Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.

159
 REVIEW ARTICLE
published: 09 November 2009
HUMAN NEUROSCIENCE doi: 10.3389/neuro.09.031.2009

The human brain in numbers: a linearly scaled-up primate brain

Suzana Herculano-Houzel*
Instituto de Ciências Biomédicas, Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brasil

Edited by:
Andreas Jeromin,
Allen Institute for Brain Science, USA
Reviewed by:
Karl Herrup,
Case Western University, USA
Robert Barton,
University of Durham, UK
*Correspondence:
Suzana Herculano-Houzel, Laboratório
de Neuroanatomia Comparada,
Instituto de Ciências Biomédicas,
Universidade Federal do Rio de
Janeiro, Rua Carlos Chagas Filho 373,
21950-902 Rio de Janeiro,
Rio de Janeiro, Brazil.
e-mail: suzanahh@gmail.com
The human brain has often been viewed as outstanding among mammalian brains: the most
cognitively able, the largest-than-expected from body size, endowed with an overdeveloped
cerebral cortex that represents over 80% of brain mass, and purportedly containing 100 billion
neurons and 10× more glial cells. Such uniqueness was seemingly necessary to justify the
superior cognitive abilities of humans over larger-brained mammals such as elephants and whales.
However, our recent studies using a novel method to determine the cellular composition of the
brain of humans and other primates as well as of rodents and insectivores show that, since
different cellular scaling rules apply to the brains within these orders, brain size can no longer
be considered a proxy for the number of neurons in the brain. These studies also showed that
the human brain is not exceptional in its cellular composition, as it was found to contain as many
neuronal and non-neuronal cells as would be expected of a primate brain of its size. Additionally,
the so-called overdeveloped human cerebral cortex holds only 19% of all brain neurons, a
fraction that is similar to that found in other mammals. In what regards absolute numbers of
neurons, however, the human brain does have two advantages compared to other mammalian
brains: compared to rodents, and probably to whales and elephants as well, it is built according
to the very economical, space-saving scaling rules that apply to other primates; and, among
economically built primate brains, it is the largest, hence containing the most neurons. These
findings argue in favor of a view of cognitive abilities that is centered on absolute numbers of
neurons, rather than on body size or encephalization, and call for a re-examination of several
concepts related to the exceptionality of the human brain.
Keywords: brain scaling, number of neurons, human, encephalization

INTRODUCTION
THE HUMAN BRAIN AS A SPECIAL BRAIN
What makes us human? Is our brain, the only one known to study
other brains, special in any way? According to a recent popular
account of what makes us unique, “we have brains that are bigger than
expected for an ape, we have a neocortex that is three times bigger
than predicted for our body size, we have some areas of the neocortex
and the cerebellum that are larger than expected, we have more white
matter” – and the list goes on (Gazzaniga, 2008). Most specialists
seem to agree (for example, Marino, 1998; Rilling, 2006; Sherwood
et al., 2006). Since ours is obviously not the largest brain on Earth, our
superior cognitive abilities cannot be accounted for by something as
simple as brain size, the most readily measurable parameter regarding
the brain. Emphasis is thus placed on an exceptionality that is, curi-
ously, not brain-centered, but rather body-centered: With a smaller
body but a larger brain than great apes, the human species deviates
from the relationship between body and brain size that applies to
other primates, great apes included, boasting a brain that is 5–7× too
large for its body size (Jerison, 1973; Marino, 1998). Recent efforts
to support this uniqueness have focused on finding genetic differ-
ences between humans and other primates (reviewed in Vallender,
2008), as well as cellular particularities such as the presence and
distribution of Von Economo neurons (Nimchinsky et al., 1999; but
see Butti et al., 2009; Hakeem et al., 2009).
To regard the human brain as unique requires considering it
to be an outlier: an exception to the rule, whatever that rule is.
This makes little sense, however, in light of evolution. If we go
to such great lengths to affirm, and teach, that evolution is the
origin of diversity in life, and to find trends and laws that apply to
kingdoms, phyla and orders as a whole, why then insist that what-
ever scaling rules apply to other primates must not apply to us?
In view of the vexing size inferiority in brain size and of the lack
of information about what our brains are actually made of – and
how that compares to other brains, particularly those of whales and
elephants – resorting to a quest for uniqueness may have seemed
as a necessary, natural step to justify the cognitive superiority of
the human brain.
Recently, a novel quantitative tool developed in our lab
(Herculano-Houzel and Lent, 2005) has finally made the num-
bers of neurons and non-neuronal cells that compose the brains
of various mammals, humans included, available for comparative
analysis. This review will focus on such a quantitative, compara-
tive analysis, with emphasis on the numbers that characterize the
human brain: what they are, how they have been viewed in the past,
and how they change our view of where the human brain fits into
the diversity of the mammalian nervous system.
THE HUMAN BRAIN IN NUMBERS
How many neurons does the human brain have, and how does
that compare to other species? Many original articles, reviews and
textbooks affirm that we have 100 billion neurons and 10 times
more glial cells (Kandel et al., 2000; Ullian et al., 2001; Doetsch,
2003; Nishiyama et al., 2005; Noctor et al., 2007; Allen and Barres,
2009), usually with no references cited. This leaves the reader with

Frontiers in Human Neuroscience www.frontiersin.org November 2009 | Volume 3 | Article 31 | 1

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Herculano-Houzel
the impression that the cellular composition of the human brain has
long been determined. Indeed, an informal survey with senior neu-
roscientists that we ran in 2007 showed that most believed that the
number of cells in the human brain was indeed already known: that
we have about 100 billion neurons, outnumbered by about 10 times
more glial cells – but none of the consulted scientists could cite an
original reference for these numbers (Herculano-Houzel and Lent,
unpublished observations). Curiously, the widespread concept that
neurons represent about 10% of all cells in the human brain might
be one of the arguments behind the popular, but mistaken, notion
that we only use 10% of our brain (Herculano-Houzel, 2002).
The reason for such lack of references is that indeed there was,
to our knowledge, no actual, direct estimate of numbers of cells
or of neurons in the entire human brain to be cited until 2009.
A reasonable approximation was provided by Williams and Herrup
(1988), from the compilation of partial numbers in the literature.
These authors estimated the number of neurons in the human brain
at about 85 billion: 12–15 billion in the telencephalon (Shariff,
1953), 70 billion in the cerebellum, as granule cells (based on Lange,
1975), and fewer than 1 billion in the brainstem. With more recent
estimates of 21–26 billion neurons in the cerebral cortex (Pelvig
et al., 2008) and 101 billion neurons in the cerebellum (Andersen
et al., 1992), however, the total number of neurons in the human
brain would increase to over 120 billion neurons.
As to the 10 times more numerous glial cells in the human
brain, that seems to be the case only in subcortical nuclei such as
the thalamus (17 glial cells per neuron) and the ventral pallidum
(12 glial cells per neuron; Pakkenberg and Gundersen, 1988). In the
gray matter of the cerebral cortex, glial cells outnumber neurons
by a factor of <2 (Sherwood et al., 2006; Pelvig et al., 2008). Given
the relatively small number of glial cells reported for the human
cerebellum, where they are outnumbered by neurons by at least
25:1 (Andersen et al., 1992), the only possible explanation for the
ubiquitous quote of 10 times more glial than neuronal cells in the
entire human brain would be the presence of nearly one trillion
glial cells in the remaining structures alone – an unlikely scenario,
since these structures represent <10% of total brain mass.
WHY BOTHER WITH CELL NUMBERS?
Across species, the number of neurons and their relative abundance
in different parts of the brain is widely considered to be a determi-
nant of neural function and, consequently, of behavior (Williams
and Herrup, 1988). Among mammals, those species with the largest
brains, such as cetaceans and primates, have a greater range and
versatility of behavior than those with the smallest brains, such as
insectivores (Jerison, 1985; Marino, 2002). Among birds, those that
are larger-brained (corvids, parrots and owls) are also considered
the most intelligent (Lefebvre et al., 2004). A recent comparison
of several parameters, including brain size, relative brain size,
encephalization, conduction velocity and estimated numbers of
neurons led two authors to conclude that the “factors that correlate
better with intelligence (across species) are the number of corti-
cal neurons and conduction velocity, as the basis for information
processing” (Roth and Dicke, 2005). Indeed, within non-human
primates, a recent meta-analysis concluded that the best predictor
of the cognitive abilities of a species is absolute brain size, not rela-
tive size nor encephalization quotient (EQ; Deaner et al., 2007).
Frontiers in Human Neuroscience www.frontiersin.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
The human brain in numbers
However, the correlation between absolute brain size and
cognitive abilities breaks down when species of similar brain size are
compared across orders. Monkeys, for instance, possess brains that
are much smaller than those of ungulates, but the higher cognitive
and behavioral flexibility of monkeys over ungulates is anecdotally
evident to any observer who compares the ingenious and complex
abilities of macaques to those of cows or horses, even though the
latter have 4–5× larger brains than macaques. For similar-sized
brains, rodents also perform more poorly than primates: With a
brain of only 52 g, the behavioral, social and cognitive repertoire of
the capuchin monkey is outstanding compared to the capybara, a
giant Amazonian rodent (MacDonald, 1981), even though the latter
has a larger brain of 75 g. This is reminiscent of the most striking and
troubling discrepancy regarding brain size and cognitive abilities:
that between humans and larger-brained species such as whales and
elephants. If the latter have brains that are up to six times larger than
a human brain, why should we be more cognitively able? Answering
this question requires a direct examination of the numbers of neu-
rons that compose the brains of humans and other species.
BRAIN AND BODY SCALING: THE TRADITIONAL VIEW
ASSUMPTION 1: BODY SIZE MATTERS
If the smaller size of the human brain compared to elephant and
whale brains (Figure 1) translates into a smaller number of neu-
rons in the human brain than in the latter, then what makes the
human brain outstanding in its cognitive abilities? In the absence
of direct estimates of numbers of neurons in these and other spe-
cies, the search for a neural correlate for human capacities has
placed emphasis on the characteristic that most undisputedly places
humans above other mammals: the EQ (Jerison, 1973). This meas-
ure is based on the observation that, across species, brain size cor-
relates with body size in a way that can be described mathematically
with a power function, thus allowing the predicted brain mass to
be calculated for any species. EQ indicates how much the observed
brain mass of a species deviates from the expected for its body
mass: an EQ of 1 indicates that the observed brain mass matches
the expected value; an EQ >1 means that brain size in that species
is larger than expected for its body mass.
Compared to mammals as a whole, humans have the largest EQ
found for any mammal, of between 7 and 8 (Jerison, 1973); even if
compared to anthropoid primates only, humans still have an EQ
of over 3, a value that is larger than that obtained for any other
FIGURE 1 | The human brain is not the largest. Brains of a human and of
an African elephant are depicted here at the same scale. Drawings by
Lorena Kaz based on images freely available from the University of
Wisconsin and Michigan State Comparative Mammalian Brain Collections
(www.brainmuseum.org).
November 2009 | Volume 3 | Article 31 | 2
 Herculano-Houzel
primate or cetacean (Marino, 1998). The position of the human
species as an outlier in the body × brain comparison is made clear
if one considers that although gorillas and orangutans overlap or
exceed humans in body size, their brains amount to only about
one-third of the size of the human brain.
There are, however, several problems with the notion that
the explanation for the superior cognitive abilities of the human
species lies in its large EQ. For one, it is not obvious how larger-
than-expected brain mass would confer a cognitive advantage. In
principle, this advantage would rely on the availability for cognitive
functions of whatever brain mass exceeds what is necessary to proc-
ess body-related information. However, according to this notion,
small-brained animals with very large EQs should be expected
to have more cognitive abilities than large-brained animals with
smaller EQs. Capuchin monkeys, for instance, have much larger
EQs than gorillas (Marino, 1998), but are outranked by these in cog-
nitive performance (Deaner et al., 2007). Absolute brain mass and
number of neurons, left out of the encephalization equation, must
clearly be taken into consideration, since the “exceeding number of
neurons” in a large brain should necessarily be larger than that in
a smaller brain of same EQ (Herculano-Houzel, 2007).
Another problem with the utility of the EQ is that the body–
brain mass relationship from which expected brain mass is derived
depends on the precise combination of species computed (Barton,
2006; Herculano-Houzel et al., 2007). We have recently found that,
compared to the linear brain × body relationship that applies to
the primate species in our sample (which consisted of simian and
prosimian primates; Herculano-Houzel et al., 2007), the human
brain deviates by only 10% from its expected size (Azevedo et al.,
2009). This conformity to the body × brain relationship that applies
to non-anthropoid primates is consistent with the observation that,
like in other non-anthropoid primates, the human brain mass rep-
resents about 2% of body mass. Given the sensitivity of EQ to the
species included and our finding that the human brain conforms
to the scaling rules that apply to other primates (see below), we
have suggested that, rather than humans having a larger brain than
expected, it is the great apes such as orangutans and, more notably,
gorillas that have bodies that are much larger than expected for
primates of their brain size (Herculano-Houzel et al., 2007).
This latter possibility of a dissociation between brain and body
development in evolution (which might be only circumstantially,
and not causally, related) constitutes a final criticism to the useful-
ness of the EQ as an index of “brain” evolution in comparative
studies: indeed, the emphasis on the body-centered EQ overlooks
the observation that, compared to other mammalian orders, primate
encephalization is the result of a shift in postcranial growth proc-
esses, not a modification of brain growth (Deacon, 1997). In the
words of Deacon, “if primates have big brains merely because they
have small bodies, we cannot presume that this represents an evolu-
tionary trend driven by cognitive demands”(p. 343). In this scenario,
however, the human brain exhibits a further modification in that it
continues to grow as though in a larger body (Deacon, 1997).
ASSUMPTION 2: BRAIN SIZE MATTERS
Brain size varies across mammals by a factor of approximately
100,000 (Tower, 1954; Stolzenburg et al., 1989). Different mam-
malian orders have traditionally been pooled together in studies
Frontiers in Human Neuroscience www.frontiersin.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
The human brain in numbers
of brain allometry, as if their brains were built according to
the same scaling rules (for example, Haug, 1987; Zhang and
Sejnowski, 2000).
Comparisons across orders that seem to invalidate the correla-
tion between numbers of neurons and cognitive ability, such as
between monkeys and ungulates, or rodents and primates, also bear
this hidden caveat: the assumption that brain size relates to number
of neurons in the brain in a similar fashion across orders. This
assumption, which was justifiable by the lack of direct estimates
of the neuronal composition of the brain of different species, is so
widespread that it implicitly or explicitly underlies most compara-
tive studies to date (for example, Haug, 1987; Finlay and Darlington,
1995; Barton and Harvey, 2000; Clark et al., 2001). The very concept
of encephalization presupposes that not only the brain scales as a
function of body size, but that all brains scale the same way, such
that the only informative (and sufficient) variable is brain size and
its deviation from the expected. However, our quantitative studies
on the cellular scaling rules that apply to different mammalian
orders have shown that this assumption is invalid and therefore
should no longer be applied (see below).
ASSUMPTION 3: PROPORTIONS AND RELATIVE SIZE MATTER
An often cited argument in favor of the uniqueness of the human
brain is its relatively large cerebral cortex, which accounts for 82%
of brain mass. Within this large cerebral cortex, a relative enlarge-
ment of the prefrontal cortex was once considered a hallmark of
the human brain, but this view has however been overthrown by
modern measurements (Semendeferi et al., 2002). Still, the distri-
bution of cortical mass in humans may differ from that in other
primates, endowing particularly relevant regions such as area 10
with relatively more neurons in the human cortex (Semendeferi
et al., 2001).
Relative size is supposed to be a meaningful indicator of relative
functional importance of a brain structure based on the assumption
that it is a proxy for relative number of neurons. For instance, the
increase in relative size of the cerebral cortex with increasing brain
size simultaneously with no systematic change in the relative size of
the cerebellum has been used as evidence that these structures are
functionally independent and have been evolving separately (Clark
et al., 2001). Such discrepancy would support the popular notion
that brain evolution equates with development of the cerebral cor-
tex, which comes to predominate over the other brain structures.
However, analysis of absolute, rather than relative, cerebral cortical
and cerebellar volumes in the same dataset leads to the opposite
conclusion: the coordinated scaling of these volumes, as well as of
the surface areas of these structures, would be evidence that the
cerebral cortex and cerebellum are functionally related and have
been evolving coordinately (Barton, 2002; Sultan, 2002).
As it turns out, however, the underlying assumption that the
relative size of a brain structure reflects the relative number of
brain neurons that it contains is flawed.
Now that numbers of neurons are available across rodents, pri-
mates and insectivores, we find that the cerebral cortex, despite
varying in relative size from 42% (in the mouse) to 82% of brain
mass (in the human), contains between 13 and 28% of all brain
neurons in 15 of 18 species studied, ranging between 13% (in
moles) and 41% (in the squirrel monkey; Herculano-Houzel et al.,
November 2009 | Volume 3 | Article 31 | 3
 Herculano-Houzel
2006, 2007; Sarko et al., 2009). Most importantly, this fractional
number of neurons in the cerebral cortex relative to the whole
brain is not correlated with the relative size of the cerebral cortex
(Figure 2). Instead, the number of neurons in the cerebral cortex
increases coordinately with the number of neurons in the cerebel-
lum (Herculano-Houzel, submitted).
A NEW VIEW: SCALING OF NEURONAL NUMBERS
CELLULAR SCALING RULES FOR RODENT, INSECTIVORE,
AND PRIMATE BRAINS
Our group has been investigating the cellular scaling rules that
apply to brain allometry in different mammalian orders using the
novel method of isotropic fractionation, which produces cell counts
derived from tissue homogenates from anatomically defined brain
regions (Herculano-Houzel and Lent, 2005). Through the estima-
tion of absolute numbers of neuronal and non-neuronal cells in
the brains of different mammalian species and their comparison
within individual orders, we have been able to determine the scal-
ing rules that apply to the brains of species spanning a wide range
of body and brain masses in rodents (Herculano-Houzel et al.,
2006), primates (Herculano-Houzel et al., 2007) and more recently
in insectivores (Sarko et al., 2009). A comparative overview of brain
mass and total number of neurons for these species can be seen
in Figure 3.
A recent issue in comparative studies of brain scaling has been
the examination of how residual variation in different parameters
relate to phylogenetic relationships once shared evolutionary com-
monalities in body or brain size are accounted for (Harvey and
Pagel, 1991; Nunn and Barton, 2000). Although such analyses of
independent contrasts are instrumental for identifying evolutionary
correlations across taxa while taking into account this phylogenetic
nonindependence, they overlook the very issue at hand here: how
the size of the brain reflects the number of neurons that it contains,
regardless of body size and of any other shared characteristics.
For this reason, the analysis reviewed here, referred to as unveiling
the “cellular scaling rules” for the brain of different mammalian
FIGURE 2 | Relative size of the cerebral cortex does not inform about the
relative number of neurons in the cortex compared to the whole brain.
Each point indicates, for a given species, the average relative cortical mass as
a percentage of total brain mass (X-axis) and the average relative number of
cortical neurons as a percentage of the total number of neurons in the brain
(Y-axis). Data from Herculano-Houzel et al. (2006, 2007); Azevedo et al. (2009);
and Sarko et al. (2009).
Frontiers in Human Neuroscience www.frontiersin.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
The human brain in numbers
orders, considers solely how brain size changes as a function of
its number of neurons within a given order, irrespective of body
size, and without any concerns regarding phylogenenetic effects
within that order, or even whether evolution of the extant species
has involved an expansion of brain size, a reduction, or both. In the
particular case of primates, we have recently extended our analysis
to another set of five primate species (Gabi et al., submitted), and
found that the same cellular scaling relationships apply to the origi-
nal dataset (Herculano-Houzel et al., 2007), to the second dataset,
and to the combined, extended dataset. This is evidence that the
cellular scaling rules considered here from a set of primate species
also extend to primates as a whole, and can be used to infer the
expected cellular composition of the human brain – even though
small variations may occur across species that might, indeed, be
due to phylogenetic interdependencies.
In the order Rodentia, we find that the brain increases in size
faster than it gains neurons, with a decrease in neuronal densities
which, in the presence of constant non-neuronal cell densities,
implies that average neuronal size increases rapidly as neurons
become more numerous (Herculano-Houzel et al., 2006). The
increase in numbers of neurons in the cerebral cortex, cerebellum
and remaining areas is concurrent with an even greater increase in
numbers of non-neurons, yielding a maximal glia/neuron ratio that
increases with brain size (Herculano-Houzel et al., 2006). These
findings corroborated previous studies describing neuronal density
decreasing and the glia-to-neuron ratio increasing with increasing
brain size across mammalian taxa (Tower and Elliot, 1952; Shariff,
1953; Friede, 1954; Tower, 1954; Hawking and Olszewski, 1957;
Haug, 1987; Reichenbach, 1989; Stolzenburg et al., 1989).
In contrast to rodent brains, which scale hypermetrically in
size with their numbers of neurons, primate brain size increases
approximately isometrically as a function of neuron number, with
no systematic change in neuronal density or in the non-neuronal/
neuronal ratio with increasing brain size (Herculano-Houzel et al.,
2007). Across insectivore species, on the other hand, the cerebellum
increases linearly in size as a function of its number of neurons (as
in primates), while the cerebral cortex increases in size hypermetri-
cally as it gains neurons (as in rodents; Sarko et al., 2009). In view
of the similar non-neuronal cell densities across species, hyper-
metric scaling of brain structure mass as a function of its number
of neurons implies a concurrent increase in the average neuronal
size (which, in the method’s definition, includes not only the cell
soma but also the entire dendritic and axonal arborizations as well
as synapses; Herculano-Houzel et al., 2006). As a consequence of
these different cellular scaling rules, shown in Table 1, a 10-fold
increase in the number of neurons in a rodent brain results in a
35-fold larger brain; in contrast, a similar 10-fold increase in the
number of neurons in a primate brain results in an increase in
brain size of only 11-fold.
NOT ALL BRAINS ARE CREATED EQUAL: COGNITIVE ABILITIES AND
NUMBERS OF NEURONS
The different cellular scaling rules that apply to rodent, primate
and insectivore brains show very clearly that brain size cannot be
used indiscriminately as a proxy for numbers of neurons in the
brain, or even in a brain structure, across orders. By maintaining
the average neuronal size (including all arborizations) invariant
November 2009 | Volume 3 | Article 31 | 4
 Herculano-Houzel
FIGURE 3 | Brain mass and total number of neurons for the mammalian
species examined so far with the isotropic fractionator. Brains are arranged
from left to right, top to bottom, in order of increasing number of neurons
according to average species values from Herculano-Houzel et al., 2006
(rodents), Herculano-Houzel et al., 2007 (non-human primates), Sarko et al.,
2009 (insectivores) and Azevedo et al., 2009 (human brain). Rodent brains face
right, primate brains face left, insectivore brains can be identified in the figure by
Frontiers in Human Neuroscience www.frontiersin.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
The human brain in numbers
their bluish hue (due to illumination conditions). All images shown to the same
scale. Primate images, except for the capuchin monkey and human brain, from
the University of Wisconsin and Michigan State Comparative Mammalian Brain
Collections (www.brainmuseum.org). Insectivore images kindly provided by
Diana Sarko, and human brain image by Roberto Lent. Rodent images from the
author. Notice that some rodent brains, such as the agouti and the capybara,
contain fewer neurons than primate brains that are smaller than them.
November 2009 | Volume 3 | Article 31 | 5
 Herculano-Houzel The human brain in numbers

as brain size changes, primate brains scale in size in a much more larger-brained capybara (3.7 billion against 1.6 billion), and also
space-saving, economical manner compared to the inflationary about four times more neurons in the cerebral cortex (1.1 billion
growth that occurs in rodents, in which larger numbers of neurons against 0.3).
are accompanied by larger neurons. The significance of the difference in scaling rules for building
The cognitive consequences of this difference, which allows brains with large numbers of neurons becomes even more obvi-
primate brains to enjoy the benefits of a large increase in numbers ous if one considers the expected number of neurons for a generic
of neurons without the otherwise associated cost of a much larger rodent brain of human-sized proportions, weighing 1.5 kg: such
increase in overall brain volume, can be glimpsed by returning to a brain would have only 12 billion neurons, and a much larger
the comparison between rodents and primates of similar brain size. number of 46 billion non-neuronal cells. This number of neurons
Now that absolute numbers of neurons can be compared across is smaller than the number of neurons estimated to exist in the
the similar-sized brains of agoutis and owl monkeys, and of capy- human cerebral cortex alone (Pakkenberg and Gundersen, 1997;
baras and capuchin monkeys (Figure 4), the expected correlation Pelvig et al., 2008), and about seven times smaller than the number
between cognitive ability and numbers of neurons is actually found of neurons predicted for a 1.5-kg brain built with the scaling rules
to hold: with 1468 million neurons, owl monkeys have almost twice that apply to primates (see below).
as many neurons in the brain as agoutis (which hold 857 million),
and about four times more neurons in the cerebral cortex than the THE CELLULAR COMPOSITION OF THE HUMAN BRAIN
agouti (442 million versus 113 million). Likewise, the capuchin The determination of the cellular scaling rules that apply to pri-
monkey brain has more than twice the number of neurons of the mate brains (Herculano-Houzel et al., 2007) enabled us to predict
the cellular composition of the human brain. According to these
Table 1 | Power law exponents that apply to the scaling of brain mass, rules, a generic primate brain of 1.5 kg should have 93 billion
or structure mass, as a function of the number of neurons they contain neurons, and 112 billion non-neuronal cells: glial cells, thus, should
in rodents, insectivores and primates. constitute at most half of all brain cells. This generic primate brain
should have a cerebral cortex of about 1.4 kg, containing 25 bil-
Rodents Insectivores Primates lion neurons, and a cerebellum weighing 120 g, with 70 billion
neurons (Table 2).
Brain mass × neurons MBR ∼ NBR1.550 MBR ∼ NBR1.016 MBR ∼ NBR1.056
Establishing whether the human brain indeed conforms to
Cortical mass × neurons MCX ∼ NCX1.744 MCX ∼ NCX1.520 MCX ∼ NCX1.077
the scaling rules that apply to other primates, however, required
Cerebellar mass × neurons MCB ∼ NCB1.372 MCB ∼ NCB1.028 MCB ∼ NCB0.990
determining its cellular composition using the same method. This
Remaining areas MRA ∼ NRA1.153 MRA ∼ NRA0.926 MRA ∼ NRA1.013
was accomplished by Azevedo et al. (2009), who found that the
mass × neurons
adult male human brain, at an average of 1.5 kg, has 86 billion
neurons and 85 billion non-neuronal cells – numbers that devi-
Data are from Herculano-Houzel et al. (2006), Sarko et al. (2009) and Herculano-
Houzel et al. (2009). Scaling laws for primate brains do not include human ate from the expected by 7 and 24% only. The human cerebral
values. cortex, with an average 1233 g and 16 billion neurons, is slightly

FIGURE 4 | Brain size is not a reliable indicator of number of neurons across orders. Because of the different cellular scaling rules that apply to rodent and
primate brains, primates always concentrate larger numbers of neurons in the brain than rodents of a similar, or even larger, brain size. Data from Herculano-Houzel
et al. (2006, 2007). Illustration by Lorena Kaz.

Frontiers in Human Neuroscience www.frontiersin.org November 2009 | Volume 3 | Article 31 | 6

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Herculano-Houzel The human brain in numbers

Table 2 | Expected values for a generic rodent and primate brains of 1.5 kg, and values observed for the human brain (Azevedo et al., 2009).

Generic rodent brain Generic primate brain Human brain

Brain mass 1500 g 1500 g 1508 g

Total number of neurons in brain 12 billion 93 billion 86 billion
Total number of non-neurons in brain 46 billion 112 billion 85 billion
Mass, cerebral cortex 1154 g 1412 g 1233 g
Neurons, cerebral cortex 2 billion 25 billion 16 billion
Relative size of the cerebral cortex 77% of brain mass 94% of brain mass 82% of brain mass
Relative number of neurons in cerebral cortex 17% of brain neurons 27% of brain neurons 19% of brain neurons
Mass, cerebellum 133 g 121 g 154 g
Neurons, cerebellum 10 billion 61 billion 69 billion
Relative size of the cerebellum 9% of brain mass 8% of brain mass 10% of brain mass

Notice that although the expected mass of the cerebral cortex and cerebellum are similar for these hypothetical brains, the numbers of neurons that they
contain are remarkably different. The human brain thus exhibits seven times more neurons than expected for a rodent brain of its size, but 92% of what would
be expected of a hypothetical primate brain of the same size. Expected values were calculated based on the power laws relating structure size and number of
neurons (irrespective of body size) that apply to average species values for rodents (Herculano-Houzel et al., 2006) and primate brains (Herculano-Houzel et al.,
2007), excluding the olfactory bulb.

below expectations for a primate brain of 1.5 kg, while the human
cerebellum, at 154 g and 69 billion neurons, matches or even slightly
exceeds the expected (Table 2).
Although not observed in the comparatively small rodent species
analyzed, the enlargement of the cerebral cortex is not, in principle,
an exclusive feature of the human brain: a similar expansion of the
mass of the cerebral cortex, relative to the whole brain, is predicted
by both the rodent and primate cellular scaling rules, irrespec-
tive of the number of neurons contained in the cortex (Table 2).
Remarkably, the human cerebral cortex, which represents 82% of
brain mass, holds only 19% of all neurons in the human brain – a
fraction that is similar to the fraction that we observed in several
other primates, rodents, and even insectivores (Figure 1). The
relatively large human cerebral cortex, therefore, is not different
from the cerebral cortex of other animals in its relative number
of neurons.
It should be noted that the unchanging proportional number
of neurons in the cerebral cortex relative to the whole brain does
not contradict an expansion in volume, function and number of
neurons of the cerebral cortex in evolution: the absolute number
of neurons in the rodent and primate cerebral cortex does increase
much faster in larger brains compared to the number of neurons
in the combined brainstem, diencephalon and basal ganglia, and is
accompanied by a similarly fast increase in the number of neurons
in the cerebellum (Figure 5).
Because of the diverging power laws that relate brain size and
number of neurons across rodents and primates, the latter can
hold more neurons in the same brain volume, with larger neuronal
densities than found in rodents. Since neuronal density does not
scale with brain size in primates, but decreases with increasing
brain size in rodents, the larger the brain size, the larger is the
difference in number of neurons across similar-sized rodent and
primate brains.
PREDICTIONS FOR GREAT APES
The finding that the same cellular scaling rules apply to humans
and non-anthropoid primate brains alike, irrespective of body size,
indicates that the brains of the great apes, which diverged from the
hominin lineage before humans, should also conform to the same
cellular scaling rules. An examination of the cellular composition
of the cerebellum of orangutans and one gorilla shows that the
sizes of the cerebellum and cerebral cortex predicted for these spe-
cies from the number of cells in the cerebellum match their actual
sizes, which suggests that the brain of these animals indeed is built
according to the same scaling rules that apply to humans and other
primates (Herculano-Houzel and Kaas, in preparation). In view of
the discrepant relationship between body and brain size in humans,
great apes, and non-anthropoid primates, these findings suggest
that the rules that apply to scaling primate brains are much more
conserved than those that apply to scaling the body. This raises the
possibility that brain mass and body mass across species are only
correlated, rather than brain mass being determined by body mass,
as presumed in studies that focus on the variation of residuals after
regression onto body size. Supportive evidence comes from the
dissociation between brain and body growth in development, in
which the former actually precedes the latter (reviewed in Deacon,
1997), and from our observation that body mass seems more free to
vary across species than brain mass as a function of its number of
neurons. In this view, it will be interesting to consider the alternative
hypothesis that body size is not a determining variable for brain
size in comparative studies of brain neuroanatomy, and particularly
not an (independent) parameter for assessing quantitative aspects
of the human brain.
DO WE HAVE THE MOST NEURONS? PREDICTIONS FOR OTHER LARGE-
BRAINED MAMMALS
The different cellular scaling rules that apply to rodents and pri-
mates strongly indicate that it is not valid to use brain size as a proxy
for number of neurons across humans, whales, elephants and other
large-brained species belonging to different mammalian orders.
One consequence of this realization is that sheer size alone, or in
relation to body size, is not an adequate parameter to qualify, or
disqualify, the human brain as “special”.
A comparison of expected numbers can nevertheless be very illu-
minating. For instance, given the cellular scaling rules that we have
observed for rodents (Herculano-Houzel et al., 2006), a hypothetical

Frontiers in Human Neuroscience www.frontiersin.org November 2009 | Volume 3 | Article 31 | 7

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Herculano-Houzel
FIGURE 5 | Numbers of neurons increase faster in the cerebral cortex and cerebellum than in the remaining brain areas (the combined brainstem,
diencephalon and basal ganglia). Data points indicate average values for individual species of rodents (Herculano-Houzel et al., 2006), primates (Herculano-Houzel
et al., 2007), including humans (Azevedo et al., 2009), and insectivores (Sarko et al., 2009).
rodent brain with 86 billion neurons, like the human brain, would
be predicted to weigh overwhelming 35 kg – a value that is way
beyond the largest known brain mass of 9 kg for the blue whale,
and probably physiologically unattainable. As mentioned above, a
generic rodent brain of human-sized proportions, weighing 1.5 kg,
would have only 12 billion neurons: in this sense, therefore, being
a primate endows us with seven times more neurons than would
be expected if we were rodents. Notice that this remarkable differ-
ence does not rely on assumptions about how brain size or cellular
composition relate to body size in the species.
A burning question is now whether cetaceans and elephants,
endowed with much larger brains than humans, also have much
larger numbers of neurons than humans. According to one estimate,
the false killer whale and the African elephant would have about 11
billion neurons in the cerebral cortex, despite their large size – and
fewer neurons than the 11.5 billion estimated by the same method
for the human cerebral cortex, though only marginally so (Roth and
Dicke, 2005). These estimates, however, were obtained by simply
multiplying cerebral cortical volume and the neuronal densities
determined for a few cortical areas, which probably do not reflect
average neuronal density in the entire cortex.
Although direct measurements of cellular composition are not
yet available from whole elephant and whale brains, it is illumi-
nating to consider how their cellular compositions would differ
depending on whether predicted from the scaling rules that apply
to rodent or to primate brains. As shown in Table 3, the difference
in numbers of neurons predicted to compose the brains of the false
killer whale and of the African elephant is 10-fold depending on
the scaling rules employed. Speculatively, the estimate of neuronal
density in the gray matter of the cerebral cortex of the whale and the
elephant at a low figure of about 7000 neurons/mm3 (Tower, 1954)
suggests that these brains conform to scaling rules that are much
closer to those that apply to rodents than to the primate scaling
rules. It may turn out, therefore, these very large brains are com-
posed of remarkably fewer neurons than the human brain, despite
their size, thanks to the distinct, economical scaling rules that apply
to primates in general (and not to humans in particular).
Frontiers in Human Neuroscience
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
The human brain in numbers
THE HUMAN BRAIN IS A LINEARLY SCALED-UP PRIMATE
BRAIN IN ITS NUMBER OF NEURONS. WHAT NOW?
COGNITIVE ABILITIES, BRAIN SIZE AND NUMBER OF NEURONS
To conclude that the human brain is a linearly scaled-up primate
brain, with just the expected number of neurons for a primate
brain of its size, is not to state that it is unremarkable in its capabili-
ties. However, as studies on the cognitive abilities of non-human
primates and other large-brained animals progress, it becomes
increasingly likely that humans do not have truly unique cognitive
abilities, and hence must differ from these animals not qualita-
tively, but rather in the combination and extent of abilities such
as theory of mind, imitation and social cognition (Marino et al.,
2009). Quantitative changes in the neuronal composition of the
brain could therefore be a main driving force that, through the
exponential combination of processing units, and therefore of
computational abilities, leads to events that may look like “jumps”
in the evolution of brains and intelligence (Roth and Dicke, 2005).
Such quantitative changes are likely to be warranted by increases
in the absolute (rather than relative) numbers of neurons in rel-
evant cortical areas and, coordinately, in the cerebellar circuits
that interact with them (Ramnani, 2006). Moreover, viewing the
human brain as a linearly scaled-up primate brain in its cellular
composition does not diminish the role that particular neuro-
anatomical arrangements, such as changes in the relative size of
functional cortical areas (for instance, Semendeferi et al., 2001;
Rilling and Seligman, 2002), in the volume of prefrontal white
matter (Schoenemann et al., 2005) or in the size of specific por-
tions of the cerebellum (Ramnani, 2006) may play in human
cognition. Rather, such arrangements should contribute to brain
function in combination with the large number of neurons in the
human brain. Our analysis of numbers of neurons has so far been
restricted to large brain divisions, such as the entire cerebral cortex
and the ensemble of brainstem, diencephalon and basal ganglia,
but an analysis of the cellular scaling of separate functional corti-
cal areas and the related subcortical structures is underway. Such
data should allow us to address important issues such as mosaic
evolution through concerted changes in the functionally related
www.frontiersin.org November 2009 | Volume 3 | Article 31 | 8
 Herculano-Houzel The human brain in numbers

Table 3 | Predicted cellular composition of whale and elephant brains if they scaled according to rodent or primate cellular scaling rules.

