing function of the difference in their pre-
ferred values (or physical separation when arranged as above), can help by ensuring that the
firing pattern takes the shape of a smooth bump1 (Fig. 14–3B). With the appropriate choice
of recurrent connections, such a network can perform optimal decoding (Latham et al., 2003),
including the situation where the representation is formed from different, unreliable sources
of information (Deneve et al., 2001).
The patterns of activation comprising a smooth bump can be thought of as a line in the N
dimensional state space au  (a1, a2 , .. aN) of the network. Each point on the line corresponds to
a different estimate of s (referred to as ŝ). Conversely, all of the possible noisy patterns of activa-
tion that end up producing the same ŝ lie on an N-1 dimensional subspace within which the
action of the recurrent connections corresponds to convergence onto the line (Fig. 14–3C).
An important aspect of these networks is that, although the recurrent connections ensure that
patterns of activation move onto the line attractor, movement along it, corresponding to chang-
ing ŝ, is not affected by the recurrent connections (because the connections between a pair of
neurons depends only on the difference in their preferred values, not what those preferred
values are).
LINE ATTRACTORS AND PATH INTEGRATION

Because the (symmetrical) recurrent connections provide no resistance to the motion of the
bump of activity along the line attractor, its position is easily moved by asymmetrical connec-
tions from each neuron to neighbors farther along the line (Skaggs et al., 1995; Zhang, 1996).
1 These patterns have low “energy” as activation is concentrated in nearby neurons, which have the
strongest interconnections.
 Computational Models of Spatial and Mnemonic Functions 723

The greater the size of the asymmetrical connections—which should correspond to the spatial
derivative of the symmetrical connections for the bump to move without changing shape
(Zhang, 1996)—compared to the symmetrical ones, the faster the movement of the bump
(Figs. 14–3 to 14–5). Thus, if the strength of the asymmetrical connections is proportional to
angular velocity, the location of the bump of activity in a ring of head direction cells tracks the
head direction of the animal—performing angular “path integration.” For a more detailed
model, see Redish et al. (1996).
As noted by Zhang (1996) and McNaughton et al. (1996), the angular path integration mod-
els of head direction cell firing can be extended to path integration models of place cell firing.
In this case, the place cells are imagined as a two-dimensional array, so the location of each
neuron corresponds to the location of its place field in the environment (Fig. 14–6). Again,
symmetrical connections decreasing in strength with the physical separation of the pre- and
postsynaptic neurons can ensure that neural activity forms a single-peaked bump over the
array. Asymmetrical connections from each neuron to its neighbors along a given direction
causes the bump to shift in that direction (Figs. 14–3 and 14–4). In this case, to perform path
integration of position, the strength of the asymmetrical connections between a pair of neu-
rons displaced in a given direction needs to be proportional to the velocity of the rat in that
direction. See Samsonovich and McNaughton (1997) for a more detailed model and Droulez
and Berthoz (1991) and Dominey and Arbib (1992) for the origins of this type of model.

Continuous Attractor Models
of Head Direction Cells
The simplest examples of the use of continuous attractors (see
Box 14–2) to model spatial representations come from mod-
els of the representation of head direction rather than loca-
tion. In many respects the literature on head-direction cells
(HDCs) is much more straightforward than that on place
cells. The overall orientation of the head-direction representa-
tion can be controlled by sensory cues in a way similar to that
of the place cell representation. Unlike the place cells, how-
ever, there have been no reports to date of HDCs changing
their preferred orientations relative to each other. Even when
the rat is disoriented or is in a symmetrical environment with-
out polarizing cues, the preferred directions of simultaneously
recorded HDCs remain in synchrony—if they rotate, all rotate
together. For this reason all models of the head-direction sys-
tem follow the same basic mechanism of a one-dimensional
continuous attractor, or “line attractor” (Skaggs et al., 1995;
Zhang, 1996) from which the two-dimensional continuous
attractor models of place cells developed (Figs. 14–3 to 14–6)
(see Box 14–2).
If HDCs are imagined laid out in a ring with each cell’s
location corresponding to its preferred direction and each is
connected to its neighbors (see Box 14–2), activity is smoothly
peaked at the current heading direction. Skaggs et al.’s model
contains two more rings of cells, with each cell receiving con-
nections from the corresponding HDC. One ring is composed
of “left rotation” cells, which project back to the HDCs to the
left (anticlockwise) of their location; the other ring is com-
posed of “right rotation” cells, which project back to HDCs to
the right (clockwise) of their location. The left rotation cells
corresponding to the current heading direction are activated
when the rat is turning left because of inputs from the vestibu-
lar system as well as the HDCs, causing the HDC activation to
move leftward. In addition to these cells, the HDCs receive
input from “sensory cells” (“visual cells” in Fig. 14–5) and so
can be associated with those sensory inputs appearing at a sta-
ble bearing during exploration of a new environment. These
sensory inputs subsequently prevent the cumulative errors that
would otherwise occur in the integration of angular velocity.
This basic model has been implemented and extended in
various ways, developing in hand with our knowledge of the
operation of the head-direction system. This is now thought
to involve a circuit from the mammillary bodies (MBs) to
anterior thalamic nuclei (ATN) to dorsal presubiculum (PS).
In this circuit, cells in the MBs code for head direction further
in the future (60–70 ms) than those in the ATN (20–30 ms),
whereas those in the PS code for current or past head direc-
tion (0 to –10 ms) (Blair and Sharp, 1995; Blair et al., 1998;
Taube, 1998; Taube and Muller, 1998) (see Chapter 11).
Various of the additional detailed properties of these systems,
such as time advances and asymmetrical responses during
turning, have been modeled more recently (e.g., Touretzky
and Redish, 1996; Blair et al., 1997; Goodridge and Touretzky,
2000). However, here I focus on the hippocampus, and return
to models of location.
Continuous Attractor Models of Place Cells
Samsonovich and McNaughton (1997) produced a detailed
model of the place cell representation as a continuous attrac-
tor, following Zhang (1996). In this model, the recurrent con-
nections in CA3 are preconfigured to provide several
continuous attractor representations of location (termed
“charts”). Each chart involves a different set of place cells, the
relative positions of whose place fields are predetermined. The
strength of the recurrent connection between two cells in a
chart is set as a Gaussian function of the proximity of their
place fields. The place cells connect with a “path integration”
(PI) system, thought to be in the subiculum, in which neurons
respond to combinations of the rat’s location and orientation
724 The Hippocampus Book
Figure 14–3. Point attractors and line attractors in neural systems
(see Box 14–2). A. Point attractor is a pattern of activation a  (a1,
a2, a3...) into which other nearby patterns evolve under the dynam-
ics of the network (determined by the pattern of connection
strengths, update rule, and so on). In some cases a function can be
defined that can only decrease under the dynamics (a Lyanpunov
function, or the energy (E) of a physical system), so that attractor
states lie at local minima of this function. B. Population encoding
and decoding. Neural populations can encode the current value of a
variables, say in the pattern of activation across neurons, each of
which is tuned to respond to a preferred value. These are often
imagined to be laid out so that the position of each neuron on the
ordinate corresponds to its preferred value. For example, a set of
head-direction (HD) cells might each be tuned to a different
(Sharp, 1996; Cacucci et al., 2004). Specifically, the place cells
connect to PI neurons representing similar locations, and the
return projections connect back to place cells representing
slightly different locations shifted along the rat’s direction of
motion. The gain of this return projection is modulated by
information relating to the rat’s speed of self-motion
(presumably carried by motor efference signals). This system,
although demanding a highly specific set of hard-wired
connections, provides a self-consistent continuous attractor
representation of location that moves automatically with self-
motion. Finally, the hippocampus also receives sensory input
so that when the rat is placed in an environment for the
first time associations between the sensory scene and the
internal representation of location can be formed, which can
then be used to reset the system periodically. Overall, the
model can be seen as a possible implementation of O’Keefe
and Nadel’s (1978, pp. 220–230) view of the role of path
(s)
(s)
(s)
“preferred” direction (top). If firing rates are noisy, it may be diffi-
cult to estimate the actual value of the variable (middle). Recurrent
connections can be organized so all other patterns of activation
evolve into smooth bump-shaped patterns of activation (below).
This process can provide an optimal way of decoding the value of
the variable from the population (the peak of the bump is the
dashed line). C. The set of smooth bump-shaped patterns of activa-
tion form a “line attractor,” a continuous set of patterns of activity
onto which other nearby patterns evolve but along which move-
ment is unimpeded. Locations along the line attractor can be
thought of as estimates of the variable (Ŝ); all of the patterns of
activity that end up at a given estimate (such as Ŝ 1) form a subspace
a(Ŝ) within which the intersection with the line is a point attractor.
(Source: Adapted from Latham et al., 2003.)
integration in supporting short-term continuity in a cognitive
map.
The Samsonovich and McNaughton model is consistent
with data showing that the stability of the place cell represen-
tation is dependent on NMDA receptors (Kentros et al., 1998)
and that place field locations remain consistent with each
other but slowly drift in the absence of anchoring sensory cues
or if the rat is consistently disoriented before each trial
(Knierim et al., 1995). In addition, the involvement of some
form of path integration is suggested by the increased influ-
ence of the boundary the rat is running from compared to the
one it is running toward (Gothard et al., 1996; O’Keefe and
Burgess, 1996; Redish et al., 2000). Note that the important
role played by path integration in this model does not, con-
versely, imply that the hippocampus is necessary for guiding
behavior in tests of path integration (e.g., Alyan and
McNaughton, 1999). The role of the CA3 recurrent collaterals

Figure 14–4. Continuous attractor networks of place and head
direction cells (HDCs). A. Emergence of a stable firing profile from
an arbitrary initial state in a network of HDCs arranged as a one-
dimensional continuous attractor. The cells are indexed by their pre-
ferred firing direction and connected by weights with a symmetrical
distribution (i.e., an even function of the difference between cell’s
tuning directions—see C, top row). B. Movement of the peak caused
by an asymmetrical component in the weight distribution (see C,
middle row). C. Distribution of connection weights in a one-dimen-
sional attractor network (bottom row), showing a symmetrical com-
ponent (top row) and an additional asymmetrical component
(middle row). Note the slight asymmetry in the combined connec-
S
γ S
S γ S
tion weights. The asymmetrical component is the spatial derivative
of the symmetrical component along the direction of drift, and its
size () determines the speed of drift of the represented head direc-
tion. D. Two-dimensional place cell network similar to the one-
dimensional HDC network, showing emergence of a stereotyped
stable firing profile from an arbitrary initial state, using symmetrical
weight distribution (a Gaussian with constant inhibitory back-
ground). E. As with the one-dimensional network, the addition of
an asymmetrical component to the connection weights causes the
represented location to drift (again, the asymmetrical component is
the spatial derivative along the direction of drift, and its size deter-
mines the speed of drift). (Source: Adapted from Zhang, 1996.)
725
726 The Hippocampus Book
Vestibular cell (right) Visual cell
Vestibular cell (left)
Rotation cell (left)
Rotation cell (right)
Head direction cell
Figure 14–5. Skaggs et al.’s (1995) model of HDCs, showing the lat-
eral connections among HDCs providing a continuous attractor
and connections from left or right rotation cells, visual inputs, and
vestibular inputs. The input from rotation cells serves the same pur-
pose as the asymmetrical component of lateral connections in
Figure 14–4. (Source: Adapted from Skaggs et al., 1995.)
in self-localization; forming the most accurate representation
of location within a continuous attractor network given con-
flicting inputs was also stressed by Redish and Touretzky
(1998). In addition, they elaborated the relation of this system
Figure 14–6. Place cell activity on a “chart.” The activity of a
population of 36 simultaneously recorded place cells shown
symbolically distributed in the box (in which the rat foraged for
food) each cell being placed at the center of its place field. Units on
horizontal axes are centimeters. The animal is located at the center
with other systems for path integration and for maintaining
orientation (Touretzky and Redish, 1996). However, no spe-
cific mechanism has been proposed for the path integration.
Because errors accumulate so rapidly in this system, it can be
expected to behave very differently from a perfect system. The
integration by place cells of visual inputs and inputs from a
recurrently connected parietal system, and its relation to
remapping, was explored further by Guazzelli et al. (2001).
Samsonovich and McNaughton suggested that remap-
ping reflects the system switching between uncorrelated
charts. This is a reasonable model of the situation after fast-
remapping, which is consistent with each chart acting as an
attractor (Wills et al., 2005), although in this case the charts
would not be preconfigured but formed by the process of fast
remapping. However, the model is not consistent with the sit-
uations in which individual place fields can move relative to
each other in response to environmental change (e.g., O’Keefe
and Burgess, 1996; Fenton et al., 2000). To fit these data
requires feedforward inputs to dominate, replacing the
model’s main feature. The model is also not consistent with
slow remapping (Lever et al., 2002) or partial remapping (e.g.,
Skaggs and McNaughton, 1998). The experimental conditions
and neural mechanisms resulting in fast versus slow remap-
ping are currently not well understood (Knierim, 2003).
Recurrent networks have also been used to model place
field directionality, as modeled in a feedforward manner
by Sharp (1991). In these models (Brunel and Trullier, 1998;
Kali and Dayan, 2000), place cell firing is initially derived from
orientation-specific sensory input at each location (referred to
as the “local view” from that location), and the dynamics of
the network are strongly dependent on the recurrent connec-
of the square and is moving to the left and toward the viewer.
The shape of the activity packet does not depend on velocity,
acceleration, future trajectory of motion, or theta frequency.
(Source: Samsonovich and McNaughton, 1997.)
 Computational Models of Spatial and Mnemonic Functions
tions in CA3. If exploration is unconstrained and random,
Hebbian learning in the recurrent collaterals of these models
results in a continuous attractor of the sort hard-wired
by Samsonovich and McNaughton and an orientation-
independent place cell representation of space. One caveat to
this is that the Hebbian learning must be modulated by nov-
elty to prevent inhomogeneous exploration from causing
highly nonuniform weight structures and unrealistic firing
patterns (Kali and Dayan, 2000). Such novelty information
has been suggested as a function of the cholinergic septal
inputs to the hippocampus (Hasselmo et al., 1996; but see
Hasselmo and Fehlau, 2001 and Lisman and Grace, 2005). As
with Sharp (1991), in these models directionally constrained
exploration results in orientation-dependent place fields.
Kali and Dayan (2000) also simulated the effects of chang-
ing the environment on the pattern of place cell firing. The
model first learns a representation in one environment, as
described above, and is then exposed to a second environ-
ment. If the second environment is sufficiently novel, the pat-
tern of sensory input is assumed to be completely different
and novelty-modulated learning is enabled. As a result, the
system successfully learns a second, unrelated, place cell repre-
sentation, corresponding to complete remapping. In simula-
tions using two similar boxes (as in Skaggs and McNaughton,
1998), similar firing in the two boxes is imposed by similar
sensory input, and the model is also shown to be capable of
storing and recalling partially overlapping representations.