Predicted from rodent rules Predicted from primate rules

FALSE KILLER WHALE, 3650 G

Neurons, whole brain 21 billion 212 billion
Mass, cerebral cortex 3000 g 3655 g
Neurons, cerebral cortex 3 billion 55 billion
Neuronal density, cortex 1000 neurons/mg* 30–80,000 neurons/mg**
Mass, cerebellum 304 g 279 g
Neurons, cerebellum 19 billion 140 billion
Neuronal density, cerebellum 63,500 neurons/mg* 400–600,000 neurons/mg*
AFRICAN ELEPHANT, 4200 G
Neurons, whole brain 23 billion 241 billion
Mass, cerebral cortex 3488 g 4245 g
Neurons, cerebral cortex 3 billion 62 billion
Neuronal density, cortex 960 neurons/mg* 30–80,000 neurons/mg***
Mass, cerebellum 347 g 318 g
Neurons, cerebellum 21 billion 159 billion
Neuronal density, cerebellum 61,200 neurons/mg* 400–600,000 neurons/mg*

Neuronal densities (*) predicted from the rodent scaling rules apply to the whole structures, including white matter. Neuronal densities predicted from the primate
scaling rules are the range observed in primate gray matter (**) (Herculano-Houzel et al., 2008), and in the primate cerebellum including white matter (***), since
neuronal density does not covary with structure size in primates (Herculano-Houzel et al., 2007). Notice the difference in predicted numbers of neurons depending
on the scaling rules applied. Given the low neuronal densities observed in the whale and elephant gray matter, of about 7000 neurons/mm3 of gray matter, it is
reasonable to speculate that the scaling rules that apply to whale and elephant brains are closer to the rules that apply to rodent brains than to the rules that apply to
primate brains. Notice also that the actual size of the elephant cerebellum, at about 1 kg (Hakeem et al., 2005), is much larger than the predicted here.

components of distributed systems, and the presumed increase in
relative number of neurons in systems that increase in importance
(Barton and Harvey, 2000; Barton, 2006).
If cognitive abilities among non-human primates scale with abso-
lute brain size (Deaner et al., 2007) and brain size scales linearly across
primates with its number of neurons (Herculano-Houzel et al., 2007),
it is tempting to infer that the cognitive abilities of a primate, and of
other mammals for that matter, are directly related to the number
of neurons in its brain. In this sense, it is interesting to realize that,
if the same linear scaling rules are considered to apply to great apes
as to other primates, then similar three-fold differences in brain size
and in brain neurons alike apply to humans compared to gorillas,
and to gorillas compared to baboons. This, however, is not to say
that any cognitive advantages that the human brain may have over
the gorilla and that the gorilla may have over the baboon are equally
three-fold – although these differences are difficult to quantify. Since
neurons interact combinatorially through the synapses they establish
with one another, and further so as they interact in networks, the
increase in cognitive abilities afforded by increasing the number of
neurons in the brain can be expected to increase exponentially with
absolute number of neurons, and might even be subject to a threshold-
ing effect once critical points of information processing are reached.
In this way, the effects of a three-fold increase in numbers of neurons
may be much more remarkable when comparing already large brains,
such as those of humans and gorillas, than when comparing small
brains, such as those of squirrel monkeys and galagos.
INTRASPECIFIC VARIABILITY IN SIZE, NUMBERS AND ABILITIES
One final caveat to keep in mind when studying scaling of numbers
of brain neurons, particularly in regard to cognition, is that rela-
tionships observed across species need not apply to comparisons
across individuals of the same species. Not only the extent of
intraspecific variation is much smaller (on the order of 10–50%)
than interspecific variation (which spans five orders of magnitude
within mammals; Tower, 1954; Stolzenburg et al., 1989), but also the
mechanisms underlying interspecific and intraspecific variation are
also likely to differ. Our own preliminary data suggest that, indeed,
variations in brain size across rats of the same age are not correlated
with variations in numbers of neurons (Morterá and Herculano-
Houzel, unpublished observations). There is no justification, there-
fore, to extend the linear correlation between brain size and number
of neurons across primates to a putative correlation across persons
of different brain sizes (which might be used, inappropriately, as
grounds for claims that larger-brained individuals have more neu-
rons, and are therefore “smarter”, than smaller-brained persons).
In fact, although men have been reported to have more neurons in
the cerebral cortex than women (Pakkenberg and Gundersen, 1997;
Pelvig et al., 2008), there is no significant correlation between brain
size and general cognitive ability within families (Schoenemann
et al., 2000). Across these individuals, other factors such as varia-
tions in number and identity of synaptic connections within and
across structures, building on a statistically normal, albeit variable,
number of neurons, and depending on genetics and life experiences
such as learning, are more likely to be determinant of the individual
cognitive abilities (see, for instance, Mollgaard et al., 1971; Black
et al., 1990; Irwin et al., 2000; Draganski et al., 2004).
CONCLUDING REMARKS: OUR PLACE IN NATURE
Novel quantitative data on the cellular composition of the human
brain and its comparison to other primate brains strongly indicate that
we need to rethink our notions about the place that the human brain
holds in nature and evolution, and rewrite some of the basic concepts

Frontiers in Human Neuroscience www.frontiersin.org November 2009 | Volume 3 | Article 31 | 9

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Herculano-Houzel The human brain in numbers

that are taught in textbooks. Accumulating evidence (Deacon, 1997;
Roth and Dicke, 2005; Deaner et al., 2007) indicates that an alterna-
tive view of the source of variations in cognitive abilities across species
merits investigation: one that disregards body and brain size and
examines absolute numbers of neurons as a more relevant parameter
instead. Now that these numbers can be determined in various brains
and their structures, direct comparisons can be made across species
and orders, with no assumptions about body–brain size relationships
required. Complementarily, however, it now becomes possible to
examine how numbers of neurons in the brain, rather than brain size,
relate to body mass and surface as well as metabolism, parameters
that have been considered relevant in comparative studies (Martin,
1981; Fox and Wilczynski, 1986; MacLarnon, 1996; Schoenemann,
2004), in order to establish what mechanisms underlie the loosely
correlated scaling of body and brain.
According to this now possible neuron-centered view, rather
than to the body-centered view that dominates the literature (see
Gazzaniga, 2008, for a comprehensive review), the human brain
has the number of neurons that is expected of a primate brain of
its size; a cerebral cortex that is exactly as large as expected for a
primate brain of 1.5 kg; just as many neurons as expected in the
cerebral cortex for the size of this structure; and, despite having a
relatively large cerebral cortex (which, however, a rodent brain of
1.5 kg would also be predicted to have), this enlarged cortex holds
just the same proportion of brain neurons in humans as do other
primate cortices (and rodent cortices, for that matter). This final
observation calls for a reappraisal of the view of brain evolution
that concentrates on the expansion of the cerebral cortex, and its
replacement with a more integrated view of coordinate evolution
of cellular composition, neuroanatomical structure, and function
of cerebral cortex and cerebellum (Whiting and Barton, 2003).
Other “facts” that deserve updating are the ubiquitous quote
of 100 billion neurons (a value that lies outside of the margin of
variation found so far in human brains; Azevedo et al., 2009), and,
more strikingly, the widespread remark that there are 10× more glial
cells than neurons in the human brain. As we have shown, glial cells
in the human brain are at most 50% of all brain cells, which is an
important finding since it is one more brain characteristic that we
share with other primates (Azevedo et al., 2009).
Finally, if being considered the bearer of a linearly scaled-up
primate brain does not sound worthy enough for the animal that
considers himself the most cognitively able on Earth, one can note
that there are, indeed, two advantages to the human brain when
compared to others – even if it is not an outlier, nor unique in any
remarkable way. First, the human brain scales as a primate brain:
this economical property of scaling alone, compared to rodents,
assures that the human brain has many more neurons than would
fit into a rodent brain of similar size, and possibly into any other
similar-sized brain. And second, our standing among primates as
the proud owners of the largest living brain assures that, at least
among primates, we enjoy the largest number of neurons from
which to derive cognition and behavior as a whole. It will now be
interesting to determine whether humans, indeed, have the largest
number of neurons in the brain among mammals as a whole.
ACKNOWLEDGMENTS
Thanks to Roberto Lent and Jon Kaas for continued support and
encouragement, to Bruno Mota for insightful discussions, and to
the colleagues who participated in the quantification of the cel-
lular composition of the brain of various species. Supported by
CNPq-Ed. Universal and FAPERJ-Jovem Cientista do Estado do
Rio de Janeiro.

REFERENCES
Allen, N. J., and Barres, B. A. (2009). Glia –
more than just brain glue. Nature 457,
675–677.
Andersen, B. B., Korbo, L., and
Pakkenberg, B. (1992). A quantitative
study of the human cerebellum with
unbiased stereological techniques.
J. Comp. Neurol. 326, 549–560.
Azevedo, F. A., Carvalho, L. R.,
Grinberg, L. T., Farfel, J. M.,
Ferretti, R. E., Leite, R. E., Jacob
Filho, W., Lent, R., and Herculano-
Houzel, S. (2009). Equal numbers
of neuronal and nonneuronal cells
make the human brain an isometri-
cally scaled-up primate brain. J. Comp.
Neurol. 513, 532–541.
Barton, R. A. (2002). How did brains
evolve? Nature 415, 134–135.
Barton, R. A. (2006). Primate brain evo-
lution: integrating comparative, neu-
rophysiological, and ethological data.
Evol. Anthropol. 15, 224–236.
Barton, R. A., and Harvey, P. H. (2000).
Mosaic evolution of brain structure in
mammals. Nature 405, 1055–1058.
Black, J. E., Isaacs, K. R., Anderson, B. J.,
Alcantara, A. A., and Greenough, W T.
(1990). Learning causes synaptogenesis,
whereas motor activity causes angio-
genesis, in cerebellar cortex of adult
rats. Proc. Natl. Acad. Sci. USA 87,
5568–5572.
Butti, C., Sherwood, C. C., Hakeem, A. Y.,
Allman, J. M., and Hof, P. R. (2009).
Total number and volume of Von
Economo neurons in the cerebral
cortex of cetaceans. J. Comp. Neurol.
515, 243–259.
Clark, D. A., Mitra, P. P., and Wang, S. S.
(2001). Scalable architecture in mam-
malian brains. Nature 411, 189–193.
Deacon, T. W. (1997). What makes the
human brain different? Annu. Rev.
Anthropol. 26, 337–357.
Deaner, R. O., Isler, K., Burkart, J., and
van Schaik, C. (2007). Overall brain
size, and not encephalization quotient,
best predicts cognitive ability across
non-human primates. Brain Behav.
Evol. 70, 115–124.
Doetsch, F. (2003). The glial identity of
neural stem cells. Nat. Neurosci. 6,
1127–1134.
Draganski, B., Gaser, C., Busch, V.,
Schuierer, G., Bogdahn, U., and
May, A. (2004). Changes in grey
matter induced by training. Nature
427, 311–312.
Finlay, B. L., and Darlington, R. B. (1995).
Linked regularities in the development
and evolution of mammalian brains.
Science 268, 1578–1584.
Fox, J. H., and Wilczynski, W. (1986).
Allometry of major CNS divisions:
towards a reevaluation of somatic
brain-body scaling. Brain Behav. Evol.
28, 157–169.
Friede, R. (1954). Quantitative share of
the glia in development of the cortex.
Acta Anat. 20, 290–296.
Gazzaniga, M. (2008). Human: The
Science Behind What Makes us
Unique. New York, Harper Collins.
Hakeem, A. Y., Hof, P. R., Sherwood, C. C.,
Switzer R. C. III, Rasmussen, L. E. L.,
and Allman, J. M. (2005). Brain of
the African elephant (Loxodonta
africana): neuroanatomy from mag-
netic resonance images. Anat. Rec.
287A, 1117–1127.
Hakeem, A. Y., Sherwood, C. C.,
Bonar, C. J., Butti, C., Hof, P. R., and
Allman, J. M. (2009). Von Economo
neurons in the elephant brain. Anat.
Rec. 292, 242–248.
Harvey, P. H., and Pagel, M. D. (1991).
Modeling physiological and anthro-
pometric variables known to vary
with body size and other confound-
ing variables. Yearb. Phys. Anthropol.
48, 141–153.
Haug, H. (1987). Brain sizes, surfaces, and
neuronal sizes of the cortex cerebri:
a stereological investigation of man
and his variability and a compari-
son with some mammals (primates,
whales, marsupials, insectivores,
and one elephant). Am. J. Anat. 180,
126–142.
Hawking, A., and Olszewski, J. (1957).
Glia/nerve cell index for cortex of the
whale. Science 126, 76–77.
Herculano-Houzel, S. (2002). Do you
know your brain? A survey on public
neuroscience literacy at the closing of
the decade of the brain. Neuroscientist
8, 98–110.
He rc u l a n o - Ho u z e l , S . ( 2 0 0 7 ) .
Encephalization, neuronal excess, and
neuronal index in rodents. Anat. Rec.
290, 1280–1287.
Herculano-Houzel, S., Collins, C. E.,
Wong, P., and Kaas, J. H. (2007).
Cellular scaling rules for primate

Frontiers in Human Neuroscience www.frontiersin.org November 2009 | Volume 3 | Article 31 | 10

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Herculano-Houzel
brains. Proc. Natl. Acad. Sci. USA 104,
3562–3567.
Herculano-Houzel, S., Collins, C. E.,
Wong, P., Kaas, J. H., and Lent, R.
(2008). The basic nonuniformity of
the cerebral cortex. Proc. Natl. Acad.
Sci. USA 105, 12593–12598.
Herculano-Houzel, S., and Lent, R. (2005).
Isotropic fractionator: a simple, rapid
method for the quantification of total
cell and neuron numbers in the brain.
J. Neurosci. 25, 2518–2521.
Herculano-Houzel, S., Mota, B., and
Lent, R. (2006). Cellular scaling rules
for rodent brains. Proc. Natl. Acad. Sci.
USA 103, 12138–12143.
Ir w in, S. A., Galvez, R., and
Greenough, W. T. (2000). Dendritic
spine structural anomalies in fragile-
X mental retardation syndrome. Cereb.
Cortex 10, 1038–1044.
Jerison, H. (1973). Evolution of the Brain
and Intelligence. New York, Academic
Press.
Jerison, H. (1985). Animal intelligence as
encephalization. Philos. Trans. R. Soc.
Lond., B, Biol. Sci. 308, 21–35.
Kandel, E. R., Schwartz, J. H., and
Jessell, T. M. (2000). Principles of
Neural Science, 4th Edn. New York,
McGraw-Hill, pp. 19–20.
Lange, W. (1975). Cell number and cell
density in the cerebellar cortex of man
and some other mammals. Cell Tissue
Res. 157, 115–125.
Lefebvre, L., Reader, S. M., and Sol, D.
(2004). Brains, innovations and evo-
lution in birds and primates. Brain
Behav. Evol. 63, 233–246.
MacDonald, D. W. (1981). Dwindling
resources and the social behavior
of capybaras (Hydrochaeris hydro-
chaeris) (Mammalia). J. Zool. Lond.
194, 371–391.
MacLarnon, A. (1996). The scaling of
gross dimensions of the spinal cord
in primates and other species. J. Hum.
Evol. 30, 71–87.
Marino, L. (1998). A comparison of
encephalization between odontocete
cetaceans and anthropoid primates.
Brain Behav. Evol. 51, 230–238.
Marino, L. (2002). Convergence of com-
plex cognitive abilities in cetaceans
and primates. Brain Behav. Evol. 59,
21–32.
Marino, L., Connor, R. C., Fordyce, R. E.,
Herman, L. M., Hof, P. R., Lefebvre,
Frontiers in Human Neuroscience
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
L., Lusseau, D., McCowan, B.,
Nimchinsky, E. A., Pack, A. A., Rendell,
L., Reidenberg, J. S., Reiss, D., Uhen, M.
D., Van der Gucht, E., and Whitehead,
H. (2009). Cetaceans have complex
brains for complex cognition. PLoS
Biol. 5:e139. doi:10.1371/journal.
pbio.0050139
Martin, R. D. (1981). Relative brain size
and basal metabolic rate in terrestrial
vertebrates. Nature 293, 57–60.
Mollgaard, K., Diamond, M. C.,
Bennett, E. L., Rosenzweig, M. R.,
and Lindner, B. (1971). Quantitative
synaptic changes with differential
experience in rat brain. Int. J. Neurosci.
2, 113–127.
Nimchinsky, E. A., Glissen, E.,
Allman, J. M., Perl, D. P., Erwin, J. M.,
and Hof, P. R. (1999). A neuronal mor-
phologic type unique to humans and
great apes. Proc. Natl. Acad. Sci. USA
96, 5268–5273.
Nishiyama, A., Yang, Z., and Butt, A.
(2005). What’s in a name? J. Anat.
207, 687–693.
Noctor, S. C., Martinez-Cerdeno, V., and
Kriegstein, A. R. (2007). Contribution
of intermediate progenitor cells to
cortical histogenesis. Arch. Neurol.
64, 639–642.
Nunn, C., and Barton, R. A. (2000).
Allometric slopes and independ-
ent contrasts: a comparative test of
Kleiber’s law in primates. Am. Nat.
156, 519–533.
Pakkenberg, B., and Gundersen, H. J. G.
(1988). Total numbers of neurons
and glial cells in human brain nuclei
estimated by the disector and the frac-
tionator. J. Microsc. 150, 1–20.
Pakkenberg, B., and Gundersen, H. J. G.
(1997). Neocortical neuron number in
humans: effect of sex and age. J. Comp.
Neurol. 384, 312–320.
Pelvig, D. P., Pakkenberg, H., Stark, A K.,
and Pakkenberg, B. (2008). Neocortical
glial cell numbers in human brains.
Neurobiol. Aging 29, 1754–1762.
Ramnani, N. (2006). The primate cortico-
cerebellar system: anatomy and func-
tion. Nat. Neurosci. 7, 511–522.
Reichenbach, A. (1989). Glia:neuron
index: review and hypothesis to
account for different values in vari-
ous mammals. Glia 2, 71–77.
Rilling, J. K. (2006). Human and
nonhuman primate brains: are they
www.frontiersin.org
allometrically scaled versions of the
same design? Evol. Anthropol. 15,
65–77.
Rilling, J. K., and Seligman, R. A. (2002).
A quantitative morphometric compar-
ative analysis of the primate temporal
lobe. J. Hum. Evol. 42, 505–533.
Roth, G., and Dicke, U. (2005). Evolution
of the brain and intelligence. Trends
Cogn. Sci. 9, 250–257.
Sarko, D. K., Catania, K. C., Leitch, D. B.,
Kaas, J. H., and Herculano-Houzel, S.
(2009). Cellular scaling rules of insec-
tivore brains. Front. Neuroanat. 3, 8.
doi:10.3389/neuro.05.008.2009
Schoenemann, P. T. (2004). Brain size
scaling and body composition in
mammals. Brain Behav. Evol. 63,
47–60.
Schoenemann, P. T., Budinger, T. F., Sarich,
V. M., and Wang, W. S. (2000). Brain
size does not predict general cogni-
tive ability within families. Proc. Natl.
Acad. Sci. USA 97, 4932–4937.
Schoenemann, P. T., Sheehan, M. J., and
Glotzer, L. D. (2005). Prefrontal white
matter volume is disproportionately
larger in humans than in other pri-
mates. Nat. Neurosci. 8, 242–252.
Semendeferi, K., Armostrong, E.,
Schleicher, A., Zilles, K., and Van
Hoesen, G. W. (2001). Prefrontal cor-
tex in humans and apes: a comparative
study of area 10. Am. J. Phys. Anthropol.
114, 224.
Semendeferi, K., Lu, A., Schenker, N., and
Damasio, H. (2002). Humans and
great apes share a large frontal cortex.
Nat. Neurosci. 5, 272–276.
Shariff, G. A. (1953). Cell counts in the pri-
mate cerebral cortex. J. Comp. Neurol.
98, 381–400.
Sherwood, C. C., Stimpson, C. D., Raghanti,
M. A., Wildman, D. E., Uddin, M.,
Grossman, L. I., Goodman, M.,
Redmond, J. C., Bonar, C. J., Erwin, J.
M., and Hof, P. R. (2006). Evolution
of increased glia-neuron ratios in the
human frontal cortex. Proc. Natl. Acad.
Sci. USA 103, 13606–13611.
Stolzenburg, J. U., Reichenbach, A., and
Neumann, M. (1989). Size and density
of glial and neuronal cells within the
cerebral neocortex of various insec-
tivorian species. Glia 2, 78–84.
Sultan, F. (2002). Analysis of mamma-
lian brain architecture. Nature 415,
133–134.
November 2009 | Volume 3 | Article 31 | 11
The human brain in numbers
Tower, D. B. (1954). Structural and func-
tional organization of mammalian
cerebral cortex: the correlation of
neurone density with brain size;
cortical neurone density in the fin
whale (Balaenoptera physalus L.) with
a note on the cortical neurone den-
sity in the Indian elephant. J. Comp.
Neurol. 101, 19–51.
Tower, D. B., and Elliot, K. A. (1952).
Activity of acetylcholine system in
cerebral cortex of various unanesthe-
tized mammals. Am. J. Physiol. 168,
747–759.
Ullian, E. M., Sapperstein, S. K.,
Christopherson, K. S., and Barres, B. A.
(2001). Control of synapse number by
glia. Science 291, 657–660.
Vallender, E. J. (2008). Exploring the
origins of the human brain through
molecular evolution. Brain Behav.
Evol. 72, 168–177.
Whiting, B. A., and Barton, R. A. (2003).
The evolution of the cortico-cerebellar
complex in primates: anatomical con-
nections predict patterns of correlated
evolution. J. Hum. Evol. 44, 3–10.
Williams, R. W., and Herrup, K. (1988).
The control of neuron number. Annu.
Rev. Neurosci. 11, 423–453.
Zhang, K., and Sejnowski, T. J. (2000).
A universal scaling law between gray
matter and white matter of cerebral
cortex. Proc. Natl. Acad. Sci. USA 97,
5621–5626.
Conflict of Interest Statement: The author
declares that the research was conducted in
the absence of any commercial or financial
relationships that could be construed as a
potential conflict of interest.
Received: 15 July 2009; paper pending
published: 05 August 2009; accepted: 29
September 2009; published online: 09
November 2009.
Citation: Herculano-Houzel S (2009) The
human brain in numbers: a linearly scaled-
up primate brain. Front. Hum. Neurosci.
3:31. doi: 10.3389/neuro.09.031.2009
Copyright © 2009 Herculano-Houzel. This
is an open-access article subject to an exclu-
sive license agreement between the authors
and the Frontiers Research Foundation,
which permits unrestricted use, distribu-
tion, and reproduction in any medium,
provided the original authors and source
are credited.
 Poo et al. BMC Biology (2016)4:
DOI 10.1186/s12915-016-0261-6

FORUM Open Access

What is memory? The present state of

the engram
Mu-ming Poo1*, Michele Pignatelli2, Tomás J. Ryan2,3, Susumu Tonegawa2,3*, Tobias Bonhoeffer4, Kelsey C. Martin5,
Andrii Rudenko6, Li-Huei Tsai6, Richard W. Tsien7, Gord Fishell7, Caitlin Mullins7, J. Tiago Gonçalves8,
Matthew Shtrahman8, Stephen T. Johnston8, Fred H. Gage8, Yang Dan9, John Long7, György Buzsáki7
and Charles Stevens8

connections due to correlated activities during memory ac-

Abstract
The mechanism of memory remains one of the great
unsolved problems of biology. Grappling with the
question more than a hundred years ago, the German
zoologist Richard Semon formulated the concept of
the engram, lasting connections in the brain that
result from simultaneous “excitations”, whose precise
physical nature and consequences were out of reach
of the biology of his day. Neuroscientists now have
the knowledge and tools to tackle this question,
however, and this Forum brings together leading
contemporary views on the mechanisms of memory
and what the engram means today.
The cellular basis of memory
Mu-ming Poo
Neurobiological studies of memory over the past century
have progressed along two relatively independent lines of
inquiry: the top-down approach examines the animal’s be-
haviors associated with memory acquisition, consolidation,
and retrieval, as well as the brain regions underlying these
processes, whereas the bottom-up approach explores the
cellular and circuit mechanisms of memory encoding and
storage by examining the patterns of neuronal firing and the
efficacy of synaptic transmission. In his monumental treatise
[1] The Organization of Behavior (1949), Donald Hebb
made a bold attempt to link these two lines of inquiry by
postulating that perceptual memory resides in specific “cell
assemblies” formed by the strengthening of interneuronal
* Correspondence: mpoo@ion.ac.cn;
1
Institute of Neuroscience, Chinese Academy of Sciences, Shanghai, China
2
RIKEN-MIT Center for Neural Circuit Genetics at the Picower Institute for
Learning and Memory, Department of Biology and Department of Brain and
Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA
02139, USA
Full list of author information is available at the end of the article
quisition. The discovery of activity-induced long-term po-
tentiation (LTP) and long-term depression (LTD) of central
synapses in the 1970s and 80s further sparked the interest of
a whole generation of neurobiologists in studying synaptic
plasticity and its relationship to memory. There is now gen-
eral consensus that persistent modification of the synaptic
strength via LTP and LTD of pre-existing connections
represents a primary mechanism for the formation of mem-
ory engrams. In addition, LTP and LTD could also lead to
the formation of new and elimination of old synapses and
thus changes in structural connectivity in the brain. Indeed,
early development of neural circuits, whereby neural activity
sculpts synaptic connectivity [2], depends on processes simi-
lar to that associated with LTP and LTD in the adult brain
and could be considered as the imprinting of memory en-
grams generated by early experience.
In this Forum, a group of experts on the cellular mecha-
nisms of memory were invited to present their views on
“what is memory”, including where and how memory en-
grams are stored, consolidated, and retrieved. Drawing on
an elegant set of studies, Michele Pignatelli, Tomás Ryan,
and Susumu Tonegawa illustrate how recently developed
techniques to tag and manipulate neurons have begun to
establish a causal link between neuronal activity, persistent
synaptic changes, and an animal’s memory-associated be-
haviors. The theme of persistent synaptic changes and their
causal role in memory is taken up by Tobias Bonhoeffer,
who summarizes the evidence that dendritic spines, where
excitatory synapses are located, represent the basic cellular
unit for memory; long-term memory is stored in a set of
spines that are formed or modified during learning and
these changes may persist throughout the animal’s life.
Based on the findings of activity-induced transcrip-
tional activation and synapse-specific local translation of
proteins, Kelsey Martin expands on the idea that the
basic building block of memory is the synapse, where

© 2016 Poo et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Poo et al. BMC Biology (2016)4:
both pre- and postsynaptic elements together with
associated glial processes form an integral unit with an
individual identity and distinct “neighborhood”. Andrii
Rudenko and Li-Huei Tsai redirect attention to the nu-
clei of engram cells, discussing the evidence that epigen-
etic alterations of the neurons activated during memory
acquisition may be involved in the long-term retention
of memory. They propose that such epigenetic modifica-
tion represents a priming event during the initial phase
of memory formation; memory retrieval would then
trigger the expression of the primed genes, leading to
protein synthesis and synaptic modification at individual
synaptic units.
Depending on the availability of cellular resources,
immediate modifications (LTP and LTD) and long-term
turnover (formation and elimination) of individual syn-
aptic units are bound to influence other units on the
same postsynaptic cell. Richard Tsien, Gord Fishell, and
Caitlin Mullins focus on the important issue of lateral
synaptic interaction and redistribution of synaptic
strength associated with LTP and LTD, from the point of
view of cellular homeostasis as well as the normalization
and signal-to-noise ratio of neuronal activities, and
propose a conceptual scheme to address the underlying
mechanisms.
The hippocampus is unique in being a key brain re-
gion for memory formation and a region in which adult
neurogenesis occurs. Associated with hippocampus-
dependent spatial memory, Tiago Gonçalves, Matthew
Shtrahman, Stephen Johnston, and Fred Gage discuss an
intriguing new dimension in the cellular mechanisms of
memory formation, whereby continuous addition of
newborn dentate gyrus neurons in the adult hippo-
campus, with their enhanced synaptic plasticity, may
contribute significantly to establishing the engram for
spatial memory.
As proposed by David Marr in his model of
hippocampus-dependent memory [3] and supported by
many experimental and clinical studies, episodic memor-
ies are transferred after acquisition from the hippocampus
to the neocortex for long-term storage. The mechanisms
underlying the transfer and consolidation of spatial mem-
ory are discussed by John Long and György Buzsaki in the
context of hippocampal and entorhinal sharp wave-
ripples. These activity patterns occur during sleep or non-
attentive brain states and are replays of neuronal firing
sequences triggered by recent experience, for example
they can be temporally compressed replayed versions of
the sequential neuronal firing seen as the animal traverses
through a particular environment. As discussed by myself
and Yang Dan, although spike timing-dependent plasticity
could offer a synaptic mechanism for storing sequence
information with intervals up to a few hundreds of
milliseconds, it remains largely unknown how neural
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 2 of 18
circuits store and recall the temporal sequence of informa-
tion up to seconds and longer, periods often associated
with episodic memory. A compression of the temporal
sequence of events such as occurs during sharp wave-
ripples in the hippocampus and neocortex offers a
potential solution.
The contributing articles of this Forum reflect the
tremendous progress made in our understanding of the
cellular building blocks of memory. There is a clear con-
sensus on where the memory engram is stored—specific
assemblies of synapses activated or formed during mem-
ory acquisition—and a substantial body of knowledge on
how the engram is generated and maintained in the
brain. However, knowing the building blocks and their
properties is far from understanding the architecture of
the “memory palace”. As Charles Stevens indicates in his
epilogue, and the readers will soon discover, many new
territories are now open for exploration.
Engram cell connectivity as a substrate for
memory storage
Michele Pignatelli, Tomás J. Ryan, and Susumu Tonegawa
The storage of information refers to the systematic
process of collecting and cataloging data so that they
can be retrieved on request.
One of the most enlightening conceptualizations of
the neural representation of stored memory information
was developed by Richard Semon, who conceived the
Engram Theory, a theory of memory traces [4]. Accord-
ing to this theory, as fortified by contemporary know-
ledge, learning activates a small ensemble of brain cells,
inducing in these cells persistent physical/chemical
changes. In addition, reactivation of these cells by
relevant recall cues results in retrieval of the specific
memory. The theory poses an important question: what
is the nature of the persistent changes?
In his seminal book published in 1949, Donald Hebb
proposed a mechanism based on synaptic plasticity as a
substrate of memory [1]. With an example of two cells
connected by an excitatory synapse, if the activation of
one cell leads to the activation of the second one, the
connection between the two cells is reinforced, a
postulate that has been confirmed experimentally [5–8].
The increase in connectivity strength within a diffuse
group of cells in a more complex feedforward circuit re-
sults in the emergence of an engram cell ensemble.
The systematic dissection of the molecular mecha-
nisms involved in synaptic plasticity has revealed that
the cascade of events underlying the plastic changes
requires two distinct phases [9, 10]. In the encoding
phase, also known as early long-term potentiation
(E-LTP), an increase in intracellular Ca2+ concentration
mediated by post-synaptic NMDA receptors elicits a
change in synaptic weight by increasing the insertion and
 Poo et al. BMC Biology (2016)4:
the conductance of AMPA receptors [11]. The dendritic
spines that support the post-synaptic machinery rapidly
increase in number [12]. In a second phase lasting a few
hours after the initial encoding period, the increased syn-
aptic weight is maintained by a protein synthesis-
dependent process known as cellular consolidation, during
which the steady state synthesis of AMPA receptors is
shifted to a higher level. This second phase is known as
late LTP (L-LTP) [9, 10] and is sensitive to protein synthe-
sis inhibitors (PSI).
To date, memory storage has mainly been investigated
by pharmacological or molecular manipulation and by
correlating synaptic changes with the strength of memory
recall. Only recently has it become possible to specifically
tag cells activated by learning. The demonstration that
these cells are part of a memory engram ensemble was
provided by a series of “optogenetic” experiments where a
learning-induced tagging strategy was employed to
express channelrhodopsin [13] in a small population of
learning-activated cells [14, 15]. The opsin allowed for the
artificial light-induced reactivation of the cellular popula-
tion labeled during learning and resulted in memory
retrieval. The same tagging strategy also verified the
reactivation of tagged cells upon presentation of
retrieval cues.
Tagging “engram” cells offers a straightforward oppor-
tunity to investigate the nature of the persistent changes
that occurred in these cells in response to learning. In a
recent study [16], “engram cells” were compared to
“non engram cells” (non-tagged cells) by ex vivo patch
clamp recordings after contextual fear conditioning
(CFC). Engram cells displayed changes in synaptic weight
typical of LTP such as high current amplitude, insertion of
AMPA receptors, high spontaneous excitatory post-
synaptic current frequency and amplitude, and increased
dendritic spine density. These changes were blocked by
the systemic injection of PSI specifically within the
consolidation window. Therefore, it is now clear that cells
recruited by learning display synaptic changes typical of
LTP and are reactivated by retrieval cues, and their
reactivation can elicit memory recall. Remarkably, how-
ever, protein synthesis-dependent L-LTP seems to be dis-
pensable for memory storage because direct optogenetic
activation of the engram cells in PSI-injected mice elicited
full memory recall under a variety of conditions.
If L-LTP is dispensable for memory storage, what
mechanism is responsible?
An integral memory engram may consist of preferential
connectivity between engram cell ensembles distributed
across multiple brain regions. In the same report [16] it
was shown ex vivo that engram cells from the dentate
gyrus established preferential connections with engram
cells in the downstream hippocampal CA3 region in a
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 3 of 18
feedforward excitatory engram cell circuit. Remarkably,
this preferential connectivity was maintained in mice
rendered amnesic by treatment with PSI within the con-
solidation window, suggesting that memory storage may
survive retrograde amnesia in the form of a neural con-
nectivity pattern. Indeed, optogenetic stimulation of DG
engram cells in vivo elicited similar cellular reactivation
patterns not only in the CA3 region but also in the
amygdala for both control and amnesic groups, thus
confirming the persistence of engram cell connectivity.
These observations support the concept of preferential
connectivity of engram cell ensembles distributed across
multiple brain regions, which is established during learn-
ing and persists despite disruption of consolidation and
thereby provides a lasting substrate for memory storage.
These data also suggest that the synaptic potentiation
observed in consolidated engram cells is necessary for
memory retrievability and not for storage [17]. While
regulation of synaptic weight provides a scalar quantity
to control information retrieval, synaptic connectivity
holds the information specificity. This is because synap-
ses that are activated during the encoding stage will dic-
tate the eventual pattern of cellular connectivity of the
upstream and downstream engram cell ensembles. This
notion is compatible with the broad view that synapses
are the basic units of information storage (see Bonhoeffer,
this Forum).
How would the specific connectivity pattern between
engram cell ensembles be formed by specific learning?
Neural connections are formed during development and
certain circuits hold the innate capability to elicit complex
behavioral reactions in response to specific perceptual cues
[18]. However, this does not seem to be the case in the hip-
pocampal formation because inactivation of the down-
stream CA1 region before CFC results in anterograde
amnesia which cannot be bypassed by direct optogenetic
stimulation of DG engram cells [16]. Thus, the memory cir-
cuit is not configured under anterograde amnesic conditions
and is not, therefore, genetically determined but requires
hippocampal activity during memory encoding. As reported
by Ryan et al. [16], learning-induced changes in connectivity
patterns are insensitive to PSI. So is E-LTP [9, 10] and this
early phase of plasticity may provide a framework to in-
vestigate the formation of new connections. For instance,
blocking NMDA receptor function should impair the
emergence of learning-induced connectivity patterns.
During the induction of LTP, existing connections can
be potentiated [19] but new connections can also emerge
[20]. A hypothetical way this might happen is through the
activation of silent synaptic connections [21]. These
synapses expressing only NMDA receptors and not
AMPA receptors can become unsilenced through AMPA
receptor insertion, a mechanism that could, in principle,
 Poo et al. BMC Biology (2016)4: Page 4 of 18

support the formation of learning-induced changes in en-
gram cell connectivity. Another possibility is that local
dendritic protein synthesis contributes to the rapid syn-
apse formation on engram cells, independently of its role
in synaptic potentiation (see Martin, this Forum).
What then maintains the learning-induced engram cell
connectivity that is initiated during encoding?
Although learning-induced synaptic potentiation would
be suppressed by PSI, the new connectivity pattern could
persist through unsilenced synaptic connections of basal
unpotentiated strength (Fig. 1). Consistent with this per-
spective, it has been recently shown that optogenetically
induced long-term depression (LTD) of amygdala cells
impaired existing conditioned fear responses but subse-
quent optogenetically induced LTP of the same cells
could restore optogenetic cue-evoked recall of the fear
memory [22].
The ability to tag cells activated by learning has
opened up new horizons in the investigation of memory

Fig. 1. Synaptic connectivity between engram cells as a mechanism for memory storage. a Cellular connectivity in a feedforward excitatory
circuit, b synaptic configuration, c dendritic spine density, and d protein synthesis state, shown in a naïve circuit, a circuit during encoding, a
circuit after consolidation, or a circuit in an amnesic condition. Engram circuit, cells, and synapses are displayed in green, non-engram in gray. In
the naïve state, the circuit displays a variety of synaptic patterns, including strong (thick gray lines) and weak synapses (thin gray lines) as well as
silent synapses (dotted lines) exclusively expressing NMDA receptors. During encoding, a network of engram cells is recruited. The preferential
connection between engram cells occurs either by potentiation of existing connections (blue dotted circles) or by unsilencing synapses (red dotted
circles). A spine density increase supports the synaptic changes. During consolidation, the steady state synthesis of AMPA receptors is shifted to a
higher level and the disruption of consolidation with protein synthesis inhibitors (PSI) results in retrograde amnesia. However, during PSI-induced
amnesia, memory storage persists within an engram-specific set of weak synaptic connections