Finally, if the second environment differs from the first only
geometrically, the same sensory inputs are assumed to be used
(but with different values, as the distances to boundaries have
changed) and there is no new learning. This results in para-
metric changes to the pattern of firing. In this case, the effect
of the recurrent collaterals is to preserve the relative spatial
relation of the place fields and so acts to maintain the posi-
tions of fields in terms of the ratio of distances between
boundaries (effectively scaling up the representation). By con-
trast, the sensory inputs are equivalent to the boundary vector
cell model (O’Keefe and Burgess, 1996; Hartley et al., 2000)
and so act to maintain the position of fields in terms of their
absolute distance from boundaries. Thus, this model can
account for both types of behavior. To fit the data in O’Keefe
and Burgess (1996), the model should be dominated by the
sensory input rather than the recurrent connections, as place
fields tend to maintain a fixed distance to environmental
boundaries in this experiment. By contrast, the apparently
similar experiment of Muller and Kubie (1987) results in
complete remapping. It is not yet clear whether models such
as that of Kali and Dayan (2000), which depend on assigning
different amounts of similarity to two environments, can cap-
ture the complexities of the data concerning remapping. For
example, the differences between fast and slow remapping
referred to above imply some learning in the feedforward con-
nections (to model slow remapping, see, e.g., Barry et al.,
2005; Fuhs and Touretzky, 2000), as well as in the recurrent
connections (to model fast-remapping). Like Sharp’s (1991)
model, these recurrent models are also inconsistent with hints
727
that place cell firing in open fields is initially nondirectional
but can become directional as a result of experience.
Finally, note that the recent discovery of grid cells in the
dorsomedial entorhinal cortex (Hafting et al., 2005) are likely
to have profound implications for ideas of hippocampal func-
tion. Each cell fires when the rat occupies multiple spatial
locations laid out on a startlingly regular hexagonal grid. The
orientation and scale of these grids seems to be constant in
different environments and between different nearby cells.
Overall, grids get larger as the recording site moves ventrolat-
erally. Because the grids of neighboring cells have the same
orientation and scale and appear to have fixed offsets, it is pos-
sible that the recurrent connections between local sets of grid
cells could be endlessly tuned to perform path integration,
irrespective of the location of the animal, providing the con-
tinuous attractor envisaged in CA3 by Zhang (1996) and
Samsonovich and McNaughton (1997). The stable association
of each grid to the environment might occur via connections
to place cells that, by virtue of generally having only a single
firing location, could be associated with the environmental
stimuli at that location. If this is correct, place cell firing could
reflect the superposition of multiple grids that all overlap at
the place field, and remapping could reflect changes in offset
in these multiple inputs (O’Keefe and Burgess, 2005, see also
Fuhs and Touretzky (2000) and McNaughton et al., (2006)).
14.3.4 Modeling Phase Coding in Place Cells
The origin of the intriguing phenomenon of the phase coding
of place cell firing with respect to the concurrent theta rhythm
of the EEG (O’Keefe and Recce, 1993) has been the subject of
several computational models. Before considering these mod-
els, I briefly review the relevant experimental findings (see
Chapter 11 for more details). The theta rhythm is a large-
amplitude oscillation of around 6 to 10 Hz of the EEG and
is present whenever the rat is moving its head through the
environment. As the rat runs through a place field, the corre-
sponding place cell tends to fire spikes with a systematic phase
relation to the theta rhythm. On entering the field, spikes
are fired at a “late” phase; as the rat passes through the field,
spikes are fired at successively earlier phases so that on exiting
the field the phase of firing may be up to 360 earlier (corre-
sponding to an “early” phase). This effect is most pronounced
when the rat runs repetitively in a directionally constrained
manner (e.g., on a linear track). Interestingly, the phase of fir-
ing correlates better with the location of the rat in the place
field than with other variables such as the time spent in the
field or the instantaneous firing rate of the cell (Huxter et
al., 2003).
The apparently tight coupling between location and phase
implies that the phase of firing might depend on the sensory
input driving the place cell in a simple feedforward way. Thus,
O’Keefe (1990, 1991) proposed that each place cell might act
as a phasor representing location relative to the centroid and
eccentricity, or slope, of a subset of the sensory cues in an
environment: the firing rate encoding proximity to the cen-
728 The Hippocampus Book
troid and the phase relative to theta encoding the bearing to
the centroid relative to the direction defined by the eccentric-
ity, or slope, of the subset of cues. The advantages of the pha-
sor representation for simple vector calculations were stressed
in this model, but its easy relation to the subsequent experi-
mental data on phase is at least as significant. Burgess et al.
(1993) suggested a direct interpretation of phase of firing: that
it could be used to separate place cells with fields ahead of the
rat from those with fields behind the rat, which can be useful
for navigation (see Section 14.4.2). This was subsequently ver-
ified experimentally (Burgess et al., 1994; Skaggs et al., 1996;
Samsonovich and McNaughton, 1997). They also demon-
strated that a phase shift in each individual cell is consistent
with theta modulation of the net activity of the population of
cells. In a simple feedforward model of place cell firing,
Burgess et al. (1994) simulated the phase of firing as a func-
tion of the rat’s position relative to the sensory cues that drove
the cell. The input to entorhinal comes from pairs of “sen-
sory” cells, each tuned to respond at a given distance from a
particular sensory cue such that the maximum input ampli-
tude occurs at the centroid of the two cues. The phase of fir-
ing of entorhinal cells was assumed to reflect the angle of the
cue-centroid from the rat: firing at a late phase when driven by
cues ahead of the rat and at an early phase when driven by
cues behind the rat. This model is broadly consistent with the
synchrony and oscillations seen in some sensory circuits (e.g.,
Nicolelis et al., 1995) but remains more qualitative than quan-
titative. Bose and Recce (2001) propose an alternative model
that also accounts for the lack of phase shift seen in place cells
when a rat runs in a running wheel (Hirase et al., 1999),
involving the dynamics of the interaction of place cells and
interneurons and the assumption that the frequency of the
theta rhythm depends on running speed on the linear track
but not in the running wheel. However, this picture became
more complicated with the subsequent claim that phase shift-
ing does occur in the running wheel at high firing rates
(Harris et al., 2002).
An appealingly simple alternative feedforward model is
possible (Mehta et al., 2000; Harris et al., 2002): The excitatory
synaptic input to a place cell might increase as the rat runs
through the place field, and the theta rhythm might reflect a
sawtooth-shaped inhibitory input (i.e., inhibition decreasing
through each cycle). In this model, the firing phase would
advance simply because the increasing excitatory input man-
ages to overcome the inhibitory input successively earlier in
each cycle. The cause of the increasing excitatory input might
reflect an exaggerated form of the asymmetry reported by
Mehta et al. (1997) or an increasing then decreasing input but
with lack of firing on the decreasing portion due to effects
such as habituation (Harris et al., 2002). These models capture
the observation that the phase shift becomes more reliable
over the first few runs of a trial, as does the asymmetry of
place fields (Mehta et al., 2002), and allows phase to be ana-
lyzed in terms of firing rate during nontranslational behaviors
such as dreaming or wheel running (Harris et al., 2002).
However, the correlation between phase and location is
stronger than that between phase and rate; and on the linear
track at least, the weaker correlation is a side effect of the
stronger one (O’Keefe and Burgess, 2005).
As with models of place cell firing, a second type of model
of the phase shift stresses the recurrent connections in contrast
to the feedforward connections. Indeed, simulations of the
Samsonovich and McNaughton model, in which net activa-
tion is made to oscillate at the theta frequency, shows some-
thing qualitatively similar to the phase shift owing to path
integration occurring within each cycle. That is, initially the
set of active place cells settles to just those place cells with
fields centered on the rat and then expands to include those
with fields centered ahead of the rat. The first quantitative
model of the phase shift was proposed by Tsodyks et al. (1996).
In this model, the recurrent connections between place cells in
CA3 are asymmetrically arranged so that each place cell proj-
ects to place cells farther along a learned path (see also Blum
and Abbott, 1996, discussed below). External input to a CA3
place cell arrives at a fixed (early) phase of theta, causing place
cell activity at this phase, which in turn propagates through
the recurrent connections to place cells with fields farther
along the path and causes them to fire. Overall activity is
inhibited at the end of each theta cycle, preventing further
propagation of activity into the next cycle. Thus, when the rat
enters a place field, the corresponding cell starts to fire at a late
phase owing to propagated activity from cells with fields ear-
lier on the path; and it fires earlier during each cycle as the rat
advances owing to activity having to propagate through fewer
cells, until finally firing at the early phase is due solely to exter-
nal inputs. Several similar mechanisms that depend on the
association of place cells firing earlier along a learned path to
those firing later along it have now been proposed (Jensen and
Lisman, 1996; Touretzky and Redish, 1996; Wallenstein and
Hasselmo, 1997). The Jensen and Lisman (1996) model makes
interesting suggestions for the gamma rhythm, in separating
the firing of cells corresponding to the current and succes-
sively farther advanced locations, and for the dynamics of
NMDA channels, in separating each route retrieval into suc-
cessive cycles of the theta rhythm. Wallenstein and Hasselmo
(1997) emphasized the role of GABAB receptors in varying the
relative influence of the inputs to CA1 from CA3 compared to
those directly from entorhinal cortex (EC) over the theta cycle:
allowing sensory (EC) input to dominate early and predictive
input from CA3 to dominate late in the cycle.
These models fit nicely with the observations of Mehta et
al. (1997, 2000): They produce a phase shift limited to 360
that is more strongly correlated with position than time, and
that is greater for well learned paths than for random explo-
ration. Interestingly, Mehta et al. (2000) suggested a similar
model based on learned asymmetry in the connections from
CA3 to CA1, the main testable difference here being that the
phase shift should not be observed in CA3. Other aspects of
these models seem less likely. First, because the initial firing
of a place cell depends on both the externally driven activity
of other cells and its propagation through the network, it
seems likely that on cell-by-cell and run-by-run bases the
 Computational Models of Spatial and Mnemonic Functions
initial phase of firing should be more variable than the (exter-
nally driven) final phase of firing. This is not the case in the
data (Skaggs et al., 1996; Huxter et al., 2003). Second, one
mechanism for producing the required asymmetrical connec-
tions is that suggested by Mehta et al., (2000). However, it has
since been found that, although the development of asymme-
try in place fields over the first few runs of a trial is prevented
by blockade of NMDA receptors, the phase shift phenomenon
is unaffected by this manipulation (Ekstrom et al., 2001).
A third type of model stresses the inherent oscillatory
nature of some cellular processes, as did Jensen and Lisman
(1996) for different reasons. Thus, O’Keefe and Recce (1993)
pointed out that the phase and amplitude characteristics of
place cell firing could be modeled as the interference pattern
between an 11-Hz external input to the cell (perhaps the sen-
sory input) and a 9-Hz external or internal oscillation corre-
sponding to theta (perhaps driven by the septal input). This
produces an oscillation of 10 Hz, corresponding to firing that
shifts in phase relative to theta and a 1-Hz envelope, one-half
cycle of which corresponds to the place field. This model has
since been extended (Lengyel et al., 2003), identifying the first
input as a voltage-controlled oscillation of the membrane
potential (e.g., Hoppensteadt, 1986) in the dendrites and the
second as an inhibitory input to the soma of fixed frequency.
The frequency of the dendritic oscillation is assumed to
increase above that of the somatic oscillation proportionally
with the strength of the dendritic input, which is assumed to
be zero outside the place field and proportional to the rat’s
running speed within it (McNaughton et al., 1983; Czurko et
al., 1999; Ekstrom et al., 2001; Huxter et al., 2003). Thus, the
two oscillations destructively interfere outside of the place
field, whereas phase of firing relative to the somatic input
within the field can shift more rapidly as the rat runs faster,
preserving the relation between phase and location. In addi-
tion, the dendritic oscillation needs to be weakly driven in
antiphase to the somatic input so in the absence of any den-
dritic input it returns to being in phase with it to ensure com-
plete destructive interference. Recent corroborative evidence
for interference models comes from the observation that the
increase in place field size along the dorsoventral axis of the
hippocampus parallels a corresponding decrease in the intrin-
sic firing frequency of place cells, reducing toward the theta
frequency in more ventral regions (Maurer et al., 2005).
Intriguingly, the full, repeating, interference pattern is
expressed in the recently discovered entorhinal grid cells
(McNaughton et al., 2006) as predicted by an inference model
(O’Keefe and Burgess, 2005).
Á Zipser’s (1986) “view field” model was built on the observa-
14.4 Hippocampus and Spatial Navigation
In this section we consider the contribution of hippocampal
spatial representations to guiding behavior. I focus on large-
scale spatial navigation, the spatial behavior most commonly
associated with the hippocampus and medial temporal lobes
729
(see Chapter 13). This is comparable to behavior in smaller
scale spaces and over shorter durations, such as visually guided
reaching, which are most commonly associated with the pos-
terior parietal lobe (e.g., Burgess et al., 1999). As with models
of spatial representation, these models can be approximately
divided into those stressing the role of feedforward connec-
tions and those stressing the role of recurrent connections.
14.4.1 Spatial Navigation: Data
Behavioral data indicate that rats learn about the spatial lay-
out of their environment during exploration in the absence of
explicit goals or rewards (e.g., Tolman, 1948). They can also
profit from being placed at the goal location without having
explored the rest of the environment (Keith and McVety,
1988). These processes are referred to as “latent learning.” Rats
also appear to be able to perform short cuts and detours.
These abilities contributed to the idea that rats form a cogni-
tive map of their environment rather than simply learning to
associate individual stimuli with responses (Tolman, 1948;
O’Keefe and Nadel, 1978). Of course, the learning of stimulus-
response associations also plays an important role in spatial
navigation, such as when the goal is directly visible or when a
well learned turn or sequence of turns is to be performed.
However, there seems to be good evidence that these types of
behavior are less dependent on the hippocampus than those
associated with cognitive mapping (O’Keefe and Nadel, 1978;
Morris et al., 1982; Packard and McGaugh, 1996). These issues
are dealt with in more detail in Chapter 13.
The other main spur to the association of the hippocampus
with a cognitive map of the rat’s environment was the discov-
ery of place cells whose response is not easily described in
terms of a simple response to a single stimulus. However, a gap
remains between the properties of place cells and the proper-
ties required of a system for spatial navigation. Two features of
place cell firing are particularly problematic. First, information
about a place in an environment (i.e., firing of the correspon-
ding place cells) can only be accessed locally (by actually visit-
ing that place). Second, place fields appear to be no more
affected by the location of the goal than by the location of any
other cue. That is, place cells tell you only where you are cur-
rently and not where to go to get to your goal (Speakman and
O’Keefe, 1990). A caveat to the first point may be indicated by
the recording of “spatial view cells” in the macaque hippocam-
pus (Rolls et al., 1997), which fire as a function of where the
monkey is looking rather than where it is physically located.