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Poo et al. BMC Biology (2016)4:
by revealing the role of the engram-specific connectivity
in the storage of information.
Spines and synapses as basic elements of memory
storage
Tobias Bonhoeffer
In his seminal work, Donald Hebb [1] proposed that the
basic mechanism by which memories are stored in the
brain is the enhancement of synaptic strength and, in
connection with that, morphological changes of the re-
spective synaptic contacts. In other words he proposed
that synapses and not cells are the basic building blocks
of memory, from a theoretical perspective a reasonable
suggestion as there are approximately 10,000–100,000
times more synapses in the brain than neurons.
By now it is well established that morphological
changes at the synaptic level occur in conjunction with
stimuli that are thought to mimic learning events. In
vitro experiments [12, 23] have shown that long-term
potentiation results in the addition of dendritic spines,
tiny protrusions which harbor synaptic contacts. Tens of
thousands of these spines decorate the dendrites of most
excitatory cells in the hippocampus and the neocortex.
And indeed, further studies showed that spines not only
come and go but also change their shape during putative
learning events [19], a suggestion that had been put for-
ward in a purely theoretical paper by Francis Crick [24].
So, it is well established that spines emerge, disappear,
and change with cellular events thought to underlie
learning processes. But is this merely a correlation or are
there ways towards showing that these events really lie
at the basis of learning and memory storage in the brain?
Recent experiments have made substantial progress in
this respect.
The first study that made a clear case that spines are
important for the long-term storage of information was
done in the visual cortex of mice [25]. In the visual cor-
tex it is well known that synaptic connections are estab-
lished or modified with changes in visual experience,
like the temporary closure of one eye. These plastic
changes are often used as a proxy for what happens dur-
ing memory formation since they share key features: it
is, for instance, a universally accepted fact in memory re-
search that information that has been acquired early in
life can be learned much more easily a second time, even
if it had been “completely forgotten” in the meantime.
This effect has been called “savings” [26] and is a hall-
mark of most memory processes. It has been shown that
the same effect occurs in the visual system [27]. Mice
were monocularly deprived for a couple of days early in
life so that the visual system adapted to this change of
the visual environment. Subsequently, animals were sub-
jected to normal vision again so that their visual cortex
reverted to normal function. If monocular deprivation
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 5 of 18
was then performed a second time, much later in life,
when normally this procedure has only a very limited
effect (if any), substantial adaptation still takes place
because of the early experience that the animal has had.
Importantly, this savings effect could be related to new
spines that emerged during the first plasticity episode
and persisted [25]. The fact that there was no growth of
additional spines during the second plasticity period,
while the functional adaptation occurred much faster
and more reliably, suggests that the persistent spines
facilitate the second adaptation [25]. Therefore, these
spines serve to “remember” the previous sensory experi-
ences the animals had. Two subsequent studies [28, 29]
further bolstered the case by showing that also in the
motor cortex the generation of new spines forms the
basis of learning motor tasks of different sorts. Interest-
ingly, in one of these studies [28] it was also found that
relearning a task occurred faster and did not involve the
generation of new spines, again arguing for persistent
spines “memorizing” specific motor tasks. Furthermore,
this study demonstrated that learning different tasks
involves different sets of spines, providing a strong
argument for spines and not cells being the relevant
entity for information storage in the brain.
These three papers were among the first to make a
strong case for a causal relationship between new
(or changing) spines and learning or information storage
in the brain. Subsequently, a number of studies further
strengthened this hypothesis. Some of them used fear con-
ditioning to show that also in this paradigm learning is
paralleled by structural changes: fear extinction and fear
conditioning are marked by the generation or removal of
spines in the frontal association cortex [30] and the
auditory cortex [31]. One particularly interesting finding
in this context is that extinction induces appearance of
spines that were eliminated upon the original fear
conditioning to the same stimulus but not to a distinct
conditioned stimulus, suggesting that the spines are again
specifically associated with extinction of one specific asso-
ciation [30]. Interestingly, also in a completely different
animal model—song learning in zebra finches—it was
shown that new spines are generated in the forebrain
nucleus HVC when an animal learns a new song from a
tutor [32].
Finally, what about the experiment that has long been
on the agenda [33], namely to specifically ablate spines
that have been generated during learning? If the above
interpretations are true, spine ablation should lead to
forgetting of the information that was learned when the
new spines were generated. First important strides in
that direction have been made in a recent experiment by
the group of Haruo Kasai [34], who specifically labeled
spines that were generated at a particular time window
immediately after learning. When these spines were later
 Poo et al. BMC Biology (2016)4:
ablated or at least reduced in size, the animal indeed
forgot the previously learned information. The learning
paradigm is so far relatively simple (rotarod learning)
but it provides a very nice indication that the generation
of new spines or their enlargement is truly causal at least
in some forms of learning.
Taken together, there is now considerable evidence
from different species as well as from different learning
paradigms that spines, and thus synapses, change when
an animal learns. Furthermore, there are convincing in-
dications that the maintenance of previously established
structural connections on the level of dendritic spines
explains the memory phenomenon of savings. Finally, if
spines are ablated, an animal forgets what it has learned
through the addition of new or stronger spine synapses.
All of these experiments seem to point strongly towards
the notion that spines or synapses (and not entire cells)
may be the smallest unit of memory storage in the brain
and it may, therefore, be most appropriate to say that
the “engram” of a memory is laid down in the set of
spines or synapses that are changed when specific infor-
mation is stored. This is of course not to say that
engrams are not visible on the level of single cells
(see preceding contribution to this Forum by Pignatelli,
Ryan, and Tonegawa); after all, the activity of cells is
determined by the complement of their synapses. Yet, the
finest resolution of the engram may only become apparent
if one truly considers everything on the basis of the
pattern of synapses or spines which are changed during a
particular memory event.
A cell biologist’s view of memory: revisiting the
neuron doctrine
Kelsey C. Martin
Over a century ago, the anatomist Ramon y Cajal used
the Golgi staining method to visualize individual cells in
the brain. He observed that the brain was composed of
discrete cells rather than of a “reticular network” of
interconnected cells (as was commonly believed at the
time). Cajal’s observations provided critical support for
the neuron doctrine, which postulates that the neuron is
the central unit of the brain. At the same time, his
drawings revealed the beautiful complexity, polarity, and
compartmentalization of neurons: cell bodies elaborating
axonal and dendritic processes, forming up to thousands
of synapses with one another. With a remarkable amount
of prescience, Ramon y Cajal also speculated that
memories were stored as increases in the numbers of
connections between neurons. This idea forms the
basis of learning-related synaptic plasticity—the idea
that memories are stored as changes in the number
and strength of synapses between neurons—a frame-
work that has endured as a model for understanding
the biology of memory.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 6 of 18
Studies of memory in rodents and goldfish performed
in the 1960s and 1970s demonstrated that long-term
memory required protein synthesis whereas short-term
memory did not [35]. Subsequent studies of learning-
related plasticity in organisms ranging from Aplysia
sensory-motor synapses to rodent hippocampal synapses
similarly demonstrated that long-lasting forms of plasti-
city can be differentiated from short-term plasticity by
their dependence on RNA and protein synthesis [36].
Molecular biological approaches led to the identification
of many genes that contribute to long-term plasticity
and memory and to the elucidation of specific patterns
of neuronal activity and specific signaling pathways that
trigger changes in RNA and protein synthesis within
neurons. These findings gave rise to the idea that activity-
dependent changes in the neuronal transcriptome and
proteome mediate and/or maintain the changes in neur-
onal structure and physiology that result in persistent
changes in synaptic strength.
Initially, studies of gene expression underlying mem-
ory focused on activity-dependent changes in transcrip-
tion in the nucleus. The discovery of immediate early
genes, such as c-fos, arc, and zif268, that were induced
during memory formation implied that the neuron was
the unit of long-term plasticity and memory [37]. Imme-
diate early genes are now widely used to map neurons
involved in memory formation, in line with the idea that
memories are encoded within networks of discrete neu-
rons in the brain.
In contrast to this neuron-centric view of memory
formation, studies of learning-related plasticity revealed
that plasticity was synapse-specific, that is, it could occur
at some but not all synapses made by an individual
neuron [38]. This finding raised questions about how
the products of gene expression, synthesized in the nu-
cleus, could be targeted to alter structure and function
at subsets of synapses within a single, highly polarized
and compartmentalized neuron. One solution to this
question came from findings that mRNAs localized to
distal dendrites and synapses [39], where their transla-
tion could be regulated by activity. These findings in-
cluded detection of polyribosomes at the base of
synapses as well as the identification of a subset of
mRNAs that localized to distal dendrites. Studies from
several labs have identified hundreds to thousands of
dendritically localized mRNAs [40–42] translation of
which could alter the structure and function of synapses,
thereby making the synapse (or neighborhood of synapses),
rather than the entire neuron, the unit of plasticity.
Local translation at synapses has been shown to regu-
late translation in a synapse-specific manner [43, 44]
and inhibiting local translation at synapses in a variety of
in vitro preparations has been shown to block long-term
learning-related plasticity [45, 46]. In one of these
 Poo et al. BMC Biology (2016)4: Page 7 of 18

preparations, the Aplysia sensory-motor culture system,
we directly tested the relationship between trans-
criptional regulation in the nucleus and translational
regulation at the synapse during synaptic plasticity and
found that stimulus-induced newly transcribed localized
mRNAs were delivered throughout the neuron but were
only translated at locally stimulated synapses [47]. The
implication of this finding is that activity-dependent
transcription sets the entire neuronal arbor in a state of
readiness to respond to local cues via regulated transla-
tion of localized mRNAs. From the perspective of
activity-dependent gene regulation, this idea shifts the
focus from the neuron as the unit of plasticity to the
synapse (or neighborhoods of synapses). In so doing, it
underscores the need to refine the current neuron-based
maps of memories by developing synaptic maps of
memory in the brain (Fig. 2).
The focus on the synapse as the unit of plasticity calls
into question aspects of the neuron doctrine. Thus,
while the neuron doctrine assumes that neurons are
separate, autonomous units, cell biological studies of
synapses indicate that all components of the synapse are
intimately interconnected, including not only the pre-
and post-synaptic compartments but also glial processes.
For example, in one study of local translation at Aplysia
sensory-motor synapses, we found stimulus-induced
local translation of a reporter depended on trans-
synaptic signals from the post-synaptic motor neuron
compartment to the presynaptic sensory neuron com-
partment [44]. Taking the idea of non-cell autonomy to
an extreme, recent studies have reported that neurons
and/or glia can transfer mRNAs and microRNAs to
neighboring neurons and/or glia [48]. This finding raises
the provocative idea that gene expression can be regulated
by transfer of genetic information from cell to cell. As
such, the idea of “local” translation serving to alter the
structure and function of synapses encoding a memory
suggests that “synaptic maps” of memory are not encoded
by discrete synaptic compartments of individual cells but
rather by complex fluid synaptic neighborhoods. These
neighborhoods include not only the products of gene
expression that are made in the nucleus of the neuron the
synapse belongs to but also signals (including RNAs) from
neighboring neurons and glia.
To add an additional layer of complexity onto this view
of memory, single cell RNA sequencing experiments have
uncovered remarkable complexity in the nerve cells and
glia within the brain [49]. These findings indicate that syn-
aptic neighborhoods are not just comprised of common
modules of pre-synaptic, post-synaptic, and glial compo-
nents but that they also exist in a multitude of distinct fla-
vors or “ethnicities” as neighborhoods. A recent electron
microscopic reconstruction of a small volume of mouse
neocortex revealed that the formation of synapses between
neurons is more dependent on cell identity than on prox-
imity [50]. The diversity of cell identities, in combination
with the idea that cell identity drives synapse formation,
indicates that generating a synaptic map of memory will
require understanding the distinct identities of the partici-
pating cells as well as the details of their cell-to-cell
interactions.
As a cell biologist interested in understanding mem-
ory, the challenges moving forward include identifying
conserved local processes that persistently alter synaptic
function and developing methods to manipulate these
processes in order to test their function in the formation
and storage of memory. Possibilities include local mech-
anisms of translational regulation at synapses, trans-

Fig. 2. Neuron versus synaptic maps of memory. The Golgi method used by Ramon y Cajal at the turn of the 19th century revealed that the
brain consisted of individual nerve cells, leading to the formulation of the neuron doctrine. In modern neurobiology, expression of soluble
fluorescent proteins allows visualization of individual neurons in the brain or, as shown here, in a hippocampal pyramidal neuron in dissociated
culture (blue) (a). Identification of immediate early genes, such as cFos, provides a means of labeling the nuclei of individual neurons that are
activated to undergo transcription following neuronal activity (violet) (b). This provides a neuron-level map of activity. However, each neuron
forms thousands of synapses, shown in c by labeling a single neuron with presynaptic (green) and postsynaptic (red) markers (yellow where the
pre- and post-synaptic elements are adjacent to one another). Note that this neuron is connected to other neurons in the culture that are not
shown. Developing a synaptic rather than a neuron-wide map of memory requires obtaining activity-dependent markers that label the activated
synapses rather than nuclei. Understanding synaptic maps will not only require understanding the identity of individual neurons but also the
details of their cell-to-cell interactions

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Poo et al. BMC Biology (2016)4:
synaptic signaling pathways (including the transfer of
RNAs via extracellular vesicles), interactions between
synaptic cell-adhesion molecules, and even extracellular
matrix dynamics. As an example of the latter, Roger
Tsien has proposed that memories may be stored in the
pattern of holes formed within the perineuronal net, a
specialized extracellular matrix structure that is formed
at the end of critical periods in the brain [51]. Taken to-
gether, recent lessons learned suggest that elucidation of
these local processes requires consideration of the syn-
aptic compartment as a local environment rather than as
a collection of separate and autonomous membrane
bound compartments. While morphology, from Golgi
stains to electron microscopy, emphasizes the boundar-
ies between cells in the brain, molecular cell biological
studies uncover a fundamental role for local cell–cell
interactions and interconnections in the encoding of
memories.
Memory as a functional consequence of
epigenetic priming in engram neurons
Andrii Rudenko and Li-Huei Tsai
Memory formation, storage, and recall constitute the
essence of human nature. The search for the mecha-
nisms underlying learning and memory has revealed the
importance of a number of molecular and cellular
processes, such as activity-dependent gene expression,
intracellular signaling cascades, and synaptic plasticity
[52, 53]. The long-lasting attempts to characterize mem-
ory localization recently resulted in identification of
specific neuronal populations—so-called engrams—that
provide a physical location for the storage and retrieval
of memory traces [14, 15, 54, 55]. Despite discovery of
the engram cells, molecular mechanisms of memory
storage remain unclear. We propose that epigenetic
alterations taking place in these cells may represent a
critical process involved in the long-term retention of
memory traces.
One well-studied example of such alterations is histone
acetylation, a covalent mark of active chromatin. Muta-
tions in CBP, a gene product necessary for the acetylation
of multiple memory genes, result in severe intellectual dis-
ability in humans as well as in mice [56–58]. Conversely,
histone deacetylase inhibitors (HDACi) were shown to re-
store histone acetylation and ameliorate cognitive deficits
in a CBP-deficient mouse model [56, 57]. Moreover,
HDACi have been found to ameliorate memory deficits in
mouse models of Alzheimer’s disease [59, 60]. In addition
to histone acetylation, several other epigenetic mecha-
nisms, including DNA methylation and hydroximethyla-
tion, have also been demonstrated to regulate memory
function [61, 62].
While inhibiting HDACs was found to be effective in
enhancing synaptic plasticity and memory, such inhibition
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 8 of 18
per se, in the absence of neural stimulation, produced very
limited results [63, 64]. Thus, HDAC inhibition appears to
convey its effects on learning and memory via facilitating
gene expression elicited by neural activity, a phenomenon
known as epigenetic priming [65, 66].
As discussed earlier in this Forum by Pignatelli, Ryan,
and Tonegawa, a recent study by Ryan et al. [16] has
elegantly demonstrated that inhibiting protein synthesis
during the memory consolidation window does not dis-
rupt memory retrieval by means of engram activation.
This exciting observation suggests that while augmented
synaptic strength may be critical for memory encoding,
some other mechanisms, potentially involving epigenomic
modifications in engram neurons, appear to be necessary
for memory trace storage. The recent discovery that long-
term memory can be re-instated following erasure of its
synaptic expression strongly supports this idea [67]. We
propose that, mechanistically, the engram cells are
marked, or tagged, not only synaptically [68] but also at
the epigenetic level, be it histone acetylation, methylation,
DNA methylation, or Topoisomerase IIβ-dependent topo-
logical changes of the DNA/chromatin, as recently sug-
gested by our work [62, 69]. Specifically, the initial phase
of memory formation would cause changes in the epigen-
etic state of the engram cells through a priming event
which may be protein synthesis-independent (for example,
epigenetic modifications to make specific genomic regions
poised for efficient transcriptional activation). Such
changes may also lead to long-lasting alterations in chro-
matin structure and function underlying the memory
consolidation process. Finally, memory retrieval would
signal the engram cells potentiating initial epigenetic
priming, including molecular events such as generation of
DNA breaks within the promoter areas of early response
genes such as c-Fos, Npas4, Nr4a1, and Egr1 [69],
triggering expression of the primed genes leading to pro-
tein synthesis and increases in the number and strength of
the synapses. Such a chain of events may explain why,
even after considerable neurodegeneration, HDAC inhib-
ition coupled with behavioral training is capable of re-
instating learning and retrieval of long-term, and even
remote, memory [59, 70]. This scenario may be possible if
engram cells, epigenetically primed by the initial learning
experience and capable of re-engaging in the chain of mo-
lecular events leading to memory retrieval, still remain in
the brain after neurodegeneration.
We should note that in these early days of functional
memory engram investigation, we still do not have
satisfactory answers to many important questions. For
example, the exact molecular and structural features of
engram-containing networks, or potential mechanisms
that might allow participation of a specific neuron in
different engram ensembles, currently remain unknown.
Moreover, although there is accumulating evidence of
 Poo et al. BMC Biology (2016)4: Page 9 of 18

the epigenetic marks/tags that would prime engram cells
for an efficient transcriptional response, the exact nature
of those marks remains unclear. Deciphering such prim-
ing signatures will help us immensely in understanding
the mechanistic basis of memory processing.
Memory mechanisms: LTP and LTD in partnership,
not merely in opposition
Richard W. Tsien, Gord Fishell, and Caitlin Mullins
How memory is stored in the brain can be effectively
queried by examining how this process is affected in
neuropsychiatric conditions. To this end, many aspects
of the emerging pathophysiology of autism spectrum
disorders (ASD) may provide insight into the tuning of
synaptic strength in memory [71]. In ASD and related
intellectual disability (ID), gene discovery points to dys-
regulation of interrelated neuronal functions, including
control of nuclear gene expression, local protein synthe-
sis in dendrites, and excitation:inhibition (E:I) coordin-
ation. These functions are linked together in feedback
loops involving electrical or chemical sensors, termed
“homeostats”. The feedback loop may malfunction as a
result of disease-causing mutations in any of the compo-
nents. Given the effects of these disorders on cognition,
it is likely no accident that the same set of functions
loom large in current thinking about memory. Indeed,
consideration of ASD and other neuropsychiatric dis-
eases provides fresh perspective on the basic underpin-
nings of memory. From this viewpoint, we offer some
thoughts about the relationship between LTP and LTD
and the way that information in the brain may be stored
and retrieved.
In an intriguing study published in 2012, Bourne and
Harris used serial section electron microscopy to com-
pare dendrites receiving either theta burst LTP or
control stimulation [72]. As expected for LTP [34, 73],
they found single spines with increased postsynaptic area
by 2 h. Surprisingly, however, this occurred concomitant
with a remarkable reduction in the number of small
spines, leaving the total postsynaptic area per unit den-
dritic length unchanged compared with control (Fig. 3a).
The constancy of total postsynaptic area per unit den-
dritic length fits well with prevailing concepts about
homeostasis and normalization, but it also raises pro-
vocative questions about the neurobiological mechanism
and organization of memory.
Two previously proposed and rather different hypoth-
eses considered here might explain the underlying basis

Fig. 3. Possible mechanisms of synaptic modification in memory storage. a Structural synaptic scaling is analyzed by serial section electron microscopy
in hippocampal tissue under control and LTP (induced via theta burst stimulation) conditions. Control and LTP dendrites have equal postsynaptic areas
(red outline) despite differences in synaptic density and size. Adapted from [72], with permission. b Representation of metaplasticity occurring, shifting
the plasticity threshold (θm). The graphs on the top row depict the distribution of synaptic responses, red shading indicating LTP and white shading LTD.
The four synapses illustrated (bottom row) reflect how the shifting of θm would impact the strength of individual synapses (intensity of red shading
correlating to strength). Stage 1 depicts a saturating event, where all of the synaptic strengths have been excited to levels above θm. Stage 2 shows a
metaplastic response to the activity levels reached in stage 1. In stage 2, θm has shifted such that some synapses are below the new threshold and
weaken in stage 3. This changed θm leads to a new stable state, illustrated in stage 4. Reproduced from [75], with permission. c Synaptic potentiation at
one spine (center, upward-pointing black arrow added for emphasis) is predicted to temporarily increase activity at adjacent spines. In response to the
increased activity, homeostatic plasticity weakens the entire area (negative feedback arrows), with closely neighboring spines being disproportionately
weakened (downward-pointing white arrows). The weakened spines allow the overall dendritic length to maintain a constant level of activity with the
central spine still maintaining a level of potentiation. Adapted from [76], with permission