14.4.2 Spatial Navigation: Feedforward Models
tion that in some circumstances place cell firing is modu-
lated by the orientation of the rat. In this model, a set of
orientation-dependent place cells, or “view-field units,”
become associated with a set of “goal units,” which encode
the direction to the goal relative to the current heading direc-
tion. So long as the appropriate cells become associated, the
730 The Hippocampus Book
population vector of directions represented by the goal units
guide the rat to the goal, as goal units driven by place cells rep-
resenting the current location fire the most strongly. However,
it is unclear how the directions to the goal would become
associated with the place units if the goal were not visible. In
a second model (the “beta coefficient” model), Zipser sug-
gested that the location of the goal relative to subsets of land-
marks is calculated and stored (as the coefficient of the linear
sum of landmark locations that is equal to the goal location).
In this model, learning at the goal location is sufficient,
although the neural mechanisms required to implement the
desired calculations are not explained. One possible route to
enabling place cells to perform this type of computation was
provided by the phasor model of O’Keefe (1991) (see above).
A simplistic version of this type of model was simulated by
Wilkie and Palfrey (1987), in which the distance of the goal
from two landmarks is stored so that navigation back to the
goal can be effected by moving so as to match these landmark
distances.
Several models have followed Zipser’s view field model in
associating places to movements or local views. See Trullier et
al. (1997) for a wider review of biologically based artificial
navigation systems. McNaughton and Nadel (1990) suggested
that routes might be learned as a chain of associations from a
local view to an action and thence to the next local view, and
so on. This model was not actually simulated, and simply stor-
ing routes is insufficient to enable spatial navigation (see
Section 14.6). Even if a given route can be correctly selected in
terms of the locations to which it leads, navigational abilities
such as generating novel shortcuts and detours are beyond a
simple route-based system. The task of accumulating route-
independent spatial information faces several problems,
including the “credit assignment” problem: deciding which
actions along a route are critical for determining whether it
eventually leads to the goal.
Brown and Sharp (1995) provided a more sophisticated
model for associating locations with actions. In their model
the possible actions in a place are represented by a left-turn
cell and a right-turn cell (in nucleus accumbens) driven by
each place cell. These “turn cells” receive modifiable connec-
tions from head-direction cells (HDCs), which support the
rat’s spatial learning. When the rat reaches the goal, connec-
tions between HDCs and turn cells are modified according to
a recency-weighted index of their simultaneous activity. Thus,
if turning left in a particular place when facing north leads
immediately to the goal, the HDC representing north becomes
more strongly associated with the left-turn cell driven by the
corresponding place cell. The recency weighting of connection
modification is designed to provide an approximate solution
to the credit assignment problem (when applied over many
trials) by effectively dividing credit according to the number
of steps within which an action leads to finding the goal, see
reinforcement learning (e.g., Dayan and Abbott, 2002) for a
more principled approach. This model successfully simulates
learning in the Morris watermaze but would not show latent
learning; performance would not be affected by whether the
rat can look around from the goal location; and navigation to
the goal would be strongly affected if stereotyped routes were
used during learning.
A related way to think about spatial navigation is to imag-
ine defining a surface over the environment such that gradient
ascent on it leads to the goal. The simplest model of this sort
is for the place cells to be connected to a goal cell via reward-
modulated Hebb-modifiable connections such that encoun-
tering a goal causes strengthening of connections to it from
the concurrently active place cells (Burgess and O’Keefe,
1996). The subsequent activity of the goal cell then increases
with the proximity of the goal, as the net activity of those
place cells with strengthened connections increases with the
proximity to the goal (Fig. 14–7). The task for the rat in find-
ing the goal is then to move in the direction that increases the
rate of firing of the goal cell representing the desired goal (Fig.
14–7B). This type of model qualitatively captures the rapid
nature of learning a goal location once place cell firing has
become established and the ability to learn simply by being at
the goal location rather than having to find it many times.
However, finding the goal would involve the rat hunting
around to determine the best direction in which to move. This
behavior, known as vicarious trial and error, can be observed
at choice points but is not common. A second problem raised
by this model is the range over which spatial information
is accessible. If there are no place cells that fire at both the
goal location and the current location of the rat, there is
no gradient in the firing rate of the goal cell (being locally
zero). Hence, this type of model requires the population of
place cells to include some cells that have nonzero firing rates
at any two points in the environment, however far apart.
Consideration of the population of place field shapes and sizes
in the environments used so far (i.e., not more than about 2 m
in diameter) indicate that this is quite probable (Hartley et al.,
2000). However, whether this is true of larger environments is
not known.
A more sophisticated model of navigation (Burgess et al.,
1994) was proposed to make two improvements on the simple
model above. The first was to be able to calculate the direction
to the goal from any subsequent location after a single visit to
the goal location without having to hunt around. The model
made use of the fact that rats placed at the goal location like to
rear up and look around and the supposition that place cells
firing at a late phase relative to the theta rhythm have fields
peaked ahead of the rat (see Section 14.3.4). The model posits
a set of goal cells for each goal, such that each cell in a set is
associated with a different head direction. The connections
from place cells to a “north” goal cell are modified when the
rat is at the goal and facing north and similarly for the goal
cells associated with other directions. Crucially, connections
are modified at “late” phases of the theta cycle so the active
place cells (i.e., those from which connections are strength-
ened) are cells with fields ahead of the rat. Therefore, after
spending at least one theta cycle at the goal facing in each
direction, a north goal cell has strong connections from place
cells with fields peaked to the north of the goal—and similarly
for south, east, and west goal cells. As a consequence, a north
goal cell’s firing rate forms a surface over the environment that
 Computational Models of Spatial and Mnemonic Functions
Figure 14–7. Simple model of place cells and navigation. A. A
“goal” cell stores a goal’s location by taking a snapshot of place cell
activity via Hebbian synaptic modification when the goal cell is
excited by the attributes of a particular goal location. Solid circles
are active place cells; open circles are inactive place cells; and solid
squares mark potentiated synapses between place cell axons and
goal cell dendrites. B. Firing rate map of the goal cell (roughly an
inverted cone) during subsequent movements of the rat codes for
is peaked to the north of the goal location—and similarly for
south, east, and west goal cells (Fig. 14–7C). This is useful for
subsequent navigation, as the rat can now use the relative fir-
ing rates of the various goal cells to indicate the direction of
the goal. If the rat is to the north of the goal, the north goal
cell has a higher firing rate than the south goal cell. More pre-
cisely, the population vector (Georgopoulos et al., 1986) of a
set of goal cells successfully indicated the direction of the rat
from the goal in simulations, and different sets of goal cells
can be used simultaneously to indicate different locations of
interest. Conversion to egocentric (e.g., left or right) move-
ment directions was envisaged to take place in the parietal
cortex or basal ganglia, given knowledge of the current head
direction (Burgess et al., 2001a).
The second improvement was to ameliorate the problems
of the range over which spatial information is available from
place cell firing. The proposed solution was to interpose a set
of subicular cells between the place cells and goal cells. Given
weaker inhibitory competition between subicular cells than
between place cells, competitive learning in the connections
from place to subicular cells during exploration causes the
subicular cells to build up larger firing fields, each effectively
composed of several place fields. This learning is goal-
independent and corresponds to latent learning in preparing
the ground for effective one-shot learning of the location of
any goals should they be encountered. Large spatial firing
fields in subicular cells are consistent with experimental data
731
the proximity of the goal (G). C. Firing rate maps of four goal cells
whose population vector codes for the goal location (G). Each is
associated with the allocentric direction ui in which the location of
its peak firing rate is displaced from G; thus, the vector sum of the
directions ui weighted by the instantaneous firing rates fi of the goal
cells (i.e., Σi fiui/Σi fi) codes for the direction of the rat from the goal,
and the net firing rate of the goal cells (i.e., Σi fi) codes for the goal’s
proximity. (Source: Adapted from Burgess and O’Keefe, 1996.)
(Sharp and Green, 1994), and larger place fields have since
been found in the ventral hippocampus (Jung et al., 1994) that
might also serve this purpose.
To model hippocampal navigation Foster et al. (2000; see
also Dayan, 1991) used “temporal difference” learning in
which the “state” corresponds to the place cell representation
of the rat’s current location, and its value reflects the expected
number of steps to reach the goal (one unit of reward is
received on reaching the goal). Temporal difference learning
(Sutton and Barto, 1988) can be implemented by simply con-
necting place cells to “actor” and “critic” units with connec-
tions that are adjusted according to a modified Hebbian rule.
Activation of the critic unit is the evaluation of the current
state (the expected future reward from that state, discounted
by distance into the future—e.g., Dayan and Abbott, 2002), a
more principled analogue of the simple goal cell above. The set
of action units, only one of which can be active at a given time,
represent movements North, South, and so on. At each step the
connection weight from a place cell to the critic unit or to the
active actor unit is adjusted by an amount proportional to the
place cell’s activity times the amount by which the reward
exceeds that expected from the change in the activity of the
critic unit (Fig. 14–8). This type of learning is consistent with
a role for dopaminergic modulation of LTP (e.g., Montague et
al., 1996; Schultz et al., 1997). Over many routes to the goal,
ideally involving performing all actions at all locations many
times, this rule causes (1) the critic to provide an accurate esti-
732 The Hippocampus Book
Figure 14–8. Learning in the actor-critic system in a watermaze.
The plots for each trial show the critic’s value function C(p) (above),
the preferred actions at various locations (below left; the length of
each arrow is related to the probability that the particular action
shown is taken), and a sample path (below right). Trial 2: After a
timed-out first trial, the critic’s value function remains zero every-
where, the actions point randomly in different directions, and a long
mation of value and (2) the appropriate actions to be associ-
ated with each state (Fig. 14–8). For a given configuration of
goal and environment, this can provide the optimal strategy,
which is not the case with the approximate recency weighting
(see Brown and Sharp, 1995, above; Blum and Abbott, 1996,
below) or, if obstacles are present, with goal cells simply indi-
cating the physical proximity of the goal (Burgess et al., 1994).
However, learning is both experience-dependent and goal-
dependent (e.g., to know to head north from a given place
results from this combination having previously led to the
goal) and so perhaps does not capture the characteristics of the
hippocampal contribution to spatial learning.
To simulate goal independent learning over many trials in
which the location of the goal changes, Foster et al. (2000)
proposed that a second system learns to form a coordinate
representation of the rat’s position by using the rat’s locally
accurate ability to estimate self-motion. The place cells are
connected to two units that learn to estimate the x and y coor-
dinates of the rat, again using temporal difference learning to
adjust connection weights. In this system, the change in a con-
nection weight to the x unit is proportional to the place cell
activation times the amount by which the change in x, as esti-
mated by self-motion, exceeds that estimated by the change in
the activity of the x unit. The explicit representation of x and
y coordinates enables accurate navigation after one exposure
to the goal and so corresponds well with latent learning.
However, it is not clear if such a representation actually exists
in the brain, or if it would necessarily be more useful than a
place field-like representation that covered the appropriate
length scales (e.g., Burgess et al., 1994).
and tortuous path is taken to the platform. Trial 7: The critic’s value
function having peaked in the northeast quadrant of the pool, the
preferred actions are correct for locations close to the platform but
not for locations further away. Trial 22: The critic’s value function
has spread across the whole pool and the preferred actions are close
to correct in most locations, so the actor takes a direct route to the
platform. (Source: Foster et al., 2000.)
14.4.3 Spatial Navigation: Feedback Models
As well as being linked with spatial representation and asso-
ciative memory (see Box 14–2), the CA3 recurrent collaterals
have also been proposed to play a role in spatial navigation.
The first model to formalize such a role was suggested by
Muller et al. (1991, 1996) and focused on the Hebb-associative
effects of LTP on the recurrent collaterals. If pre- and postsy-
naptic firing within a short time interval leads to a small
increase in synaptic strength, the firing of place cells as the rat
moves around an environment leads to the strength of a con-
nection between two place cells depending on the proximity
of their place fields. This occurs simply because the greater the
overlap between place fields the more often they fire near-
coincidentally during random exploration. Muller et al.
(1996) showed that after extensive exploration the synaptic
strengths represent a “cognitive graph,” each approximately
representing the minimum path length between the centers of
the place fields of the cells it connects. Their model proposes
that the rat navigates by moving through the place fields of the
cells most strongly connected to the cells with fields at the cur-
rent and destination positions. This mechanism, reminiscent
of a resistive grid (Connelly et al., 1990), works well but relies
on a graph search. It is not easy to imagine how such a process
could be implemented in a biologically plausible fashion,
given the apparent lack of influence of the goal location on the
firing of place cells (Speakman and O’Keefe, 1990). One sug-
gestion (Gorchetchnikov and Hasselmo, 2002) is that activa-
tion corresponding to the goal location occurs in entorhinal
cortex while activation corresponding to the current location
 Computational Models of Spatial and Mnemonic Functions
occurs in CA3. Activation in each region spreads along the
available paths (although more slowly in CA3 than entorhinal
cortex) until a commonly activated location is detected in
CA1. This location represents the next immediate destination
for the rat, although how this information is interpreted in
terms of whether to turn left or right is not described.
In a related model, Blum and Abbott (1996) make use of
the temporal asymmetry of LTP to strengthen recurrent con-
nections from CA3 place cells that fire early on the rat’s path
to those that fire later along it. This causes the activation of
place cells to spread backward along the path (see also Mehta
et al., 1997, 2000 and models of spatial representation, above).
If the firing of place cells is interpreted by systems down-
stream of CA3 as representing the location of the rat, this shift
in firing (backward along the path) would be interpreted as a
shift in the location of the rat forward along the path. Blum
and Abbott suggested that navigation along a previously per-
formed route could be undertaken by moving from the cur-
rent location (e.g., read from CA1) to the shifted location
represented in CA3, although how this could be implemented
was not described. To enable navigation to a goal location, the
rules for synaptic change were modified to be proportional to
the amount of pre- and postsynaptic activation weighted by
how recently it occurred prior encountering the goal (i.e.,
modification similar to that suggested by Brown and Sharp,
1995, above).
It is interesting to note that the symmetrical pattern of
connection strengths learned in Muller et al.’s model resem-
bles that used in continuous attractor models of place cell fir-
ing and so also serves to produce a consistent pattern of
activity to represent each location (see above). By contrast, the
additional asymmetry in connection strengths learned in
Blum and Abbott’s model serve to shift the represented loca-
tion along the learned route. Indeed, the latter property has
been shown to allow the represented location to move
smoothly along the route over time, suggested as a model for
mental replay during sleep (Redish and Touretzky, 1998). The
goal-independent encoding of spatial proximity (as in the
symmetrical connections of Muller et al.’s cognitive graph) or
of previously traveled routes (as in the asymmetrical connec-
tion of the later models) correspond to latent learning.
However, the way these models incorporate a new goal loca-
tion necessarily requires many trips to the goal and thus prob-
ably falls short of a rat’s abilities. Another drawback associated
with these models is that none makes clear the details of how
the rat’s brain might deduce the direction it should move in or
if it would be able to generate a short-cut or detour. They also
assume place fields of fixed relative location but might still
make some interesting predictions regarding the locus of
search in environments that had changed in shape or size. To
build up a true distance metric in complex environments
would take a long time, in common with the reinforcement
learning approaches (see Foster et al., 2000, above).
These models can be viewed in the context of a more gen-
eral set of higher-level algorithms for navigation based on
directed graphs. Lieblich and Arbib (1982) described a “world
733
graph” in which locations are represented as nodes connected
by (asymmetrical) edges that represent the movement neces-
sary to get from one node to the next. In terms of the naviga-
tion of autonomous agents, Scholkopf and Mallot (1995)
described a “view graph” in which the local view or sensory
perception at given locations form the nodes and the actions
required to get from one view to the next form the edges. See
also McNaughton and Nadel (1990) for a description of how
a view graph might be implemented in the hippocampus.
Mallot and Gillner (2000) argued that such view graphs are a
good model for human navigation despite their simplicity.
One key requirement for building a world graph is to be able
to decide whether to assign a new node to a location. This can
be done on the basis of its familiarity (see Touretzky and
Redish, 1996; Kali and Dayan, 2000, for models relating to
this) or possibly on the basis of the sequence of actions that
lead back to a location. Lieblich and Arbib further suggested
that the nodes of a world graph might also represent a loca-
tion’s motivational valence and thus become a general model
for goal-directed behavior, even though its creation might
correspond to latent learning.
Á
14.5 Hippocampus and Associative
or Episodic Memory
In contrast to the vast amount of animal work linking hip-
pocampal damage to deficits in spatial memory, the major
impairment noted in humans following bilateral damage to
the hippocampus is amnesia: a much more general impair-
ment in memory. The extent of this impairment into various
subdivisions of memory and into information acquired prior
to the damage is a contentious issue (see Chapter 12). Here I
briefly review the data on human hippocampal function in
memory, introduce the generic model for it derived from
Marr’s seminal paper in 1971, and discuss various details and
developments of this model over the years.
14.5.1 Hippocampus and Memory: Data
Substantial bilateral damage to the hippocampus and medial
temporal lobes almost invariably leads to amnesia, character-
ized as a drop in the memory component of the intelligence
quotient (MIQ) of at least 20 points relative to full-scale IQ.
Because only a relatively small number of cases of damage
restricted to the hippocampus have been studied, it is difficult
to draw general conclusions regarding its role in memory, in
contrast to the roles of the surrounding cortical areas.
However, some general points can be made (see Spiers et al.,
2001 and Chapter 12 for details). They include a ubiquitous
deficit in memory for personally experienced events that
occur after the lesion (i.e., an “anterograde” deficit in
“episodic” memory) and spared procedural and working
memory. The extent of retrograde amnesia (loss of memory
for information acquired prior to the lesion) appears to vary
734 The Hippocampus Book
across patients and possibly across types of information
(Nadel and Moscovitch, 1997). Memory loss can extend over
the entire lifetime or can be restricted to shorter periods prior
to the damage; but it does seem to be relatively limited in the
case of lesions to the fornix. More controversial findings
include relative sparing of semantic memory (memory for
facts) and of familiarity-based recognition in some cases of
focal hippocampal damage. Relative sparing of recognition
memory is consistent with findings in monkeys showing that
this type of memory is perhaps more strongly dependent on
nearby cortical areas (e.g., Gaffan, 1994; Zola-Morgan et al.,
1994; Aggleton and Brown, 1999; Baxter and Murray, 2001).
Finally, it should be noted that the human hippocampus, par-
ticularly in the right hemisphere, is also involved in spatial
navigation (Burgess et al., 2002).
14.5.2 Marr’s Hippocampo-neocortical
Model of Long-term Memory
Most modeling work on the role of the hippocampus in mem-
ory can be considered part of a long tradition reaching back to
Marr (1971). In this section I sketch the main components of
Marr’s model, which provide a common framework for all of
the subsequent models, and indicate how these components
correspond to various aspects of the data on human memory
(see also Willshaw and Buckingham, 1990; Burgess et al.,
2001a). I refer to this as the generic hippocampo-neocortical
model (Fig. 14–9).
In Marr’s (1971) model, events in the outside world are
represented by patterns of activity in neocortical areas. The
role of the hippocampus is to store these representations over
the short term so relevant events can be categorized and
stored for the long term in neocortex (see Marr’s 1970 model
of the cerebral neocortex). This is achieved by mapping the
Figure 14–9. Generic hippocampo-neocortical model of long-term
memory. Strong connections and active cells are shown in black.
Relatively dense recurrent connections and sparse representations
in the hippocampus enable efficient pattern completion. Connec-
tions between neocortex and hippocampus allow the hippocampal
representation of an event to be associated with its sensory details,
including reactivation of the representations in various neocortical
areas dealing with different sensory modalities. Abstracted seman-
tic representations may also be learned over time in the neocortex.
The recurrent connections in each neocortical area allow unimodal
recognition. (Source: Burgess et al., 2001a.)
hippocampus
neocortex
neocortical representation of an event into a “simple repre-
sentation” in hippocampus, with modifiable connections to
and from the hippocampus storing the mappings between the
full representation and the simple representation. The CA3
recurrent collaterals are modified to store the simple repre-
sentation as an associative memory—so if a simple represen-
tation is incompletely activated, the “collateral effect” results
in the full representation being recovered. Thus, partial acti-
vation of the neocortical representation of an event can lead
to complete reactivation of its simple representation in the
hippocampus, which in turn can reactivate the entire neocor-
tical representation. In Marr’s view, the capacity of the hip-
pocampal system should be enough to store a day’s events so
the process of categorization and long-term storage in neo-
cortex can take place during the following night’s sleep. Note
that this now seems at odds with the much larger extent of ret-
rograde amnesia following hippocampal lesions, and the pos-
sibility that some types of information (e.g., episodic, in
contrast to semantic) remain forever dependent on the hip-
pocampus (Nadel and Moscovitch, 1997). Marr further sug-
gested that the simple representations need reflect only those
parts of the event through which they are addressed, and that
they should be sparsely encoded to reduce possible interfer-
ence between representations. The sparseness of a representa-
tion refers to the fraction of neurons that are active: If this
fraction is low, the chance of the same neuron being active in
the representation of different events is small, and interference
between representations is minimized.
It should be noted that the proposed capacity of the hip-
pocampus has varied widely in subsequent models, as memo-
ries remain hippocampus-dependent for more than 1 day and
possibly never become fully independent of it (see below).
The issue of “capacity” itself is also confounded by the need to
specify how to divide continuous experience into discrete
binary patterns. Another confusing issue is the correspon-
dence between Marr’s rather abstract generic model of mem-
ory and the processes of retrieval in human memory. For
example, it is not clear whether the model refers to recogni-
tion or recollection, where the incomplete retrieval cue comes
from, or whether the model applies more to memory for some
kinds of information than others. With regard to retrieval
cues, the prefrontal cortex probably plays an important role in
the strategic organization of retrieval, taking this issue beyond
the scope of this chapter (e.g., Roberts et al., 1998). By the end
of Section 14.5 some of the other issues are resolved, and oth-
ers are discussed further in Section 14.6.
14.5.3 Associative Memory
and the Hippocampus
Much of the development and analysis of the generic model
has followed from basic research on the associative properties
of feedforward networks (e.g., Willshaw et al., 1969) and
recurrent networks (Kohonen, 1972; Gardner-Medwin, 1976;
Hopfield, 1982) based on Hebbian learning (Fig. 14–10).
Much of the initial development of the model concentrated on
matching the major anatomical properties of the hippocam-
 Computational Models of Spatial and Mnemonic Functions 735

Heteroassociation Autoassociation
0 1 0
0 0 0
1 1 0 B
0 0 1
1 1 1
1 0 1
x3 x2 x1 1 0 1 0 1
1 0 0 0 0
0 1 0 1 1
1 0 1 0 1
A 0
0
1
1
1
0
1
1
0
0
y1 y2 y3 x2 x1
axon
dendrite unpotentiated synapse
input neuron potentiated synapse
output neuron detonator synapse
inhibitory interneuron inhibitory synapse

y inputs

C y inputs
1 0 0 1 1 0 y3 D 1
1
1
0
0 0 0
0 1 1
1
0
y4
y3 E 0 0 1 0 1 1 x2
1 0 1 1 0 0 x1
0 0 1 0 1 1 y2 0 0 1 0 1 1 y2
1 1 0 1 0 0 y1 1 1 0 1 0 0 y1 0 1 1 0 1 1 0 0
0 0 0 0 0 0 0 0

x inputs
0 1 0 0 0 1 0 1 1 0 0 1 0 0 0 1 0 1 1 1 1 1 0 1 1 1 1
0 0 0 0 0 0 0 0 0 1 0 0 0 1 1 0 0 0 1 0 1 1 0 1 1 0 0
x inputs
x inputs

1 1 0 1 0 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 0 0 0 1 0 1 1
0 0 1 1 1 0 1 0 0 1 0 0 1 1 1 0 1 0 1 1 0 0 0 1 0 1 1
1 1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 1 1 x2 x1
1 0 1 1 1 0 1 1 0 0 1 0 1 1 1 0 1 1 0
x3 x2 x1 x4 x3 x2 x1 PATTERN COMPLETION
(0 0 1 0 0 1) is part of x2
CORRECT RECALL SATURATION (0 0 1 0 0 1) · C = (1 0 2 1 2 2)
x3 = (0 0 1 0 1 1) x4 = (0 1 1 1 0 0) (1 0 2 1 2 2)/ 2 = (0 0 1 0 1 1) = x2
x3 · C = (3 2 2 3 3 2) x4 · C = (3 3 1 2 1 3)
(3 2 2 3 3 2)/ 3 = (1 0 0 1 1 0) = y3 (3 3 1 2 1 3)/ 3 = (1 1 0 0 0 1) = y4 ERROR CORRECTION
BUT (0 0 0 1 1 1) is a corrupted x2
PATTERN COMPLETION x3 · C = (3 3 2 3 3 2) (0 0 0 1 1 1) · C = (1 0 3 1 2 2)
(0 0 1 0 0 1) is part of x3 (3 3 2 3 3 2)/ 3 = (1 1 0 1 1 0) =/ y3 (1 0 3 1 2 2)/ 3 = (0 0 1 0 0 0) =/ x2
(0 0 1 0 0 1) · C = (2 1 1 2 2 1) BUT
(2 1 1 2 2 1)/ 2 = (1 0 0 1 1 0) = y3 (1 0 3 1 2 2)/ 2 = (0 0 1 0 1 1) = x2
Figure 14–10. Biological implementation of associative memory in binary vectors (e.g., x1, y1) are presented to the system for storage.
the hippocampus. A. Heteroassociative network associates pattern A Hebbian learning rule is used: A synaptic connection is strength-
of activity y1 with input x1, and y2 with x2, and so on to form an ened (set to 1 from 0) given pre- and postsynaptic activity (i.e., both
associative memory. (The matrix of connection weights shown in A inputs set to 1). Because the net input to a cell is the sum of inputs
is the result of successive presentations of input pattern x1 and out- multiplied by the strength of the connection mediating it, the net
put pattern y1, x2 and y2, x3 and y3 to an initially blank matrix.) input to a cell corresponds to the number of active inputs arriving
This enables input x1 to reproduce activation y1. Even an incom- via strengthened connections. To fire, the net input to a cell must
plete or corrupted version of x1 can reproduce y1 (see C). Storing equal the number of currently active inputs. Thus, retrieval of a vec-
too much information leads to “saturation” of the system and tor given its corresponding paired associate is achieved by matrix-
retrieval failures (see D). B. Autoassociative network associates pat- vector multiplication (e.g., pattern y3 in A is extracted by multiply-
tern x1 with itself, x2 with itself, and so on to form an autoassocia- ing the matrix rows by corresponding elements of vector x3 and
tive memory. This enables an incomplete or corrupted pattern to be summing the columns) followed by integer division by the number
cleaned up (see E). The essential functional components of these of active bits in the input vector. Provided not too many patterns
networks are a set of powerful “detonator” synapses that can impose have been stored, any unique subset of an x vector can recall the cor-
the pattern of activity to be stored, a set of extensively connected rect y vector. The autoassociative cases (B and E) work as the
inputs with modifiable synapses, and a set of inhibitory interneu- heteroassociative cases (A, C, D) with output y1  input x1,
rons whose role is to set a divisive threshold for the activity of the y2  x2, and so on. (Source: Adapted from McNaughton and
cells. C–E. Details of associative memory using the correlation Nadel, 1990.)
matrix formalism (Willshaw et al., 1969; Kohonen, 1972). Pairs of
736 The Hippocampus Book

pus (see Chapter 3), with constraints on the representations

(a) Hippocampal anatomy
and learning mechanisms indicated by functional analysis
of associative memory (see Figs. 14–10 to 14–12). The first CA1
major attempt of this sort (McNaughton and Morris, 1987) 4
CA3
highlighted the potential contribution of the dentate gyrus 3
(DG). 5 2
First, the much larger number of projection cells in the DG 1 Dentate gyrus
(around one million granule cells in the rat) than in either the
entorhinal cortex (EC) (around 200,000 layer II cells project
into the hippocampus) or region CA3 (around 300,000
pyramidal cells) indicate that it could be used for “pattern sep- 6
aration” (Amaral et al., 1990). This means that distinct (i.e., V III II
Entorhinal
nonoverlapping) patterns of activation are created in the DG cortex
despite similarity in patterns of EC activation representing
similar events. This process is also referred to as orthogonal- (b) Computational model
ization. Thus pattern completion in CA3, so important for Medial septum
retrieving the representation of a familiar event, does not lead Regulation of
to a similar novel event also causing retrieval of the represen- learning dynamics
tation of the familiar event. Of course, one cannot have per- 7 ACh 8 3
fect pattern completion of the representation of an event from Region CA3
Region CA1 4
partial cues and perfect pattern separation of it from the rep- Hetero- Autoassociative
Comparison associative recall
resentations of similar events in the same system. One mech- recall
2
anism must fail once the partial cue is less similar to the old
Self-organization5 6 Dentate gyrus
event than the new event. Self-organization
Second, the specific nature of the various synaptic inputs Hippocampus
1
to CA3 pyramidal cells suggests different functions for them.