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Poo et al. BMC Biology (2016)4:
of Bourne and Harris’ observations. In reviewing
evidence for a “sliding threshold” concept [74], synaptic
plasticity with flexible rules (metaplasticity), Deisseroth
et al. [75] discussed a scheme that could fit with Bourne
and Harris’ results (Fig. 3b). If a stretch of dendrite
receiving strong inputs undergoes LTP but then autore-
gulates the threshold dividing LTD and LTP, it would
attain a new equilibrium that fractionates the synapses
into stronger and weaker subpopulations (Fig. 3b). Also
pertinent to Bourne and Harris’ results, Rabinowitch
and Segev [76] focused on lateral coordination of synap-
tic strength, invoking unknown mechanisms of local
regulation on dendritic branches. In their hypothesis,
LTP at spines receiving strong synaptic input is yoked
together by a compensatory mechanism that weakens, or
even eliminates, nearby neighbors (Fig. 3c).
Although these papers make no reference to each other,
they converge on a shared view of synaptic plasticity,
jointly supported by biological evidence and theoretical
rationale: a synapse may undergo Hebbian strengthening
or weakening as an individual entity but, over time, it can
also behave as a connected entity, operating in coordin-
ation with other synapses in the same neuron or dendritic
branch. Further, LTP and LTD can cooperate to redis-
tribute synaptic weight. This notion differs from the trad-
itional analogy between synapses and digital information
storage devices, in which bits are stored and retrieved
independently. On the other hand, coordination amongst
multiple synapses, made by different inputs, provides
benefits with regard to issues of normalization and signal-
to-noise.
What can be said about the molecular mechanism(s)
that might allow or even drive the coexistence of LTP
and synaptic weakening in close proximity along a
stretch of dendrite? This question has received much
less attention than strengthening and weakening of the
same synapses (see, for example, an elegant demonstra-
tion of LTP and LTD in direct opposition [22]). In the
spirit of this Forum, we list here multiple possibilities for
such lateral interaction:
1. Concentration of the excitatory neurotransmitter
glutamate ([Glu]) must fall off with increasing distance
from a strong input and could play some role in
lateral interactions. Whereas NMDA receptors are
essential for most forms of postsynaptically expressed
LTP and are driven by high [Glu], lower levels of [Glu]
would be sensed by the more sensitive metabotropic
glutamate receptors (mGluR) at nearby synapses and
could foster LTD mechanisms.
2. Different programs of local protein synthesis may
be triggered by NMDARs and mGluRs and support
LTP (e.g., increased synthesis of AMPA receptors)
or LTD (e.g., upregulation of Arc). As we propose
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 10 of 18
elsewhere [71], mutual inhibition between such
translational programs could help enforce a sharp
threshold dividing LTP and LTD, whereby local
signals mediate LTP proximally and LTD at a
distance. Interestingly, defects in either program
can give rise to ASD.
3. βCaMKII (beta calcium/calmodulin-dependent
protein kinase II) is well-suited to serve as a local
sensor of activity-dependent rises in Ca2+ because
of its high Ca2+/CaM sensitivity. Hence, βCaMKII
can also serve as an arbiter, dictating the decision
between exo- and endocytosis of AMPARs to pro-
mote LTP/LTD. High [Ca2+] activates βCaMKII,
increasing AMPAR exocytosis [77]. In contrast,
low [Ca2+] leads to CaM-free, kinase-deactivated
βCaMKII interacting with Arc, facilitating AMPAR
endocytosis [78].
4. AMPA receptors can also be redistributed along the
dendritic length by coordination of exocytosis of
AMPARs and LTP at one site with endocytosis of
AMPARs and LTD in flanking regions. A frank
lateral transfer of AMPARs [79] could support a
coupling of LTP/LTD at nearby dendritic spines.
Many studies on the cell biology of the dendrite take
on different meaning if considered in this light. Also, an
organizational principle of “robbing Peter to pay Paul”
might engender marked strengthening of some synapses
(rich getting richer) at the expense of others (poor
getting poorer), thus contributing to empirical observa-
tions of a highly skewed distribution of synaptic weights
[80–82]. It remains to be seen, however, whether the
observations of Bourne and Harris are relevant to
behavioral memory in vivo and whether the tradeoff of
synaptic strength is predominantly a local, dendritic
branch-based phenomenon as suggested by Rabinowitch
and Segev [76] or also strongly dependent on regulation
at the neuron-wide level [75, 83]. We have spoken little
about the temporal dimension but the static snapshots
in Fig. 3 are clearly shorthand for complex dendritic dy-
namics. Whatever the molecular/subcellular mechanism
and dynamics, tradeoffs in synaptic strength will have a
strong influence on the way that memory traces are
stored and retrieved. As the debate continues about
memory engrams and persistent changes at the level of
synapses, synaptic neighborhoods, ensembles of cells,
and even whole circuits, we would do well to consider
redistribution in net synaptic strength as an underlying
mechanism, rather than merely net increases or de-
creases. Perhaps we should modify silicon-based notions
of memory units with independent read-write capabil-
ities and embrace assemblies—not just “cell assemblies”
[1, 17] but also “synaptic assemblies”—in their full
spatiotemporal glory.
 Poo et al. BMC Biology (2016)4:
A role for adult-born neurons in pattern
separation
J. Tiago Gonçalves, Matthew Shtrahman,
Stephen T. Johnston, and Fred H. Gage
Memory involves the complex interplay between form-
ing representations of novel objects or events and devel-
oping generalizations of similar experiences. Distinct
instances of similar events must be discriminated—for
example, being able to find your car at work despite
parking in a different spot every day—but, at the same
time, experiencing just a fragment of a familiar experi-
ence, such as a particular smell, can trigger specific
memories from your childhood. The interplay between
forming distinct memories and generalizing events is
conceptualized to involve two separate processes: pat-
tern separation and pattern completion.
The dentate gyrus (DG) is the information gateway of
the hippocampal formation and as such plays a crucial
role in hippocampal function, including the formation of
episodic memories. Additionally, the DG is one of only
two regions within the mammalian brain that generates
new neurons throughout the life span of an individual in
both rodents and humans [84–86]. There is increasing
evidence that adult-born dentate granule cells (DGCs)
are important for fine pattern separation of similar but
distinct events [87, 88].
The DG as a pattern separator
The elegantly delineated anatomy of the hippocampus
has long served as a substrate for theories about
memory formation. It is grossly delineated into a series
of interconnected loops within four major subregions.
Specifically, the DG of the hippocampus receives excita-
tory input from the entorhinal cortex (EC) via the per-
forant path [89]. The DG then projects through the
mossy fiber tract to form powerful synaptic connections
with CA3 neurons, which in turn project to CA1, form-
ing the classic trisynaptic pathway. In addition, the DG
projects to and receives input from local inhibitory inter-
neurons, most notably from the hilus.
In rodents, the DG contains approximately four- to
fivefold more neurons than up- or downstream EC and
CA3, respectively. Thus, input from relatively few cells is
processed by a much larger neural network within the
DG before generating a condensed output. However,
only a small percentage of DGCs is activated in response
to a given event [90, 91]. Based on these characteristics,
modeling studies postulate that the DG is a competitive
network that can function as a pattern separator by par-
tially de-correlating inputs [3, 92]. One prediction from
this theory is that the DG is critical for forming memor-
ies of events that are similar but not identical to each
other. This prediction is supported by accumulating evi-
dence from both high-resolution functional MRI studies
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 11 of 18
in humans and studies of “behavioral pattern separation”
in rodents [93, 94], where subjects discriminate between
similar environments or sensory stimuli presented at dif-
ferent times.
What mechanistic role do adult-born DGCs play in
behavioral pattern separation?
Although the gross anatomical connectivity of the DG
suggests a role in pattern separation, the DG’s ability to
perform this function appears to be further enhanced via
its ability to incorporate new neurons. Adult-born DGCs
undergo a lengthy process of morphological and physio-
logical maturation before they fully integrate into the
local hippocampal network [95]. Each immature DGC
enters a critical period of greater plasticity approximately
4 to 6 weeks after it is born. These immature DGCs ex-
hibit greater excitability [96], receive less inhibition from
local interneurons [97], are more broadly tuned to input
stimuli [98], and exhibit greater synaptic plasticity [99]
than mature cells. Therefore, immature adult-born
DGCs may perform unique computational tasks critical
for hippocampal function, in particular behavioral pat-
tern separation [87, 100].
The physiological properties of immature neurons at
the single-cell level, however, might yield counterin-
tuitive predictions at the circuit level. While hyperexcit-
ability and broad tuning of immature neurons may allow
the hippocampus to encode novel stimuli, their broad
tuning would also seem to suggest that a hippocampal
circuit rich in immature neurons would fire more fre-
quently and indiscriminately to various environmental
inputs. Yet, this explanation appears to conflict with the
role of the DG in pattern separation. Therefore, it has
recently been proposed that, though hyperexcitable
themselves, these immature neurons’ key contributions
could be to suppress overall DG activity, maintain net-
work sparseness, and thereby decrease interference be-
tween memory representations [101–103] as illustrated
in Fig. 4. Individual immature neurons may be more
excitable and broadly tuned but it is hypothesized that
these cells in turn activate inhibitory interneurons in the
DG and hilus, resulting in greater inhibition of mature
DGCs. Developing a unified picture of how single-cell
physiological features combine to determine circuit-level
responses and behavior presents a great challenge for
future work.
Future steps for understanding the role of adult-born
neurons in the DG
Previous efforts to link DG function to cellular and mo-
lecular mechanisms have been hindered by technical
limitations. First, although extracellular recording of
DGCs has been used to monitor the dynamics of neur-
onal firing patterns, sparse activity within the DG makes
 Poo et al. BMC Biology (2016)4:
CA1
CA2
DG
Without Neurogenesis
CA3
EC inputs Event 1
(Molecular Layer) Event 2
DGCs
(Granule Layer)
Interneurons
(Hilus/Sub-granular zone)
Output to CA3
(Mossy fibers)
Fig. 4. Immature adult-born neurons improve pattern-separation in the DG by enhancing feedback inhibition. Two events are encoded in separate
but partially overlapping populations of activated DGCs in the DG (red and green, with overlap in yellow). DGCs receive strong inhibitory
inputs from interneurons (purple) in the hilus and sub-granular zone. It is hypothesized that hyperactive immature adult-born DGCs (blue)
drive these interneurons, enhancing feedback inhibition from the hilus, which results in decreased overlap of activated DGCs and output to CA3,
thereby improving pattern separation
it difficult to locate neurons, identify neuronal subtypes,
and monitor the activity of a large population of DGCs
[104]. Second, immunohistochemical analysis of imme-
diate early gene expression, such as c-fos, can assess ac-
tivity across a large population but only at limited time
points, providing a limited snapshot of DG activity.
Nevertheless, advances in in vivo recording techniques
promise to eventually enable the monitoring of DG cells
during behaviorally relevant tasks. In vivo imaging
methods [105] in particular would be perfectly suited for
recording DG activity—hundreds of cells can be moni-
tored simultaneously and cellular subtypes identified
through the use of genetic markers. Combined with
modern techniques to manipulate activity in adult-born
neurons, imaging large populations of neurons in vivo
would permit testing the hypothesis that immature and
mature DGCs work cooperatively to maintain the nor-
mal sparse network activity of the DG. These methods
also have the potential to provide further insights into the
mechanisms behind memory formation and recall, revolu-
tionizing our understanding of the features encoded by
both mature and immature DGCs and of how responsive-
ness to those features changes with learning.
Mechanisms for memorizing temporal sequence
and interval
Mu-ming Poo and Yang Dan
The temporal sequence and interval of events are
essential elements of episodic and procedural memories.
Where and how sequence and interval information is
stored in the brain remains a mystery. If modifications
of synaptic connectivity are the cellular substrates for
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 12 of 18
With Neurogenesis
Event 1
Event 2
newly born
neurons
Improved Pattern Separation
(decreased overlap)
memory storage, then the challenge is to understand
how distributed synaptic changes within neural circuits
represent the sequence and interval of previously experi-
enced sensory or motor events.
In his physiological postulate for perceptual memory,
Hebb [1] proposed that reverberating activity generated
by perceptual experience along distinct neuronal path-
ways could provide temporary storage of the experience.
Repeated firing of a specific assembly of neurons in a
particular temporal sequence could then serve to im-
print the memory by modifying synaptic connections
among the cells and re-activation of the assembly in this
sequence represents recall of the memory.
The popular version of Hebb’s postulate refers to synaptic
modifications based on correlated pre- and postsynaptic fir-
ing (“cells that fire together wire together”), and it received
experimental validation with the discovery of long-term po-
tentiation (LTP) [5] and long-term depression (LTD) [106]
induced by high- and low-frequency presynaptic stimula-
tion, respectively—high-frequency stimulation results in
postsynaptic spiking and thus LTP due to correlated
activity, whereas low-frequency stimulation fails to do so,
leading to LTD. The discovery of spike timing-dependent
LTP and LTD [107, 108] led to further revision of Hebb’s
learning rule—the sequence of pre- and postsynaptic spik-
ing, rather than simple coincidence of activity, is critical
for determining whether the synapse is strengthened or
weakened. The defined time windows for spike timing-
dependent plasticity (STDP) [8, 109, 110] and its presence
at many excitatory synapses [111] suggest that temporal
information may be stored via STDP, and Hebb’s assembly
could be established by sequence-dependent synaptic
 Poo et al. BMC Biology (2016)4:
strengthening or weakening. In addition to the well-
studied sequence replay in the hippocampus (see
contribution by Long and Buzsáki in this Forum), this
hypothesis was supported by a finding that repetitive
visual stimulation with a unidirectional moving spot that
evokes sequential spiking of the cortical neurons enhances
their sequential firing in response to a flashed stimulus, in
a manner that depends on the speed of the conditioning
stimulus and activation of NMDA receptors in the visual
cortex [112].
The time window for STDP, normally quite narrow
(~20 ms) for LTP and wider and more variable (~20–
100 ms) for LTD, also imposes a limit on the time inter-
vals between pre- and postsynaptic spiking that can leave
a long-term imprint at the synapse. Could the interval be
extended via polysynaptic excitation within a network? A
simple test of this idea was performed in a random synap-
tic network formed by dissociated hippocampal neurons
in culture. Repetitive pair-pulse stimulation at a fixed
interval was applied to a single input neuron and changes
in the efficacy of polysynaptic connections within the net-
work were measured. The result showed that intervals up
to a few hundred milliseconds could lead to distributed
long-term modifications (both LTP and LTD) of polysyn-
aptic pathways within the network [113]. While this study
provides proof of principle for encoding long intervals
through polysynaptic delays, does the natural neuronal
network in the brain contain serially connected groups of
synchronously firing neurons (synfire chain) [114] to
encode long sequences and temporal intervals? The find-
ing of sequential firing of neurons in the premotor area
HVC of songbirds during each song syllable (which
lasts >100 ms [115]) indicates the existence of synfire
chains in the brain, the formation of which underlies song
learning [116]. Spike sequences lasting for several seconds
have also been observed in the mammalian neocortex [117].
In addition to STDP in polysynaptic networks, mem-
ories of time intervals on the order of seconds are also
likely to involve other mechanisms. In a study in zebra-
fish larvae, Sumbre et al. [118] applied a unidirectional
moving visual stimulus repeatedly at a fixed temporal
interval (several seconds) and recorded the population
activity of tectal neurons with Ca2+ imaging. They found
that following cessation of the visual stimulation, se-
quential firing of the tectal neurons resembling that
evoked by the moving stimulus reappeared at the same
time interval as the conditioning stimuli, and this
rhythmic reappearance lasted for up to 20 s. This post-
conditioning spontaneous rhythmic firing of tectal neu-
rons reflects short-term memory of a specific rhythm
with time intervals of seconds. Interestingly, since the
tectal activity did not persist continuously through the
several seconds of interval, it is unlikely mediated by
reverberation of activity in the polysynaptic network.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 13 of 18
The underlying circuit mechanism and the location of
memory storage in this case remain unknown.
Remembering the association between events sepa-
rated by intervals ranging from seconds to tens of sec-
onds has been referred to as temporal associative
learning, and has been found in recent studies to depend
critically on entorhinal cortex (EC)–hippocampal circuits
[119–121]. Pathway-selective inhibition and optogenetic
activation of the EC–hippocampal circuit showed that
layer II “island cells” and layer III neurons of the EC con-
trol temporal associative learning [119, 120]; in particular,
persistent activity of medial EC layer III neurons may play
a key role [121]. Recent studies have also shown the exist-
ence of hippocampal “time cells” that fire at particular mo-
ments in a temporally structured experience [122–124],
suggesting a function that parallels that of place cells in
spatial memory. The sequential activation of such time
cells in the hippocampus may reflect temporally struc-
tured inputs from the cortex and other brain regions or
alternatively firing chain generated within local hippo-
campal circuits by repetitive sequential activity-induced
strengthening of synaptic connections.
Theories and models of sequence learning and interval
timing have been proposed [125–127] but very few
studies have directly addressed the circuit and synaptic
mechanisms [128, 129]. Since the cellular and synaptic
building blocks of memory are increasingly well charac-
terized and the technology for manipulating specific cells
is becoming available (see other contributions to this
Forum), the storage mechanisms for sequence and inter-
val information are now amenable to fruitful exploration.
Understanding these mechanisms is a pre-requisite for
further studies of circuit mechanisms underlying higher
cognitive functions involving complex temporal informa-
tion processing such as human language, the ultimate
challenge to neuroscience.
Hippocampal sharp wave-ripple: a repetitive
mechanism to support single trial learning
John Long and György Buzsáki
Memories are not “imprinted” immediately but evolve
over time. Memory has many forms and supportive
brain mechanisms. While learning complex skills and
habits, such as walking elegantly in high-heel shoes or
riding a unicycle, may require tens to thousands of repe-
titions, consciously remembering episodic information,
such as recalling one’s first date, often requires only a
single trial. The brain can achieve such a feat by deploy-
ing a mechanism that repeats segments of the original
episode subconsciously hundreds to thousands of times
after the experience. Such repetitions occur during non-
attentive, off-line states of brain operation in the form of
hippocampal sharp wave-ripples (SPW-Rs). SPW-Rs
operate as a time-compressing mechanism, which can
 Poo et al. BMC Biology (2016)4:
transfer information from the hippocampus to numerous
regions of the neocortex. SPW-Rs also serve as a pre-
conscious “mixer” of existing knowledge and recently
acquired information. Thus, acquiring episodic memory
is a two-stage process, initiated by a rapid encoding
mechanism during attentive waking and followed by a
protracted consolidation process during “off-line” states
of the brain [130].
SPW-Rs are self-generated hippocampal patterns
The SPW-R complex is the most synchronous and
phylogenetically preserved pattern in the mammalian
brain, associated with enhanced transient excitability in
the hippocampus and its partner structures. These
super-synchronous bursts arise when the release of
subcortical neuromodulators within the hippocampus is
decreased—for example, during consummatory behav-
iors and slow wave sleep. A SPW-R is a superposition of
two events, the sharp wave-related population burst
(SPW), which emerges in the strongly recurrent system
of the CA2/3 regions, and the fast ripple oscillation,
which is dominant in the CA1 output circuit of the
hippocampus. The synchronous discharge of CA3 pyr-
amidal cells excites primarily the mid-apical dendrites of
the CA1 region and the inward currents brought about
by this transient depolarization process (40–150 ms)
manifest extracellularly as the local field potential SPW.
The response of target circuits to the strong SPW-
related depolarization is a fast oscillatory balancing act
between principal cells and perisomatic inhibitory
interneurons, resulting in the LFP ripple and the phase-
locked discharge of sequentially active neurons [130]. It
is this waking history-dependent, sequential firing of a
large fraction of hippocampal cells, and consequent re-
cruitment of neocortical neurons, that makes SPW-Rs a
candidate biomarker for memory consolidation.
Compressed replay of experience during SPW-Rs
serves memory
The restructuring of hippocampal–cortical networks
through synaptic plasticity is necessary for the formation
of new episodic memories. Neurons participating in
SPW-R events are organized to fire sequentially and the
orderly structure of these events reflects a temporally
compressed version of the sequential neuronal firing pat-
terns observed in the waking animal [131, 132] (Fig. 5;
“replay”). For example, the sequences of “place cells”
[133] in a novel environment are formed from a combin-
ation of relatively fast-firing and slow-firing groups of
pyramidal neurons. Conspicuously, the former neurons
exhibit relatively unchanging temporal dynamics while
the latter are highly plastic. The greater plasticity of
slow-firing pyramidal neurons is evidenced by their
greater gain in place specificity during maze exploration
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 14 of 18
and their increased SPW-R-related recruitment during
sleep following waking experience [134]. These sequences
are not confined to the hippocampus–entorhinal system;
the SPW-R output also brings about sequential activations
in neocortical circuits, such as the prefrontal cortex [135],
leading to systems-level consolidation of the memory
trace. Even more direct evidence for the role of SPW-R
sequences in memory consolidation is provided by close-
loop truncation of SPW-Rs. Selective elimination of the
hundreds of SPW-R-assisted replays of recently learned
sequences during post-learning sleep prevented rats from
becoming proficient at a spatial-reference memory task
[136]. This supports the hypothesis that single-trial learn-
ing is made possible by repeated off-line SPW-R replays of
the wake-experienced episode.
Memory and planning: retrospective and prospective roles
for SPW- R
A particularly striking feature of SPW-Rs is that the
neuronal sequences contained in them can propagate
both forward and backward (Fig. 5) relative to waking
experience. Before the beginning of a journey upon a fa-
miliar maze, the upcoming sequences are preplayed dur-
ing SPW-Rs in a forward manner; that is, prior to a run,
the CA1 pyramidal cells fire in a sequence consistent
with the upcoming trajectory of the animal in the maze.
At the end of the journey, again, the same neurons are
reactivated during SPW-Rs but now in a reversed direc-
tion as the hippocampus recapitulates the landmarks
passed by the animal, though in a time-compressed
manner. These findings support the hypothesis that for-
ward replay events play a role in “planning” upcoming
trajectories [137]. This hypothesis is further supported
by several recent experiments that demonstrate that the
routes chosen by the animal in two-dimensional envi-
ronments or between maze corridors can be predicted
by the spike sequence content of SPW-Rs [138]. The
subconscious route-priming role of SPW-Rs is also sup-
ported by other experiments, which show that truncat-
ing SPW-Rs prior to a choice results in a spatial working
memory deficit [139]. Because the neurons participating
in SPW-Rs sequences are drawn from a diverse pool of
log-scale firing rate distributed neurons, with varying
coding, biophysical, circuit, and plasticity properties,
these events can transit a vast array of preexisting and
new information to downstream cortical partners [134].
To date, memory mechanisms in experimental animals
are typically studied in the framework of spatial naviga-
tion [133]. However, it is important to emphasize that
mechanisms of memory and planning have evolved from
mechanisms of navigation in the physical world. There-
fore, neuronal algorithms underlying navigation in real
and mental space, as well as place memory and episodic
memory, are fundamentally the same [140].
 Poo et al. BMC Biology (2016)4:
Fig. 5. Place cell sequences experienced during behavior (middle panel) are replayed in both forward (left panel) and reverse (right panel) direction
during awake SPW-R. The rat is moving from left to right on a familiar track. Spike trains for place fields of 13 CA3 pyramidal cells (color ticks, spikes
of individual neurons) on the track are shown before (forward replay; left red box), during (middle), and after (reverse replay; right blue box) a single
traversal. The CA1 local field potential is shown on top (black traces) and the animal’s velocity is shown below. Reproduced from [137]
Where is the study of Hebbian memory
mechanisms going?
Charles F. Stevens
BMC Biology brought together some of the leaders in
the learning and memory field to identify, from eight
separate perspectives, some of the currently most im-
portant questions and approaches for understanding
memory formation, consolidation, and retrieval. My job
is to relate the current state of the field to how, in my
opinion, it might evolve.
With rare, and highly valued, exceptions, a field de-
velops organically from where it started and from the
questions that grew out of each advance along the way.
Although several of the contributions to this Forum re-
late key ideas from earlier eras (Cajal, Golgi, and Semon)
to modern views, my starting point for the questions
that drive learning and memory research will be Hebb’s
1949 “firing - > wiring” proposal [1] for learning (cited in
the introduction to this Forum and in four of the
contributions).
Although Hebb’s book was always well known to
psychologists, it had essentially no impact on neurophysi-
ology (as neuroscience was known in those days) for about
a quarter of a century because the longest lasting form of
synaptic plasticity known before 1973 (the date of Bliss and
Lomo’s LTP paper [5]) was post-tetanic potentiation (PTP)
and this form of plasticity lasted, at most, only a few
minutes. The only suggested brain mechanism for memory
storage had been “reverberating circuits” and Von
Neumann gave a compelling argument against the plausi-
bility of this idea in his 1957 lectures at Yale on “The
computer and the brain” [141]. Neurophysiologists were all
very interested in the cellular basis of memory—everyone
agreed it was a central problem—but no one had any idea
how to study it.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 15 of 18
As with any novel finding at odds with current know-
ledge and concepts, Bliss and Lomo’s LTP paper [5] did not
make an immediate splash. Indeed, during the next decade
only a hand-full of researchers worked on LTP. But over
the succeeding decades the study of LTP grew explosively
to become a large and important sub-field of neuroscience,
some would even say a preoccupation. As this Forum
shows, we have learned an enormous amount about synap-
tic plasticity in the two-thirds of a century since Hebb’s The
Organization of Behavior. But where will we go now?
First, I need to confess that my record for predicting
the future is abysmal: I have failed close to 100 % of the
time. This time, however, I am trying a new strategy to
see if I can move up from abysmal to, say, just terrible.
This new approach is to see where the questions that
drive our research have come from and then ask what
we might have missed in asking these questions. Because
of space constraints, I will focus on a single question:
what might Hebbian learning be missing?
Although Behaviorism is pretty much dead [142], there
is absolutely no question that reinforcement plays a key
role in what behaviors an animal selects. And, as anyone
who has experienced a traumatic event can tell you, the
circumstances associated with the event are committed
to an enduring memory (what were you doing when you
heard about the 9/11/2001 terrorist attacks in the United
States?). Of course, everyone is aware of this, yet the
focus on Hebb’s rule tends to neglect these other types
of learning. Major exceptions to this are the climbing
fibers in the cerebellum and the mossy fibers in the CA3
region of the hippocampus, where a single impulse or
brief burst will fire the target neuron and is believed to
potentiate the other synapses that are active at the same
time. For both of these cases, Hebb’s rule is sufficient to
determine what needs to be learned.
 Poo et al. BMC Biology (2016)4:
So my specific prediction is that future scientists work-
ing on LTP/LTD (or STDP) will also include reward
(positive or negative) related mechanisms in their inves-
tigations of synaptic plasticity at the molecular, cellular,
circuit, and behavioral levels.
Where can we look to get guidance on wedding Hebb’s
rule to reinforcement learning? One of the simplest and
best studied examples of alternative learning rules is the
fruit fly olfactory system, where flies can learn to
approach or avoid any odor if that odor is paired with
reward (sugar water) or punishment (electric shock). A
good example from Glenn Turner’s laboratory [143] is
an electrophysiological study showing how a specific
dopamine neuron can decrease the synaptic strength of
inputs into a specific mushroom body output neuron
(MBON). The inputs into the MBON are synapses from
2000 mushroom body Kenyon cells and each odor acti-
vates a specific subpopulation of the Kenyon cells, about
100/2000. These 100 cells constitute a tag for that odor
[144], a unique subpopulation of Kenyon cells whose fir-
ing stands in for the odor’s combinatorial code generated
by odorant receptor neurons in the fly’s antennae. When
the odor is paired with a shock, the specific dopamine
neuron whose synapses overlap with the MBON’s den-
dritic arbor and with the Kenyon cell’s axonal arbor is
activated. Firing of this dopamine neuron is what causes
the Kenyon cell synapses to weaken and this change in
synaptic strength is responsible for the avoidance of the
odor associated with punishment.
Most scientists who study LTP/LTD/STDP use mam-
mals as their experimental animals. It seems, however,
that the dopamine system is, to a large degree, evolu-
tionarily conserved even though the fly circuits are
generally less complex than the corresponding ones in
vertebrates. I am not aware of a review article that docu-
ments the insect/vertebrate parallels in the dopamine
systems but the literature on the amygdala, a vertebrate
center for valence learning, is vast [145] and can be
compared to the large fly literature devoted to mush-
room body learning and memory [146].
We can see over the next few years if my success for
predicting future trends in science has improved.
Acknowledgements
Kelsey C. Martin
I thank S.L. Zipursky, S. Van Driesche, and J.T. Braslow for their comments, K.
Otis for the photomicrograph in Fig. 2, and funding from R01NS045324.
Author details
1
Institute of Neuroscience, Chinese Academy of Sciences, Shanghai, China.
2
RIKEN-MIT Center for Neural Circuit Genetics at the Picower Institute for
Learning and Memory, Department of Biology and Department of Brain and
Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA
02139, USA. 3Howard Hughes Medical Institute, Massachusetts Institute of
Technology, Cambridge, MA 02139, USA. 4MPI of Neurobiology,
Munich-Martinsried, Germany. 5Department of Biological Chemistry and
Department of Psychiatry and Biobehavioral Studies, David Geffen School of
Medicine, BSRB 390B, 615 Charles E. Young Dr. South, University of California,
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 16 of 18
Los Angeles, CA 90095, USA. 6Picower Institute for Learning and Memory,
Department of Brain and Cognitive Sciences, Massachusetts Institute of
Technology, Cambridge, MA 02139, USA. 7The Neuroscience Institute, School
of Medicine and Center for Neural Science, New York University, New York,
NY 10016, USA. 8Salk Institute for Biological Studies, Laboratory of Genetics,
10010 N. Torrey Pines Road, La Jolla, CA 92037, USA. 9HHMI, Department of
Molecular and Cell Biology, University of California, Berkeley, USA.
References
1. Hebb DO. The organization of behavior; a neuropsychological theory.
New York: Wiley; 1949.
2. Katz LC, Shatz CJ. Synaptic activity and the construction of neural circuits.
Science. 1996;274:1133–8.
3. Marr D. Simple memory: a theory for archicortex. Philos Trans R Soc Lond B
Biol Sci. 1971;262(841):23–81.
4. Semon RW. The mneme. 1921.
5. Bliss TV, Lomo T. Long-lasting potentiation of synaptic transmission in the
dentate area of the anaesthetized rabbit following stimulation of the
perforant path. J Physiol. 1973;232(2):331–56.
6. Malinow R. Transmission between pairs of hippocampal slice neurons:
quantal levels, oscillations, and LTP. Science. 1991;252(5006):722–4.
7. Markram H, Lübke J, Frotscher M, Sakmann B. Regulation of synaptic efficacy
by coincidence of postsynaptic APs and EPSPs. Science. 1997;275(5297):213–5.
8. Bi GQ, Poo MM. Synaptic modifications in cultured hippocampal neurons:
dependence on spike timing, synaptic strength, and postsynaptic cell type.
J Neurosci. 1998;18(24):10464–72.
9. Govindarajan A, Kelleher RJ, Tonegawa S. A clustered plasticity model of
long-term memory engrams. Nat Rev Neurosci. 2006;7(7):575–83.
10. Mayford M, Siegelbaum SA, Kandel ER. Synapses and memory storage.
Cold Spring Harb Perspect Biol. 2012;4(6):a005751.
11. Shi SH, Hayashi Y, Petralia RS, Zaman SH, Wenthold RJ, Svoboda K, et al.
Rapid spine delivery and redistribution of AMPA receptors after synaptic
NMDA receptor activation. Science. 1999;284(5421):1811–6.
12. Engert F, Bonhoeffer T. Dendritic spine changes associated with
hippocampal long-term synaptic plasticity. Nature. 1999;399(6731):66–70.
13. Zhang F, Wang L-P, Boyden ES, Deisseroth K. Channelrhodopsin-2 and
optical control of excitable cells. Nat Methods. 2006;3(10):785–92.
14. Liu X, Ramirez S, Pang PT, Puryear CB, Govindarajan A, Deisseroth K, et al.
Optogenetic stimulation of a hippocampal engram activates fear memory
recall. Nature. 2012;484(7394):381–5.
15. Ramirez S, Liu X, Lin P-A, Suh J, Pignatelli M, Redondo RL, et al. Creating a
false memory in the hippocampus. Science. 2013;341(6144):387–91.
16. Ryan TJ, Roy DS, Pignatelli M, Arons A, Tonegawa S. Memory. Engram cells
retain memory under retrograde amnesia. Science. 2015;348(6238):1007–13.
17. Tonegawa S, Pignatelli M, Roy DS, Ryan TJ. Memory engram storage and
retrieval. Curr Opin Neurobiol. 2015;35:101–9.
18. Root CM, Denny CA, Hen R, Axel R. The participation of cortical amygdala in
innate, odour-driven behaviour. Nature. 2014;515(7526):269–73.
19. Matsuzaki M, Honkura N, Ellis-Davies GCR, Kasai H. Structural basis of long-term
potentiation in single dendritic spines. Nature. 2004;429(6993):761–6.
20. Le Bé J-V, Markram H. Spontaneous and evoked synaptic rewiring in the
neonatal neocortex. Proc Natl Acad Sci U S A. 2006;103(35):13214–9.
21. Liao D, Hessler NA, Malinow R. Activation of postsynaptically silent synapses
during pairing-induced LTP in CA1 region of hippocampal slice. Nature.
1995;375(6530):400–4.
22. Nabavi S, Fox R, Proulx CD, Lin JY, Tsien RY, Malinow R. Engineering a
memory with LTD and LTP. Nature. 2014;511(7509):348–52.
23. Maletic-Savatic M, Malinow R, Svoboda K. Rapid dendritic morphogenesis
in CA1 hippocampal dendrites induced by synaptic activity. Science.
1999;283(5409):1923–7.
24. Crick F. Do dendritic spines twitch? Trends Neurosci. 1982;5:44–6.
25. Hofer SB, Mrsic-Flogel TD, Bonhoeffer T, Hubener M. Experience leaves a
lasting structural trace in cortical circuits. Nature. 2009;457(7227):313–7.
26. Ebbinghaus H. Über das Gedächtnis. Untersuchungen zur experimentellen
Psychologie. Leipzig: Duncker & Humblot; 1885.
27. Hofer SB, Mrsic-Flogel TD, Bonhoeffer T, Hubener M. Prior
experience enhances plasticity in adult visual cortex. Nat Neurosci.
2006;9(1):127–32.
 Poo et al. BMC Biology (2016)4:
28. Xu T, Yu X, Perlik AJ, Tobin WF, Zweig JA, Tennant K, et al. Rapid formation
and selective stabilization of synapses for enduring motor memories.
Nature. 2009;462(7275):915–9.
29. Yang G, Pan F, Gan WB. Stably maintained dendritic spines are associated
with lifelong memories. Nature. 2009;462(7275):920–4.
30. Lai CS, Franke TF, Gan WB. Opposite effects of fear conditioning and
extinction on dendritic spine remodelling. Nature. 2012;483(7387):87–91.
31. Moczulska KE, Tinter-Thiede J, Peter M, Ushakova L, Wernle T, Bathellier B, et al.
Dynamics of dendritic spines in the mouse auditory cortex during memory
formation and memory recall. Proc Natl Acad Sci U S A. 2013;110(45):18315–20.
32. Roberts TF, Tschida KA, Klein ME, Mooney R. Rapid spine stabilization and
synaptic enhancement at the onset of behavioural learning. Nature.
2010;463(7283):948–52.
33. Hubener M, Bonhoeffer T. Searching for engrams. Neuron. 2010;67(3):363–71.
34. Hayashi-Takagi A, Yagishita S, Nakamura M, Shirai F, Wu YI, Loshbaugh AL,
et al. Labelling and optical erasure of synaptic memory traces in the motor
cortex. Nature. 2015;525(7569):333–8.
35. Davis HP, Squire LR. Protein synthesis and memory: a review. Psychol Bull.
1984;96(3):518–59.
36. Goelet P, Castellucci VF, Schacher S, Kandel ER. The long and the short of
long-term memory–a molecular framework. Nature. 1986;322(6078):419–22.
37. Alberini CM. Transcription factors in long-term memory and synaptic
plasticity. Physiol Rev. 2009;89(1):121–45.
38. Bliss TV, Collingridge GL. A synaptic model of memory: long-term
potentiation in the hippocampus. Nature. 1993;361(6407):31–9.
39. Wang DO, Martin KC, Zukin RS. Spatially restricting gene expression by local
translation at synapses. Trends Neurosci. 2010;33(4):173–82.
40. Cajigas IJ, Tushev G, Will TJ, tom Dieck S, Fuerst N, Schuman EM. The local
transcriptome in the synaptic neuropil revealed by deep sequencing and
high-resolution imaging. Neuron. 2012;74(3):453–66.
41. Miyashiro K, Dichter M, Eberwine J. On the nature and differential
distribution of mRNAs in hippocampal neurites: implications for neuronal
functioning. Proc Natl Acad Sci U S A. 1994;91(23):10800–4.
42. Poon MM, Choi SH, Jamieson CA, Geschwind DH, Martin KC. Identification
of process-localized mRNAs from cultured rodent hippocampal neurons.
J Neurosci. 2006;26(51):13390–9.
43. Aakalu G, Smith WB, Nguyen N, Jiang C, Schuman EM. Dynamic visualization of
local protein synthesis in hippocampal neurons. Neuron. 2001;30(2):489–502.
44. Wang DO, Kim SM, Zhao Y, Hwang H, Miura SK, Sossin WS, et al. Synapse-
and stimulus-specific local translation during long-term neuronal plasticity.
Science. 2009;324(5934):1536–40.
45. Kang H, Schuman EM. A requirement for local protein synthesis in neurotrophin-
induced hippocampal synaptic plasticity. Science. 1996;273(5280):1402–6.
46. Martin KC, Casadio A, Zhu H, Yaping E, Rose JC, Chen M, et al. Synapse-specific,
long-term facilitation of aplysia sensory to motor synapses: a function for local
protein synthesis in memory storage. Cell. 1997;91(7):927–38.
47. Kim S, Martin KC. Neuron-wide RNA transport combines with netrin-mediated
local translation to spatially regulate the synaptic proteome. eLife. 2015;4.
48. Rajendran L, Bali J, Barr MM, Court FA, Kramer-Albers EM, Picou F, et al. Emerging
roles of extracellular vesicles in the nervous system. J Neurosci. 2014;34(46):15482–9.
49. Zeisel A, Munoz-Manchado AB, Codeluppi S, Lonnerberg P, La Manno G,
Jureus A, et al. Brain structure. Cell types in the mouse cortex and hippocampus
revealed by single-cell RNA-seq. Science. 2015;347(6226):1138–42.
50. Kasthuri N, Hayworth KJ, Berger DR, Schalek RL, Conchello JA, Knowles-Barley S,
et al. Saturated reconstruction of a volume of neocortex. Cell. 2015;162(3):648–61.
51. Tsien RY. Very long-term memories may be stored in the pattern of holes in
the perineuronal net. Proc Natl Acad Sci U S A. 2013;110(30):12456–61.
52. Kandel ER. The molecular biology of memory: cAMP, PKA, CRE, CREB-1,
CREB-2, and CPEB. Mol Brain. 2012;5:14.
53. Kandel ER, Dudai Y, Mayford MR. The molecular and systems biology of
memory. Cell. 2014;157(1):163–86.
54. Garner AR, Rowland DC, Hwang SY, Baumgaertel K, Roth BL, Kentros C, et al.
Generation of a synthetic memory trace. Science. 2012;335(6075):1513–6.
55. Reijmers LG, Perkins BL, Matsuo N, Mayford M. Localization of a stable
neural correlate of associative memory. Science. 2007;317(5842):1230–3.
56. Alarcon JM, Malleret G, Touzani K, Vronskaya S, Ishii S, Kandel ER, et al.
Chromatin acetylation, memory, and LTP are impaired in CBP+/- mice: a
model for the cognitive deficit in Rubinstein-Taybi syndrome and its
amelioration. Neuron. 2004;42(6):947–59.
57. Korzus E, Rosenfeld MG, Mayford M. CBP histone acetyltransferase activity is
a critical component of memory consolidation. Neuron. 2004;42(6):961–72.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 17 of 18
58. Rubinstein JH, Taybi H. Broad thumbs and toes and facial abnormalities. a
possible mental retardation syndrome. Am J Dis Child. 1963;105:588–608.
59. Fischer A, Sananbenesi F, Wang X, Dobbin M, Tsai LH. Recovery of learning and
memory is associated with chromatin remodelling. Nature. 2007;447(7141):178–82.
60. Ricobaraza A, Cuadrado-Tejedor M, Marco S, Perez-Otano I, Garcia-Osta A.
Phenylbutyrate rescues dendritic spine loss associated with memory deficits
in a mouse model of Alzheimer disease. Hippocampus. 2012;22(5):1040–50.
61. Miller CA, Sweatt JD. Covalent modification of DNA regulates memory
formation. Neuron. 2007;53(6):857–69.
62. Rudenko A, Dawlaty MM, Seo J, Cheng AW, Meng J, Le T, et al. Tet1 is critical for
neuronal activity-regulated gene expression and memory extinction. Neuron.
2013;79(6):1109–22.
63. Bredy TW, Wu H, Crego C, Zellhoefer J, Sun YE, Barad M. Histone modifications
around individual BDNF gene promoters in prefrontal cortex are associated
with extinction of conditioned fear. Learn Mem. 2007;14(4):268–76.
64. Haggarty SJ, Tsai LH. Probing the role of HDACs and mechanisms of
chromatin-mediated neuroplasticity. Neurobiol Learn Mem. 2011;96(1):41–52.
65. Scandura JM, Roboz GJ, Moh M, Morawa E, Brenet F, Bose JR, et al. Phase 1
study of epigenetic priming with decitabine prior to standard induction
chemotherapy for patients with AML. Blood. 2011;118(6):1472–80.
66. Graff J, Tsai LH. The potential of HDAC inhibitors as cognitive enhancers.
Annu Rev Pharmacol Toxicol. 2013;53:311–30.
67. Chen S, Cai D, Pearce K, Sun PY, Roberts AC, Glanzman DL. Reinstatement
of long-term memory following erasure of its behavioral and synaptic
expression in Aplysia. eLife. 2014;3:e03896.
68. Frey U, Morris RG. Synaptic tagging and long-term potentiation. Nature.
1997;385(6616):533–6.
69. Madabhushi R, Gao F, Pfenning AR, Pan L, Yamakawa S, Seo J, et al. Activity-
induced DNA breaks govern the expression of neuronal early-response
genes. Cell. 2015;161(7):1592–605.
70. Tsai LH, Graff J. On the resilience of remote traumatic memories against
exposure therapy-mediated attenuation. EMBO Rep. 2014;15(8):853–61.
71. Mullins C, Fishell G, Tsien RW. Unifying views of autism spectrum disorders:
a consideration of autoregulatory feedback loops. Neuron. 2016. In press.
72. Bourne JN, Harris KM. Nanoscale analysis of structural synaptic plasticity.
Curr Opin Neurobiol. 2012;22(3):372–82.
73. Yuste R, Bonhoeffer T. Morphological changes in dendritic spines associated
with long-term synaptic plasticity. Annu Rev Neurosci. 2001;24:1071–89.
74. Bienenstock EL, Cooper LN, Munro PW. Theory for the development of
neuron selectivity: orientation specificity and binocular interaction in visual
cortex. J Neurosci. 1982;2(1):32–48.
75. Deisseroth K, Bito H, Schulman H, Tsien RW. Synaptic plasticity: a molecular
mechanism for metaplasticity. Curr Biol. 1995;5(12):1334–8.
76. Rabinowitch I, Segev I. Two opposing plasticity mechanisms pulling a single
synapse. Trends Neurosci. 2008;31(8):377–83.
77. Shin SM, Zhang N, Hansen J, Gerges NZ, Pak DT, Sheng M, et al. GKAP
orchestrates activity-dependent postsynaptic protein remodeling and
homeostatic scaling. Nat Neurosci. 2012;15(12):1655–66.
78. Okuno H, Akashi K, Ishii Y, Yagishita-Kyo N, Suzuki K, Nonaka M, et al. Inverse
synaptic tagging of inactive synapses via dynamic interaction of Arc/Arg3.1
with CaMKIIbeta. Cell. 2012;149(4):886–98.
79. Li L, Li Y, Tsien RW. Maintaining local receptor homeostasis along neuronal dendrites
via endo-exocytic coupling. Poster abstract# 704.14/D18. San Diego: Society for
Neuroscience.
80. Song S, Sjostrom PJ, Reigl M, Nelson S, Chklovskii DB. Highly nonrandom features
of synaptic connectivity in local cortical circuits. PLoS Biol. 2005;3(3):e68.
81. Thiagarajan TC, Lindskog M, Malgaroli A, Tsien RW. LTP and adaptation to
inactivity: overlapping mechanisms and implications for metaplasticity.
Neuropharmacology. 2007;52(1):156–75.
82. Buzsaki G, Mizuseki K. The log-dynamic brain: how skewed distributions
affect network operations. Nat Rev Neurosci. 2014;15(4):264–78.
83. Turrigiano GG, Nelson SB. Homeostatic plasticity in the developing nervous
system. Nat Rev Neurosci. 2004;5(2):97–107.
84. Altman J, Das GD. Autoradiographic and histological evidence of postnatal
hippocampal neurogenesis in rats. J Comp Neurol. 1965;124(3):319–35.
85. Eriksson PS, Perfilieva E, Bjork-Eriksson T, Alborn AM, Nordborg C, Peterson DA,
et al. Neurogenesis in the adult human hippocampus. Nat Med.
1998;4(11):1313–7.
86. Spalding KL, Bergmann O, Alkass K, Bernard S, Salehpour M, Huttner HB,
et al. Dynamics of hippocampal neurogenesis in adult humans. Cell.
2013;153(6):1219–27.
 Poo et al. BMC Biology (2016)4:
87. Clelland CD, Choi M, Romberg C, Clemenson Jr GD, Fragniere A, Tyers P,
et al. A functional role for adult hippocampal neurogenesis in spatial
pattern separation. Science. 2009;325(5937):210–3.
88. Sahay A, Scobie KN, Hill AS, O'Carroll CM, Kheirbek MA, Burghardt NS, et al.
Increasing adult hippocampal neurogenesis is sufficient to improve pattern
separation. Nature. 2011;472(7344):466–70.
89. Amaral DG, Scharfman HE, Lavenex P. The dentate gyrus: fundamental
neuroanatomical organization (dentate gyrus for dummies). Prog Brain Res.
2007;163:3–22.
90. Chawla MK, Guzowski JF, Ramirez-Amaya V, Lipa P, Hoffman KL, Marriott LK,
et al. Sparse, environmentally selective expression of Arc RNA in the upper
blade of the rodent fascia dentata by brief spatial experience. Hippocampus.
2005;15(5):579–86.
91. Jung MW, McNaughton BL. Spatial selectivity of unit activity in the hippocampal
granular layer. Hippocampus. 1993;3(2):165–82.
92. O'Reilly RC, McClelland JL. Hippocampal conjunctive encoding, storage, and
recall: avoiding a trade-off. Hippocampus. 1994;4(6):661–82.
93. Bakker A, Kirwan CB, Miller M, Stark CE. Pattern separation in the human
hippocampal CA3 and dentate gyrus. Science. 2008;319(5870):1640–2.
94. Gilbert PE, Kesner RP, Lee I. Dissociating hippocampal subregions: double
dissociation between dentate gyrus and CA1. Hippocampus. 2001;11(6):626–36.
95. Esposito MS, Piatti VC, Laplagne DA, Morgenstern NA, Ferrari CC, Pitossi FJ,
et al. Neuronal differentiation in the adult hippocampus recapitulates
embryonic development. J Neurosci. 2005;25(44):10074–86.
96. Schmidt-Hieber C, Jonas P, Bischofberger J. Enhanced synaptic plasticity
in newly generated granule cells of the adult hippocampus. Nature.
2004;429(6988):184–7.
97. Li Y, Aimone JB, Xu X, Callaway EM, Gage FH. Development of GABAergic
inputs controls the contribution of maturing neurons to the adult
hippocampal network. Proc Natl Acad Sci U S A. 2012;109(11):4290–5.
98. Marin-Burgin A, Mongiat LA, Pardi MB, Schinder AF. Unique processing
during a period of high excitation/inhibition balance in adult-born neurons.
Science. 2012;335(6073):1238–42.
99. Ge S, Yang CH, Hsu KS, Ming GL, Song H. A critical period for enhanced
synaptic plasticity in newly generated neurons of the adult brain. Neuron.
2007;54(4):559–66.
100. Nakashiba T, Cushman JD, Pelkey KA, Renaudineau S, Buhl DL, McHugh TJ,
et al. Young dentate granule cells mediate pattern separation, whereas old
granule cells facilitate pattern completion. Cell. 2012;149(1):188–201.
101. Ikrar T, Guo N, He K, Besnard A, Levinson S, Hill A, et al. Adult neurogenesis
modifies excitability of the dentate gyrus. Front Neural Circuits. 2013;7:204.
102. Piatti VC, Ewell LA, Leutgeb JK. Neurogenesis in the dentate gyrus: carrying
the message or dictating the tone. Front Neurosci. 2013;7:50.
103. Sahay A, Wilson DA, Hen R. Pattern separation: a common function for new
neurons in hippocampus and olfactory bulb. Neuron. 2011;70(4):582–8.
104. Leutgeb JK, Leutgeb S, Moser MB, Moser EI. Pattern separation in the
dentate gyrus and CA3 of the hippocampus. Science. 2007;315(5814):961–6.
105. Danielson NB, Kaifosh P, Zaremba JD, Lovett-Barron M, Tsai J, Denny CA,
et al. Distinct contribution of adult-born hippocampal granule cells to
context encoding. Neuron. 2016;90(1):101–12.
106. Ito M. Long-term depression. Annu Rev Neurosci. 1989;12:85–102.
107. Levy WB, Steward O. Temporal contiguity requirements for long-term
associative potentiation/depression in the hippocampus. Neuroscience.
1983;8(4):791–7.
108. Markram H, Lubke J, Frotscher M, Sakmann B. Regulation of synaptic efficacy
by coincidence of postsynaptic APs and EPSPs. Science. 1997;275(5297):213–5.
109. Bell CC, Han VZ, Sugawara Y, Grant K. Synaptic plasticity in a cerebellum-like
structure depends on temporal order. Nature. 1997;387(6630):278–81.
110. Zhang LI, Tao HW, Holt CE, Harris WA, Poo M. A critical window for
cooperation and competition among developing retinotectal synapses.
Nature. 1998;395(6697):37–44.
111. Dan Y, Poo MM. Spike timing-dependent plasticity: from synapse to
perception. Physiol Rev. 2006;86(3):1033–48.
112. Xu S, Jiang W, Poo MM, Dan Y. Activity recall in a visual cortical ensemble.
Nat Neurosci. 2012;15(3):449–455;S441–2.
113. Bi G, Poo M. Distributed synaptic modification in neural networks induced
by patterned stimulation. Nature. 1999;401(6755):792–6.
114. Abeles M. Local cortical circuits: an electrophysiological study. Berlin;
New York: Springer-Verlag; 1982.
115. Long MA, Jin DZ, Fee MS. Support for a synaptic chain model of neuronal
sequence generation. Nature. 2010;468(7322):394–9.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Page 18 of 18
116. Okubo TS, Mackevicius EL, Payne HL, Lynch GF, Fee MS. Growth and
splitting of neural sequences in songbird vocal development. Nature.
2015;528(7582):352–7.
117. Harvey CD, Coen P, Tank DW. Choice-specific sequences in parietal cortex
during a virtual-navigation decision task. Nature. 2012;484(7392):62–8.
118. Sumbre G, Muto A, Baier H, Poo MM. Entrained rhythmic activities of
neuronal ensembles as perceptual memory of time interval. Nature.
2008;456(7218):102–6.
119. Suh J, Rivest AJ, Nakashiba T, Tominaga T, Tonegawa S. Entorhinal cortex
layer III input to the hippocampus is crucial for temporal association
memory. Science. 2011;334:1415–20.
120. Kitamura T, Pignatelli M, Suh J, Kohara K, Yoshiki A, Abe K, et al. Island cells
control temporal association memory. Science. 2014;343:896–901.
121. Kitamura T, Macdonald CJ, Tonegawa S. Entorhinal-hippocampal neuronal
circuits bridge temporally discontiguous events. Learn Mem. 2015;22:438–43.
122. Eichenbaum H. Time cells in the hippocampus: a new dimension for
mapping memories. Nat Rev Neurosci. 2014;15:732–44.
123. Manns JR, Howard MW, Eichenbaum H. Gradual changes in hippocampal
activity support remembering the order of events. Neuron. 2007;56:530–40.
124. Pastalkova E, Itskov V, Amarasingham A, Buzsáki G. Internally generated cell
assembly sequences in the rat hippocampus. Science. 2008;321:1322–7.
125. Buhusi CV, Meck WH. What makes us tick? Functional and neural
mechanisms of interval timing. Nat Rev Neurosci. 2005;6(10):755–65.
126. Gallistel CR, Gibbon J. Time, rate, and conditioning. Psychol Rev.
2000;107(2):289–344.
127. Mauk MD, Buonomano DV. The neural basis of temporal processing.
Annu Rev Neurosci. 2004;27:307–40.
128. Jazayeri M, Shadlen MN. A neural mechanism for sensing and reproducing
a time interval. Curr Biol. 2015;25(20):2599–609.
129. Xu M, Zhang SY, Dan Y, Poo MM. Representation of interval timing by
temporally scalable firing patterns in rat prefrontal cortex. Proc Natl Acad
Sci U S A. 2014;111(1):480–5.
130. Buzsaki G. Hippocampal sharp wave-ripple: a cognitive biomarker for
episodic memory and planning. Hippocampus. 2015;25(10):1073–188.
131. Dupret D, O'Neill J, Pleydell-Bouverie B, Csicsvari J. The reorganization and
reactivation of hippocampal maps predict spatial memory performance.
Nat Neurosci. 2010;13(8):995–1002.
132. Wilson MA, McNaughton BL. Reactivation of hippocampal ensemble
memories during sleep. Science. 1994;265(5172):676–9.
133. O’Keefe J, Nadel L. The hippocampus as a cognitive map. Oxford: Oxford
University Press; 1978.
134. Grosmark AD, Buzsaki G. Diversity in neural firing dynamics supports both
rigid and learned hippocampal sequences. Science. 2016;351(6280):1440–3.
135. Peyrache A, Khamassi M, Benchenane K, Wiener SI, Battaglia FP. Replay of
rule-learning related neural patterns in the prefrontal cortex during sleep.
Nat Neurosci. 2009;12(7):919–26.
136. Girardeau G, Benchenane K, Wiener SI, Buzsaki G, Zugaro MB. Selective
suppression of hippocampal ripples impairs spatial memory. Nat Neurosci.
2009;12(10):1222–3.
137. Diba K, Buzsaki G. Forward and reverse hippocampal place-cell sequences
during ripples. Nat Neurosci. 2007;10(10):1241–2.
138. Pfeiffer BE, Foster DJ. Hippocampal place-cell sequences depict future paths
to remembered goals. Nature. 2013;497(7447):74–9.
139. Jadhav SP, Kemere C, German PW, Frank LM. Awake hippocampal sharp-
wave ripples support spatial memory. Science. 2012;336(6087):1454–8.
140. Buzsaki G, Moser EI. Memory, navigation and theta rhythm in the
hippocampal-entorhinal system. Nat Neurosci. 2013;16(2):130–8.
141. Von Neumann J, Kurzweil R. The computer and the brain. Yale University
Press; 2012.
142. Miller GA. The cognitive revolution: a historical perspective. Trends Cogn
Sci. 2003;7(3):141–4.
143. Hige T, Aso Y, Modi MN, Rubin GM, Turner GC. Heterosynaptic plasticity
underlies aversive olfactory learning in Drosophila. Neuron. 2015;88(5):985–98.
144. Stevens CF. What the fly's nose tells the fly's brain. Proc Natl Acad Sci U S A.
2015;112(30):9460–5.
145. Janak PH, Tye KM. From circuits to behaviour in the amygdala. Nature.
2015;517(7534):284–92.
146. Perisse E, Burke C, Huetteroth W, Waddell S. Shocking revelations and
saccharin sweetness in the study of Drosophila olfactory memory. Curr Biol.
2013;23(17):R752–63.
 1998 Nature America Inc. • http://medicine.nature.com
ARTICLES

Neurogenesis in the adult human hippocampus

PETER S. ERIKSSON1,4, EKATERINA PERFILIEVA1, THOMAS BJÖRK-ERIKSSON2, ANN-MARIE ALBORN1,

CLAES NORDBORG3, DANIEL A. PETERSON4 & FRED H. GAGE4

Department of Clinical Neuroscience, Institute of Neurology 1, Department of Oncology2, Department of Pathology3,

Sahlgrenska University Hospital, 41345 Göteborg, Sweden
4
Laboratory of Genetics, The Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla,
California 92037, USA
Correspondence should be addressed to F.H.G.

The genesis of new cells, including neurons, in the adult human brain has not yet been demon-
strated. This study was undertaken to investigate whether neurogenesis occurs in the adult
human brain, in regions previously identified as neurogenic in adult rodents and monkeys.
Human brain tissue was obtained postmortem from patients who had been treated with the
thymidine analog, bromodeoxyuridine (BrdU), that labels DNA during the S phase. Using im-
munofluorescent labeling for BrdU and for one of the neuronal markers, NeuN, calbindin or neu-
1998 Nature America Inc. • http://medicine.nature.com

ron specific enolase (NSE), we demonstrate that new neurons, as defined by these markers, are
generated from dividing progenitor cells in the dentate gyrus of adult humans. Our results further
indicate that the human hippocampus retains its ability to generate neurons throughout life.

Loss of neurons is thought to be irreversible in the adult human one intravenous infusion (250 mg; 2.5 mg/ml, 100 ml) of bro-
brain, because dying neurons cannot be replaced. This inability modeoxyuridine (BrdU) for diagnostic purposes11. One patient
to generate replacement cells is thought to be an important diagnosed with a similar type and location of cancer, but with-
cause of neurological disease and impairment. In most brain re- out BrdU treatment, was included as a control. A thymidine ana-
gions, the generation of neurons is generally confined to a dis- log, BrdU is incorporated into the DNA of dividing cells and can
crete developmental period. Exceptions are found in the dentate be detected immunohistochemically in their progeny5,12,13.
gyrus and the subventricular zone of several species that have
been shown to generate new neurons well into the postnatal and Cell genesis and survival in the adult human dentate gyrus
adult period1–6. Granule neurons are generated throughout life The number of surviving labeled, proliferating progenitors was
from a population of continuously dividing progenitor cells re-
siding in the subgranular zone of the dentate gyrus in the rodent a b
brain5. ‘Newborn’ neurons generated from these progenitor cells
migrate into the granule cell layer, differentiate, extend axons
and express neuronal marker proteins7–10.
We examined whether progenitor cells reside in the adult
human hippocampus and whether new neurons are born within
the dentate gyrus of the adult human brain. Postmortem tissue
from the hippocampus and the subventricular zone of caudate
nucleus was obtained from cancer patients (n = 5) who received
c d
Fig. 1 Newly generated cells can be detected in the adult human brain in
patients previously treated with BrdU. a, The hippocampal region of the
adult human brain immunoperoxidase-stained for the neuronal marker
NeuN. b, The hippocampal dentate gyrus granule cell layer (GCL) visualized
with immunoperoxidase staining for NeuN. c, Differential interference con-
trast photomicrograph showing BrdU-labeled nuclei (arrows) in the dentate
granule cell layer (GCL). d, Differential interference contrast photomicro-
graph showing a BrdU-labeled nucleus (arrow) in the human dentate GCL.
BrdU-positive nuclei have a rounded appearance and resemble the chro- e f
matin structure of mature granule cells and are found within the granule cell
layer. e, Differential interference contrast photomicrograph showing BrdU-
positive cells (arrows) adjacent to the ependymal lining in the subventricular
zone of the human caudate nucleus. Cells with elongated nuclei resembling
migrating cells are in the rat subventricular zone (SVZ). f, Differential inter-
ference contrast photomicrograph showing BrdU-positive cells (arrows) with
round to elongated nuclei in the subventricular zone of the human caudate
nucleus. All scale bars represent 50 µm.

NATURE MEDICINE • VOLUME 4 • NUMBER 11 • NOVEMBER 1998 1313

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 1998 Nature America Inc. • http://medicine.nature.com
ARTICLES
a Number of BrdU-labeled cells in SGZ
BrdU-labeled cells (mm3)
Days post-
BrdU injection
Age at death (years)
Fig. 2 Quantitation of newly generated cells in the adult human hip-
pocampus. The density of BrdU immunoperoxidase-stained cells in the sub-
granular zone (SGC) (a) and the granule cell layer (GCL) (b) and the hilus
(c) was determined in 5 to 7 sections per patient (mean number of BrdU-
positive cells per mm3 sample volume ± s.e.m.) The corresponding patient’s
age at the time of death and interval as BrdU infusion are given for each
BrdU-treated patient (n = 5).
1998 Nature America Inc. • http://medicine.nature.com
quantified using immunohistochemical staining for BrdU and
unbiased counting techniques14,15,16. BrdU-labeled cells were
quantified in the granule cell layer and the subgranular zone of
the dentate gyrus and in the hilus (area CA4; Fig. 1a). In all
BrdU-treated patients, the granule cell layer contained BrdU-pos-
itive, round- to oval-shaped nuclei with the typical morphology
of granule cell neuron nuclei (Fig. 1b–d). The morphological ap-
pearance and localization of these nuclei were similar to those of
the nuclei of cells found in the dentate gyrus of adult mouse, rat
and marmoset monkey5,6,17. No BrdU-immunoreactive profiles
could be detected in the control patient that had not received
BrdU. The presence of BrdU-positive nuclei in a brain region that
is neurogenic in many other mammalian and primate species in-
dicates that proliferating neural progenitor cells are present in
the adult human dentate gyrus.
The number of BrdU-positive cells in the granule cell layer, the
subgranular zone and hilus varied between individuals, as may
be expected because of the varying post-infusion intervals and
different ages of the patients (Fig. 2a–c). There is an apparent de-
cline in the number of cells that are detected in the patients that
had the longest interval between BrdU treatment and histologi-
cal assessment. This decline may indicate a progressive death of
the newly generated cells over time. However, the small sample
weakens any quantitative conclusion.
BrdU-labeled cells co-express neuronal markers
To determine the cell fate of the BrdU-immunoreactive cells, we
did triple immunofluorescent labeling for BrdU and cell-specific
markers, including glial fibrillary acidic protein (GFAP), a marker
for astroglia, and one of the neuronal markers, NeuN 18,19, cal-
bindin20 or neuron specific enolase21 (NSE). Confocal microscopy
was used to determine the phenotype of the BrdU-positive cells
(Figs. 3 and 4). No evidence for cross-reactivity between GFAP-
immunoreactivity and any of the three neuronal markers was
detected in these sections. GFAP-immunopositive processes were
observed to surround neurons and their processes, but did not
co-express in the same cell (Fig. 3e–h). Autofluorescence was ob-
served in this tissue, but the nonspecific emission artifact was
easily detected by its contribution to multiple wavelengths (Fig.
3a–d). We took care to evaluate potential sources of autofluores-
1314
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
b Number of BrdU-labeled cells in GCL
BrdU-labeled cells (mm3)
Days post-
BrdU injection
Age at death (years)
c Number of BrdU-labeled cells in hilus
BrdU-labeled cells (mm3)
Days post-
BrdU injection
Age at death (years)
cence in the human tissue by directly imaging untreated sections
and sections from the control patient both with and without the
pretreatment for BrdU-immunohistochemical detection in the
absence of antibodies against BrdU. The autofluorescence was
present even in completely untreated sections and was restricted
to the cytoplasm of neuronally marked cells or to endothelial
cells and red blood cells and could be clearly distinguished from
BrdU-immunoreactive profiles (Fig. 3). The majority of the
NeuN-positive neurons that double-labeled with BrdU were lo-
cated within or near the granule cell layer and had the morpho-
logical characteristics of granule cell neurons: round or oval
nuclei with small- to medium-sized cell bodies (Fig. 3). Confocal
microscopic Z-series analyses of BrdU-positive nuclei unambigu-
ously identified double-labeling with NeuN and demonstrated
the existence of newly generated (BrdU-labeled) granule cell neu-
rons in the dentate gyrus of the adult human brain (Fig. 3d–h).
The phenotypic expression of these BrdU-positive neurons in
adult humans was equivalent to the expression found in adult
rodents5,10,12,17 (Fig. 3i–l). The average fraction of BrdU-NeuN dou-
ble-labeled cells in the granule cell layer was 22.0 ± 2.4% among
the BrdU-treated patients.
Neurons double-labeled with BrdU and NSE could also be de-
tected in all BrdU-treated subjects (Fig. 4e–f). The average frac-
tion of BrdU-labeled cells that were also NSE-labeled in the
granule cell layer was 22.7 ± 2.8%. Cells that double-labeled for
BrdU and calbindin were also observed and provided further sup-
port for the occurrence of neurogenesis in the adult SGZ, GCL
and hilus (Fig. 4a–d). The average fraction of BrdU-labeled cells
that were also calbindin-positive in the granule cell layer was 7.9
± 2.2% among the BrdU-treated patients. The dentate gyrus from
all individuals contained a fraction of BrdU-positive cells that
were also GFAP-positive, 18.1 ± 1.8%, and had the typical mor-
phology of star-shaped glial cells with small, irregularly shaped
nuclei and small cell bodies with thin, GFAP-positive processes
(Fig. 4c–d). The dentate gyrus from all subjects contained BrdU-
NATURE MEDICINE • VOLUME 4 • NUMBER 11 • NOVEMBER 1998
 1998 Nature America Inc. • http://medicine.nature.com
ARTICLES

Fig. 3 Newly generated cells

in the adult human dentate a b c d
gyrus can express a neuronal
phenotype. Simultaneous de-
tection of immunofluorescent
labels for NeuN (a; scale bar
represents 25 µm), BrdU (b)
and GFAP (c) for detection of
astrocytes and a merge of
these signals in x-, y- and z-
registration (d) examined by
confocal microscopy showed e f g h
that BrdU-labeled nuclei could
specifically co-express NeuN
without expressing GFAP (ar-
rows in a–d). In addition to
specific BrdU-labeling, some
neurons contained nonspecific
fluorescence in the green and
red emission, reflecting their
accumulation of lipofuchsin
granules (arrowheads in a–d). i j k l
Red blood cells and endothelial
cells, present in several small
1998 Nature America Inc. • http://medicine.nature.com

blood vessels, also emit non-

specific green and red fluores-
cence (small blood vessel
indicated by arrowheads,
larger blood vessels indicated
by asterisks in e–h). The speci-
ficity of BrdU-NeuN co-expres-
sion in three-dimensions is demonstrated by a series of focal planes above human dentate gyrus is equivalent to BrdU-labeled neurons in the adult
(e,f) and below (g, h) the focal plane shown in d (arrows indicate the rat dentate gyrus (i–l; scale bar represents 25 µm). The rat tissue (i–l) is at
same cell as in d). The appearance of adult BrdU-positive neurons in the a different magnification than the human tissue (a–h).

positive cells that were immunonegative for both neuronal and
glial markers. These immunonegative cells likely represent quies-
cent undifferentiated cells, newborn cells of a phenotype not ex-
amined here and/or a pool of asymmetrically dividing
progenitor cells. These cells were characterized by small, round-
to-oval BrdU-positive nuclei and the absence of cell-specific im-
munoreactivity.
To further confirm the presence of neurogenesis, we double-la-
beled using antibodies against BrdU and either NSE, calbindin or
NeuN with chromagens for brightfield optics (alkaline phos-
phatase (Vector Blue) for BrdU and 3-amino-9-ethylcarbazole
(AEC) for the neuronal markers). Although confocal microscopy
could not be used for imaging, the brightfield chromagens had
the advantage of not fading with examination and not con-
tributing autofluorescence to the image. Examination of bright-
field staining confirmed that BrdU-positive cells in the adult
human hippocampus could express a neuronal phenotype (Fig.
5a–e) morphologically indistinguishable from adult rodent
BrdU-positive neurons (Fig. 5f and g), strongly supporting our
conclusion that neurogenesis occurs in the dentate granule cell
layer of the adult human brain.
BrdU labeling in the subventricular zone of adult human brain
Another neurogenic region, the subventricular zone (SVZ) adja-
cent to the caudate nucleus, was examined for the presence of
BrdU-positive cells. Tissue samples from all BrdU-treated pa-
tients contained BrdU-positive cells within the SVZ (Fig. 1e–f).
BrdU-positive cells did not co-express the cell-specific markers
GFAP and NeuN (data not shown). The morphology of BrdU-la-
beled nuclei within the SVZ was small and round-to-oval, resem-
bling that of progenitor cells in the rat SVZ (ref. 5). This finding
supports the idea that the human SVZ contains progenitor cells
and that these cells are required to migrate from the SVZ before
they differentiate22. We were unable to detect any BrdU-im-
munoreactive cells in tissue from the control patient who had
not received BrdU treatment, supporting the interpretation that
the BrdU staining we report here reflects the persistence in the
adult human brain of cell genesis.
Discussion
Our study demonstrates that cell genesis occurs in human brains
and that the human brain retains the potential for self-renewal
throughout life. Although earlier studies in adult primates have
been unsuccessful in showing neurogenesis in the dentate
gyrus23,24, a recent report has demonstrated neurogenesis in
three-year-old marmoset monkeys6. Although the number of
BrdU-labeled cells entering the neuronal lineage seems to be
lower in the human hippocampus than in marmosets, those
monkeys6 were considerably younger, even in relative terms,
than the humans examined here (average age of 64.4 ± 2.9
years). Therefore, we conclude that, as in rodents5,25, neurogene-
sis in the human dentate gyrus continues throughout life.
Although our results demonstrate that cells in the adult brain
undergo cell division and that some of the newly generated cells
can survive and differentiate into cells with morphological and
phenotypic characteristics of neurons, we have not proven that
these newly generated cells are functional. We also do not yet
know the biological significance of cell genesis in the adult
human brain. However, this does provide a basis to investigate a
newly discovered type of ‘neuroplasticity’ in humans, one based

NATURE MEDICINE • VOLUME 4 • NUMBER 11 • NOVEMBER 1998 1315

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 1998 Nature America Inc. • http://medicine.nature.com
ARTICLES

Fig. 4 Newly
generated cells in
a b c d
the adult human
dentate gyrus can
express additional
neuronal pheno-
types. The cal-
cium-binding
protein, calbindin,
is expressed by
certain neuronal
populations, in-
cluding dentate granule neurons, in vivo. a, Fluorescent labeling of calbindin (red) e f
and GFAP (blue) discriminated between granule neurons and astrocytes. Scale bar
represents 10 µm. The arrow indicates a newly generated, BrdU-labeled calbindin-
positive neuron shown with the label-colors merged in b. c, Another calbindin-posi-
tive neuron (arrow) that co-expresses BrdU (d). Newly generated cells may also
differentiate into astrocytes (c,d; arrowheads). BrdU-labeled cells can also express
neuron specific enolase (NSE, shown in red; arrows in e and f) without expressing
GFAP (blue).
1998 Nature America Inc. • http://medicine.nature.com

on addition of neurons, that has not been previously considered.
Studies in rodents have shown that the adult hippocampus con-
tains progenitor cells that can be expanded in vitro and grafted
back into the adult brain, where they can respond to regional
cues by differentiation into site-specific phenotypes, including
neurons26,27. The presence of progenitor cells in the human den-
tate gyrus, reported here, indicates that these cells also may be
used for in vitro and in vivo studies of cell differentiation and pos-
sibly subsequent transplantation studies. Furthermore, environ-
mental stimulation can influence the rate of neurogenesis in the
adult and senescent rodent dentate gyrus 12,17. The potential to reg-
ulate human neurogenesis should prove to be an interesting area
of investigation.
Methods
Autopsy material. Human hippocampal tissue was obtained at autopsy
with the full consent of each family. All patients were diagnosed with squa-
mous cell carcinomas at the base of the tongue, in the larynx or in the phar-
ynx. All patients received bromodeoxyuridine (BrdU) (250 mg) dissolved in
saline and given as an intravenous infusion (2.5 mg/ml, 100 ml). The BrdU
was given to the patients to assess the proliferative activity of the tumor
cells, expressed as BrdU-labeling index. No signs of macroscopic or micro-
scopic metastases were found in autopsy material from the cerebrum in any
of the patients. No anti-cancer therapy was administered before or during
BrdU administration to any of the patients.
Tissue preparation. The hippocampal formation and ventricular zone were
dissected out and post-fixed in 4% paraformaldehyde for 24 hours and then
transferred into 30% sucrose solution until equilibrated. The hippocampi
were sectioned (slices 40 µm in thickness) in the coronal plane on a sliding
microtome and stored at –20 °C in a cryoprotecting buffer containing 25%
ethylene glycol, 25% glycerin and 0.05 M phosphate buffer. For compari-
son, sections derived from BrdU-injected adult rats and mice that had been
intracardially perfused with 4% paraformaldehyde were processed in paral-
lel with the human tissue.
Histology. Immunohistochemical detection of BrdU requires a pre-treatment
of the sections to denature DNA (ref. 5). All staining was done on free-floating
sections and the blocking buffer contained both 3% normal donkey serum
and 3% normal human serum (Sigma). For sections stained only for BrdU, a

Fig. 5 Brightfield
double-immunohis-
a b d e
tochemical demon-
stration of neuronal
phenotype. a, Stain-
ing with the neu-
ronal marker NeuN c
labels both dentate
granule cells (top
inset, shown in b)
and hilar neurons
(bottom inset,
shown in c). Scale bar represents 100 µm. Combining NeuN staining with differential f g
interference contrast optics demonstrated that NeuN labeling included the entire nu-
cleus and perikaryal cytoplasm and extended into proximal portions of major den-
drites (arrowheads) in both granule cells (b; scale bar represents 20 µm) and hilar
neurons (c, scale bar represents 20 µm). Newly generated cells could be found in the
dentate granule layer when detected with antibodies against BrdU in conjunction
with NeuN staining (d, scale bar represents 25 µm). Dark blue-stained BrdU-labeled
nuclei can co-express the neuronal marker NeuN shown in red (e, scale bar represents
10 µm). The appearance of BrdU-labeled cells in adult human dentate gyrus is equiv-
alent to that of the adult rat dentate gyrus (f, scale bar represents 25 µm; and g, scale
bar represents 10 µm.).

1316 NATURE MEDICINE • VOLUME 4 • NUMBER 11 • NOVEMBER 1998

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 1998 Nature America Inc. • http://medicine.nature.com
mouse-anti-BrdU antibody (Boehringer; diluted 1:400) was used in combina-
tion with the avidin-biotin complex method using a biotinylated donkey anti-
mouse-IgG antibody (Vector Laboratories, Burlingame, California; diluted
1:167) and reacted with a diaminobenzidine (DAB) chromagen.
Immunofluorescent double- and triple-labeling was done as de-
scribed5,12. For multiple immunostaining, BrdU was detected with a rat anti-
BrdU antibody (Harlan Sera-Lab, Loughborough, England; diluted 1:500)
and visualized with a FITC-conjugated secondary donkey anti-rat antibody
(Jackson ImmunoResearch, West Grove, Pennsylvania; diluted 1:250). For
neuronal phenotype markers, sections were incubated with one of the fol-
lowing antisera: rabbit anti-calbindin antiserum (SWant, Bellinzona,
Switzerland; diluted 1:1,000), or mouse anti-NeuN antiserum (from R.
Mullen18; diluted 1:20), or rabbit anti-NSE antiserum (Polysciences,
Warrington, Pennsylvania; diluted 1:800). These neuronal markers were vi-
sualized by using the species-appropriate Cy3-conjugated secondary anti-
body (Jackson ImmunoResearch, West Grove, Pennsylvania; diluted 1:250).
GFAP was detected in the same sections using a guinea pig anti-GFAP anti-
serum (Advanced Immunochemicals, Long Beach, California; diluted
1:250) and visualized using a Cy5-conjugated donkey anti-guinea pig anti-
body (Jackson ImmunoResearch, West Grove, Pennsylvania; diluted1:250).
For double immunostaining using brightfield chromagens, sections were
incubated with a pooled solution of rat anti-BrdU (Harlan Sera-Lab,
Loughborough, England; diluted 1:500) and mouse anti-NeuN (from R.
Mullen18; diluted 1:100) antisera. After being rinsed and blocked, sections
1998 Nature America Inc. • http://medicine.nature.com
were incubated first with a biotinylated donkey anti-rat antibody (Jackson
ImmunoResearch, West Grove, Pennsylvania; diluted 1:250), followed by
incubation with an alkaline phosphatase avidin–biotin substrate and then
reaction with the blue chromagen (Vector Blue; Vector Laboratories,
Burlingame, California). After being rinsed and blocked further, the sections
then were incubated with a biotinylated donkey anti-mouse antibody
(Jackson ImmunoResearch, West Grove, Pennsylvania; diluted 1:250) fol-
lowed by incubation with a peroxidase avidin-biotin substrate and then re-
action with the red chromagen (3-amino-9-ethylcarbazole or AEC; Vector
Laboratories, Burlingame, California). Sections were mounted and cover-
slipped with an aqueous mounting medium.
Fluorescent signals were imaged using a confocal laser scanning micro-
scope equipped with a krypton/argon mixed gas laser (Bio-Rad MRC1024;
Bio-Rad, Richmond, California) using individual collection of wavelengths
to minimize artifactual detection of non-specific fluorescent emission. At
least 20 BrdU-positive cells within multiple immunofluorescently stained
sections from each patient were examined on the confocal microscope to
determine the phenotype of BrdU-labeled cells. Brightfield images were de-
rived from conventional photomicrographs digitized by using a slide scan-
ner. Figures were composed by using Adobe Photoshop (Adobe Systems,
Mountain View, California) with image adjustments confined to creating an
equivalent signal distribution between panels.
Quantitation. The numbers of BrdU-positive cells in the granule cell layer,
the subgranular layer and the hilus, and their corresponding sample vol-
umes were determined in five to seven immunoperoxidase-stained coronal
sections 40 µm in thickness, at least 240 µm apart, from each patient. Area
estimations were done on an IBAS image analysis system. The section thick-
ness of 40 µm (microtome setting) was used in the dissector estimation of
volume. The number of BrdU-positive cells was counted within the granule
cell layer (GCL) and two cell diameters below the granule cell layer, ignor-
ing the cells in the uppermost focal plane and focusing through the thick-
ness of the section (optical disector principle; refs. 14,15,16) to avoid
oversampling errors. The subgranular zone (SGZ) was defined as the area
from one cell diameter within the GCL from the hilus-GCL border and two
cell diameters below the hilus-GCL border (Fig. 1). Quantitative analysis of
BrdU immunoperoxidase-stained sections was done on a Nikon microscope
equipped with a video camera. The results are expressed as BrdU-positive
cells per sample volume per section.
Acknowledgments
We thank G. Kempermann, T. Palmer, E. Brandon, M.-C. Senut, K. Sakurada,
L. Chehabeddine and M. L. Gage for their comments, and H. van Praag for the
contribution of rat tissue. In addition, we thank L. Kitabayashi for technical as-
sistance. This study was supported by grants from the Swedish Medical
NATURE MEDICINE • VOLUME 4 • NUMBER 11 • NOVEMBER 1998
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
ARTICLES
Research Council (project no. K98-12X-12535-01A), Faculty of Medicine,
University of Göteborg, the Gunvor and Josef Anérs Stiftelse, the John and Brit
Wennerströms Stiftelse for Neurologisk Forskning, the Rune and Ulla Amlövs
Stiftelse for Neurologisk and Reumatologisk Forskning, NHR-fonden, Stiftelsen
Göteborgs MS förenings forsknings och byggnadsfond, Stiftelsen Handlanden
Hjalmar Svenssons Forskningsfond, Göteborgs Läkaresällskap, Hjärnfonden,
The Swedish Society of Medicine, Stiftelsen Lars Hiertas Minne, Stiftelsen Assar
Gabrielssons Fond and the Edit Jacobssons Fond and from NIA and NINDS
and the Alzheimer’s Association (F.H.G.) and the American Federation for
Aging Research (D.A.P.).
RECEIVED 9 SEPTEMBER; ACCEPTED 13 OCTOBER 1998
1. Altman, J. &. Das, G.D. Autoradiographic and histological evidence of postnatal
hippocampal neurogenesis in rats. J. Comp. Neurol. 124, 319–335 (1965).
2. Altman, J. & Das, G.D. Postnatal neurogenesis in the guinea-pig. Nature 214,
1098–1101 (1967).
3. Caviness, V.S. Time of neuron origin in the hippocampus and dentate gyrus of nor-
mal and reeler mutant mice: an autoradiographic analysis. J. Comp. Neurol. 151,
113–120 (1973).
4. Gueneau, G., Privat, A., Drouet, J. & Court, L. Subgranular zone of the dentate
gyrus of young rabbits as a secondary matrix. A high-resolution autoradiographic
study. Dev. Neurosci. 5, 345–358 (1982).
5. Kuhn, H.G., Dickinson-Anson, H. & Gage, F.H. Neurogenesis in the dentate gyrus
of the adult rat: Age-related decrease of neuronal progenitor proliferation. J.
Neurosci. 16, 2027–2033 (1996).
6. Gould, E., Tanapat, P., McEwen, B.S., Flugge, G. & Fuchs, E. Proliferation of granule
cell precursors in the dentate gyrus of adult monkeys is diminished by stress. Proc.
Natl. Acad. Sci. USA 95, 3168–3171 (1998).
7. Kaplan, M.S. & Bell, D.H. Mitotic neuroblasts in the 9-day-old and 11-month-old
rodent hippocampus. J. Neurosci. 4, 1429–1441 (1984).
8. Kaplan, M.S. & Hinds, J.W. Neurogenesis in the adult rat: electron microscopic
analysis of light radioautographs. Science 197, 1092–1094 (1977).
9. Stanfield, B.B. & Trice, J.E. Evidence that granule cells generated in the dentate
gyrus of adult rats extend axonal projections. Exp. Brain Res. 72, 399–406 (1988).
10. Cameron, H.A., Woolley, C.S., McEwen, B.S. & Gould, E. Differentiation of newly
born neurons and glia in the dentate gyrus of the adult rat. Neuroscience 56,
337–344 (1993).
11. Dolbeare, F. Bromodeoxyuridine: a diagnostic tool in biology and medicine, Part I:
Historical perspectives, histochemical methods and cell kinetics. Histochem. J. 27,
339–369 (1995).
12. Kempermann, G., Kuhn, H.G. & Gage, F.H. Genetic influence on neurogenesis in
the dentate gyrus of adult mice. Proc. Natl. Acad. Sci. USA 94, 10409–10414
(1997).
13. del Rio, J.A. & Soriano, E. Immunocytochemical detection of 5’-bromodeoxyuridine
incorporation in the central nervous system of the mouse. Dev. Brain Res. 49,
311–317 (1989).
14. Gundersen, H.J. et al. The new stereological tools: disector, fractionator, nucleator
and point sampled intercepts and their use in pathological research and diagnosis.
APMIS 96, 857–881 (1988).
15. West, M.J. Regionally specific loss of neurons in the aging human hippocampus.
Neurobiol. Aging 14, 287–293 (1993).
16. Coggeshall, R.E. & Lekan, H.A. Methods for determining numbers of cells and
synapses: a case for more uniform standards of review. J. Comp. Neurol. 364, 6–15
(1996).
17. Kempermann, G., Kuhn, H.G. & Gage, F.H. More hippocampal neurons in adult
mice living in an enriched environmen Nature 386, 493–495 (1997).
18. Mullen, R.J., Buck, C.R. & Smith, A.M. NeuN, a neuronal specific nuclear protein in
vertebrates. Development 116, 201–211 (1992).
19. Wolf, H.K. et al. NeuN: A useful neuronal marker for diagnostic histopathology. J.
Histochem. Cytochem. 44, 1167–1171 (1996).
20 Sloviter, R.S. Calcium-binding protein (calbindin-D28k) and parvalbumin immuno-
cytochemistry: localization in the rat hippocampus with specific reference to the se-
lective vulnerability of hippocampal neurons to seizure activity. J. Comp. Neurol.
280, 183–196 (1989).
21. Schmechel, D., Marango, P.J., Zis, A.P., Brightman, M. & Goodwin, F.K. Brain eno-
lases as specific markers of neuronal and glial cells. Science 199, 313–315 (1978).
22. Kirschenbaum, B. et al. In vitro neuronal production and differentiation by precur-
sor cells derived from the adult human forebrain. Cereb. Cortex 4, 576–589 (1994).
23. Rakic, P. Limits of neurogenesis in primates. Science 227, 1054–1056 (1985).
24. Eckenhoff M.F. & Rakic, P. Nature and fate of proliferative cells in the hippocampal
dentate gyrus during the life span of the rhesus monkey. J. Neurosci. 8, 2729–2747
(1988).
25. Kempermann, G., Kuhn, H.G. & Gage, F.H. Experience-induced neurogenesis in
the senescent dentate gyrus. J. Neurosci. 18, 3206–3212 (1998).
26. Gage, F.H. et al. Survival and differentiation of adult neuronal progenitor cells trans-
planted to the adult brain. Proc. Natl. Acad. Sci. USA 92, 11879–11883 (1995).
27. Suhonen, J.O., Peterson, D.A., Ray, J. & Gage, F.H. Differentiation of adult hip-
pocampus-derived progenitors into olfactory neurons in vivo. Nature 383, 624–627
(1996).
1317
 An in vivo correlate of exercise-induced neurogenesis
in the adult dentate gyrus
Ana C. Pereira*†, Dan E. Huddleston*†, Adam M. Brickman*†, Alexander A. Sosunov‡, Rene Hen§, Guy M. McKhann‡,
Richard Sloan§, Fred H. Gage¶, Truman R. Brown储, and Scott A. Small*†**
*The Taub Institute for Research on Alzheimer’s Disease and the Aging Brain, Departments of †Neurology, ‡Neurosurgery, §Psychiatry, and 储Radiology,
Columbia University College of Physicians and Surgeons, New York, NY 10032; and ¶The Salk Institute, La Jolla, CA 92037
Contributed by Fred H. Gage, December 30, 2006 (sent for review November 26, 2006)
With continued debate over the functional significance of adult
neurogenesis, identifying an in vivo correlate of neurogenesis has
become an important goal. Here we rely on the coupling between
neurogenesis and angiogenesis and test whether MRI measurements
of cerebral blood volume (CBV) provide an imaging correlate of
neurogenesis. First, we used an MRI approach to generate CBV maps
over time in the hippocampal formation of exercising mice. Among all
hippocampal subregions, exercise was found to have a primary effect
on dentate gyrus CBV, the only subregion that supports adult neu-
rogenesis. Moreover, exercise-induced increases in dentate gyrus CBV
were found to correlate with postmortem measurements of neuro-
genesis. Second, using similar MRI technologies, we generated CBV
maps over time in the hippocampal formation of exercising humans.
As in mice, exercise was found to have a primary effect on dentate
gyrus CBV, and the CBV changes were found to selectively correlate
with cardiopulmonary and cognitive function. Taken together, these
findings show that dentate gyrus CBV provides an imaging correlate
of exercise-induced neurogenesis and that exercise differentially
targets the dentate gyrus, a hippocampal subregion important for
memory and implicated in cognitive aging.
hippocampus 兩 in vivo imaging 兩 cerebral blood volume 兩 angiogenesis
T he hippocampal formation, a brain circuit made up of separate
but interconnected subregions (1), is vital for memory function
(2) and is targeted by the aging process (3). The dentate gyrus is the
only hippocampal subregion that supports neurogenesis in the adult
brain (4–6). Nevertheless, because neurogenesis can only be as-
sessed in postmortem tissue, its functional significance remains
undetermined. With this limitation in mind, we have explored
different imaging approaches applicable to rodents and humans
that might provide an in vivo correlate of neurogenesis.
Although imaging radioligands designed to bind newly dividing
cells is an attractive approach, positron emission tomography
imaging suffers inherently poor resolution and cannot visualize the
dentate gyrus. Additionally, radiolabeling newborn cells introduces
potential safety concerns. For these reasons, we have focused on
MRI technologies instead. Notably, a coupling has been established
between neurogenesis and angiogenesis (7, 8). The process of
angiogenesis, in turn, gradually gives rise to the formation of new
blood vessels, increasing regional microvascular density (9–12).
Importantly, vascular density can be measured in vivo with imaging
techniques that map regional blood volume. Numerous studies have
established a tight relationship between angiogenesis and regional
blood volume (13–17), including in the brain where regional
angiogenesis is coupled to regional cerebral blood volume (CBV)
(18–26).
Because CBV can be measured with MRI, we hypothesized that
a regionally selective increase in hippocampal CBV might provide
an imaging correlate of neurogenesis. This hypothesis was tested in
mice in which in vivo imaging and postmortem analysis can be
performed on the same subjects. We used an MRI approach that
can generate hippocampal CBV maps repeatedly and safely over
time in mice while simultaneously assessing multiple hippocampal
subregions (27). We exploited the capabilities of this approach to
5638 –5643 兩 PNAS 兩 March 27, 2007 兩 vol. 104 兩 no. 13
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
establish how hippocampal CBV maps change during physical
exercise, a potent stimulant of dentate gyrus neurogenesis (28), and
tested whether CBV changes correlate with postmortem measure-
ments of neurogenesis. Once these findings were established in
mice, we were interested in determining how exercise affects the
hippocampal CBV maps of humans. To accomplish this goal, we
first optimized an MRI approach (29) previously shown to be
capable of generating hippocampal CBV maps in non-human
primates (30) and then charted exercise-induced CBV changes in
the human hippocampal formation.
Results
Exercise Selectively Increases Dentate Gyrus CBV in Mice and Corre-
lates with Neurogenesis. In designing our experimental protocol, we
were guided by the observation that angiogenesis-induced sprout-
ing of new blood vessels progresses through different stages,
forming gradually over time (9). Accordingly, mice were allowed to
exercise for 2 weeks, the period during which neurogenesis reaches
its maximum increase, and bromodeoxyuridine (BrdU), a marker of
newly born cells, was injected daily during the second week. To
capture the predicted delayed effect in CBV, mice were kept alive
for 4 more weeks, then killed and processed for BrdU labeling.
Hippocampal CBV maps (Fig. 1b) assessing the entorhinal cortex,
the dentate gyrus, and the CA3 and CA1 subfields were generated
four times over the 6-week experiment: at preexercise baseline and
at weeks 2, 4, and 6. A control group following the identical protocol
but without exercise was imaged in parallel.
A repeated-measure ANOVA was used to analyze the imaging
data set (Fig. 1a). A group–time interaction was found only for the
dentate gyrus, showing that exercise was associated with a selective
increase in dentate gyrus CBV (F ⫽ 5.0, P ⫽ 0.034). As shown by
simple contrasts, the effect was driven by a maximum increase that
emerged 2 weeks after the cessation of exercise, from week 2 to
week 4 (F ⫽ 5.9, P ⫽ 0.021) (Fig. 1a). The entorhinal cortex was the
only other hippocampal subregion whose CBV increased appre-
ciably over time, although not achieving statistical significance (Fig.
1a). Although exercise might potentially affect CBV by increasing
metabolism and cerebral blood flow, previous studies (31, 32) have
shown that exercise-induced changes in metabolism should mani-
fest during, not after, the exercise regimen. Thus, the observed
spatiotemporal profile with which CBV emerged fits better with a
model of exercise-induced angiogenesis in the dentate gyrus.
In agreement with previous studies (33), the exercise group was
found to have greater BrdU labeling compared with the nonexer-
cise group (F ⫽ 9.8, P ⫽ 0.004) (Fig. 2a). More than 90% of
Author contributions: G.M.M., R.S., F.H.G., and S.A.S. designed research; A.C.P., D.E.H.,
A.M.B., and A.A.S. performed research; R.H. and T.R.B. contributed new reagents/analytic
tools; and A.C.P., F.H.G., and S.A.S. wrote the paper.
The authors declare no conflict of interest.
Abbreviation: CBV, cerebral blood volume.
**To whom correspondence should be addressed. E-mail: sas68@columbia.edu.
This article contains supporting information online at www.pnas.org/cgi/content/full/
0611721104/DC1.
© 2007 by The National Academy of Sciences of the USA
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0611721104
 Fig. 1. Exercise selectively increases dentate gyrus CBV in mice. (a) Exercise had a selective effect on dentate gyrus CBV. Bar graphs show the mean relative CBV
(rCBV) values for each hippocampal subregion in the exercise group (filled bars) and the nonexercise group (open bars) over the 6-week study. The dentate gyrus
was the only hippocampal subregion that showed a significant exercise effect, with CBV peaking at week 4, whereas the entorhinal cortex showed
a nonsignificant increase in CBV. (b) An individual example. (Left) High-resolution MRI slice that visualizes the external morphology and internal architecture
of the hippocampal formation. (Center) Parcellation of the hippocampal subregions (green, entorhinal cortex; red, dentate gyrus; dark blue, CA3 subfield; light

NEUROSCIENCE
blue, CA1 subfield). (Right) Hippocampal CBV map (warmer colors reflect higher CBV).