Entorhinal cortex
The input from DG comes from a small number (around 46) Afferent input
of very large synapses around the soma. A much large number Current Opinion in Neurobiology
of connections are received farther up the dendrites from
within CA3 (up to 12,000), and up to 3750 connections are Figure 14–12. Hippocampal anatomy (a) and a computational
received from the EC onto the distal apical dendrites. It was model of hippocampal episodic memory function (b). Numbers
suggested (McNaughton and Morris, 1987) that the powerful label synaptic connections mediating various functions in the model.
input from DG (via “detonator synapses”) serve to impose a 1: Synapses of the perforant path fibers projecting from entorhinal
new pattern of activity to be learned. A strong input is cortex layer II to the dentate gyrus undergo sequential self-organiza-
required to impose a new pattern of activity in CA3 in the face tion to form sparse, less overlapping representations of entorhinal
of the interference due to feedback via the recurrent connec- activity patterns. 2: Mossy fibers projecting from dentate gyrus to
tions, which tend to cause the system to return to a previously region CA3 transfer dentate gyrus activity to CA3. 3: Excitatory
stored pattern of activity. Once the representation of a new recurrent connections in region CA3 mediate autoassociative encod-
ing and retrieval of the features of episodic memories. 4: Schaffer
event has been imposed, the Hebbian modification of both
collaterals from region CA3 to CA1 encode and retrieve associations
the recurrent connections and the connections from EC can
between CA3 activity and activity patterns induced by entorhinal
input to region CA1. 5: Perforant path input to region CA1 under-
Figure 14–11. Anatomy of the inputs to CA3 pyramidal cells, goes self-organization, forming new representations of entorhinal
showing the approximate number of synapses onto each cell in cortex input for comparison with recall from CA3. 6: Projections
the rat. (Adapted from Treves and Rolls, 1992.) from region CA1 to deep layers of the entorhinal cortex store associ-
ations between CA1 activity and entorhinal cortex activity, allowing
CA1 representations in CA1 to activate the associated patterns in entorhi-
nal cortex. 7: Output from region CA1 to the medial septum regu-
Perforant path lates cholinergic modulation. 8: Cholinergic modulation from the
<
~ 3750 synapses medial septum sets appropriate dynamics for encoding in hip-
Recurrent collaterals pocampus. (Source: Hasselmo and McClelland, 1999.)
< 12000 synapses
~
Mossy fibres occur. The large number of recurrent synapses per cell allow
~ 46 synapses
large autoassociative memory capacity, and the large number
of synapses in the input from EC allow large heteroassociative
memory capacity in associating the EC representation to the
CA3 CA3 representation (Treves and Rolls, 1992) (Fig. 14–11).
 Computational Models of Spatial and Mnemonic Functions
Third, the requirements of Hebbian learning in the CA3
recurrent connections and the connections from EC to CA3,
but not in the connections from the DG, are consistent with
the physiology of these various connections. The synapses in
the former two pathways are thought to be capable of NMDA
receptor-dependent LTP, whereas the mossy fiber connections
from DG show only non-Hebbian modification (see Chapter
10). Equally, the divisive normalization required by associative
nets (Willshaw et al., 1969) (see Box 14–2) is consistent with
the action of interneurons providing inhibition by opening
ion channels near the soma-to shunt input current in the den-
drites (Fig. 14–10). Further analysis of autoassociative mem-
ory indicates that “progessive recall” improves performance
(Gardner-Medwin, 1976). With this model, inhibition is
slowly reduced during retrieval so the first few cells that
become active are the most likely to be correct, and feedback
from their activation decreases the chances of subsequent
erroneous activation. Such periodic fluctuation of inhibition
(or, equivalently, the cells’ firing threshold) may provide a
functional interpretation for the theta rhythm.
In Marr’s model an incomplete or incorrect neocortical
representation feeds into the hippocampus, where the correct
hippocampal representation is retrieved and feeds back to
complete or correct the neocortical representation. This sim-
ple picture has been elaborated to include separate input
and output representations in the entorhinal cortex (in the
superficial layers and deep layers, respectively). The processes
of storage and retrieval obviously do not start and end at the
entorhinal cortex. The processes of feedforward cueing of
a representation, its pattern completion and feedback can
be modeled as occurring sequentially in the cortical areas
between sensory cortex and entorhinal cortex (Rolls, 1996).
The principal difference between the processes in the hip-
pocampus and EC and those in lower cortical areas are
the additional pattern separation provided by the DG and the
greater autoassociative power of the much longer-range recur-
rent collaterals in CA3 compared to cortex. The long-range
recurrent collaterals in CA3 also provide the best opportunity
to associate information from different sensory modalities.
Many of the above ideas are reviewed or developed further
in the literature (e.g., Amit, 1989; McClelland et al., 1995;
Rolls and Treves, 1997; Hasselmo and McClelland, 1999;
Redish, 1999).
14.5.4 Hippocampal Representation,
Context, and Novelty
The above considerations of sparseness and pattern separa-
tion regarding the hippocampal representation of an event
raise the issue of how these representations relate to the vari-
ous elements of its content and context. First note that the
conflicting processes of pattern separation and pattern com-
pletion serve to define the similarity space of retrieval (i.e.,
which dimensions a retrieval cue can vary along but still
retrieve the event and which dimensions serve to discriminate
events). In the limit of complete orthogonalization in the den-
737
tate gyrus, the hippocampal representations are independent
of the details of the events and can be thought of as simply an
index for them (Teyler and DiScenna, 1986). However, as the
hippocampal representation must initially be activated by the
neocortical representation, the overall memory system is still
“content addressable” and its behavior (e.g., pattern separa-
tion or pattern completion) depends on how different aspects
of an event and its context contribute to the activation of its
hippocampal representation. This relates to Marr’s observa-
tion that the hippocampal representation should include
those aspects of an event used for its retrieval.
In a simple associative memory, all elements of the repre-
sentation of the event are equally associated with all other ele-
ments. However, as implied by Marr, it seems plausible that
some aspects of an event are better able to cue associative
retrieval than others, and other aspects of an event can be
associatively retrieved more easily than others. Thus, the “sim-
ple representations” envisaged for the hippocampus would
reflect some aspects more than others. A related suggestion is
that the hippocampal representation reflects efficient com-
pression of the neocortical representations—extracting the
distinguishing features of each event (Gluck and Myers, 1996).
For example, the name of someone you met only once is often
a good cue to recalling the meeting but can be difficut to
retrieve, whereas the location of the meeting is often both a
good cue and relatively easy to retrieve. Similar to someone’s
name, the sequential position of an item in a list is easier to
use as a cue than it is to retrieve (Jones, 1976). Thus, even the
simplest associative model of memory should include asym-
metrical associations between the elements of an event.
One of the distinguishing features of episodic memory is
the ability to retrieve the ongoing context within which the
event occurs (Gardiner and Java, 1993; Tulving, 1993;
Knowlton and Squire, 1995), and one suggestion for the role
of the human hippocampus is that it provides the spatiotem-
poral context for episodic memory (O’Keefe and Nadel,
1978). Theoretical analyses of associative memory have also
made the distinction between the content of the event and its
context (e.g., Raaijmakers and Shiffrin, 1981) or between the
record of the event and the “header” or index term used to ref-
erence it (e.g., Morton et al., 1985). One possibility is that the
hippocampus serves to associate the content of the event with
its context (e.g., Wallenstein and Hasselmo, 1997) (see Section
14.6). A related possibility is that the role of the hippocampus
is to generate a representation of temporal context itself (e.g.,
slowly varying patterns of activity generated by its own recur-
rent dynamics) (Levy, 1996). This type of model coincides
with a considerable psychological literature of models of
memory in which retrieval of stored items is held to depend,
at least in part, on their association with a representation of
context that changes slowly with time or experience. In some
of these models the context representation changes inde-
pendently of the to-be-remembered items (Mensink and
Raaijmakers, 1988; Burgess and Hitch, 1992; Davelaar et al.,
2005), whereas in others the context representation is derived
from the items themselves (Howard and Kahana, 2001).
738 The Hippocampus Book
Retrieval corresponds to finding the item most strongly asso-
ciated with a re-presented context.
I briefly describe the operation of one of these models, the
temporal context model, as I return to it in the section on
models attempting to draw links between the medial tempo-
ral roles in spatial and episodic memory. As the with other
models of episodic memory described above, item representa-
tions are associated with context representations such that a
given item is retrieved according to the similarity between the
context at retrieval and that associated with the item.
However, in this model the context representation is derived
from the presented or retrieved items themselves, becoming a
recency-weighted sum of the context arising from each item.
After presentation of the list, the context vector is most simi-
lar to the most recent items, producing the well known
recency effect. In addition, when an item is presented, it affects
the context vector with which subsequent items are associ-
ated. Thus, recall of a given item changes the context vector so
that retrieval of immediately subsequent items in the list is
more likely (by making the context vector more similar to how
it was when those items were presented). This leads to an
asymmetry such that forward associations in the list are
stronger than backward associations, as is often found with
free recall.
Consideration of the different requirements of encoding
and retrieval also raises some interesting questions.
Notwithstanding McNaughton and Morris’s suggestion that
“detonator synapses” from the dentate gyrus can impose a
new pattern of activation on CA3 despite the retrieval-related
feedback from recurrent connections, there must be some
mechanism to determine whether the system should be in
encoding or retrieval mode (McNaughton and Morris, 1987).
One proposal is that the supply of acetylcholine (ACh) from
the medial septum switches the hippocampus between encod-
ing and retrieval modes (Hasselmo et al., 1995; Murre, 1996;
Wallenstein and Hasselmo, 1997). In this model, increased
ACh increases the rate of synaptic modification of the recur-
rent connections in CA3 and suppresses the synaptic trans-
mission of intrinsic activity (within CA3 and from CA3 to
CA1), enabling new patterns of activity to be stored. The level
of delivery of ACh is determined by the novelty of the neo-
cortical input, as represented by direct activation of CA1 from
EC, compared to the most similar previously stored event, as
represented by the input to CA1 from CA3 after it settles to a
stored state. Specifically, if both CA3 and entorhinal cortex
inputs to CA1 are matching (as with a familiar stimulus),
strong activation of CA1 activates interneurons in the medial
septum, which decrease the activity of the cholinergic cells
projecting to the hippocampus (Fig. 14–12).
Two types of evidence support this model. First, by empha-
sizing the connections between the hippocampus and the
medial septum, the model begins to address data showing the
importance of the fornix, the large fiber bundle connecting
the hippocampus with the medial septum and other subcorti-
cal structures. In the model, sectioning the fornix prevents the
learning of new memories due to lack of ACh, which corre-
sponds well to reviews of the effects of damage to the fornix in
neuropsychological patients (e.g., Spiers et al., 2001) (see
Chapter 12). Second, blocking the ACh receptors prior to
encoding (by injecting scopolamine) seems to impair recall
more strongly than recognition (Hasselmo and Wyble, 1997).
This is also the case in the model, assuming that CA3 serves to
associate events and their contexts because recall is more
reliant on these associations than recognition. However, dis-
ruption of the hippocampus might also impair recall more
than recognition in many other models (e.g., due to disrupt-
ing associations with context) (see Section 14.6). In addition,
more recent work has begun to include the role of dopamine
in novelty and encoding (Lisman and Grace, 2005).
14.5.5 Consolidation and Cross-modal
Binding of Events in Memory
Ever since the initial reports of dense anterograde amnesia but
weaker or temporally graded retrograde amnesia after bilat-
eral medial temporal lobectomy (Scoville and Milner, 1957)
(see Chapter 12 and the caveats in Section 14.5.2), researchers
have considered how the hippocampus contributes to the
long-term consolidation of memories. One mechanism for
consolidation, consistent with the preservation of informa-
tion acquired prior to hippocampal damage, is that the hip-
pocampus enables information to be stored elsewhere in the
brain after which it is no longer needed for retrieval. Marr’s
(1971) model follows this view, suggesting that the day’s
events are stored in the hippocampus and that this informa-
tion is used to allow long-term categorization and storage in
the neocortex. Note that processes of temporal consolidation
might also occur in the hippocampus (i.e., without transfer of
information from one region to another).
The experimental data regarding the gradient of retrograde
amnesia (i.e., the sparing of memories acquired sufficiently
long enough before the damage) is inconsistent and remains
controversial in animals (see Chapter 13) and humans (see
Chapter is 12) (Spiers et al., 2001). One problem in the human
data is that the amnesia often extends back to childhood,
implying that the hippocampus stores several decades worth
of information. Several computational arguments have also
been put forward for the transfer of information out of the
hippocampus for consolidation in the neocortex. Marr (1971)
suggested that a short-term buffer is needed to store informa-
tion online (i.e., the hippocampus), whereas only information
deemed relevant to the animal’s future needs to be incorpo-
rated into the animal’s long-term store of knowledge and
must be first appropriately categorized with respect to it. The
hippocampal store of unprocessed experience necessarily has
a limited capacity, and the process of abstracting relevant
information from this experience to expand a long-term data-
base requires extensive offline processing, perhaps during
sleep (see also McClelland et al., 1995).
A second argument concerns the anatomical convergence
of information from difference sensory modalities at the hip-
pocampus. Thus, in the absence of long-range connections
between different sensory cortical areas, associations between
the elements of an event, such as its sight, sound, and smell,
 Computational Models of Spatial and Mnemonic Functions
cannot be formed in the lower-level cortices. Damasio (1989)
suggested that “convergence zones” must exist where these
associations could be formed. Several models have extended
this idea to include rapid learning of a hippocampal, or
medial temporal lobe, representation reciprocally connected
to all the unimodal cortical representations of the event which
allows them to be associated with each other (Alvarez and
Squire, 1994; Murre, 1996; Moll and Miikkulainen, 1997).
These models also suggest that, after multiple rehearsals of a
memory driven by the hippocampus, long-range associations
can be learned directly between the unimodal representations,
finally making the stored information independent of the
hippocampus.
The arguments regarding data abstraction and anatomical
convergence are represented in a model by Kali and Dayan
(2004). In this model the hippocampus serves to aid the learn-
ing of a hierarchical semantic system by being able to reinstate
the top-level (i.e., entorhinal/ perirhinal/ parahippocampal)
representation of events during sleep. The semantic system
contains reciprocal connections between higher cortical areas
and those immediately below them in the hierarchy but no
direct connections between areas at the same level. Aided by
the hippocampus, the neocortical system learns to form an
associative representation of patterns of activation in lower
cortical areas. This is achieved by the higher level representa-
tions learning a “generative model” of representations lower in
the hierarchy (e.g., Hinton and Sejnowski, 1999). Thus
higher-level representations can be cued by input from a sub-
set of lower cortical areas and can then cause pattern comple-
tion in all lower areas via the reciprocal connections. During
further simulations, Kali and Dayan (2004) noted that regular
reactivation of episodic hippocampal representations is
required to maintain them in register with the slowly chang-
ing semantic representations.
A third argument for consolidation refers to the effects of
interference. In some learning systems—such as using a fixed
feedforward structure and the error back-propagation learn-
ing rule (Rumelhart et al., 1986) to form associations consis-
tent with a set of examples—new information can interfere
with previously stored information, sometimes “catastrophi-
cally” such that all information is lost. This problem can also
occur with associative memories (Fig. 14–10). McClelland et
al. (1995) proposed a solution to this in which long-term neo-
cortical learning involves random interleaved re-presentation
of all previous knowledge along with newly acquired knowl-
edge to produce a single integrated neocortical representation
of semantic knowledge. Note, however, that this mechanism
requires the temporary store to have capacity for the entire
data set. Other solutions include associative memories with
continuous but bounded connection weights (Hopfield,
1982) and “constructive” algorithms in which the addition of
new information is accompanied by the addition of new pro-
cessing units (Gallant, 1986; Mezard and Nadal, 1989;
Fahlman and Lebiere, 1990; Frean, 1990). The latter algo-
rithms may relate to the recent observation of neurogenesis in
the adult dentate gyrus that is related to learning (Shors et al.,
2001).
739
A reinterpretation of the experimental data regarding the
gradient of retrograde amnesia associated with medial tempo-
ral lobe damage led Nadel and Moscovitch (1997) to a differ-
ent conclusion regarding consolidation. Noticing that in many
instances retrograde amnesia extends back over a much longer
time than that envisaged by Marr, they proposed that the hip-
pocampus remains necessary for the retrieval of detailed
episodic or spatial information (but for evidence that early
spatial memories become hippocampus-independent see
Teng and Squire, 1999; Rosenbaum et al., 2000). They pro-
posed a new model to account for both the common occur-
rence of a temporal gradient in memory for other types of
information and the possibility of partial damage to the hip-
pocampus in some cases. In this model, whenever a memory
is rehearsed or reactivated, a new hippocampal representation
is formed, again connected to all of the neocortical represen-
tations. The result is that although a complete lesion of the
medial temporal lobe impairs retrieval of all memories, the
older the memory, the more robust it is to partial damage by
virtue of being represented in multiple locations. Some evi-
dence for the reactivation of specific memories comes from
the recently revived study of reconsolidation (Nader et al.,
2000). In these experiments re-presentation of the context of
an event (usually application of an electric shock) appears
to render the memory for the event labile in the sense
that protein synthesis is again required for long-term storage
of the memory, as is the case when the event is first experi-
enced.
14.5.6 Hippocampal Contributions to Various
Types of Memory and Retrieval
A distinction has been made between retrieval of episodic in-
formation, retrieval of semantic information, and familiarity-
based recognition (see Chapter 12). Episodic retrieval is
characterized by the ability to retrieve detailed information
about the event and its context. Semantic retrieval is charac-
terized by factual knowledge without retrieval of the indi-
vidual events and contexts in which it was acquired.
Familiarity-based recognition depends purely on an unattrib-
utable feeling of familiarity associated with a stimulus in the
absence of detailed information about the event and context in
which it was encountered. Of these three processes, the hip-
pocampus is claimed to be primarily involved in episodic
retrieval (e.g., Aggleton and Brown, 1999) (see Chapter 12).
The hippocampo-neocortical model suggests that semantic
knowledge is abstracted from combinations of hippocampal
memories of unique events, but eventually becomes inde-
pendent of the hippocampus. This is consistent with memory
for the unique content and context of a specific event (episodic
memory) depending on the hippocampus but semantic mem-
ory depending on other areas of the temporal lobe (e.g.,
Graham and Hodges, 1997). It is also consistent with the sug-
gestion of Nadel and Moscovich (1997) that semantic memo-
ries show a temporal gradient of retrograde amnesia, whereas
detailed episodic memories do not. The model is less obvi-
ously consistent with the observation that semantic informa-
740 The Hippocampus Book
tion can be acquired despite early bilateral hippocampal
pathology (Vargha-Khadem et al., 1997). However, the possi-
bility of partial sparing of the hippocampus (Squire and
Zola, 1998) or the use of external rehearsal of information
(Baddeley et al., 2001) might provide explanations for these
developmental cases within the framework of the model.
The distinction between episodic retrieval and familiarity-
based recognition also has interesting parallels in the
hippocampo-neocortical model. It has been argued that the
hippocampus is required to provide associations between rep-
resentations in disparate cortical areas, whereas associations
within each area can be formed locally. Thus, effects mediated
by the familiarity of single stimuli might be supported by the
association of elements within each of the neocortical areas,
independent of the hippocampus. By contrast, correct recogni-
tion of a pair of cross-modal associates among equally familiar
distractors would require the hippocampus. Evidence indicates
that simple recognition memory does not depend on the hip-
pocampus but on nearby neocortical areas (Zhu et al., 1996;
Baxendale et al., 1997; Vargha-Khadem et al., 1997; Murray and
Mishkin, 1998; Aggleton and Brown, 1999; Wan et al., 1999;
Holdstock et al., 2000; Baddeley et al., 2001: but see also Manns
and Squire, 1999; Zola et al., 2000). In contrast, there is some
evidence that recognition of cross-modal associations is
impaired by bilateral damage restricted to the hippocampus
(Vargha-Khadem et al., 1997; Holdstock et al., 2000). More
extensive unilateral damage may also impair the binding of ele-
ments within the same modality (Kroll et al., 1996). The logi-
cal extension of this idea is that episodic memory requires full
recollection of an event and its context in all of its multimodal
detail and so requires an intact hippocampus.
Much of the analysis of the differential role of the hip-
pocampus in episodic retrieval (or “recollection”), and
familiarity-based recognition has focused on the idea that
these two processes contribute independently to the perform-
ance of recognition memory tests (e.g., Yonelinas, 2002). In
forced choice and yes–no recognition paradigms, the recollec-
tive component is assumed to be “all or nothing” and “high-
threshold.” That is, the stimulus is recalled in great detail or
not at all, and a novel foil is never falsely recalled. By contrast,
the familiarity-based process is more like a signal detection
problem: The subject guesses whether the item is familiar,
informed by a noisy measure of familiarity (see Chapter 13).
The hippocampus has been associated with the recollective
component (Aggleton and Brown, 1999; Yonelinas et al., 2002)
and so should provide a high-threshold, all-or-nothing mech-
anism in recognition memory tests. One method for detecting
this component involves the receiver-operator characteristic
(ROC) curve: the plot of the proportion of previously seen
items that are correctly identified (“hits”) versus the propor-
tion of new items incorrectly identified (“false alarms”) as a
function of the subject’s confidence in their response.
Recollection is marked by high-confidence hits in the absence
of high-confidence false alarms, producing a positive intercept
in the ROC curve. See Chapter 13 and Rugg and Yonelinas
(2003) for a review.
The hippocampal contribution to recognition memory
(via “recollection”) can be modeled by using the stimulus-
driven medial temporal neocortical representation as the cue
to retrieval of a stored pattern of activation in CA3 and com-
paring the retrieved activation to the stimulus-driven activa-
tion in the entorhinal cortex (Norman and O’Reilly, 2003). By
explicitly retrieving an entire stored pattern, the process is all-
or-nothing, and even foils that resemble presented patterns
and are not falsely recognized because the retrieval product
still differs from the stimulus-driven activation. The use of
sparse hippocampal representations, orthogonalized via the
dentate gyrus, serves to prevent interference between different
stored events. By contrast, familiarity-based recognition is
modeled in terms of “sharpening” of the neocortical represen-
tation. With this model, although a new item is represented by
weak activation of a large number of neocortical neurons,
repeated presentation of the item results in strong activation
of a smaller number of neocortical neurons via a competitive
learning mechanism. Activation of the neocortical neurons
that fire in response to a given stimulus can then be used as a
measure of familiarity-based recognition. One advantage of
this scheme is that repeated presentations of some items in a
list does not impair recognition of the other items in the list,
as seen experimentally (Ratcliff, 1990), which is not true of
some other psychological models of recognition memory
(Norman, 2003). A characteristic of this model is that hip-
pocampal damage specifically impairs recognition memory
when related lures (novel items that resemble previously seen
items) are included in the test. The familiarity-based recogni-
tion mechanism would produce false alarms to these stimuli,
whereas the hippocampal recollection mechanism would not.
Some evidence consistent with this hypothesis has been found
(Holdstock et al., 2002).
Á
14.6 Reconciling the Hippocampal
Roles in Memory and Space
In this section I discuss some of the models that have
attempted to draw together the common strands present in the
plethora of hippocampal models reviewed above. One feature
common to most of the spatial and nonspatial models is use of
the autoassociative properties of the recurrent network in
CA3. Even feedforward models of spatial representation (see
Section 14.3.2) usually note that the various sets of cells used
to represent different environments might be stored in this
way (e.g., McNaughton and Nadel, 1990; Burgess et al., 1994;
Kali and Dayan, 2000).
Many other models concern the common functional
requirement of storing and retrieving a sequence of patterns of
activation. For example, Redish and Touretzky (1998) used
replay of a spatial route as a demonstration of episodic recall.
In Levy’s (1996) model, the slowly varying patterns of activity
(resulting from the presence of asymmetrical connections in
CA3) (see Section 14.3.3) might be used as a context repre-
 Computational Models of Spatial and Mnemonic Functions
sentation, associations to which could store sequences of
memories. In a rat moving through its environment, it is
argued that they might also resemble place fields, although it
is not clear that this would be the case for a trajectory that
crosses itself. Jensen and Lisman’s (1996, 2005) model of long-
term memory and working memory, involving association of
one item to the next and preservation of serial order during
each theta cycle, is also applied to model the storage of routes
as sequences of place fields.
However, spatial navigation is not just a subset of episodic
memory; simply storing routes and sequences of places would
not enable accurate navigation, detours, or shortcuts. These
require a metric or at least a representation of proximity (see
Section 14.4.2). Put another way, specific constraints apply to
spatial information that are not necessarily required of a more
general memory system. What then is the relation between the
constraints involved in spatial and episodic memory such that
a unique structure, the hippocampus, might support both?
O’Keefe and Nadel (1978) argued that the spatial function
seen in the rat hippocampus is augmented by a linear sense of
time in humans, which allows the human hippocampus to
supply the spatiotemporal context of events required by
episodic memory. This idea has been taken up by two types of
model: one investigating the provision of temporal context
and the other the provision of spatial context.
Following the models of Levy (1996) and Jensen and
Lisman (1996) described above, Wallenstein et al. (1998)
investigated the idea that the hippocampus could provide a
signal representing temporal context as a solution to the need
to form associations between temporarily discontiguous
events. Associating events that occur much farther apart in
time than the time scale required for pre- and postsynaptic
activity to induce LTP (i.e., around 100 ms) (see Chapter 10)
obviously requires some kind of bridging mechanism. One
indication that the hippocampus is involved in this process
comes from “trace conditioning,” in which the animal must
learn to respond at a fixed delay after the disappearance of a
cue, because the timing of the response is disrupted by hip-
pocampal lesions. This contrasts with “delay conditioning” in
which the cue does not disappear, for which there is no hip-
pocampal lesion effect (see Chapter 13). In Wallenstein et al.
(1998), as with Levy (1996), unsupervised learning in the hip-
pocampus in the presence of temporally correlated inputs cre-
ates patterns of activity that vary more slowly in time than the
original stimuli (Fig. 14–13). Associations between the various
states of this temporal context signal can allow the original
stimuli to be reproduced in order. Indeed, this type of mech-
anism provides a good model for short-term serial recall
(Burgess and Hitch, 2005). This model can be extended to
spatial navigation in terms of associations between spatially
discontiguous events, but note that, as with storing temporal
sequences to aid navigation, temporal contiguity alone is
insufficient to generate metric information.
Howard et al. (2005) attempted to draw parallels between
the roles of the hippocampal region in both spatial and
episodic memory by extending the “temporal context model”
741
(TCM) of free recall (Howard and Kahana, 2001), described
above, to model the noisy place-specific firing of some cells
in entorhinal cortex (Quirk et al., 1992; Frank et al., 2000;
Fyhn et al., 2004). To model free recall, the TCM forms a
context representation derived from a recency-weighted sum
of the contributions to context from the nature of the list
items themselves. Howard et al. proposed that, in terms of
spatial memory, these context contributions might be
encoded by neurons with firing rates reflecting running speed
in particular head directions. If each context neuron per-
formed a recency-weighted sum of one of these inputs, they
would be effectively supporting an approximate route mem-
ory, or “path integration.” In simulations, these cells show
noisy place-specific firing, rather like some cells in the
entorhinal cortex, as they fire most following a series of
movements along their preferred direction (e.g., a cell with a
west head direction input fires most along the west edge of the
environment). They also show “retrospective coding”: that
is, modulation of firing by the recent history of movements, as
found by Frank et al. (2000). In addition, cells whose firing
rate reflects the time integral of their input have been found
in entorhinal cortex (Egorov et al., 2002), and systematic
errors in path integration have been found that are consistent
with recency weighting (Etienne et al., 1996). However, the
model predicts that entorhinal cell firing is directionally selec-
tive and always peaks at the edge of the environment, which
may not be the case (Fyhn et al., 2004). This idea of entorhi-
nal processing may also have to be reviewed in the light of the
discovery of grid cells (Hafting et al., 2005) (see Section
14.3.3).
The idea of using spatial context to index memory was
combined with Marr’s hippocampo-neocortical model to
produce a simple navigation system (Recce and Harris, 1996).
In this model, perceptual representations of the environment
are stored in parietal cortex and retrieved via an index repre-
sentation in the hippocampus corresponding to the location
of the animal. This allows retrieval of the correct perceptual
map for a given location. Navigation additionally requires that
the location of an encountered goal be added to the perceptual
maps (the mechanism for this is not specified). A more
detailed model of long-term memory and retrieval of a spatial
environment (Becker and Burgess, 2001) also serves as a
model of memory for the spatial context of an event (Burgess
et al., 2001a) (Fig. 14–14). This model explicitly makes use of
the constraints associated with spatial information, and our
detailed knowledge of how it is represented in the brain. The
representations of spatial layout in long-term memory are
assumed to be allocentric (e.g., independent of the orientation
of the person) in contrast to the egocentric short-term repre-
sentations involved in perception, action, and working mem-
ory (Goodale and Milner, 1992; Burgess et al., 1999; Milner et
al., 1999). By extension from spatial models of the hippocam-
pus (Hartley et al., 2000), the geometry of the environment is
encoded in (parahippocampal) boundary vector cells bidirec-
tionally connected to (hippocampal) place cells. The place
cells are connected up to form a continuous attractor. This
742 The Hippocampus Book
Figure 14–13. Context field development. Each rectangle shows a
subset of 50 simulated pyramidal cells firing across time, each
action potential represented by a vertical line. A unique group of
five cells firing at the same time encodes a single item in a 13-item
sequence (see the top portion of each rectangle). This can be
thought of as afferent activation of CA3 pyramidal cells due to a
specific pattern of sensory events. A. Note the background firing
during the first learning trial that does not encode sequence items
directly. This stems from activity at recurrent excitatory synapses.
Repeated exposure to the same sequence can lead to enhanced
acts as a spatial memory system in that partial activation of
boundary vector cells (representing the occurrence of partic-
ular large objects at particular distances and bearings) causes
activation of the corresponding place cells, which then acti-
vate the boundary vector cells representing the remaining
landmarks in the environment. The location of a specific
event is represented by a goal cell, as in the simple model of
navigation (Burgess and O’Keefe, 1996) (see Section 14.4.2).