BrdU-positive cells colabeled for NeuN, a neuron-specific marker
(Fig. 2a). To examine the relationship between neurogenesis and
CBV, the repeated-measures model was again used and included
BrdU as a covariate. A significant time-by-BrdU interaction was
observed only for dentate gyrus CBV (F ⫽ 3.3, P ⫽ 0.039), driven
primarily by changes from week 2 to week 4 (F ⫽ 8.8, P ⫽ 0.006).
As shown by a direct analysis, this effect reflected a positive
correlation between BrdU and changes in CBV from week 2 to
week 4 (␤ ⫽ 0.58, P ⫽ 0.001) (Fig. 2b). Of note, when BrdU was
included as a covariate in the ANOVA, the group–time effect
observed in the dentate gyrus was no longer significant, confirming
that neurogenesis accounted for the exercise effect on CBV. Visual
inspection of the relationship between changes in dentate gyrus
CBV and BrdU (Fig. 2b) suggested that a quadratic vs. a linear
model better characterized the relationship, which was confirmed
by curve estimation analysis (linear model, R2 ⫽ 0.34 and P ⫽ 0.001;
quadratic model, R2 ⫽ 0.59 and P ⬍ 0.0001). Thus, the association
between changes in dentate gyrus CBV and BrdU exists primarily
when CBV increases with exercise (Fig. 2b). Finally, to test whether
neurogenesis was required for the observed increases in dentate
gyrus CBV we relied on a previously described x-irradiation ap-
proach that selectively blocks dentate gyrus neurogenesis (55). Four
mice received x-irradiation, and three mice were sham controls.
After a 3-month recovery period, all mice were imaged by using the
exercise protocol. A repeated-measure ANOVA applied to the
CBV data set revealed a group (x-ray vs. sham)-by-time interaction
only for the dentate gyrus (F ⫽ 7.6, P ⫽ 0.04), showing that
irradiation blocked the exercise-induced increases in CBV [sup-
porting information (SI) Fig. 5].
Exercise Selectively Increases Dentate Gyrus CBV in Humans and
Correlates with Aerobic Fitness and Cognition. Once we established
that dentate gyrus CBV provides a correlate of exercise-induced
neurogenesis, we were interested in testing whether this effect is
observed in exercising humans. CBV maps of the human hippocam-
pal formation were generated by using our previously reported
MRI approach, specifically tailored for imaging the primate hip-
pocampal formation (30). Eleven healthy subjects (mean age ⫽ 33,
ranging from 21–45 years; two males and nine females) participated
in the study, completing a 3-month aerobic exercise regimen.
Hippocampal CBV maps were generated before and after exercise.
CBV values were reliably measured for all hippocampal subregions,
except the CA3 subregion (Fig. 3b). Compared with experimental
animals, in humans it is impossible to control the interindividual
differences in physical activity performed during daily life. There-
fore, before and after exercise, we measured the maximum volume
of oxygen consumption (VO2max), the gold standard measure of
exercise-associated aerobic fitness (34, 35) to quantify individual
differences in degree of exercise. Cognitive performance was
assessed by using a modified Rey Auditory Verbal Learning Test
(36), whose design allows cognition to be tested across different
learning trials and during delayed recall, recognition, and source

Pereira et al. PNAS 兩 March 27, 2007 兩 vol. 104 兩 no. 13 兩 5639

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Fig. 2. Exercise-induced increases in dentate gyrus CBV correlate with neurogenesis. (a) (Left) Exercising mice were found to have more BrdU labeling compared
with the no-exercise group. (Right) As shown by confocal microscopy, the majority of the new cells were colabeled with NeuN (red, BrdU labeling; green, NeuN;
yellow, BrdU/NeuN double labeling). (b) (Left) A significant linear relationship was found between changes in dentate gyrus CBV and BrdU labeling. (Right) A
quadratic relationship better fits the data. The vertical stippled line in each plot splits the x axis into CBV changes that decreased (left of line) versus those that
increased (right of line) with exercise.
memory. Ten subjects were cognitively assessed after exercise, eight
of whom were assessed at preexercise baseline.
A repeated-measures ANOVA used to analyze the imaging data
showed that the dentate gyrus was the only hippocampal subregion
whose CBV significantly increased over time (F ⫽ 12, P ⫽ 0.006)
(Fig. 3a). As in mice, the entorhinal cortex was the only other
hippocampal subregion whose CBV increased appreciably over
time, although not achieving statistical significance (F ⫽ 4.3, P ⫽
0.064) (Fig. 3a). As a group, VO2max values significantly increased
over time (F ⫽ 11.6, P ⫽ 0.007) (Fig. 4a), and, to confirm that the
imaged changes were directly related to exercise and not simply
caused by a test–retest effect, we found that individual differences
in dentate gyrus CBV were correlated to individual changes
in VO2max (␤ ⫽ 0.662, P ⫽ 0.027) (Fig. 4b). Importantly,
a correlation between CBV and VO2max was not observed for
any other hippocampal subregion, including the entorhinal cortex
(Fig. 4b), confirming that exercise has a selective effect on dentate
gyrus CBV.
Cognitively, individuals performed significantly better on trial 1
learning (F ⫽ 7.0, P ⫽ 0.027) after exercise, with a trend toward
improvement on all-trial learning (F ⫽ 5.0, P ⫽ 0.053) and delayed
recall (F ⫽ 5.0, P ⫽ 0.057). There was no effect on delayed
recognition (F ⫽ 0.19, P ⫽ 0.67) or source memory (F ⫽ 0.15, P ⫽
0.25) (Fig. 4a). To test that cognitive improvement was related to
exercise per se, we found that individual changes in trial 1 learning
were correlated with individual changes in VO2max (␤ ⫽ 0.660, P ⫽
0.037). However, because only 8 of the 10 subjects completed
preexercise cognitive testing, we repeated the analysis with postex-
cercise cognitive performance scores. Again, we found that changes
in VO2max correlated exclusively with postexercise trial 1 learning
(␤ ⫽ 0.70, P ⫽ 0.026) (Fig. 4b). Additional analyses showed that the
correlation between changes in VO2max and cognition was selec-
tive to trial 1 learning (Fig. 4b), thereby confirming that, despite
apparent increases in other cognitive measures (i.e., delayed rec-
ognition, as shown in Fig. 4a), this particular measure was selec-
tively influenced by exercise.
5640 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0611721104
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Finally, we examined the relationship between cognition and
CBV. Among all hippocampal subregions, the correlation between
improvements in trial 1 performance and increases in dentate gyrus
CBV was the only one that trended toward significance (␤ ⫽ 0.62,
P ⫽ 0.052). Because of the missing preexercise data, we repeated
all of the analyses comparing changes in CBV with postexercise
cognition, finding an exclusive correlation between postexercise
trial 1 learning and dentate gyrus CBV (␤ ⫽ 0.63, P ⫽ 0.026)
(Fig. 4b).
Discussion
Taken together, our findings show that, within the hippocampal
formation, exercise targets the dentate gyrus with regional selec-
tivity. In addition, our results identify that, in mice, dentate gyrus
CBV is an imaging correlate of exercise-induced neurogenesis.
Because of its pleiotropic effect on the brain (37, 38), the
subregional selectivity that exercise was found to have within the
hippocampal formation was an unexpected finding. This observa-
tion is particularly interesting in light of studies suggesting that the
dentate gyrus is a hippocampal subregion differentially vulnerable
to the aging process, as shown in humans (39, 40), nonhuman
primates (30, 41), and rodents (30, 42), and that dentate gyrus
dysfunction contributes to cognitive aging (30, 40). Interestingly, a
growing number of studies have established that exercise amelio-
rates age-related cognitive decline (43–46). Thus, the effect that
exercise was observed to have on the dentate gyrus likely contrib-
utes to the reported cognitive benefits exercise has on the aging
brain.
Because neurogenesis couples with angiogenesis (7, 8) and
angiogenesis, in turn, couples with CBV (18–26), we hypothesized
that measures of CBV might provide an in vivo correlate of
neurogenesis. This hypothesis is based in part on our previous
findings that exercise increases both neurogenesis and angiogenesis
in young adult rodents (43). We have confirmed this prediction with
our high-resolution MRI techniques, which are capable of assessing
individual hippocampal subregions, showing that dentate gyrus
Pereira et al.
 Fig. 3. Exercise selectively increases dentate gyrus CBV in humans. (a) Exercise had a selective effect on dentate gyrus CBV. Bar graph shows the mean relative
CBV (rCBV) values for each hippocampal subregion before exercise (open bars) and after exercise (filled bars). As in mice, the dentate gyrus was the only
hippocampal subregion that showed a significant exercise effect, whereas the entorhinal cortex showed a nonsignificant increase in CBV. (b) An individual
example. (Left) High-resolution MRI slice that visualizes the external morphology and internal architecture of the hippocampal formation. (Center) Parcellation
of the hippocampal subregions (green, entorhinal cortex; red, dentate gyrus; blue, CA1 subfield; yellow, subiculum). (Right) Hippocampal CBV map (warmer
colors reflect higher CBV).
CBV selectively correlates with underlying neurogenesis in mice.
Blocking the exercise-induced CBV effect with irradiation, a neuro-
genesis suppressant, further confirms that neurogenesis underlies
the observed increases in CBV. However, in the absence of direct
measurements, we have not confirmed that angiogenesis is
the intermediate factor that accounts for this relationship. Mech-
anistically, there are many molecules that may be mediating
the exercise-induced coupling between neurogenesis and angio-
genesis (47), including VEGF (8) or exercise-induced changes in
BDNF (48).
The remarkable similarities between the exercise-induced CBV
changes in the hippocampal formation of mice and humans suggest
that the effect is mediated by similar mechanisms. Of course, in
contrast to mice, it is impossible to directly confirm whether the
changes in dentate gyrus CBV observed in humans are a reflection
of neurogenesis. Nevertheless, previous rodent studies have shown
that levels of neurogenesis are coupled with individual differences
in degree of exercise (28). Therefore, together with the cross-
species similarities on hippocampal CBV, the observation that
exercise-induced changes in dentate gyrus CBV correlate with
VO2max supports the interpretation that the human effect is
mediated, at least in part, by neurogenesis.
In any case, our results show that in vivo imaging can predict levels
of neurogenesis in living mice. Furthermore, as demonstrated, the
MRI technologies used in these studies are capable of measuring
meaningful changes in dentate gyrus CBV over time in both
humans and mice. By providing evidence in support of exercise-
induced neurogenesis and by introducing the tools needed to
achieve this goal, this study shows that it is possible to isolate the
specific profile of cognitive changes that are neurogenesis-
dependent. Furthermore, the imaging tools presented here are
uniquely suited to investigate potential pharmacological modula-
tors of neurogenesis, testing their role in treating depression (49)
and in ameliorating the cognitive decline that occurs in all of us as
we age (43, 50).
Methods
Exercise. Mice. A total of 46 C57BL/6 mice (7 weeks old) were used
(23 exercising and 23 nonexercising animals). The experimental
mice were placed in cages with running wheels (Lafayette Instru-
ment Company, Lafayette, IN). The animals ran voluntarily for
2 weeks. MRI images were acquired at the following time points:
week 0 (baseline), week 2 (when exercise was stopped), week 4,
and week 6. The thymidine analog BrdU marker was injected i.p.
for 7 consecutive days (60 mg䡠kg⫺1䡠day⫺1) during the second week
of the experiment. At week 6, the animals were anesthetized and
killed in accordance with institutional guidelines.
Pereira et al.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Human. Eleven subjects (mean age, 33 ranging from 21–45 years; 2
males and 9 females) who fulfilled the American Heart Association
criteria for below-average aerobic fitness (VO2max, ⬍43 for men
and ⬍37 for women) were recruited (51). The 11 enrolled subjects
engaged in an exercise training protocol for 12 weeks at Columbia
University Fitness Center at a frequency of four times a week. Each
exercise session lasted ⬇1 h: 5 min of low-intensity warm-up on a
treadmill or stationary bicycle, 5 min of stretching, 40 min of aerobic
training, and 10 min for cool down and stretching. During the 40
min of aerobic activity, subjects were permitted to select from
cycling on a stationary ergometer, running on a treadmill, climbing
on a StairMaster, or using an elliptical trainer.
VO2max was measured by a graded exercise test on an Ergoline
800S electronic-braked cycle ergometer (SensorMedics, Anaheim,
CA). Each subject began exercising at 30 W for 2 min, and the work
rate was continually increased by 30 W each 2 min until VO2max
criteria (respiratory quotient of ⱖ 1.1, increases in ventilation
without concomitant increases in VO2, maximum age-predicted
heart rate, and/or volitional fatigue) were reached. Minute venti-
NEUROSCIENCE
lation was measured by a pneumotachometer connected to a FLO-1
volume transducer module (Physio-Dyne Instrument Corporation,
Quogue, NY). Percentages of expired oxygen (O2) and carbon
dioxide (CO2) were measured by using a paramagnetic O2 analyzer
and an infrared CO2 analyzer connected to a computerized system
(MAX-1; Physio-Dyne Instrument Corporation). These analyzers
were calibrated against known medical grade gases. The highest
VO2 value attained during the graded exercise test is considered
VO2max.
In Vivo Imaging. Mice. Mice were imaged with a 9.4-T Bruker
scanner (AVANCBV 400WB spectrometer; Bruker NMR, Bil-
lerica, MA) by following a previously described protocol (27).
Briefly, axial T2-weighted images were optimally acquired with a
fast sequence (time to repeat/effective echo time ⫽ 2,000 ms/70 ms;
30-mm i.d. birdcage radio frequency probe; shielded gradient
system, 100 g/cm; rapid acquisition with relaxation enhancement
factor, 16; field of view, 19.6 mm; acquisition matrix, 256 ⫻ 256; no.
of slices, 8; slice thickness, 0.6 mm; slice gap, 0.1 mm; number of
excitations, 28). Five sets of images were acquired sequentially, each
requiring 16 min. The first two sets were precontrast. Gadodiamide
was then injected (13 mmol/kg i.p.) through a catheter placed
intraperitoneally before imaging. The last three sets corresponded
to the postcontrast images. To prevent head motion and reduce
anxiety, the animals were anesthetized with isoflurane gas [1.5%
(vol/vol) for maintenance at 1 liter per minute of air flow) via a nose
cone. Isoflurane was chosen because it induces minimal cerebral
hemodynamic change (52). Monitoring of the heart rate, respira-
tory rate, and oxygen saturation was performed during the whole
PNAS 兩 March 27, 2007 兩 vol. 104 兩 no. 13 兩 5641
 pre
a 40 20
post

VO2max
30

T o ta l C o rre c t
15
20

*
10
10
pre post

5
Learning Learning Recall Recall Delay Delay
(trial 1) (all-trial) (5 min) (90 min) Recog Source

b 10 10 2.5
VO2max (diff)

DG CBV (diff)
8 2.0
VO2max (diff)

6 6 1.5
4 4 1.0
2 2 0.5
0 0 0.0
-2 -2 -0.5
-0.5 0.0 0.5 1.0 1.5 2.0 2.5 4 6 8 10 12 14 16 4 6 8 10 12 14 16
DG CBV (diff) Trial 1 learning (post) Trial 1 learning (post))
10 10
2.0
VO2max (diff)

8 8
VO2max (diff)

EC CBV (diff)

6 6 1.0
4 4
2 2 0.0

0 0 -1.0
-2 -2

-1.0 0.0 1.0 2.0 14 15 16 17 18 19 20 4 6 8 10 12 14 16

EC CBV (diff) Delayed recog (post) Trial 1 learning (post)
Fig. 4. Exercise-induced increases in dentate gyrus CBV correlate with aerobic fitness and cognition. (a) (Left) VO2max, the gold-standard measure of
exercise-induced aerobic fitness, increased after exercise. (Right) Cognitively, exercise has its most reliable effect on first-trial learning of new declarative
memories. (b) (Left) Exercise-induced changes in VO2max correlated with changes in dentate gyrus (DG) CBV but not with other hippocampal subregions,
including the entorhinal cortex (EC), confirming the selectivity of the exercise-induced effect. (Center) Exercise-induced changes in VO2max correlated with
postexercise trial 1 learning but not with other cognitive tasks, including delayed recognition. (Right) Postexercise trial 1 learning correlated with exercise-
induced changes in dentate gyrus CBV but not with changes in other hippocampal subregions, including the entorhinal cortex.

procedure. Relative CBV was mapped as changes of the transverse thickness, 4 mm) were acquired oriented perpendicular to the
relaxation rate (⌬R2) induced by the contrast agent. When the hippocampal long-axis before and 4 min after i.v. administration of
contrast agent reaches uniform distribution, CBV maps can be the contrast gadolinium (0.1 mmol/kg). The difference between
measured from steady-state T2-weighted images as CBV ⬁ ⌬R2 ⫽ precontrast and postcontrast images was used to access the regional
ln(Spre/Spost)/TE, where TE is the effective echo time, Spre is the CBV map. To control for differences in levels of contrast admin-
signal before the contrast administration, and Spost is the signal after istration, cardiac output, and global blood flow, the derived differ-
the contrast agent reaches steady-state. To control for differences ences in signal intensity were normalized to the maximum 4-pixel
in levels of contrast administration, cardiac output, and global blood signal value of the sagittal sinus (29). For each subject, the pre-
flow, the derived maps were normalized to the maximum 4-pixel contrast scan was used to identify the slice with the best visualiza-
signal value of the posterior cerebral vein. Visualized anatomical tion of the external morphology and internal architecture of the
landmarks were used together with standard atlases to identify hippocampal formation. Visualized anatomical landmarks were
the localization of four hippocampal subregions: the dentate gyrus, used together with standard atlases to identify the general locale of
the CA3 subfield, the CA1 subfield, and the entorhinal cortex (53). four subregions: the dentate gyrus, the CA1 subfield, the subicu-
The normalized CBV measurements from each subregion were lum, and the entorhinal cortex (30). The normalized CBV mea-
used for group data analysis. surements from each subregion were used for group data analysis.
Human. Subjects were imaged with a 1.5-T scanner Intera scanner
(Philips, Amsterdam, The Netherlands). As previously described Microscopy. Immunohistochemistry. Free-floating, 40-␮m coronal
(30), coronal T1-weighted images (repetition time, 20 ms; echo sections were used in the determination of BrdU labeling. DNA
time, 6 ms; flip angle, 25°; in-plane resolution, 0.86 ⫻ 86 mm; slice denaturation was conducted by incubation for 1 h at 2 M HCl at

5642 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0611721104 Pereira et al.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 37°C, followed by washing in boric buffer (pH 8.5). After washing,
sections were incubated for 30 min in 10% H2O2 to eliminate
endogenous peroxidases. After blocking with 3% normal donkey
serum in 0.2% Triton X-100, sections were incubated with mono-
clonal anti-BrdU (1:600; Roche, Basel, Switzerland) overnight at
4°C. Sections were then incubated for 1 h at room temperature with
the secondary antibody (biotinylated donkey anti-mouse; Jackson
ImmunoResearch Laboratories, West Grove, PA) followed by
amplification with an avidin–biotin complex (Vector Laboratories,
Burlingame, CA), and visualized with diaminobenzidine (Sigma, St.
Louis, MO). For double-immunolabeling, free-floating sections
were incubated in a mixture of primary antibodies, anti-BrdU
(1:600; Roche), and anti-NeuN (1:500; Chemicon, Temecula, CA),
raised in different species overnight. For visualization, Alexa Fluor-
conjugated appropriate secondary antibodies (1:300; Molecular
Probes, Eugene, OR) raised in goats were used for 1 h at room
temperature. Blocking serum and primary and secondary anti-
bodies were applied in 0.2% Triton X-100 in PBS. Sections
for fluorescent microscopy were mounted on slides in
VECTASHIELD (Vector Laboratories). For control of the spec-
ificity of immunolabeling, primary antibodies were omitted and
substituted with appropriate normal serum. Slides were viewed with
a confocal microscope [Radiance 2000 (Bio-Rad, Hercules, CA)
mounted on a E800 microscope (Nikon, Tokyo, Japan)]. The
images presented are stacks of six to 16 optical sections (step, 1 mm)
that were collected individually (in the green and red channels) or
simultaneously with precaution against cross-talk between chan-
nels. The images were processed with PhotoShop 7.0 (Adobe
Systems, San Jose, CA) without contrast and brightness changes in
split images.
Quantitation of BrdU labeling. Every sixth section throughout the
hippocampus was processed for BrdU immunohistochemistry. Ten
sections were used for each animal. All BrdU-labeled cells in the
dentate gyrus (granule cell layer and subgranular zone) were
counted under a light microscope by an experimenter blinded to the
study code. The average number of BrdU-labeled cells was then
derived for each animal.
1. Amaral DG (1993) Curr Opin Neurobiol 3:225–229.
2. Squire LR, Stark CE, Clark RE (2004) Annu Rev Neurosci 27:279–306.
3. Erickson CA, Barnes CA (2003) Exp Gerontol 38:61–69.
4. Altman J, Das GD (1965) J Comp Neurol 124:319–335.
5. Kaplan MS, Hinds JW (1977) Science 197:1092–1094.
6. Eriksson PS, Perfilieva E, Bjork-Eriksson T, Alborn AM, Nordborg C, Peterson DA, Gage
FH (1998) Nat Med 4:1313–1317.
7. Palmer TD, Willhoite AR, Gage FH (2000) J Comp Neurol 425:479–944.
8. Louissaint A, Jr, Rao S, Leventhal C, Goldman SA (2002) Neuron 34:945–960.
9. Carmeliet P (2003) Nat Med 9:653–660.
10. Risau W (1998) Kidney Int Suppl 67:S3–S6.
11. McDonald DM, Choyke PL (2003) Nat Med 9:713–725.
12. D’Amore PA, Thompson RW (1987) Annu Rev Physiol 49:453–464.
13. Mayr NA, Hawighorst H, Yuh WT, Essig M, Magnotta VA, Knopp MV (1999) J Magn Reson
Imaging 10:267–276.
14. Taylor JS, Tofts PS, Port R, Evelhoch JL, Knopp M, Reddick WE, Runge VM, Mayr N
(1999) J Magn Reson Imaging 10:903–907.
15. Bremer C, Mustafa M, Bogdanov A, Jr, Ntziachristos V, Petrovsky A, Weissleder R (2003)
Radiology 226:214–220.
16. Kerwin W, Hooker A, Spilker M, Vicini P, Ferguson M, Hatsukami T, Yuan C (2003)
Circulation 107:851–856.
17. van Dijke CF, Brasch RC, Roberts TP, Weidner N, Mathur A, Shames DM, Mann JS,
Demsar F, Lang P, Schwickert HC (1996) Radiology 198:813–818.
18. Dennie J, Mandeville JB, Boxerman JL, Packard SD, Rosen BR, Weisskoff RM (1998)
Magn Reson Med 40:793–799.
19. Cha S, Johnson G, Wadghiri YZ, Jin O, Babb J, Zagzag D, Turnbull DH (2003) Magn Reson
Med 49:848–855.
20. Pathak AP, Rand SD, Schmainda KM (2003) J Magn Reson Imaging 18:397–403.
21. Dunn JF, Roche MA, Springett R, Abajian M, Merlis J, Daghlian CP, Lu SY, Makki M
(2004) Magn Reson Med 51:55–61.
22. Lin TN, Sun SW, Cheung WM, Li F, Chang C (2002) Stroke 33:2985–2991.
23. Sugahara T, Korogi Y, Kochi M, Ikushima I, Hirai T, Okuda T, Shigematsu Y, Liang L, Ge
Y, Ushio Y, Takahashi M (1998) Am J Roentgenol 171:1479–1486.
24. Maia AC, Jr, Malheiros SM, da Rocha AJ, da Silva CJ, Gabbai AA, Ferraz FA, Stavale JN
(2005) Am J Neuroradiol 26:777–783.
25. Aronen HJ, Pardo FS, Kennedy DN, Belliveau JW, Packard SD, Hsu DW, Hochberg FH,
Fischman AJ, Rosen BR (2000) Clin Cancer Res 6:2189–2200.
26. Jiang Q, Zhang ZG, Ding GL, Silver B, Zhang L, Meng H, Lu M, Pourabdillah-Nejed DS,
Wang L, Savant-Bhonsale S, et al. (2006) NeuroImage 32:1080–1089.
27. Moreno H, Hua F, Brown T, Small S (2006) NMR Biomed 19:535–543.
28. van Praag H, Kempermann G, Gage FH (1999) Nat Neurosci 2:266–270.
Pereira et al.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Cognitive testing. Declarative memory was measured with a version
of the Rey Auditory Verbal Learning Test (36) modified to increase
variability in memory performance among healthy young adults.
Twenty nonsemantically or phonemically related words were pre-
sented over three learning trials in which the test administrator read
the word list and the subject free-recalled as many words as possible.
Administration of the three learning trials was immediately fol-
lowed by one learning trial of a distracter list and then a short
delayed free recall of the initial list. After a 90-min delay period,
subjects were asked to freely recall words from the initial list and
then to freely recall items from the distracter list. After a 24-hr delay
period, subjects were contacted by telephone and asked to freely
recall items from the initial list and then from the distracter list.
They were then administered a forced-choice recognition trial in
which they were required to identify the 20 words from the initial
learning trial among semantically and phonemically related words,
as well as words from the distracter trial. Finally, a source memory
trial was administered in which subjects were read a list containing
only words from the initial learning list and from the distracter list
and were asked to identify from which list each word came. Two
forms of the verbal learning test were created, and the administra-
tion order was counterbalanced. As in previous studies (54), we
measured words correctly recalled on the first trial of the initial
learning trials, the average number of words recalled across the
three learning trials, the number of words from the initial learning
trial that were correctly recalled after a short delay (⬍5 min), the
number of words from the initial learning trial that were correctly
recalled after a 90-min delay, the number of items correctly
identified on the recognition trial, and the correct number of items
identified on the source memory trial.
We thank Kenneth Hess and Fan Hau for help in mouse imaging and Clay
Lacefield for helping deliver x-irradiation. This work was supported by
grants from the Defense Advanced Research Projects Agency, the National
Institutes of Health (AG025161), the New York Heart Association, the
Lookout Foundation, and the James S. McDonnell Foundation.
29. Lin W, Celik A, Paczynski RP (1999) J Magn Reson Imaging 9:44–52.
NEUROSCIENCE
30. Small SA, Chawla MK, Buonocore M, Rapp PR, Barnes CA (2004) Proc Natl Acad Sci USA
101:7181–7186.
31. Vissing J, Andersen M, Diemer NH (1996) J Cereb Blood Flow Metab 16:729–736.
32. Ide K, Secher NH (2000) Prog Neurobiol 61:397–414.
33. van Praag H, Christie BR, Sejnowski TJ, Gage FH (1999) Proc Natl Acad Sci USA
96:13427–13431.
34. Mitchell JH, Sproule BJ, Chapman CB (1958) J Clin Invest 37:538–547.
35. Wagner PD (1996) Annu Rev Physiol 58:21–50.
36. Rey A (1964) L’Examen Clinique en Psychologie (Presses Univ de France, Paris).
37. Cotman CW, Berchtold NC (2002) Trends Neurosci 25:295–301.
38. Dishman RK, Berthoud HR, Booth FW, Cotman CW, Edgerton VR, Fleshner MR,
Gandevia SC, Gomez-Pinilla F, Greenwood BN, Hillman CH, et al. (2006) Obesity (Silver
Spring) 14:345–356.
39. West MJ, Coleman PD, Flood DG, Troncoso JC (1994) Lancet 344:769–772.
40. Small SA, Tsai WY, DeLaPaz R, Mayeux R, Stern Y (2002) Ann Neurol 51:290–295.
41. Gazzaley AH, Siegel SJ, Kordower JH, Mufson EJ, Morrison JH (1996) Proc Natl Acad Sci
USA 93:3121–3125.
42. Shetty AK, Turner DA (1999) Exp Neurol 158:491–503.
43. van Praag H, Shubert T, Zhao C, Gage FH (2005) J Neurosci 25:8680–8685.
44. Kramer AF, Hahn S, Cohen NJ, Banich MT, McAuley E, Harrison CR, Chason J, Vakil E,
Bardell L, Boileau RA, Colcombe A (1999) Nature 400:418–419.
45. Chan AS, Ho YC, Cheung MC, Albert MS, Chiu HF, Lam LC (2005) J Am Geriatr Soc
53:1754–1760.
46. Weuve J, Kang JH, Manson JE, Breteler MM, Ware JH, Grodstein F (2004) J Am Med Assoc
292:1454–1461.
47. Emanueli C, Schratzberger P, Kirchmair R, Madeddu P (2003) Br J Pharmacol 140:614–619.
48. Neeper SA, Gomez-Pinilla F, Choi J, Cotman CW (1996) Brain Res 726:49–56.
49. Santarelli L, Saxe M, Gross C, Surget A, Battaglia F, Dulawa S, Weisstaub N, Lee J, Duman
R, Arancio O, et al. (2003) Science 301:805–809.
50. Small SA, Stern Y, Tang M, Mayeux R (1999) Neurology 52:1392–1396.
51. Fletcher GF, Balady G, Froelicher VF, Hartley LH, Haskell WL, Pollock ML (1995)
Circulation 91:580–615.
52. Lei H, Grinberg O, Nwaigwe CI, Hou HG, Williams H, Swartz HM, Dunn JF (2001) Brain
Res 913:174–179.
53. Paxinos G, Franklin K (2001) The Mouse Brain in Stereotaxic Coordinates (Academic,
Philadelphia).
54. Van der Elst W, van Boxtel MP, van Breukelen GJ, Jolles J (2005) J Int Neuropsychol Soc
11:290–302.
55. Santarelli L, Saxe M, Gross C, Surget A, Battaglia F, Dulawa S, Weisstaub N, Lee J, Duman
R, Arancio O, et al. (2003) Science 301:805–809.
PNAS 兩 March 27, 2007 兩 vol. 104 兩 no. 13 兩 5643

11. Y. Montet et al., Nature Struct. Biol. 4, 523 (1997).
12. L. Holm, C. Sander, Proteins 19, 165 (1994).
13. T. Shanmugasundaram, G. K. Kumar, H. G. Wood,
Biochemistry 27, 6499 (1988).
14. S. Menon, S. W. Ragsdale, J. Biol. Chem. 274, 11513
(1999).
15. S. W. Ragsdale, J. E. Clark, L. G. Ljungdahl, L. L. Lundie,
H. L. Drake, J. Biol. Chem. 258, 2364 (1983).
16. S. W. Ragsdale, in Enzyme-Catalyzed Electron and
Radical Transfer, A. Holzenburg, N. Scrutton, Eds.
(Plenum, New York, 2000), vol. 35, pp. 487–518.
17. J. W. Peters, W. N. Lanzilotta, B. J. Lemon, L. C.
Seefeldt, Science 282, 1853 (1998).
18. Y. Nicolet, C. Piras, P. Legrand, C. E. Hatchikian, J. C.
Fontecilla-Camps, Structure 7, 13 (1999).
19. W. K. Russell, C. M. V. Stalhandske, J. Q. Xia, R. A.
Scott, P. A. Lindahl, J. Am. Chem. Soc. 120, 7502
(1998).
20. S.-I. Hu, H. L. Drake, H. G. Wood, J. Bacteriol. 149,
440 (1982).
21. M. Kumar, S. W. Ragsdale, J. Am. Chem. Soc. 114,
8713 (1992).
22. C. Harford, B. Sarkar, Acc. Chem. Res. 30, 123 (1997).
RESEARCH ARTICLES
24. S. Nagashima et al., Nature Struct. Biol. 5, 347
(1998).
25. U. Ermler, W. Grabarse, S. Shima, M. Goubeaud, R. K.
Thauer, Science 278, 1457 (1997).
26. R. K. Thauer, Microbiology UK 144, 2377 (1998).
27. D. P. Barondeau, P. A. Lindahl, J. Am. Chem. Soc. 119,
3959 (1997).
28. S. Jaron, N. J. Blackburn, Biochemistry 38, 15086 (1999).
29. A. Szulc, D. Meyerstein, H. Cohen, Inorg. Chim. Acta
270, 440 (1998).
30. E. L. Maynard, C. Sewell, P. A. Lindahl, J. Am. Chem.
Soc. 123, 4697 (2001).
31. J. Seravalli, M. Kumar, S. W. Ragsdale, Biochemistry
41, 1807 (2002).
32. J. R. Andreese, A. Schaupp, C. Neuraute, A. Brown,
L. G. Ljungdahl, J. Bacteriol. 114, 743 (1973).
33. Z. Otwinowski, W. Minor, Methods Enzymol. 276,
307 (1997).
34. I. Uson, G. M. Sheldrick, Curr. Opin. Struct. Biol. 9,
643 (1999).
35. A. T. Brünger et al., Acta Crystallogr. D54, 905 (1998).
36. C. R. Kissinger, D. K. Gehlhaar, D. B. Fogel, Acta
37. K. Cowtan, Joint CCP4 and ESF-EACBM Newsletter on
Protein Crystallography 31, 34 (1994).
38. K. D. Cowtan, K. Y. Zhang, Prog. Biophys. Mol. Biol.
72, 245 (1999).
39. D. E. McRee, J. Struct. Biol. 125, 156 (1999).
40. T. A. Jones, J.-Y. Zou, S. W. Cowen, M. Kjeldgaard,
Acta Crystallogr. A47, 110 (1991).
41. R. A. Laskowski, M. W. McArthur, D. S. Moss, J. M.
Thornton, J. Appl. Crystallogr. 26, 283 (1993).
42. M. Carson, Methods Enzymol. 277, 493 (1997).
43. J. A. Christopher, thesis, Texas A&M University
(1998).
44. M. L. Youngblood, D. W. Margerum, Inorg. Chem. 19,
3068 (1980).
45. P. Strop et al., Biochemistry 40, 651 (2001).
46. We thank T. Earnest and G. McDermott at Advanced
Light Source (ALS) for help with data collection.
Support has been provided by the Templeton Foun-
dation (C.L.D.) and National Institutes of Health grant
R01-GM39451 (S.W.R.). The data collection facilities
at ALS are funded by the U.S. Department of Energy,
Office of Basic Energy Sciences.

Downloaded from www.sciencemag.org on November 14, 2015

23. W. Huang et al., Structure 5, 691 (1997). Crystallogr. D55, 484 (1999). 8 July 2002; accepted 3 September 2002

Shadows Cast by Retinal Blood nance columns. The relevance of these

experiments, however, can be questioned

Vessels Mapped in Primary on the grounds that suture of the eyelids is

an artificial manipulation, creating an ex-
treme imbalance in the stimulation of each
Visual Cortex retina. It has not been shown that under
natural circumstances, competition be-
Daniel L. Adams* and Jonathan C. Horton tween the two eyes has an important influ-
ence on the appearance of the ocular dom-
The mammalian eye is a remarkable optical device, but its design is not inance columns.
perfect. The blood vessels that supply the inner retina are located in front Cortical representation of retinal ves-
of the photoreceptor layer, blocking access to light. Their shadows create sels. We now show that visual deprivation
a pattern of blindness in the field of vision that corresponds precisely to the occurs routinely in normal primates, affecting
location of the largest vessels in the eye. We show here that in squirrel ocular dominance patterns. Helmholtz dis-
monkeys, focal deprivation by blood vessels leads to rewiring of the eye’s covered while plotting his blind spot that he
geniculocortical projections, imprinting an image of the retinal vascular tree could discern the proximal stumps of large
onto the primary visual cortex. This process illustrates vividly that local blood vessels emerging from the optic disc
imbalances in neuronal activity can influence column formation during (15). With refinement of perimetric tech-
normal development. niques, subsequent investigators have plotted
these so-called angioscotomas over wide por-
In many higher species, the geniculate affer- geniculate afferents driven by each eye. tions of the visual field (16, 17). The hemo-
ents serving right and left eyes are segregated These small imbalances may be amplified by globin in red blood cells absorbs light, cast-
in the brain in layer 4 of primary (striate, V1) competitive synaptic interactions between ing a shadow from blood vessels in the inner
visual cortex into a mosaic of interdigitated geniculocortical afferents to generate a ma- retina onto photoreceptors located under-
inputs called ocular dominance columns (1). ture system of ocular dominance columns neath. We have found a representation of
The formation of these columns has been (6–9). angioscotomas in striate cortex of the squirrel
scrutinized in the hope of elucidating the Doubt has been cast on this explanation monkey (Saimiri sciureus).
processes that guide the development of the for column development by recent experi- The squirrel monkey was reported original-
cerebral cortex (2). Some experiments show ments showing that ocular dominance ly to lack ocular dominance columns (18, 19).
an essential role for neuronal activity, be- columns appear in ferrets despite early bi- They are present, in fact, but often appear poor-
cause infusion of tetrodotoxin into both eyes lateral enucleation (10). This finding sug- ly developed and irregular (20). To examine
seems to prevent the emergence of ocular gests that column formation is ordained by their pattern more closely, we enucleated one
dominance columns (3). According to this intrinsic molecular cues, and not by pat- eye and stained striate cortex for the mitochon-
idea, the correlated discharge of neighboring terns of neuronal activity. Although neuro- drial enzyme cytochrome oxidase (CO) (21–
ganglion cells, perhaps generated by sponta- nal activity has a disputed role in column 23). The ocular dominance columns are re-
neous waves of retinal activity (4, 5), induces induction, it certainly can influence column vealed by this approach because metabolic ac-
local cortical imbalances in the density of morphology at a later stage in development. tivity is reduced in the missing eye’s columns.
Raising a newborn animal with monocular After perfusion, striate cortex was stripped from
eyelid suture, to mimic congenital cataract, the white matter, flattened in a single piece, and
Beckman Vision Center, University of California, San
Francisco, San Francisco, CA 94143– 0730, USA.
causes shrinkage of the deprived eye’s sectioned tangentially to the surface. Flatmount
columns and expansion of the normal eye’s preparations enhance the likelihood of detecting
*To whom correspondence should be addressed at
Beckman Vision Center, University of California, San
columns (11–14 ). Here, an imbalance in subtle metabolic patterns, such as the optic disc
Francisco, 10 Kirkham Street, San Francisco, CA neuronal activity has a profound influence representation, which lies buried in the calcar-
94143– 0730, USA. E-mail: dadams@itsa.ucsf.edu on the shape and size of the ocular domi- ine sulcus (24).