Bidirectional connections between place cells and these
“event” cells allows for both navigation to the location (via the
synaptic potentiation between cells firing in the background and
cells that encode sequence items owing to a simple Hebbian learn-
ing rule. B. After the fourth learning trial, this repeated potentiation
leads to a condition where background cells begin to respond to the
appearance of contiguous segments of the entire sequence. The por-
tion of the full sequence to which the cell responds is called the
“context field” of the cell. Because the context fields overlap, the
entire sequence can be reconstructed by interdigitating them in the
proper order. (Source: Wallenstein et al., 1998.)
feedforward connections) and retrieval of its spatial context
(via the feedback connections).
Retrieval of allocentric spatial information into a format
appropriate for visual imagery requires that it be translated
into an egocentric reference frame. Accordingly, in the model,
the parahippocampal boundary vector cell representation
(arranged in terms of distances and bearings) is translated
successively into trunk-centered then head-centered represen-
tations by networks designed for this purpose to model poste-
rior parietal cortex (e.g., Pouget and Sejnowski, 1997).
 Computational Models of Spatial and Mnemonic Functions
Figure 14–14. A. Functional architecture of the model for encoding
and retrieving the spatial context of an event. Neurons shown in
gray illustrate a possible pattern of activation corresponding to an
imagined location with extended landmarks nearby to the west, far
away to the north, and at intermediate distances east and south; the
imagined westerly heading direction means that these landmarks
are imagined as far to the right, near straight ahead, and at an inter-
mediate distance to the left (and behind, but this is not in view).
(Source: Burgess et al., 2001a.) B. Simulation of retrieving spatial
information in the Milan square experiment of Bisiach and Luzzatti
(1978). i: Training consists of simulated exploration of the square
(shaded area, north is up). The system is cued to imagine being near
the cathedral (i.e., the perirhinal cell for the texture of building 1
and parahippocampal cell for a building at a short distance north
are activated), and the hippocampal-parahippocampal-perirhinal
system settles. ii: The hippocampus settles to a location in the
743
northwest corner of the square (hippocampal cell activity shown
as the brightness of the pixel corresponding to the location of each
cell’s place field). iii: The parahippocampus correctly retrieves the
locations of the other buildings (parahippocampal cell activity
shown as the brightness of the pixel for the location encoded by
each cell, relative to the subject at the center). The line indicates that
the imagined head direction is south. iv: Medial parietal (MP) cell
activity. The parahippocampal map has been correctly rotated given
head direction south (straight ahead is up); stars indicate a direc-
tion of inspection to the left, circles to the right. v: Perirhinal (PR)
cell activations correctly showing building 5 to the left and building
7 to the right. vi: Effect of a right parietal lesion on the medial pari-
etal representation (note lack of activation on the left) and perirhi-
nal activations (vii). Note the decrease in activation of building 5
when inspection is to the left. (Source: Adapted from Becker and
Burgess, 2001.)
744 The Hippocampus Book
Likewise, forming an allocentric representation from egocen-
tric perception requires translation in the reverse direction.
Visual imagery occurs in medial parietal areas using the final
head-centered representation with the intermediate represen-
tations buffered in retrosplenial cortex. This model is consis-
tent with the observation of hemispatial neglect in imagery
following parietal damage (e.g., Bisiach and Luzzatti, 1978)
and functional imaging of the retrieval of spatial context
(Burgess et al., 2001b). It also provides an explanation for the
involvement of Papez’s circuit (see Chapter 3) in supporting
both episodic memory (see Chapter 12) and the representa-
tion of head direction (see Chapter 11). Finally it implies a
strong link between episodic recollection (and its neural
bases) and mental imagery.
Á barbital test. Epilepsia 38:998–1007.
14.7 Conclusions
As in other areas of neuroscience, the ability to specify a pro-
posed mechanism of hippocampal function in terms of a
computational model has been invaluable in many ways. First,
it reduces ambiguity and “hypothesis drift.” Second, it enables
quantitative simulation of the resulting behavior. At the very
least this can serve as a demonstration of the viability of a
given proposal. Beyond this, it enables quantitative predic-
tions to be made at the levels of cells, systems, and behavior
regarding the effects of experimental manipulations. Some of
the models reviewed here have been used to make such pre-
dictions and have also contributed to the design of experi-
ments to test them. Many questions for future research are
prompted by these models, not least the possibility of com-
paring or combining the influences of spatial and temporal
contexts as retrieval cues. Because any worthwhile theory
must make potentially falsifiable predictions, computational
modeling has a crucial role to play in the development of the
field. Investigating the link between the properties of cells
interacting in a complex system such as the brain to the result-
ing behavior of the animal would become almost impossible
without the aid of computational models.
Á ACKNOWLEDGMENTS
I am pleased to acknowledge useful discussions with Peter Latham
and John O’Keefe, and the support of the Medical Research Council,
UK.
Á REFERENCES
Aggleton JP, Brown MW (1999) Episodic memory, amnesia, and
the hippocampal-anterior thalamic axis. Behav Brain Sci 22:
425–490.
Alvarez P, Squire LR (1994) Memory consolidation and the medial
temporal lobe: a simple network model. Proc Natl Acad Sci USA
91:7041–7045.
Alyan S, McNaughton BL (1999) Hippocampectomized rats are
capable of homing by path integration. Behav Neurosci 113:
19–31.
Amaral DG, Ishizuka N, Claiborne B (1990) Neurons, numbers and
the hippocampal network. Prog Brain Res 83:1–11.
Amit DH (1989) Modelling brain function. Cambridge, UK:
Cambridge University Press.
Amit DJ (1992) Modeling brain function: the world of attractor neural
networks. Cambridge, UK: Cambridge University Press.
Baddeley AD, Vargha-Khadem F, Mishkin M (2001) Preserved recog-
nition in a case of developmental amnesia: implications for the
acquisition of semantic memory? J Cogn Neurosci 13:357–369.
Barry C, Lever C, Hayman R, Hartley T, Burton S, O’Keefe J, Jeffery
KJ, Burgess N (2006) The boundary vector cell model of place
cell firing and spatial memory. Rev Neurosci 17:71–97.
Baxendale SA, Van Paesschen W, Thompson PJ, Duncan JS, Shorvon
SD, Connelly A (1997) The relation between quantitative MRI
measures of hippocampal structure and the intracarotid amo-
Baxter MG, Murray EA (2001) Opposite relationship of hippocampal
and rhinal cortex damage to delayed nonmatching-to-sample
deficits in monkeys. Hippocampus 11:61–71.
Becker S, Burgess N (2001) A model of spatial recall, mental imagery
and neglect. Adv Neural Inform Proc Syst 13:96–102.
Bienenstock EL, Cooper LN, Munro PW (1982) Theory for the devel-
opment of neuron selectivity: orientation specificity and binoc-
ular interaction in visual cortex. J Neurosci 2:32–48.
Bisiach E, Luzzatti C (1978) Unilateral neglect of representational
space. Cortex 14:129–133.
Blair HT, Sharp PE (1995) Anticipatory head direction signals in
anterior thalamus: evidence for a thalamocortical circuit that
integrates angular head motion to compute head direction. J
Neurosci 15:6260–6270.
Blair HT, Lipscomb BW, Sharp PE (1997) Anticipatory time intervals
of head-direction cells in the anterior thalamus of the rat: impli-
cations for path integration in the head-direction circuit. J
Neurophysiol 78:145–159.
Blair HT, Cho J, Sharp PE (1998) Role of the lateral mammillary
nucleus in the rat head direction circuit: a combined single unit
recording and lesion study. Neuron 21:1387–1397.
Blum KI, Abbott LF (1996) A model of spatial map formation in the
hippocampus of the rat. Neural Comput 8:85–93.
Bose A, Recce M (2001) Phase precession and phase locking of hip-
pocampal pyramidal cells. Hippocampus 11:204–215.
Bostock E, Muller RU, Kubie JL (1991) Experience-dependent modi-
fications of hippocampal place cell firing. Hippocampus
1:193–205.
Brown MA, Sharp PE (1995) Simulation of spatial learning in the
Morris watermaze by a neural network model of the hippocam-
pal formation and nucleus accumbens. Hippocampus 5:171–188.
Brunel N, Trullier O (1998) Plasticity of directional place fields in a
model of rodent CA3. Hippocampus 8:651–665.
Burgess N, Hartley T (2002) Orientational and geometric deter-
minants of place and head-direction. In: Neural informa-
tion processing systems 14, pp 165–172. Cambridge, MA: MIT
Press.
Burgess N, Hitch GJ (1992) Toward a network model of the articula-
tory loop. J Mem Language 31:429–460.
Burgess N, Hitch GJ (2005) Computational models of working mem-
ory: putting long term memory into context. Trends Cogn Sci
9:535–541.
 Computational Models of Spatial and Mnemonic Functions
Burgess N, O’Keefe J (1996) Neuronal computations underlying the
firing of place cells and their role in navigation. Hippocampus
6:749–762.
Burgess N, O’Keefe J, Recce M (1993) Using hippocampal ‘place cells’
for navigation, exploiting phase coding. In: Neural information
processing systems 5 (Hanson SE, Cowan JD, Giles CL, eds), pp
929–936. San Francisco: Morgan Kaufmann.
Burgess N, Recce M, O’Keefe J (1994) A model of hippocampal func-
tion. Neural Netw 7:1065–1081.
Burgess N, Jeffery KJ, O’Keefe J (1999) Integrating hippocampal and
parietal functions: a spatial point of view. In: The hippocampal
and parietal foundations of spatial cognition (Burgess N, Jeffery
KJ, O’Keefe J, eds), pp 3–29. Oxford, UK: Oxford University
Press.
Burgess N, Maguire E, O’Keefe J (2002) The human hippocampus
and spatial and episodic memory. Neuron 35:625–641.
Burgess N, Becker S, King JA, O’Keefe J (2001a) Memory for events
and their spatial context: models and experiments. Philos Trans
R Soc Lond B Biol Sci 356:1493–1503.
Burgess N, Maguire EA, Spiers HJ, O’Keefe J (2001b) A temporopari-
etal and prefrontal network for retrieving the spatial context of
lifelike events. Neuroimage 14:439–453.
Cacucci F, Lever C, Wills TJ, Burgess N, O’Keefe J (2004) Theta-
modulated place-by-direction cells in the hippocampal forma-
tion in the rat. J Neurosci 24:8265–8277.
Cohen MA, Grossberg S (1983) Absolute stability of global pattern
formation and parallel memory storage by competitive neural
networks. IEEE Trans Syst Man Cybern 13:815–821.
Connelly CI, Burns JB, Weiss R (1990) Path planning using Laplace’s
equation. In: Proceedings of the 1990 IEEE international confer-
ence on robotics and automation, pp 2102–2106.
Cressant A, Muller RU, Poucet B (1997) Failure of centrally placed
objects to control the firing fields of hippocampal place cells. J
Neurosci 17:2531–2542.
Czurko A, Hirase H, Csicsvari J, Buzsaki G (1999) Sustained activa-
tion of hippocampal pyramidal cells by ‘space clamping’ in a
running wheel. Eur J Neurosci 11:344–352.
Damasio AR (1989) The brain binds entities and events by multire-
gional activation from convergence zones. Neural Comput
1:123–132.
Davelaar EJ, Goshen-Gottstein Y, Ashkenazi A, Usher M (2005) A
context activation model of list memory: dissociating short-
term from long-term recency effects. Psychol Rev 112:34–53.
Dayan P (1991) Navigating through temporal difference. In: Neural
information processing systems 3 (Lippman RP, Moody JE,
Touretzky DS, eds), pp 464–470. San Mateo, CA: Morgan
Kaufmann.
Dayan P, Abbott LF (2002) Computational neuroscience. Cambridge,
MA: MIT Press.
Deneve S, Latham PE, Pouget A (2001) Efficient computation and
cue integration with noisy population codes. Nat Neurosci
4:826–831.
Dominey PF, Arbib MA (1992) A cortico-subcortical model for gen-
eration of spatially accurate sequential saccades. Cereb Cortex
2:153–175.
Droulez J, Berthoz A (1991) A neural network model of sensoritopic
maps with predictive short-term memory properties. Proc Natl
Acad Sci USA 88:9653–9657.
Egorov AV, Hamam BN, Fransen E, Hasselmo ME, Alonso AA (2002)
Graded persistent activity in entorhinal cortex neurons. Nature
420:173–178.
745
Ekstrom AD, Meltzer J, McNaughton BL, Barnes CA (2001) NMDA
receptor antagonism blocks experience-dependent expansion of
hippocampal “place fields.” Neuron 31:631–638.
Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman
EL, Fried I (2003) Cellular networks underlying human spatial
navigation. Nature 425:184–188.
Etienne AS, Maurer R, Seguinot V (1996) Path integration in mam-
mals and its interaction with visual landmarks. J Exp Biol
199:201–209.
Fahlman S, Lebiere C (1990) The cascade-correlation learning archi-
tecture. In: Neural information processing systems 2 (Touretzky
DS, ed), pp 524–532. San Francisco: Morgan Kaufmann.
Fenton AA, Csizmadia G, Muller RU (2000) Conjoint control of hip-
pocampal place cell firing by two visual stimuli. I. The effects of
moving the stimuli on firing field positions. J Gen Physiol
116:191–209.
Foster DJ, Morris RG, Dayan P (2000) A model of hippocampally
dependent navigation, using the temporal difference learning
rule. Hippocampus 10:1–16.
Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the
hippocampus and entorhinal cortex. Neuron 27:169–178.
Frean MR (1990) The Upstart algorithm: a method for constructing
and training feedforward neural networks. Neural Comput
2:198–209.
Fuhs MC, Touretzky DS (2000) Synaptic learning models of map sep-
aration in the hippocampus. Neurocomputing 32:379–384.
Fuhs MC, Touretzky DS (2006) A spin glass model of path integra-
tion in rat medial entorhinal cortex. J Neurosci 26:4266–4276.
Fyhn M, Molden S, Witter MP, Moser EI, Moser MB (2004) Spatial
representation in the entorhinal cortex. Science 305:1258–1264.
Gaffan D (1994) Dissociated effects of perirhinal cortex ablation,
fornix transection and amygdalectomy: evidence for multiple
memory systems in the primate temporal lobe. Exp Brain Res
99:411–422.
Gallant SI (1986) Three constructive algorithms for network learn-
ing. In: Proceedings of the 8th annual conference of the Cognitive
Science Society, pp 652–660.
Gardiner JM, Java RI (1993) In: Theories of memory (Collins A,
Gathercole S, Morris P, eds), pp 168–188. Hillsdale, NJ: Erlbaum.
Gardner-Medwin AR (1976) The recall of events through the learn-
ing of associations between their parts. Proc R Soc Lond B Biol
Sci 194:375–402.
Georgopoulos AP, Schwartz AB, Kettner RE (1986) Neuronal popula-
tion coding of movement direction. Science 233:1416–1419.
Gluck MA, Myers CE (1996) Integrating behavioral and physio-
logical models of hippocampal function. Hippocampus 6:
643–653.
Goodale MA, Milner AD (1992) Separate visual pathways for percep-
tion and action. Trends Neurosci 15:20–25.
Goodridge JP, Touretzky DS (2000) Modeling attractor deformation
in the rodent head-direction system. J Neurophysiol 83:
3402–3410.