572 18 OCTOBER 2002 VOL 298 SCIENCE www.sciencemag.org

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

Ocular dominance is expressed most strong-
ly in layer 4C of striate cortex. CO montage of
this layer revealed a fuzzy pattern of fine ocular
dominance columns (Fig. 1). The left eye’s
optic disc representation appeared as a large,
homogeneous zone of CO activity in the right
cortex. It stained darkly, because the left eye
was enucleated in this animal. In the periphery,
a large pale zone corresponded to the represen-
tation of the far nasal retina of the left eye
(monocular crescent).
Radiating from the optic disc representa-
tion were more than a dozen, thin, dark pro-
files. They represented the blood vessels em-
anating from the optic disc of the left eye.
This was confirmed by matching each ele-
ment in the cortical pattern to its correspond-
ing retinal blood vessel (Fig. 1, C and D).
How do angioscotoma representa-
tions form? Angioscotoma representations
cannot form before birth, because the uterine
environment is dark. We propose that after
birth (Fig. 2A), the right cortex, for instance,
receives binocular input organized into nas-
cent columns, except at the representation of
the left optic disc, where input comes exclu-
RESEARCH ARTICLES
sively from the right eye. Once the baby blind spot representation (Fig. 2C). This exper-
monkey opens its eyes (Fig. 2B), light enters iment proved that cortical angioscotoma repre-
through the pupil and shadows are cast by sentations are generated by anatomical rear-
retinal blood vessels. Ganglion cells in the rangement of geniculocortical inputs. It should
left eye that receive input from photorecep- be emphasized that in three control animals,
tors beneath blood vessels are not driven with intact vision in both eyes, no CO pattern
effectively by visual stimulation. In turn, was visible in layer 4C. Removal of one eye is
these ganglion cells fail to activate lateral required to reveal the cortical angioscotoma
geniculate neurons. Consequently, the axon representations with CO.
terminals of these geniculate fibers retract in Angioscotomas were represented in stri-
striate cortex, and the territory they relinquish ate cortex of 9 of 12 squirrel monkeys.
is taken over by afferents serving the right Shadows generated in the contralateral eye
eye. After the critical period, geniculocortical
afferents become frozen in place. When one
eye is enucleated (Fig. 2C), CO reveals the
ocular dominance columns, which have been
rearranged to correspond to the cortical rep-
resentation of retinal blood vessels.
To confirm this hypothesis, we double-la-
beled the ocular dominance columns in this
animal (and two others) by injecting an anatom-
ical tracer, [3H]proline, into the remaining right
eye (25, 26). The proline and CO patterns were
identical (compare Figs. 1A and 3). As expect-
ed, the angioscotoma representations in striate
cortex were solidly labeled, as intensely as the

Fig. 2. CO staining of angioscotoma repre-

sentations. (A) At birth, CO staining is uni-
form. Angioscotomas are absent, but the
right eye already monopolizes the left eye’s
blind spot representation. (B) After light ex-
posure, vessels begin to cast shadows, gen-
erating angioscotoma representations. Their
pattern becomes immutable at the end of the
critical period. (C) Removal of one eye induc-
Fig. 1. Representation of angioscotomas in striate cortex. (A) Flatmount montage stained for es an undulating pattern of CO activity, cor-
CO after removal of the left eye to reveal the ocular dominance columns. The dark lines responding to the ocular dominance columns.
emerging from the central dark oval correspond to the blood vessels of the left eye. (B) Fundus The angioscotomas appear silhouetted
photograph showing optic disc and retinal blood vessels. (C) Drawing of vessel representations, against this textured background, because
color-coded to match the retinal drawing. Dashed line denotes circumference of retinal they stain solidly. Note the tendency for the
photograph in (B), projected onto the cortex. MC, monocular crescent; BS, blind spot. (D) other eye to “frost” the vessel and blind spot
Drawing of retina, with represented portions of vessels coded by color. Star denotes the fovea. representations.

www.sciencemag.org SCIENCE VOL 298 18 OCTOBER 2002 573

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

predominated, because blood vessels are
generally larger in nasal retina, especially
near the optic disc. A few vessel represen-
tations, however, could be traced across the
vertical meridian (V1/V2 border) and into
the opposite cortex (Figs. 3 and 4). Their
contrast was reversed, compared with other
Fig. 3. Autoradiograph pre-
pared after tracer injection into
the remaining right eye, show-
ing label (bright regions) dis-
tributed in a pattern identical
to that in Fig. 1A. The congru-
ence of the patterns confirms
that CO staining after monoc-
ular enucleation precisely re-
flects the distribution of
geniculocortical inputs. The an-
gioscotomas and the blind spot
are solidly labeled, as predicted
in Fig. 2C. Arrow denotes an
unlabeled angioscotoma, aris-
ing from a temporal vessel in
the right eye.
RESEARCH ARTICLES
angioscotomas in the same cortex, because
they originated from the temporal retina of
the ipsilateral eye.
Each angioscotoma representation was
fringed by cortex that exhibited opposite CO
contrast (Fig. 5). This “frosting” enhanced
the visibility of the angioscotoma pattern by
highlighting each vessel representation. Ap-
parently, when retraction of geniculate affer-
ents occurs during the critical period, the
displaced afferents form a solid phalanx
along the edges of the representation of each
angioscotoma. The displaced afferents dom-
inate in these flanking zones, expelling those
from the other eye. This is noteworthy, be-
cause neither eye enjoys any obvious com-
petitive advantage in visual stimulation right
next to an angioscotoma.
Precision of the visual field represen-
tation. The biggest vessels represented in
the cortex were retinal veins near the optic
disc. They attained a diameter of nearly 100
␮m (0.62°). The smallest were vessels ap-
proaching the fovea, less than 30 ␮m wide
(0.17°). To calculate shadow widths, we mea-
sured exit pupil diameter, distance from exit
pupil to retinal vessel, retinal vessel diameter,
and distance from retinal vessel to photore-
ceptors (27, 28). Many smaller vessels failed
to eclipse the pupil completely and therefore
produced no umbra. For example, a 28-␮m-
diameter vessel 7.5° from the fovea cast only
penumbra, measuring 83 ␮m in width (Fig.
6). From an expression for cortical magnifi-
cation factor along an isoeccentricity ring

Fig. 4. Vessels in nasal,

compared with temporal,
retina are represented
more extensively. (A) Ret-
inal photomontage. (B)
CO staining after enucle-
ation of the right eye.
Most angioscotomas are
dark, because they repre-
sent vessels of the right
eye’s nasal retina, but a
few are pale, representing
the left eye’s temporal
retina. (C) Drawing of the
right fundus. The dotted
lines outline the left eye’s
blind spot and major infe-
rior fundus vessels (see
retinal montage, inset).
Their temporal segments
account for the two pale
angioscotoma representa-
tions visible in the lower
left cortex. (D) Drawing of
CO pattern. Abbreviations
as in Fig. 1.

574 18 OCTOBER 2002 VOL 298 SCIENCE www.sciencemag.org

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

(derived below), we calculated that an 83-␮m
shadow should produce a cortical representa-
tion 737 ␮m wide. In fact, the vessel’s corti-
cal profile measured only 350 ␮m (see vessel
11, Fig. 4D), indicating that just the central
RESEARCH ARTICLES
⬃50% of the shadow was sufficiently dense
to induce remodeling of geniculocortical af-
ferents. This portion of the shadow was only
three to four cones wide.
Generation of a retinotopic map. The
angioscotoma representations provided a
simple means for construction of a visual
field map in striate cortex (Fig. 7). First,
dozens of prominent landmarks, such as ves-
sel crossings or bifurcations, were identified
in the retina and projected onto a tangent
screen to determine their eccentricity. These
data were used to calibrate a system of rings
and rays, which was superimposed on an
image of the nasal retina. This image was
transformed by computer to match retinal
features precisely to their cortical representa-
tion (29), thereby generating a retinotopic
map. This map is truly retinotopic, as op-
posed to visuotopic, and therefore immune to
errors introduced by eye movements or inac-
curate plotting of receptive fields (30).
Areal cortical magnification factor, Ma,
was derived by measuring the area (mm2 ) of
each compartment in the map and dividing by
the area (degrees squared) of its correspond-
ing visual field zone. The data were plotted as
a function of eccentricity (Fig. 7D) for four
hemispheres from two monkeys. Cortical
magnification factor in the squirrel monkey
was similar to the value reported in the ma-
caque (31), showing that these evolutionarily
divergent primate species emphasize macular
vision to the same extent.
Discussion. Retinal blood vessels give
rise to angioscotomas in the visual field, just
as the optic disc creates a blind spot. How-
ever, the mechanism is quite different. The

Fig. 5. Vessel representations are flanked by

CO activity of opposite contrast. A vertically
oriented dark angioscotoma representation
(Fig. 4D) is surrounded by pale staining. Mean
optical density of each column of pixels is
plotted below.

Fig. 6. Optics of shadow formation. Toluidine

blue–stained retinal cross section (Fig. 4C),
showing the shadow cast by a blood vessel in
the ganglion cell layer with a 28-␮m lumen
diameter, when fully inflated. Pairs of lines,
projected from opposite margins of a 1.9-mm
pupil located 8.4 mm in front of the retina,
show the boundaries of the shadow obscuring
photoreceptors. The umbra (gray triangle) Fig. 7. Generation of a retinotopic map. Ring and ray pattern is superimposed on the right retina,
does not reach the photoreceptor layer in using landmarks plotted in vivo. Correspondence points (yellow squares) are placed on matching
this example, because the vessel is too small. features of the retina and cortex. Blue lines are placed on segments between correspondence
Gradient indicates penumbra, which reached points. They include blood vessels, vertical meridian (V1/V2) border, horizontal meridian, and the
a maximum optical density of 0.45. Open boundary between binocular and monocular cortex (measured in vivo at 72°). Warping of the retina
arrow shows the light-recipient boundary be- onto the cortex produces a retinotopic map. Areal cortical magnification factor (mean ⫾ SEM) for
tween cone inner and outer segments. four cortices is graphed versus eccentricity.

www.sciencemag.org SCIENCE VOL 298 18 OCTOBER 2002 575

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

optic disc produces a hole, devoid of gangli-
on cells, in the fabric of the retina. The other
eye fills in the cortical gap, visible as a solid
oval amid a sea of ocular dominance col-
umns. This occurs in utero, before visual
experience has any influence (32).
Angioscotomas, on the other hand, arise
because a few photoreceptors are condemned
to a life of idleness owing to their location
beneath blood vessels. The retina is otherwise
normal. Deprivation by vascular shadows
induces local remodeling of geniculate affer-
ents, in a cortical distribution that corre-
sponds to the trajectory of blood vessels with-
in the retina. The process is analogous to
amblyopia induced by a congenital cataract.
A lens opacity, however, provokes global
shrinkage of geniculate afferents in the cortex
(33). Our results show that amblyopia can be
exquisitely focal within an eye. It probably
occurs underneath retinal blood vessels in
primates other than squirrel monkeys, includ-
ing humans.
Angioscotoma representations in the cor-
tex provide a natural example of how visual
activity can influence the pattern formed by
ocular dominance columns. They are essen-
tially nothing more than long, threadlike oc-
ular dominance columns, generated because
deprivation in one eye tips the local compe-
tition for cortical territory in favor of the
other eye. The deprived eye is compensated
by preferential assignment of the closest ad-
jacent territory—as reflected by the “frost-
ing” lining the angioscotoma representations
(Fig. 5). This phenomenon presumably re-
flects an innate tendency to allot cortical
territory in a reciprocal manner (6–8, 34, 35).
The global periodicity of the columns result-
ing from this process is probably under ge-
netic control (36). However, our data show
that local imbalances in neuronal activity can
determine the precise layout of the columns
in any given region of cortex.
We have searched in vain for angioscotoma
representations in striate cortex of the macaque
and human. In general, these species have larg-
er ocular dominance columns than squirrel
monkeys. In the macaque, we have projected
the retinal vasculature onto striate cortex, to
compare the predicted size of individual vessel
representations with the width of ocular domi-
nance columns. The ocular dominance columns
are much larger. We speculate that because the
columns are larger, their pattern is resistant to
the influence of angioscotomas during develop-
ment. In the macaque, the amount of cortex
deprived by angioscotomas is simply too small
to compel the rearrangement of ocular domi-
nance columns. Nonetheless, shouldn’t an an-
gioscotoma carve a narrow swath through any
column, no matter how big? Apparently, this is
prevented by the higher degree of segregation
of geniculocortical afferents in macaques, com-
pared with squirrel monkeys. In squirrel mon-
576 18 OCTOBER 2002 VOL 298 SCIENCE www.sciencemag.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
RESEARCH ARTICLES
keys, the fine grain and weak segregation of the
ocular dominance columns create perfect
conditions for the formation of angioscotoma
representations.
Amblyogenic shadows just a few cones
wide were represented in the cortex, reflecting
the remarkable precision of the visual field map
in layer 4C (37). In other layers, angioscotomas
were not visible, because of the coarser resolu-
tion of the visual field map and the intermin-
gling of ocular channels. The representation of
angioscotomas was most extensive at eccentric-
ities from 4° to 30°. Within this range, an ideal
balance exists between cortical magnification
factor and blood vessel size. Beyond 30°, vessel
size and magnification factor both decrease, so
most angioscotoma representations disappear.
Toward the fovea, blood vessels rapidly taper.
Their reduction is offset by a steep rise in
cortical magnification factor, but nevertheless,
angioscotoma representations abruptly peter out
at 3° to 4°. This occurs because vessels near the
fovea contain too little blood to cast a suffi-
ciently dense shadow.
Angioscotomas are rarely noticed in
our vision because they are filled in perceptual-
ly, just like the blind spot. Remarkably, an-
gioscotomas amount collectively to a greater
portion of the visual field than the blind spot.
Nature has struck a compromise in the design of
the eye, by allowing blood vessels to obscure
some photoreceptors. At least the largest vessels
in the retina are prohibited from approaching the
fovea, so that we are scarcely aware of their
presence.
References and Notes
1. D. H. Hubel, T. N. Wiesel, Nature 221, 747 (1969).
2. L. C. Katz, C. J. Shatz, Science 274, 1133 (1996).
3. M. P. Stryker, W. A. Harris, J. Neurosci. 6, 2117
(1986).
4. M. Meister, R. O. L. Wong, D. A. Baylor, C. J. Shatz,
Science 252, 939 (1991).
5. R. Mooney, A. A. Penn, R. Gallego, C. J. Shatz, Neuron
17, 863 (1996).
6. C. von der Malsburg, D. J. Willshaw, Exp. Brain Res.
Suppl. 1, 463 (1976).
7. N. V. Swindale, Proc. R. Soc. London Ser. B 208, 243
(1980).
8. K. D. Miller, J. B. Keller, M. P. Stryker, Science 245, 605
(1989).
9. G. J. Goodhill, Network 9, 419 (1998).
10. J. C. Crowley, L. C. Katz, Nature Neurosci. 2, 1125
(1999).
11. D. H. Hubel, T. N. Wiesel, S. LeVay, Philos. Trans. R.
Soc. London Ser. B 278, 377 (1977).
12. C. J. Shatz, M. P. Stryker, J. Physiol. (London) 281, 267
(1978).
13. N. V. Swindale, F. Vital-Durand, C. Blakemore, Proc.
R. Soc. London Ser. B 213, 435 (1981).
14. M. C. Crair, E. S. Ruthazer, D. C. Gillespie, M. P.
Stryker, Neuron 19, 307 (1997).
15. H. von Helmholtz, in Handbook of Physiological Op-
tics (Optical Society of America, Menasha, WI, 1924),
vol. 2, p. 28.
16. J. N. Evans, Am. J. Ophthalmol. 9, 489 (1926).
17. U. Schiefer et al., Vision Res. 39, 1897 (1999).
18. M. S. Livingstone, J. Neurosci. 16, 2086 (1996).
19. A. E. Hendrickson, J. R. Wilson, M. P. Ogren, J. Comp.
Neurol. 182, 123 (1978).
20. J. C. Horton, D. R. Hocking, J. Neurosci. 16, 5510
(1996).
21. Experiments were performed on 12 adult squirrel
monkeys from the California Regional Primate Cen-
ter, Davis, CA. All procedures were approved by the
University of California, San Francisco, Committee on
Animal Research. Each animal was anesthetized with
2% isoflurane and paralyzed with succinylcholine (45
mg/kg, intravenously). The animal was placed in a
stereotaxic frame mounted on a tripod to orient the
visual axis perpendicular to a large tangent screen.
Photographs were taken of both retinae and land-
marks were projected onto a tangent screen with a
fundus camera to pinpoint their eccentricity. Retinal
magnification factor was 161 ␮m per degree in the
central 24°. One eye was then enucleated under
sterile conditions. Buprenorphine (0.03 mg/kg) was
administered for 48 hours after surgery for analgesia.
After a survival time of at least 10 days, the animal
was killed with sodium pentobarbital (150 mg/kg)
and perfused with 1% paraformaldehyde. Flatmounts
were cut at 30 to 40 ␮m, air-dried on slides, reacted
for CO, and montaged with Photoshop 6.0.
22. M. T. T. Wong-Riley, Brain Res. 171, 11 (1979).
23. J. C. Horton, Philos. Trans. R. Soc. London Ser. B 304,
199 (1984).
24. J. C. Horton, D. R. Hocking, J. Neurosci. 16, 7228
(1996).
25. Two millicuries of L-[2,3,4,5-3H]proline were recon-
stituted in 20 ␮l of saline and injected into the
vitreous of one eye. Alternate cortical sections were
processed for CO or autoradiography.
26. T. N. Wiesel, D. H. Hubel, D. M. K. Lam, Brain Res. 79,
273 (1974).
27. R. A. Applegate, A. Bradley, W. A. van Heuven, Invest.
Ophthalmol. Vis. Sci. 31, 2088 (1990).
28. Entrance pupil size was measured by photography
with a 100-mm macro-lens in five squirrel mon-
keys aged 1 week; it was 1.9 ⫾ 0.03 mm (SEM),
with a range of 1.7 to 2.0 mm. This value was used
for exit pupil size, because for small pupils, the
difference between entrance and exit pupil size is
negligible. Distance from pupil to retina was mea-
sured in three newborn squirrel monkeys with A-
mode ultrasound; it was 8.4 ⫾ 0.12 mm (SEM).
Adult retinae were sectioned at 1 ␮m in Epon-
Araldite and the lumen diameter of identified
blood vessels was measured. Distance from retinal
vessel to photoreceptors was also measured in
these sections. Correction was made for retinal
growth and tissue shrinkage.
29. About 50 to 100 correspondence points were used to
warp the nasal retina onto the cortex, using Elastic
Reality 3.0 (www.avid.com). No angioscotoma repre-
sentations were present centrally, so points were
obtained from Cowey’s microelectrode mapping of
the central 4° (30). Linear cortical magnification fac-
tor along polar rays (Mp) and along eccentricity rings
(Me) were calculated independently: Mp ⫽ 9.5
(0.95 ⫹ E)⫺1.03 and Me ⫽ 188 (6.45 ⫹ E)⫺1.85, where
E is eccentricity.
30. A. Cowey, J. Neurophysiol. 27, 366 (1964).
31. D. C. Van Essen, W. T. Newsome, J. H. Maunsell,
Vision Res. 24, 429 (1984).
32. J. C. Horton, D. R. Hocking, J. Neurosci. 16, 1791
33. 㛬㛬㛬㛬
(1996).
, J. Neurosci. 17, 3684 (1997).
34. D. G. Jones, R. C. Van Sluyters, K. M. Murphy, J. Neu-
rosci. 11, 3794 (1991).
35. G. J. Goodhill, Biol. Cybern. 69, 109 (1993).
36. M. Kaschube, F. Wolf, T. Geisel, S. Löwel, J. Neurosci.
22, 7206 (2002).
37. G. G. Blasdel, D. Fitzpatrick, J. Neurosci. 4, 880
(1984).
38. This work was supported by the National Eye Institute
(grant R01 EY10217 and core grant EY02162), That
Man May See, The Bunter Fund, and a Lew W. Wasser-
man Merit Award from Research to Prevent Blindness.
We thank D. Hocking and I. Wood for help with tissue
processing and R. Troyer for assistance with animal care
and surgery. We thank the California Regional Primate
Research Center, which is supported by NIH Base Grant
RR00169. L. Sincich, N. Swindale, and G. Westheimer
provided valuable comments on the manuscript.
10 June 2002; accepted 29 July 2002
 Shadows Cast by Retinal Blood Vessels Mapped in Primary Visual
Cortex
Daniel L. Adams and Jonathan C. Horton
Science 298, 572 (2002);
DOI: 10.1126/science.1074887

This copy is for your personal, non-commercial use only.

If you wish to distribute this article to others, you can order high-quality copies for your
colleagues, clients, or customers by clicking here.
Permission to republish or repurpose articles or portions of articles can be obtained by
following the guidelines here.

The following resources related to this article are available online at

Downloaded from www.sciencemag.org on November 14, 2015

www.sciencemag.org (this information is current as of November 14, 2015 ):

Updated information and services, including high-resolution figures, can be found in the online
version of this article at:
http://www.sciencemag.org/content/298/5593/572.full.html
A list of selected additional articles on the Science Web sites related to this article can be
found at:
http://www.sciencemag.org/content/298/5593/572.full.html#related
This article cites 32 articles, 18 of which can be accessed free:
http://www.sciencemag.org/content/298/5593/572.full.html#ref-list-1
This article has been cited by 23 article(s) on the ISI Web of Science
This article has been cited by 12 articles hosted by HighWire Press; see:
http://www.sciencemag.org/content/298/5593/572.full.html#related-urls
This article appears in the following subject collections:
Development
http://www.sciencemag.org/cgi/collection/development

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the
American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright
2002 by the American Association for the Advancement of Science; all rights reserved. The title Science is a
registered trademark of AAAS.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 The Journal of Neuroscience, January 1993, 13(l): 87-103

Plasticity in the Frequency Representation of Primary Auditory

Cortex following Discrimination Training in Adult Owl Monkeys

G. H. Recanzone,’ C. E. Schreiner, and M. M. Merzenich

Coleman Laboratory, Departments of Otolaryngology and Physiology and Keck Center for Integrative Neuroscience,
University of California at San Francisco, San Francisco, California 94143-0732

Previous studies have shown that the tonotopic organization tial cochlear lesions(Robertson and Irvine, 1989; Irvine et al.,
of primary auditory cortex is altered subsequent to restricted 1991). Tuning properties of single cortical neurons have been
cochlear lesions (Robertson and Irvine, 1989) and that the shownto be altered by conditioning paradigms(Olds et al., 1972;
topographic reorganization of the primary somatosensory Disterhoft and Stuart, 1976; Kitzes et al., 1978; Ryugo and
cortex is correlated with changes in the perceptual acuity Weinberger, 1978) and pharmacological manipulations (Ashe
of the animal (Recanzone et al., 1992a-d). Here we report et al., 1989; McKenna et al., 1989; Metherate and Weinberger,
an increase in the cortical area of representation of a re- 1990; seeWeinberger et al., 1990, for review). Although the
stricted frequency range in primary auditory cortex of adult cortical tonotopic map wasnot explicitly studied, the functional
owl monkeys that is correlated with the animal’s perfor- plasticity at the singleneuron level suggeststhat the tonotopic
mance at a frequency discrimination task. representation of the cochlea in the cortex may be altered as a
Monkeys trained for several weeks to discriminate small result of the behavioral training.
differences in the frequency of sequentially presented tonal Reorganization of cortical representationshasbeenarguedto
stimuli revealed a progressive improvement in performance be continuous throughout life and to reflect an individual’s abil-
with training. At the end of the training period, the tonotopic ity to acquire new skills and behaviors (Merzenich et al., 1988,
organization of Al was defined by recording multiple-unit 1990). Improvements in performance have been demonstrated
responses at 70-258 cortical locations. These responses for both simple auditory detection and discrimination tasks
were compared to those derived from three normal monkeys (Zwislocki et al., 1958;Sinnott et al., 1985; Prosenet al., 1990),
and from two monkeys that received the same auditory stim- and similar improvements in performance on tactually based
uli but that were engaged in a tactile discrimination task. taskshave recently beendemonstratedto parallel changesin the
The cortical representation, the sharpness of tuning, and the functional organization of somatosensorycortical areas3a and
latency of the response were greater for the behaviorally 3b (Recanzone et al., 1992a-d). The demonstrationsthat au-
relevant frequencies of trained monkeys when compared to ditory cortical neurons can alter their frequency-specific re-
the same frequencies of control monkeys. The cortical area sponseproperties by either peripheral dennervation or behav-
of representation was the only studied parameter that was ioral conditioning suggestedto us that the types of cortical
correlated with behavioral performance. These results dem- reorganization seenin the somatosensorycortex also occur in
onstrate that attended natural stimulation can modify the primary auditory cortex. Specifically, we hypothesized that the
tonotopic organization of Al in the adult primate, and that functional organization of AI is altered as a consequenceof an
this alteration is correlated with changes in perceptual acu- animal’s improvement in auditory discrimination ability. To
ity. test this hypothesis, three adult owl monkeys were trained at
[Key words: auditory cortex, primate, plasticity, behavior, an auditory frequency discrimination task for several weeks.
physiology, discrimination training] The tonotopic representationofthe primary auditory cortex was
defined electrophysiologically after significant improvements in
Studieswithin the somatosensorysystemhave shown that par-
performance were measured.The responseproperties of these
tial peripheral dennervation and restricted natural stimulation
neurons were then compared to those of normal monkeys, as
can result in a reorganization of the topographic representation well asto thoseof monkeys that had received the sameauditory
of the body surfacein the cerebral cortex (seeMerzenich et al.,
stimulation but were engagedin an unrelated tactile task. The
1990; Recanzone and Merzenich, 1992). Similarly, reorgani-
electrophysiologicaldata were then analyzed to determineif any
zation of the topographic representation of the cochlea in the
simple measureof the cortical representationof the frequencies
primary auditory cortex has been demonstrated following par- used in the behavioral paradigm was correlated with the im-
Received Nov. 27, 1991; revised June 23, 1992; accepted July 9, 1992. proved behavioral performance.
We thank G. T. Hradek, M. L. Sutter, and R. E. Beitel for their important
contributions to this work, W. M. Jenkins for insightful discussions regarding the Materials and Methods
behavioral procedures, and Bret E. Peterson for the computer software used in
the data analysis. Funding was provided by NIH Grants NS-10414, GM-0774, Anima/s. Ten adult owl monkeys ofboth sexes were used. Animals were
and ONR NO00 14-9 I-J- 13 17, Hearing Research Inc., and the Coleman Fund. judged to have normal hearing based on evoked potential recordings
Remint reauests should be addressed to M. M. Meaenich at the above address. (ABR). The outer ear canal was determined to be free of obstruction
Ali o;her co;espondence should be addressed to G. H. Recanzone, Laboratory and infection. Monkeys were given free access to water and were food
of Sensorimotor Research, Building 10, Room lOClO1, National Institutes of deprived for a 21-22 hr period prior to each psychophysical testing
Health, Bethesda, MD 20892. session. Supplemental fruit was given at the end of each testing session
Copyright 0 1993 Society for Neuroscience 0270-6474/93/l 30087-17$05.00/O to maintainbodyweightbetween85-95%adlib. Thecareandtreatment

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 88 Recanzone et al. * Auditory Cortex Plasticity with Training
of these monkeys were in accordance with the NIH Guide for the Care
and Use of Laboratory Animals (revised 1987).
Psychophysics. The psychophysical apparatus and methods have been
described in detail (Recanzone et al., I99 I). Five monkeys were trained
to detect a difference in the frequency of sequentially presented pairs of
tone pips. Monkeys were trained in an acoustically transparent test cage
housed within a single-walled acoustic chamber lined with echo-atten-
uating foam. Auditory stimuli were delivered in the free-field from a
single speaker (Realistic model l4- 1996) located directly over the ani-
mal’s head. All aspects of the stimulus generation and data collection
were automated and under the control of a microcomputer.
Monkeys were trained to initiate a trial by contacting a hand mold
(monkeys OMI, OM2, OM3, CMI) or bar located outside of the test
cage (OM4 and OM5), or by pulling a lever located inside the test cage
(CM2). In each case, this observing response located the animal’s head
in a stereotyped location and orientation. Head position was monitored
continuously with a video camera. Tone pip pairs were presented every
650 msec. Each tone pip had a duration of I50 msec (3 msec rise/fall),
with the onset of the second pip occurring 50 msec after the offset of
the first. The frequency of the first tone pip was constant for a given
frequency discrimination task. Four different frequency ranges were
tested: 2.5 kHz task (25 I I Hz standard), 3 kHz task (3 I88 Hz standard),
5 kHz task (4765 Hz standard), and 8 kHz task (7950 Hz standard).
Each monkey was extensively trained on at least one of these four tasks,
and most were also tested on at least one other task on a few sessions
spaced throughout the training period. The comparison, or Sl, stimulus
consisted of both tone pips of the pair being equal in frequency (the
standard). For the target, or S2, stimulus, the frequency of the second
tone pip was different from the frequency of the standard. Monkeys
were required to maintain contact with the mold, bar, or lever through-
out the presentations of the SI stimuli, and to release contact upon
detection ofthe S2 stimulus within a 450 msec reward window beginning
100 msec after S2 stimulus onset. Failures to release within this reward
window were scored as a “miss” and were punished by a l-5 set time-
out. Releases before this time were scored as a “false-positive” and also
resulted in a time-out. Releases within the reward window were scored
as a “hit” and the monkey received a 45 mg banana-flavored food pellet
(Bio Serve). Six to ten different S2 stimulus frequencies were presented
on at least 20 trials during a given session. These frequencies were chosen
such that the monkey received a reward on 65-75% of all trials. Each
monkey would perform 400-750 trials in a daily session.
The number of Sl stimulus presentations (bins) on a given trial was
pseudorandomly determined to be between 2 and I I. The probability
that the next stimulus would be an S2 (approximately 0.3) did not vary
during the course of a single trial. In the condition that I I Sl stimuli
were presented, the 12th stimulus was always the S2. These trials were
not included in the data analysis, as they were predictably the S2 based
on time alone. Trials in which the monkey made a response in bins I
and 2 were similarly omitted as they were always the S I. The perfor-
mance (P) for each S2 frequency was defined by
P = H.S.
where H is the hit rate for that frequency (H = # correct responses/#
presentations) and S is the safe rate correction factor (S = I ~ false-
positive rate). The psychometric function for each session was deter-
mined by plotting P as a function of the frequency difference between
the Sl and S2 stimuli (delta frequency: AF). Threshold was defined from
this function as the S2 frequency at which P = 0.50. This is a reliable
measure for discrimination performance that is essentially unaffected
by false-positive rates below 15% (see Recanzone et al., 199 I, 1992a).
The slope of the psychometric function was defined as the slope of the
line connecting the data points immediately above and below threshold.
Threshold was defined in all sessions in which at least one S2 stimulus
had a performance value below 0.50 and the false-positive rate for the
session was below 15%. The psychophysical data for several different
S2 frequencies were subjected to signal detection analysis as described
by Green and Swets (I 966) for five consecutive training sessions spaced
throughout the total training period (see Results). This analysis measures
the difference in the response between the Sl and S2 stimuli in units of
standard deviation and is independent of changes in the subject’s in-
ternal criteria.
Three classes of control trials vertified that responses were based on
the perception of the frequency of the auditory stimulus. On approxi-
mately 2% of the trials, the S2 stimulus was not presented until after
I I S I presentations (described above). On these trials, a release response
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
after the 12th stimulus always resulted in a reward; however, the per-
formance at any given frequency was comparable regardless of the stim-
ulus presentation bin. On 0.5-l .O% of the trials, the S2 stimulus gen-
erated by the computer had a delta frequency of zero (S I = S2). Monkeys
under stimulus control never responded in this condition. Finally, mon-
keys did not respond on trials in which the speaker was disconnected
yet all other aspects of the trials were identical.
In order to prevent intensity cues as a function of frequency conse-
quent to changes in the location of standing waves present in the free
field, the intensity of each tone pip was varied independently over a
IO-I 2 dB range. The intensity of the stimuli centered in this range was
measured at three or four different frequencies at I2 separate locations
throughout the sound field for each of the four tasks (Table I). These
locations were chosen to sample the area occupied by the monkey’s
head during a session and formed a cube of approximately 5 cm/side,
with four additional measurements within this cube. There was no
statistically significant difference in the intensity measured throughout
the sound field between any of the four tested frequencies for each task
(two-tailed t test; p > 0. IO). Furthermore, repeated measures throughout
the course of this study showed no statistical differences over time. The
spectral content of these signals was measured in the free field to de-
termine the potential contributions of each harmonic of the signal. The
highest-intensity harmonics were always 40-50 dB less than the intensity
of the fundamental. Finally, the head position of each monkey was
viewed continuously with a video camera on each session, and analysis
of head position with the behavioral response was made from several
videotaped sessions in each monkey. This analysis showed no consistent
relationship between the location or small movements of the monkey’s
head with the presentation of stimuli or with the behavioral response.
In addition to the five monkeys trained at the auditory frequency
discrimination task, one monkey (CMI) was trained to detect a change
in the frequency of a sinusoidal tactile stimulus (monkey E-2 of Re-
canzone et al., 1992a+l). The same auditory stimuli as used in the 8
kHz task were presented simultaneous to the presentation of the tactile
stimulus, but did not give any indication of reward availability. Monkey
CM2 was trained to perform a tactile vibration detection task in the
presence of the same auditory stimuli used in the 5 kHz task. These
monkeys represent passive-stimulation controls.
Electrophysiology. The primary auditory cortex was defined electro-
physiologically in the seven monkeys trained in a psychophysical task
and in three normal owl monkeys (N I, N2, and N3) that had not been
trained at any task. Anesthesia was induced with a halothane (3%) : nitrous
oxide (72%) : oxygen (25%) gaseous mixture. The femoral vein was can-
nulated and sodium pentobarbital was given intravenously to maintain
an areflexic state of anesthesia (28 mg/kg induction, l-5 mg/hr main-
tenance). Penicillin G (30.000 U/24 hr) and atronine sulfate (0. I me/
kg/ I2 hi) were given i&ramuscularly. kctated R’inger’s solution wirh
5% dextrose was continuously infused (2-5 ml/hr). Heart rate, respi-
ration rate, and other vital signs were monitored. Core body temperature
was maintained at 37°C with a thermostatically controlled heating blan-
ket. A craniotomy exposed the relevant cortical region, the dura was
removed, and a well of silicon oil was constructed to bathe the cerebral
cortex continuously. A 40x image of the cortical surface was taken
either photographically or with a digitized video image. All electrode
penetrations were parallel to each other. The electrode insertion points
were reproduced onto the image of the cortical surface with reference
to the vasculature. Electrode penetrations away from the sulcus were
made only to a single depth corresponding to the middle cortical layers
(500-900 pm below the cortical surface).
The stimulus generation and data acquisition methods have been
described in detail elsewhere (Schreiner and Mendelson, 1990). Briefly,
single tone pips of 50 msec duration (3 msec on/offramp) were generated
by a microprocessor (TMS32010, I6 bit D/A converter at 120 kHz,
low pass filtered at 35 kHz) and presented monaurally via a calibrated
headphone (STAX 54) connected to a sound delivery tube sealed into
the contralateral ear canal. Characteristic frequency (CF’) and threshold
were defined audiovisually at all recording locations in all monkeys.
Frequency-response areas (FRAs) were defined for all locations in mon-
keys OMI, OM3, and CMI, and in all three normal monkeys. FRAs
were reconstructed from the response to a single stimulus at each of 45
frequencies spanning a 2-5 octave range centered at the approximate
characteristic frequency (CF) and equidistant on a logarithmic scale.
Each frequency was presented at each of I5 levels spanning a 70 dB
range in 5 dB steps (675 total stimuli). QlOdB and Q40dB were defined
from these frequency response areas as the CF divided by the bandwidth
 The Journal of Neuroscience, January 1993, 13(l) 89

Table 1. Mean and SD of sound intensities (dB SPL) measured at 12 locations within the sound field
of the apparatus