Gorchetchnikov A, Hasselmo ME (2002) A model of hippocampal
circuitry mediating goal-driven navigation in a familiar envi-
ronment. Neurocomputing 44–46:423–427.
Gothard KM, Skaggs WE, McNaughton BL (1996) Dynamics of mis-
match correction in the hippocampal ensemble code for space:
interaction between path integration and environmental cues. J
Neurosci 16:8027–8040.
Graham KS, Hodges JR (1997) Differentiating the roles of the hip-
pocampus complex and the neocortex in long-term memory
746 The Hippocampus Book
storage: evidence from the study of semantic dementia and
Alzheimer’s disease. Neuropsychology 11:77–89.
Guazzelli A, Bota M, Arbib MA (2001) Competitive Hebbian learn-
ing and the hippocampal place cell system: modeling the inter-
action of visual and path integration cues. Hippocampus
11:216–239.
Hafting T, Fyhn M, Molden S, Moser MB, Moser EI (2005)
Microstructure of a spatial map in the entorhinal cortex. Nature
436:801–806.
Harris KD, Henze DA, Hirase H, Leinekugel X, Dragoi G, Czurko A,
Buzsaki G (2002) Spike train dynamics predicts theta-related
phase precession in hippocampal pyramidal cells. Nature
417:738–741.
Hartley T, Burgess N, Lever C, Cacucci F, O’Keefe J (2000) Modeling
place fields in terms of the cortical inputs to the hippocampus.
Hippocampus 10:369–379.
Hasselmo ME, Fehlau BP (2001) Differences in time course of ACh
and GABA modulation of excitatory synaptic potentials in slices
of rat hippocampus. J Neurophysiol 86:1792–1802.
Hasselmo ME, McClelland JL (1999) Neural models of memory. Curr
Opin Neurobiol 9:184–188.
Hasselmo ME, Schnell E, Barkai E (1995) Dynamics of learning and
recall at excitatory recurrent synapses and cholinergic modula-
tion in rat hippocampal region CA3. J Neurosci 15:5249–5262.
Hasselmo ME, Wyble BP (1997) Free recall and recognition in a net-
work model of the hippocampus: simulating effects of scopo-
lamine on human memory function. Behav Brain Res 89:1–34.
Hasselmo ME, Wyble BP, Wallenstein GV (1996) Encoding and
retrieval of episodic memories: role of cholinergic and
GABAergic modulation in the hippocampus. Hippocampus
6:693–708.
Hebb DO (1949) The organisation of behavior. New York: Wiley.
Hertz J, Krogh A, Palmer R (1990) Introduction to the theory of neural
computation. New York: Perseus Books.
Hinton GE, Sejnowski TJ (1999) Unsupervised learning. Cambridge,
MA: MIT Press.
Hirase H, Czurko HH, Csicsvari J, Buzsaki G (1999) Firing rate and
theta-phase coding by hippocampal pyramidal neurons during
‘space clamping.’ Eur J Neurosci 11:4373–4380.
Hodgkin AL, Huxley AF (1952) A quantitative description of mem-
brane current and its application to conduction and excitation
in nerve. J Physiol 117:500–544.
Holdstock JS, Mayes AR, Cezayirli E, Isaac CL, Aggleton JP, Roberts
N (2000) A comparison of egocentric and allocentric spatial
memory in a patient with selective hippocampal damage.
Neuropsychologia 38:410–425.
Holdstock JS, Mayes AR, Roberts N, Cezayirli E, Isaac CL, O’Reilly
RC, Norman KA (2002) Under what conditions is recognition
spared relative to recall after selective hippocampal damage in
humans? Hippocampus 12:341–351.
Hopfield JJ (1982) Neural networks and physical systems with emer-
gent collective computational abilities. Proc Natl Acad Sci USA
79:2554–2558.
Hoppensteadt FC (1986) An introduction to the mathematical proper-
ties of neurons. Cambridge, UK: Cambridge University Press.
Hori E, Tabuchi E, Matsumura N, Tamura R, Eifuku S, Endo S,
Nishijo H, Ono T (2003) Representation of place by monkey
hippocampal neurons in real and virtual translocation.
Hippocampus 13:190–196.
Howard M, Kahana MJ (2001) A distributed representation of tem-
poral context. J Math Psychol 46:269–299.
Howard MW, Fotedar MS, Datey AV, Hasselmo ME (2005) The tem-
poral context model in spatial navigation and relational learn-
ing: toward a common explanation of medial temporal lobe
function across domains. Psychol Rev 112:75–116.
Huxter J, Burgess N, O’ Keefe J (2003) Independent rate and tempo-
ral coding in hippocampal pyramidal cells. Nature 425:828–832.
Jensen O, Lisman JE (1996) Hippocampal CA3 region predicts mem-
ory sequences: accounting for the phase precession of place
cells. Learn Mem 3:279–287.
Jensen O, Lisman JE (2000) Position reconstruction from an ensem-
ble of hippocampal place cells: contribution of theta phase cod-
ing. J Neurophysiol 83:2602–2609.
Jensen O, Lisman JE (2005) Hippocampal sequence-encoding
driven by a cortical multi-item working memory buffer. Trends
Neurosci 28:67–72.
Jones VJ (1976) A fragmentation hypothesis of memory: cued recall
of pictures and of sequential position. J Exp Psychol Gen
105:277–293.
Jung MW, Wiener SI, McNaughton BL (1994) Comparison of spatial
firing characteristics of units in dorsal and ventral hippocampus
of the rat. J Neurosci 14:7347–7356.
Kali S, Dayan P (2000) The involvement of recurrent connections in
area CA3 in establishing the properties of place fields: a model.
J Neurosci 20:7463–7477.
Kali S, Dayan P (2004) Off-line replay maintains declarative memo-
ries in a model of hippocampal-neocortical interactions. Nat
Neurosci 7:286—294.
Keith JR, McVety KM (1988) Latent place learning in a novel envi-
ronment and the influence of prior training in rats.
Psychobiology 16:146–151.
Kentros C, Hargreaves E, Hawkins RD, Kandel ER, Shapiro M, Muller
RV (1998) Abolition of long-term stability of new hippocampal
place cell maps by NMDA receptor blockade. Science
280:2121–2126.
Knierim JJ (2003) Hippocampal remapping: implications for
spatial learning and navigation. In: The neurobiology of spatial
behaviour (Jeffery KJ, ed), pp 226–239. Oxford, UK: Oxford
University Press.
Knierim JJ, Kudrimoti HS, McNaughton BL (1995) Place cells, head
direction cells, and the learning of landmark stability. J Neurosci
15:1648–1659.
Knowlton BJ, Squire LR (1995) Remembering and knowing: two dif-
ferent expressions of declarative memory. J Exp Psychol Learn
Mem Cogn 21:699–710.
Kohonen T (1972) Correlation matrix memories. IEEE Trans Comp
C-21:353–359.
Kroll NE, Knight RT, Metcalfe J, Wolf ES, Tulving E (1996) Cohesion
failure as a source of memory illusions. J Mem Language
35:176–196.
Latham PE, Deneve S, Pouget A (2003) Optimal computation with
attractor networks. J Physiol Paris 97:683–694.
Lengyel M, Szatmary Z, Erdi P (2003) Dynamically detuned oscilla-
tions account for the coupled rate and temporal code of place
cell firing. Hippocampus 13:700–714.
Leutgeb JK, Leutgeb S, Treves A, Meyer R, Barnes CA, McNaughton
BL, Moser MB, Moser EI (2005) Progressive transformation of
hippocampal neuronal representations in “morphed” environ-
ments. Neuron 48:345–358.
Lever C, Wills T, Cacucci F, Burgess N, O’Keefe J (2002) Long-term
plasticity in the hippocampal place cell representation of envi-
ronmental geometry. Nature 416:90–94.
 Computational Models of Spatial and Mnemonic Functions
Levy WB (1996) A sequence predicting CA3 is a flexible associator
that learns and uses context to solve hippocampal-like tasks.
Hippocampus 6:579–590.
Lieblich I, Arbib MA (1982) Multiple representations of space iun-
derlying behavior. Behav Brain Sci 5:627–659.
Lisman JE, Grace AA (2005) The hippocampal-VIA Loop: control-
ling the entry of information into long-term memory. Neuron
46:703–713.
Ludvig N, Tang HM, Gohil BC, Botero JM (2004) Detecting location-
specific neuronal firing rate increases in the hippocampus of
freely-moving monkeys. Brain Res 1014:97–109.
Mallot HA, Gillner S (2000) Route navigating without place recogni-
tion: what is recognised in recognition-triggered responses?
Perception 29:43–55.
Manns JR, Squire LR (1999) Impaired recognition memory on the
doors and people test after damage limited to the hippocampal
region. Hippocampus 9:495–499.
Marr D (1970) A theory for cerebral cortex. Proc R Soc Lond B Biol
Sci 176:161–234.
Marr D (1971) Simple memory: a theory for archicortex. Philos Trans
R Soc Lond B Biol Sci 262:23–81.
Maurer AP, Van Rhoads SR, Sutherland GR, Lipa P, McNaughton
BL (2005) Self-motion and the origin of differential spatial
scaling along the septo-temporal axis of the hippocampus.
Hippocampus 15:841–852.
McClelland JL, McNaughton BL, O’Reilly RC (1995) Why there are
complementary learning systems in the hippocampus and neo-
cortex: insights from the successes and failures of connectionist
models of learning and memory. Psychol Rev 102:419–457.
McClelland JL, Rumelhart DE (1986) Parallel distributed pro-
cessing: explorations in the microstructure of cognition. Vol 2.
Psychological and biological models. Cambridge, MA: MIT
Press.
McNaughton BL, Barnes CA, O’Keefe J (1983) The contributions of
position, direction, and velocity to single unit activity in the
hippocampus of freely-moving rats. Exp Brain Res 52:41–49.
McNaughton BL, Barnes CA, Gerrard JL, Gothard K, Jung MW,
Knierim JJ, Kudrimoti H, Qin Y, Skaggs WE, Suster M, Weaver
KL (1996) Deciphering the hippocampal polyglot: the hip-
pocampus as a path integration system. J Exp Biol 199:173–185.
McNaughton BL, Battaglia FP, Jensen O, Moser EI, Moser MB (2006).
Path integration and the neural basis of the ‘cognitive map’. Nat
Rev Neurosci 7:663–678.
McNaughton BL, Morris RG (1987) Hippocampal synaptic enhance-
ment and information storage within a distributed memory sys-
tem. Trends Neurosci 10:408–415.
McNaughton BL, Nadel L (1990) Hebb-Marr networks and the neu-
robiological representation of action in space. In: Neuroscience
and connectionist theory (Gluck MA, Rumelhart DE, eds), pp
1–63. Hillsdale, NJ: Lawrence Erlbaum.
Mehta MR, Barnes CA, McNaughton BL (1997) Experience-
dependent, asymmetric expansion of hippocampal place fields.
Proc Natl Acad Sci USA 94:8918–8921.
Mehta MR, Quirk MC, Wilson MA (2000) Experience-dependent
asymmetric shape of hippocampal receptive fields. Neuron
25:707–715.
Mehta MR, Lee AK, Wilson MA (2002) Role of experience and oscil-
lations in transforming a rate code into a temporal code. Nature
417:741–746.
Mensink GJ, Raaijmakers JG (1988) A model for interference and for-
getting. Psychol Rev 95:434–455.
747
Mezard M, Nadal J-P (1989) Learning in feedforward layered net-
works: the Tiling algorithm. J Physics A 22:2191–2203.
Milner AD, Dijkerman HC, Carey DP (1999) Visuospatial processing
in a case of visual form agnosia. In: The hippocampal and parietal
foundations of spatial cognition (Burgess N, Jeffery KJ, O’Keefe J,
eds), pp 443–466. Oxford, UK: Oxford University Press.
Moll M, Miikkulainen R (1997) Convergence-zone episodic memory:
analysis and simulations. Neural Networks 10:1017–1036.
Montague PR, Dayan P, Sejnowski TJ (1996) A framework for mes-
encephalic dopamine systems based on predictive Hebbian
learning. J Neurosci 16:1936–1947.
Morris RGM, Garrud P, Rawlins JN, O’Keefe J (1982) Place naviga-
tion impaired in rats with hippocampal lesions. Nature
297:681–683.
Morton J, Hammersley RH, Bekerian DA (1985) Headed records: a
model for memory and its failure. Cognition 20:1–23.
Muller RU, Kubie JL (1987) The effects of changes in the environ-
ment on the spatial firing of hippocampal complex-spike cells. J
Neurosci 7:1951–1968.
Muller RU, Kubie JL, Saypoff R (1991) The hippocampus as a cogni-
tive graph (abridged version). Hippocampus 1:243–246.
Muller RU, Stead M, Pach J (1996) The hippocampus as a cognitive
graph. J Gen Physiol 107:663–694.
Murray EA, Mishkin M (1998) Object recognition and location
memory in monkeys with excitotoxic lesions of the amygdala
and hippocampus. J Neurosci 18:6568–6582.
Murre JM (1996) TraceLink: a model of amnesia and consolidation
of memory. Hippocampus 6:675–684.
Nadel L, Moscovitch M (1997) Memory consolidation, retrograde
amnesia and the hippocampal complex. Curr Opin Neurobiol
7:217–227.
Nader K, Schafe GE, LeDoux JE (2000) The labile nature of consoli-
dation theory. Nat Rev Neurosci 1:216–219.
Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun
LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S
(2002) Requirement for hippocampal CA3 NMDA receptors in
associative memory recall. Science 297:211–218.
Nicolelis MA, Baccala LA, Lin RC, Chapin JK (1995) Sensorimotor
encoding by synchronous neural ensemble activity at multiple
levels of the somatosensory system. Science 268:1353–1358.
Norman KA (2003) Episodic memory, computational models of. In:
Encyclopedia of cognitive science (Nadel L, ed), pp 15–23.
London: Nature Publishing.
Norman KA, O’Reilly RC (2003) Modeling hippocampal and
neocortical contributions to recognition memory: a comple-
mentary-learning-systems approach. Psychol Rev 110:611–646.
O’Keefe J (1990) A computational theory of the hippocampal cogni-
tive map. Prog Brain Res 83:301–312.
O’Keefe J (1991) The hippocampal cognitive map and navigational
strategies. In: Brain and Space (Paillard J, ed), pp 273–295.
Oxford, UK: Oxford University Press.
O’Keefe J, Burgess N (1996) Geometric determinants of the place
fields of hippocampal neurons. Nature 381:425–428.
O’Keefe J, Burgess N (2005) Dual phase and rate coding in hip-
pocampal place cells: theoretical significance and relationship to
entorhinal grid cells. Hippocampus 15:853–866.
O’Keefe J, Conway DH (1978) Hippocampal place units in the freely
moving rat: why they fire where they fire. Exp Brain Res
31:573–590.
O’Keefe J, Nadel L (1978) The hippocampus as a cognitive map.
Oxford, UK: Oxford University Press.