Task S I stimulus Below threshold Near threshold Above threshold

2.5 kHz 64.27 k 2.1 64.44 YE2.89 64.17 zk 2.93 65.21 f 2.93
3 kHz 57.65 i 3.97 59.55 i 3.01 60.16 f 2.90 57.15 f 3.78
5 kHz 54.12 i 2.93 55.39 k 3.08 56.51 f 3.41 53.55 f 3.88
8 kHz 71.60 & 4.69 71.81 f 4.35 73.13 f 4.71

Measurements were taken at the mid-range intensity used in the behavioral task. In the 2.5 kHz, 3 kHz, 5 kHz, and 8
kHz tasks, the frequencies measured under “SI stimulus” were 251 I Hz, 3188 Hz, 4765 Hz, and 7950 Hz; “Below
threshold” were 2527 Hz, 3275 Hz, 4849 Hz, and 8036 Hz; “Near threshold” were 2534 Hz, 3367 Hz, 4939 Hz; and
“Above threshold” were 26 13 Hz, 364 I Hz, 52 12 Hz, and 837 I Hz, respectively

10 dB and 40 dB above minimum threshold, respecttvely, tollowmg 5 12 +. The specific statistical test was an unpaired one-tailed I test unless
methods described by Schreiner and Mendelson (1990) and Sutter and otherwise indicated. p values ~0.05 were considered to be statistically
Schreiner (199 1). The minimum latency was defined as the time from significant.
stimulus onset to the earliest consistent response for all 15 intensity
levels presented for the three frequencies nearest CF (45 stimuli total).
Electrodes were parylene-insulated tungsten wires with impedances Results
of approximately 1 MR at 1 kHz (Microprobe, Inc). Neural signals were
amplified, band-pass filtered (I-10 kHz), and displayed on an oscillo- Psychophysics
scope and audio amplifier. Spike activity was isolated from this signal Improvement in performance with training. The auditory fre-
with a window discriminator (BAK DIS- 1) set to accept all waveforms quency discrimination abilities of each of the five trained mon-
greater than two times the neural noise level with the time window set keys improved progressively with training. The improvement
for the peak of the largest responses. This eliminated evoked potentials
in threshold with training is shown for two representation mon-
and included approximately one to five independent waveforms of neu-
ral origin. The number and arrival time ofeach event relative to stimulus keys in Figure 1. This improvement was characterized by a
onset were recorded and stored in a computer (DEC 1 l/73). relatively short and steep phase in which large improvements
At the start of each experiment, the angle of the electrode was set to were recorded between sessions, followed by a longer period in
be approximately parallel with the angle of the sulcus. Our objective which smaller gains were observed. These observations are sum-
was to define tuning curves limited to neurons in the middle cortical
layers, which characteristically have the “best” responses as defined by marized for each of the five monkeys in Figure 2. The discrim-
vigorous activity with a short latency and well-defined tuning curves. ination threshold was averaged over five sessions at different
Three long penetrations were made approximately 300 pm apart and periods roughly equally spaced throughout the several week
located along a line perpendicular to the sulcus. In these penetrations, training period. In almost every case, the largest gains were
the neural response to acoustic stimuli was defined audiovisually in 500 measured between the initial five sessions and the five sessions
wrn steps. By comparing the neural responses at equivalent distances
along these three tracks, we were able to verify that the angle of the taken after approximately one-third of the total number of ses-
penetration was roughly parallel to the sulcus. Electrode tracks near the sions. The decrease in threshold was generally less throughout
sulcus were continued until an unambiguous border with another cor- the remainder of training, although significant decreases were
tical field (marked by an abrupt shift in characteristic frequency) was measured in every monkey (p < 0.05). Final mean discrimi-
encountered, or until neural activity could no longer be evoked with
acoustic stimuli. These penetrations were commonly 3-4.5 mm in length. nation thresholds were slightly better for the 2.5 kHz task than
At the conclusion of the electrophysiology experiments, electrolytic the 8 kHz task (see Table 2). By contrast, the final thresholds
lesions were made at selected locations by passing 10 FA ofdirect current for the 5 kHz and the 3 kHz tasks in monkeys OM4 and OM5
for 1O-20 sec. Carbon particle marks were made by insertion of a tung- were much greater, even though these monkeys also showed
sten electrode that had been coated with india ink at four to seven
locations surrounding the investigated cortical tissue. The animals were significant improvements with training.
given a lethal injection of barbiturate, and then perfused intracardially Previous studies in owl monkeys on tactile frequency dis-
with normal saline at 37°C followed by one oftwo fixatives in phosphate crimination performance have shown that improvements in
buffer: 10% formalin or 1.5% paraformaldehyde with 1.5% glutaral- minimally detectable frequency differences resulted from (1) a
dehyde. Frozen sections were cut at 100 pm thickness in the frontal leftward shift in the psychometric function and (2) an increase
plane. The sections were parallel to the electrode penetrations as indi-
cated by the carbon particle marks. Alternate sections were stained with in the slope of the psychometric function (Recanzone et al.,
cresyl violet or for endogenous cytochrome oxidase activity (Wong- 1992a). Representative psychometric functions from one ses-
Riley et al., 1978). The angle of the electrode penetrations and the sion within each of the four periods of auditory training shown
distances between recording locations were reconstructed with reference in Figure 2 are reproduced for monkeys OM 1 (A) and OM3 (B)
to the electrolytic lesions and carbon particle marks. AI maps represent in Figure 3. Consistent with the results from the tactile task, the
reconstructions of all recording locations that corresponded to the mid-
dle cortical layers III and IV, approximately 500-900 pm below the pial performance at each S2 frequency increased in a roughly se-
surface. These maps are presented in two dimensions by projecting the quential fashion, with the performance at large frequency dif-
location of the recording site along the radial direction to a point where ferences improving before the performance at smaller frequency
it intersected an imaginary line drawn 700 pm below the cortical surface. differences. For example, delta frequencies (AF values) above
This technique limits the distortion that results from projecting record-
ing sites to the surface of the crown of the sulcus. Results from recording 300 Hz for OM 1 had improved to near maximum performance
locations in other cortical layers are not reported here. by session 36, whereas the performance at the AF of 80 Hz did
Cortical areas of representation were measured by defining borders not begin to improve until after this session. Similar examples
as equidistant between neighboring recording locations. The approxi- could be cited for each monkey.
mate error in these area measurements at interpenetration distances The slope of the psychometric function also increased
used in these experiments is on the order of lo-20% (see Merzenich et
al., 1987; Recanzone et al., 1992b). Statistical analysis of all behavioral throughout the training period. The average slopes of the psy-
and electrophysiological data was done using the software STATVIEW chometric function at threshold over five sessions are shown

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 90 Recanzone et al. - Auditory Cortex Plasticity with Training

1000 .12

*1
OMI 8 kHz Task 0 Initial

800
.lO 1 n Mid 1
0 Mid 2

200

0
0 20 40 60
Session
400 Monkey

300 -
1 OM5 3 kHz Task Figure 2. Decrease in frequency discrimination threshold with train-
ing. The discrimination thresholds taken over five continuous sessions
are shown (mean and SE) for each monkey trained at the auditory
frequency discrimination task. Units are given in AFIF. Initial sessions
are taken as sessions l-5; Final sessions are the five sessions immedi-
ately preceding the electrophysiology experiment. Mid I and Mid 2 were
l 3 kHz Task chosen to divide the total training period into four roughly equal periods.
g 0 5 kHz Task Error bars not drawn were smaller than the symbol size. Stars indicate
statistically significant differences between the starred symbol and the
9 immediately preceding training period. OMZ-2.5 kHz is shown in two
2 200 - periods because training at the 2.5 kHz task was interrupted by training
on the 8 kHz task (see Fig. 6). Sessions shown for this monkey are l-
5, 30-34, 57-61, 88-92, and 107-l 11.
z
E performance at each S2 frequency and an increase in the slope
100 - of the psychometric function at threshold.
One way in which decreases in threshold could occur is by a
shift in the monkey’s internal criteria by which it makes a re-

I
1 I I I I Table 2. Final thresholds measured for all behaviorally trained
0 20 40 60 80 monkeys
Session Task Monkey # Sessions AFIF
Figure Progressive improvement in the frequency discrimination
1. 2.5 kHz OM2 61/24 0.005/0.010
threshold measured on daily sessions. The delta frequency with a per- 2.5 kHz OM3 97 0.013
formance of 0.50 is plotted for all sessions in which at least one S2
frequency had a performance below 0.50 (minimum 20 presentations), 3 kHz OM5 80 0.028
and the overall false-positive rate was below 15%. Solid circles denote 3 kHz OM4 6 0.045
the frequency range tested on repeated sessions (8 kHz and 3 kHz for 5 kHz OM4 75 0.038
A and B, respectively), and open circles represent sessions using an
5 kHz OM5 2 0.040
untrained frequency range (5 kHz in B).
8 kHz OM1 61 0.015
8 kHz OM2 26 0.016
for each monkey in Figure 4 for the same training periods as
8 kHz OM3 4 0.041
shown in Figure 2. These slopes increased significantly between
Passive, 8 kHz CM1 131 -
at least two of the training periods measured, and between the
Passive, 5 kHz CM2 75 -
slopes measured in the initial and final training sessions for all
monkeys except OMS, which did not reach statistical signifi- Threshold measurements reoresent the mean of the final five sessions for trained
cance (p = 0.091). Thus, consistent with observations in the tasks, or all sessions for tadks with six or less total training sessions. The two
measurements for OM2 represent the two training periods for the 2.5 kHz task
tactile frequency discrimination task, the improvement in per- that bracketed the training at the 8 kHz task. No auditory thresholds were derived
formance was marked by a roughly sequential increase in the in the passive-stimulation animals.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 The Journal of Neuroscience, January 1993, 73(l) 91

A OMI 8 kHz Task

initial
1001--o---0- Session 3
Session 36
18
Mid 1
Mid 2
Final
80 I Session 61

lb0 Id00
Delta Frequency (Hz)
Monkey
Figure 4. Slope ofthe psychometric function at threshold in all trained
monkeys. The mean and SE of the slope of the psychometric function
B at threshold are shown for the four training periods as defined in Figure
OM3 2.5 kHz Task 2 for all monkeys. Stars indicate statistically significant differences.

of 20-90%) in any of the five owl monkeys studied (p > 0.05

in all cases).Finally, the measureof d’ for at least three S2
frequencieswasanalyzed over the four training periods for each
monkey. In every monkey there was at least one statistically
significant increasein the measureof d’ betweentraining periods
for each frequency tested (and no significant decreases).The
- ---- results from the two monkeys that had the highest discrimi-
nation thresholdsare shownin Figure 5. Even in thesemonkeys,
there wasa sequentialincreasein the measureof d’ with training.
-O- Session 5 These results indicate that the improvements in performance
m- + Session 27 as reflected in the measuresof threshold and d’ reflect a true
20 * Session 58 increasein perceptual acuity with training.
+ Session 95 Transference. Previous studieshad indicated that training one
sensorysurfacein a perceptual task conferslimited gainsin that
o-1-o IO0 1006
task to nearby sensory surfaces(seeJames, 1890; Craig, 1988;
Recanzone et al., 1992a). In each monkey excluding OM 1, the
performance at one other frequency rangewas testedat various
Delta Frequency (Hz) times throughout the training period (Table 2). The most ex-
tensively studied animal wasOM2, which wastrained initially
Figure 3. Representative examples of psychometric functions taken on the 2.5 kHz task, was then shifted to the 8 kHz task, and
during the training periods (see Fig. 2). The broken line defines threshold
(50% performance). Psychometric functions shifted to smaller frequency was finally returned to the 2.5 kHz task (Fig. 6). The charac-
differences and appeared to become steeper during the course of training teristic decreasein threshold wasobserved for both the 2.5 kHz
in all studied monkeys. task (solid circles)and the 8 kHz task (open circles).The thresh-
olds measuredon the 2.5 kHz task increasedduring the course
of training at the 8 kHz task. The thresholds then decreased
sponse(seeGreen and Swets, 1966; Recanzoneet al., 1992a). again when the task was shifted back to 2.5 kHz. A second
The increasedslopeof the psychometric functions arguesagainst feature to note is that the 8 kHz task does not show a rapid
this explanation. A second line of evidence is that the false- improvement in performance at the initial period of training
positive rate was maintained below 15% and did not signifi- (compare Fig. 6 with Fig. 1, and seeFig. 2).
cantly change over the course of training. Furthermore, there If a particular frequency outside the normal training range
was no significant correlation between the false-positive rate wastested on two or three individual sessionsthat were widely
and the hit rate measuredfor any AF near threshold (hit rates spacedin time, there was no improvement in performance for

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 92 Recanzone et al. - Auditory Cortex Plasticity with Training

A 1000
1
OM2 Decrease in Threshold
5
4
1
1
OM4 5 kHz Task

303 Hz
1
d3
+ 2.5kTask
2 181 Hz --o- 8kTask

151 Hz
1

0 I r I I
110 20 40 60 80 100 120
Session
Sessions Sessions Sessions Sessions
l-5 23-27 48-52 71-75 Figure 6. Improvement in frequency discrimination threshold at two
different Sl frequencies for monkey OM2. OM2 was initially trained at
the 2.5 kHz task (solid circles), then changed to the 8 kHz task (open
B circles), and finally returned to the 2.5 kHz task. The 2.5 kHz task was
tested intermittently during the consecutive training sessions at the 8
5 kHz task and vice versa.
OM5 3 kHz Task
in frequency than the first. The psychometric functions from
4 thesethree sessionsare shownin Figure 8 (left), alongwith three
305 Hz functions usingan increasedfrequency S2 stimulus(+aF, right)
taken at a similar period of training in the samemonkey. The
3 183 Hz
122 Hz
d’ 100 OM4 3 kHz Task
2

80

1 I

SeAions Sessions Sessions Sessions

5-9 25-29 51-55 76-80
-
Figure 5. Measures of d’ for three frequencies as a function of training.
The points for a single frequency are connected by a solid line. Error
bars indicate SE; stars indicate statistically significant differences from
the preceding measure.

+ Session 27
that frequency. This is shown by the performance at 5 kHz in 20 -CF Session 49
a monkey trained at the 3 kHz task (OM5; open circles, Fig.
-C Session 71
lB), and by the psychometric functions derived at 3 kHz for a
monkey trained at the 5 kHz task (OM4; Fig. 7). In every ex-
ample, even though there was a significant improvement be-
tween thesesessionsfor the trained frequencies,there waslittle 0 I 1
differencein the performanceat frequenciesthat werenot trained. 30 100 1000
Essentially all training wasconducted with the S2 frequencies
above the S1 frequency (+ AQ. A somewhatdifferent effect was Delta Frequency (Hz)
observedby occasionallytestingthe performancewith decreased Figure 7. Psychometric functions at three 3 kHz task testing sessions
S2 frequencies(-AFs). On three sessions,OM4 waspresented for the monkey trained on the 5 kHz task (OM4). The broken line
with S2 stimuli in which the secondtone of the pair was lower represents performance of 50% (threshold).

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 The Journal of Neuroscience, January 1993, 73(l) 93

OM 4 5 kHz Task
Performance
100

80

60

Session 35 -0- -U Session 25

Session 59 -O- + Session 52
Session 70 -M- t Session 72
Figure 8. Representative psychomet-
ric functionsfor +AF (right) and -AF
(left) tasksfor the 5 kHz standardfor
1 monkevOM4. Thev-axisintercentsthe
I
-I+ 1000 delta frequencyof -+50 Hz. N&e the
-1000 -300 -100 50 100 300
increasein performancefor the +ti
task and the decrease of performance
Delta Frequency (Hz) for the -AF taskwith training.

functions for the +AF values show the improvement in per- and CM2, and in three behaviorally naive owl monkeys (Nl,
formance as seen in other monkeys. The functions for -AF N2, and N3). Neural responsesin laminae III and IV of AI were
values, however, show a worsening of performance and an in- marked by short-latency (approximately 1O-l 2 msec)responses
crease in threshold as a function of session. The worst perfor- to a relatively narrow band of frequencies.In general, the low
mance at the -AF task was seen on session 70, in which there
were 10 intervening sessions using the increasing delta frequen- Effects of Intensity on Threshold
cies. This greatest decrease in performance coincides with an
increase in the performance for the +AF stimuli. The same test 200 f
on the 3 kHz task was done in monkey OM5 near the end of
the training period, and the threshold for the -AF’ task (AFIF
= 0.035) wasgreater than that observedfor the +aFtask (U/F
= 0.028).
=
ilOO
1 1
Efict of stimulus intensity. The effects of transference for
other frequencies raise the question of how the performance
would be affected for other intensitiessincethe training occurred
within a limited intensity range of only lo-12 dB, To address
this question, psychometric functions were derived at different
intensities in monkeys OM4 and OM5 near the end of the ii! 50 - -O- OM4-5kHz
training period. There was a slight decreasein frequency dis-
g -O- OM5-3kHz
crimination thresholds with increasesin intensity (Fig. 9) but
these two parameters were not correlated (p > 0.15 in both
cases).

Electrophysiology 30
Functional of AZ. Basic aspectsof the functional
organization Intensir Level “YdB SP:
organization of AI were determined for both the left and right Figure 9. Thresholdmeasured at the trainedfrequencies
asa function
hemispheresin behaviorally trained monkey OMl and in one of stimuluslevelfor two monkeys,OM4 (solid circles) andOM5 (open
hemisphere of monkeys OM3 and OM4. Sufficient electro- circles). All dataweretakenfromsinglesessions.Theintensityrepresents
themeanintensitymeasured in thesoundfieldoccupiedby themonkey’s
physiological data were not obtained from trained monkeys head(seeMaterialsandMethods).Stimuluslevelvariedindependently
OM2 and OM5 and will not be discussedfurther. AI was also for eachtonepip over a rangeof approximately15-20%of the mean
defined in the two passively stimulated control monkeys CM1 valueplotted.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 94 Recanzone et al. - Auditory Cortex Plasticity with Training
frequencies were represented on the crown of the gyrus and the
higher frequencies were represented down the lateral bank of
the sulcus. Neural responses were clustered into “isofrequency
bands” where all neurons had similar CFs. These bands were
usually oriented oblique to the sulcus but could also be roughly
parallel to it. Within the depths of the sulcus, the neural re-
sponses would either become very broadly tuned with a higher
threshold and longer latency, or become unresponsive to acous-
tic stimulation. These regions corresponded with the cytoar-
chitectonic borders of AI as seen histologically, and were char-
acteristic of other auditory cortical fields in this primate species
(see Imig et al., 1977; Merzenich et al., 1991).
A central finding of these studies is that the number of re-
cording locations and the cortical area of representation for the
frequencies used in the behavioral task were both increased
when compared to those frequencies not used in the training.
The reconstructed frequency representations from four owl
monkeys are shown in Figure 10. Recording locations at which
the CF was within the range of frequencies presented in the
behavioral tasks are indicated by frequency-specific shading.
The total cortical area representing these frequency ranges var-
ied from animal to animal. Some but not all of the differences
appeared to be related to the training task. A given frequency
was usually represented in a relatively restricted region, but not
necessarily continuously. For example, the representation of the
frequencies used in the 2.5 kHz task occupied several patches
in the trained monkey OM3. By contrast, only a few locations
represented these specific frequencies in monkeys not trained
at that frequency. Similarly, the greatest representation of the
frequencies used in the 5 kHz task were seen in OM4, even
though there was only a limited improvement in performance
and a relatively small cortical area of representation (Fig. 1OD).
This effect was not seen in the passively stimulated monkey
CM2, which received the same acoustic stimulation as OM4,
where only a single cortical location represented the frequencies
used in the 5 kHz task at CF (Fig. 1Oc). No more than two
locations with a CF of 5 kHz were encountered in any of the
other monkeys.
The tuning curve data are summarized for all monkeys by
plotting the number of cortical locations having a CF within a
500 Hz range (Fig. 11). In Figure 11, the bin size was adjusted
to include all the frequencies used in the behavioral tasks (in-
dicated by arrows). In normal owl monkeys, even though all
frequencies are represented, there is a consistent underrepre-
sentation of frequencies between 3 and 8 kHz (Fig. 11A) and
most monkeys showed a relatively large number of recording
locations with CFs above 15 kHz. These monkeys also had a
large number of locations with a CF near 2 kHz, but those CFs
were not in the range used in the 2.5 kHz behavioral task. Each
individual monkey showed idiosyncratic increases of particular
frequencies, for example, approximately 10 kHz for monkey
N2, or approximately 2 kHz for monkey N3.
By contrast, the numbers of locations with CFs in the range
of frequencies used in the behavioral task were greatest in mon-
keys trained at that frequency range. In the trained monkey, the
number of locations near 8 kHz was approximately 3-l 0 times
greater in both hemispheres compared to normal and control
monkeys (Fig. 1 IB). Monkey CMl, which received the same
acoustic stimuli as OM 1 but was not required to attend to these
stimuli, had an essentially normal representation of these fre-
quencies. The frequency range used in the 2.5 kHz task was
represented by more locations in the monkey trained at that
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
task than in any other monkey. The same was true for the
frequency range used in the 5 kHz task (Fig. 1lc).
The larger number of cortical locations that represented the
trained frequencies could not be explained by an increase in the
sampling density over that region, as the representation of these
frequencies and surrounding areas was defined at equivalent
density in all monkeys. This increase was better explained by
an increase in the cortical area of representation of those fre-
quencies. Figure 12 shows the cortical areas associated with each
of the three different frequency ranges for all monkeys. In every
case, the area of representation for the behaviorally relevant
frequencies for a trained monkey (solid bars) was significantly
greater than the cortical areas for those frequencies in all other
monkeys, including those that were passively stimulated (hatched
bars).
Response properties at individual locations. The tuning prop-
erties of the neurons at each recording location were investigated
by recording FRAs at all AI recording locations of monkeys
OMl, OM3, CMl, and in each of the three normal monkeys.
Measures of QlOdB were made at each location; examples are
plotted in Figure 13. In every monkey, the values for QlOdB
tended to increase with increasing frequency, consistent with
reports from other species (see Pelleg-Toiba and Wollberg, 1989;
Schreiner and Mendelson, 1990). There was a considerable range
of QlOdB values for any given frequency in all monkeys. The
Q 1OdB measures for neurons in trained monkeys were greater
(tuning bandwidths were narrower) for the behaviorally trained
frequencies when compared to normal or passive stimulation
control monkeys. The mean Q 1OdB and Q40dB for all monkeys
over the frequency range used in the 2.5 or 8 kHz tasks are
shown in Figure 14A. The left and right hemispheres of OMl
are considered independently, and the data from the three nor-
mal monkeys are pooled. Measures of Q 1OdB were statistically
significantly greater over the behaviorally trained frequencies
in trained monkeys when compared to the three normal con-
trols. The QlOdB of the passive control animals was also ele-
vated over that of normal control monkeys. Statistically signif-
icant increases in Q40dB were noted only in the right hemisphere
of OMl for the 8 kHz frequencies.
In addition to the increase in QlOdB, there was also an in-
crease in the minimum latency of the response for the behav-
iorally trained hemispheres in each of the three cases (Fig. 14B).
Response latencies in the passive stimulation monkey were not
significantly different from those in control monkeys. The vari-
ance of the latencies was very low in all monkeys, as indicated
by the small SE bars. The distribution of the variance in the
latency measure was not statistically significantly different be-
tween the normal monkeys and any others (p > 0.1).
Correlation of behavioral performance with neural responses.
The main objective of these studies was to determine (1) if the
functional organization of the primary auditory cortex was al-
tered consequent to training the animal in an acoustic discrim-
ination task, and (2) if the resulting cortical representations were
correlated with the behavioral performance. The three measures
showing significant changes in trained monkeys described above,
the increased cortical area, increased QlOdB (narrower band-
widths near threshold), and increased latency, were subjected
to regression analysis. For these analyses, behavioral perfor-
mance was taken as the threshold measured at the end of the
training sessions for the trained monkeys, and paired with the
electrophysiological measures derived from those same mon-
keys. To estimate the behavioral performance of “untrained”
 The Journal of Neuroscience, January 1993, 13(l) 95

Normal Owl Monkey CM2 Passive Stimulation at 5 kHz

l-2 kHz

c 2-4KHZ
, 4-a kHz

: e 6-16kHz

16-32kHz

kHz

l :: :$a
+-;
+‘-
m 16-32kHz16

B OM3 Trained on 2.5kHz Task

O-1 kHz l
OM4 Trained on 5 kHz Task

24 kHz \

’ 16-32 kHz

.
. .
.
.

2.5k frequencies
5k frequencies
6m 8k frequencies

Figure 10. Cortical representations of CF in AI of four adult owl monkeys. Thin lines define boundaries of cortical locations with CFs within one
octave. Stippled regions encircle cortical locations where neurons were recorded with CFs in the frequency range used in the 2.5 kHz task, solid
regions represent frequencies used in the 5 kHz task, and hatched regions represent the frequency range used in the 8 kHz task. The cortical areas
representing a given frequency range were approximated by connecting the 50% distance values to the neighboring recording sites with CFs outside
the given frequency range. P/uses denote recording sites with neuronal responses not consistent with properties of AI neurons. A is from a
representative normal owl monkey (N2), B is from a monkey trained at 2.5 kHz (OM3), C shows the monkey passively stimulated with the
frequencies used in the 5 kHz task (CM2), and D shows the representation of the monkey trained at the 5 kHz task (OM4).

monkeys, the meanthreshold of training sessions5-l 0 for mdn-
keys OM3 and OM 1 were usedfor the 2.5 kHz and 8 kHz tasks,
respectively. Thesesessionswere usedbecausethey showedless
variability than the initial five sessions
(1-5) and probably better
reflect the monkey’s true initial discrimination abilities (seeDis-
cussion).For the “untrained” behavioral performance of the 3
kHz and 5 kHz tasks, the mean thresholdsfor the two and six
sessionsat thesetasks for monkeys OM4 and OMS were used,
respectively. These“untrained” behavioral measureswere con-
sidered with the electrophysiological results pooled from the
untrained monkeys. The values of QlOdB, Q40dB, and mini-
mum latency showed no significant correlation with these be-

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 96 Recanzone et al. - Auditory Cortex Plasticity with Training

A 30:

n Nl
. N2
. N3

Characteristic Frequency (kHz)

w OMl Rightt 8kHz Trained

l OMl Left 8kHz Trained
* CM1 8kHz Passive

Characteristic Frequency (kHz)

C

v) 25 :
6 l OM3 2.5kHz Trained
‘fij 20 - n OM4 5kHz Trained
;: 15 : A CM2 5kHz Passive

-2 10 -
6
#!!fJ :
Figure II. The number of cortical lo-
cations plotted by CF in 500 Hz bins. 0 -
Bin size was adjusted to incorporate all
frequencies used in the behavioral task
into a single bin (indicated by arrows).
The three normal monkeys are shown
in A, the two hemispheres from monkey
OM 1 and the passive-stimulation con-
trol monkey CM1 are shown in B, and
the monkeys trained at the 5 kHz and
2.5 kHz task (OM4 and OM3) and the
passively stimulated monkey CM2 are
shown in C. Characteristic Frequency (kHz)

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 The Journal of Neuroscience, January 1993, 13(l) 97

u 20- Normal
2.5 kHz Task $ 18-
2
g 16-
g 14-
4 12-
m
-0 lo-
0
- 8

.l 1 10 100
OM3 OWL OMlR OM4 CM2 CM1 Ni N2 N3 Mean
2.5k 8k Sk 5k Untrained Characteristic Frequency (kHz)

-cl 20
2 18 3 Trained + ‘:
16
8 kHz
Task
14
12
Oh43 OWL OMlR OM4 CM2 CM1 Nl N2 N3 MeWl IO
2.5k 8k Bk 5k untrained
8
6
8 kHz Task 4
2
0
.l 1 10 100
Characteristic Frequency (kHz)

- X
OM3 OMlL OMlR OM4 CM2 CM1 Nl N2 N3 Mean
2.5k Bk Bk 5k untrained Passive .
Monkey Stimulation 0
Control a
Figure 12. Cortical area of representation of the frequency ranges used
in the behavioral study. Solid bars indicate the area of representation
of the trained frequency for that particular monkey; hatched bars rep-
resent areas of the frequencies presented to the passive-stimulation con-
trol monkeys. Crosses represent hemispheres where no cortical locations
were recorded with a CF within that frequency range. The column
“Mean Untrained” represents the mean and SD for all hemispheres
excluding the trained hemispheres.

havioral measures(JJ= 0.44, 0.21, and 0.91, respectively). By

contrast, the correlation analysis of behavioral threshold with
cortical area revealed that these two parameters were signifi-
cantly correlated (Fig. 15; Y= 0.88; p < 0.01; df = 6).
Discussion
Improvements in behavioral performance in an auditory fre-
quency discrimination task wererecordedin five adult owl mon-
keys. There was a roughly sequentialimprovement in perfor-
mance at individual S2 frequencies, which was expressed
shift of the psychometric function and an increasein the slope
of the function near threshold. The improvement in perfor-
mance was much lessfor untrained frequency ranges,and the
performance wasindependent of stimuluslevel between 25 and
65 dB SPL. Electrophysiologicalmapping experiments revealed
-.i i ld 106
Characteristic Frequency (kHz)
Figure 13. Values of Q 1OdB as a function of CF. Normal owl monkeys
N 1, N2, and N3 are pooled and shown in the top panel, both hemi-
spheres of trained monkey OMl in the middle panel, and the passive
stimulation control monkey CM 1 in the bottom panel. Each point rep-
resents the CF and Q 1OdB at a single recording location.
in a
a largercortical areaof representationof the behaviorally trained
frequencies, sharper tuning as measuredby Q 1OdB but not
Q40dB, and longer responselatenciesover the frequency range
used in the behavioral task. The cortical area of CF represen-
tation was the only measuredparameter that was significantly

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 98 Recanzone et al. * Auditory Cortex Plasticity with Training

A 0.70
Sharpness of Tuning
10 0.60 . 0

1
Trained-Left Hemisphere m
Trained-Right Hemisphere
Normal Owl Monkey Control c]
Passive-Stimulation Control m ; 0.40 y = -12.591x + 0.641
ps 0.05 *
: 0.30
2 6- -F!
F 0.20
2
0 - s
0.10
4-
0
0.01 0.02 0.03 0.04 0.05
Behavioral Threshold (AF / F)
Figure 15. Correlation analysis of the cortical area of representation
ofthe behaviorally trained frequencies at CF with the behavioral thresh-

2.5k 8k
h2.5k 8k
old. The equation describes the line with the root meansquare
to the data (r = 0.882, P < 0.01, df = 6).
bestfit

correlated with behavioral frequency discrimination perfor-

QlO dB Q40 dB mance.
Behavioral thresholdsmeasuredin this study were similar to
those described for other nonhuman primates for the 2.5 kHz
B Minimum Latency at CF and 8 kHz tasks(Sinnott et al., 1985, 1987; Prosenet al., 1990)
137 but were higher than those recorded in humans (Wier et al.,
1977; Sinnott et al., 1985, 1987). The frequency discrimination
12 - thresholdsmeasuredfor the 3 kHz and 5 kHz taskswere greater
ns than those measured for this frequency range in Old-World
11 - monkeys. That may be the result of an attenuation of these
middle frequencies in this particular species,for example, by
physical constraints imposedby the outer or middle ear. These
10 - higher-frequency discrimination thresholds are consistentwith
the relatively high behaviorally measureddetection thresholds
g 9- in the frequency rangeof 3-6 kHz in this species(Beecher, 1974),
E the relatively smallcortical representationof this frequency range
a- in AI of normal owl monkeys (Imig et al., 1977; presentresults),
and the higher responsethresholds of AI neurons over this
7- frequency range in this and other New-World monkey species
(Aitkin et al., 1986; Merzenich et al., 1991).
6- The electrophysiologicalexperiment revealed a tonotopic or-
ganization of owl monkey AI similar to that previously shown
5- - - in this species(Imig et al., 1977) as well as other primates
(Merzenich and Brugge, 1972; Aitken et al., 1986). The re-
2.5k 8k sponsesof neuronal clusters to these tonal stimuli were com-
Figure 14. Comparison of sharpness of tuning (A) and minimum la- monly very brisk, with an initial short-latency burst of activity.
tency (B) for different training conditions. Data from each experimental In spite of the globally smooth tonotopic organization in AI,
and passive-stimulation hemisphere are shown as a single bar, with the the representationof very smallfrequency rangeswasoften seen
data from the three normal monkeys combined (open bars). Stars in-
dicate statistically significant differences from the normal hemispheres. to be patchy for both the experimental and control hemispheres.
Data are pooled for all recording locations with CFs between ! .9-3.1 Part of the explanation is that we were basing the area mea-
kHz (2.5 kHz task) and 6.6-9.9 kHz (8 kHz task). A, Mean and SE of surementson the CF at threshold, and considered only a very
Q 1OdB (left) and Q40dB (right) for all studied monkeys. B, Minimum small frequency range.It is likely that this patchy representation
latency of the response for the 2.5 kHz (left) and 8 kHz tasks (right). would be more continuous if consideredfrom the view of either
a higher-intensity stimulus or over a larger frequency range.
However, the presenceof patches of representation of some
frequenciesmay accurately reflect the tonotopic organization at

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159


this level of detail, as a similarly discontinuous representation
of restricted frequency ranges can be seen in the detailed maps
of AI described in the marmoset (Aitkin et al., 1986).
Neural mechanisms of frequency discrimination. The finding
that the cortical area of CF representation was correlated with
the behavioral performance does not restrict the site of fre-
quency discrimination to that location. Most nuclei throughout
the central auditory pathways could contribute to these percep-
tual distinctions. Studies have indicated that large lesions that
include AI do not effect learned frequency discriminations in
cats (Butler et al., 1957; Goldberg and Neff, 196 l), but deficits
have been shown in other behavioral paradigms (Meyer and
Woolsey, 1952; Thompson, 1960). These studies suggest that
the sequence and timing of stimuli, or perhaps the difficulty of
the task, affect the necessary contribution of auditory cortical
fields regarding this behavior. The behavioral task used in these
experiments cannot be directly compared to those cited above,
yet were considered by human observers to be quite difficult
and to require vigilant attention in order to perform correctly.
An alternative view is that the pairs of tone pips presented
sequentially in the present study more closely match auditory
pattern discrimination paradigms, which do show deficits fol-
lowing auditory cortical lesions (Diamond and Neff, 1957). A
third point to bear in mind is that the limited data available
suggest that thalamic areas with frequency-specific responses
(ventral medial geniculate nucleus) are less affected by classical
conditioning paradigms than the cerebral cortex, although the
responses of the more broadly tuned neurons located in the
magnocellular division of the medial geniculate are affected by
classical conditioning paradigms (see Weinberger et al., 1990,
for review). The cortical representations described in this report
may be a reflection of tonotopically reorganized subcortical
regions, partially reorganized subcortical and cortical areas, or
resulting from reorganization only at the cortical level.
Regardless of the fact that we cannot identify the site(s) of
reorganization, the differences in the response properties of AI
neurons between trained and untrained monkeys do provide
some insights into plausible neural mechanisms of frequency
discrimination. Frequency discrimination abilities could poten-
tially be based on either the rate of neuronal discharge, the
temporal information encoded within the discharge, and/or the
spatial location/distribution of the neurons responding to the
different frequencies. Our finding of an increased spatial rep-
resentation of behaviorally relevant frequencies suggests that
the spatial distribution of central neurons at least contributes
to this behavior. One simple hypothesis is that the separation
in the spatial locations of appropriately tuned neurons could
provide the necessary information to form the basis of the dis-
crimination. Increases in the representation of a frequency range
would increase the spatial resolution of those frequencies. The
fact that the bandwidth of the tuned response was not signifi-
cantly different at 40 dB above threshold suggests that this larger
representation was maintained at higher intensities, for exam-
ple, those used in the behavioral paradigm. However, the cor-
tical area measure at CF described here is almost certainly an
oversimplification. By combining the several islands of cortical
representation into a single measure, we may have diluted the
contributing portion. Studies in the cat have shown that several
different parameters of both tuning and timing of the neural
response are topographically organized along the isofrequency
axis (Mendelson et al., 1988; Schreiner et al., 1988; Schreiner
and Mendelson, 1990; Imig et al., 1991; Sutter and Schreiner,
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
The Journal of Neuroscience, January 1993, 13(l) 99
199 l), and preliminary data suggest that this is also the case in
the owl monkey (Merzenich et al., 1991). It is plausible that
only a subset of the total area of representation is contributing
to the processing of frequency distinctions, while other parts of
the representation are processing other stimulus parameters.
Finally, the neural responses were not tested with respect to the
their temporal representation of the stimuli. Studies in somato-
sensory cortex have shown that the occurrences of neural dis-
charges are temporally more coherent in monkeys trained at a
tactile frequency discrimination task when compared to unstim-
ulated and passively stimulated monkeys (Recanzone et al.,
1992d). The temporal coding of the stimuli could provide ad-
ditional information regarding frequency.
The finding that Q 1OdB did increase significantly was unex-
pected, as the behavioral stimuli were not presented at these
low intensities. The disparity between Q 1OdB and Q40dB sug-
gests that these two parameters are not equivalent, consistent
with observations in cat (Schreiner and Mendelson, 1990; Sutter
and Schreiner, 199 1). QlOdB measures have been reported to
covary with latency, and both were observed to increase in the
trained monkeys. The longer latencies in the trained monkeys
may have resulted from increased neural processing at subcor-
tical levels, or perhaps from a suppression of normally shorter-
latency inputs that are involved in processing other aspects of
the acoustic signal. These two parameters were not correlated
with the behavioral performance, as the higher Q 1OdB (narrow-
er bandwidths) and latency measures were not matched with
the lowest thresholds.
Finally, both hemispheres were investigated in only one mon-
key. The hemisphere studied in each of the other trained mon-
keys was selected as the hemisphere contralateral to the ear
nearest the speaker in the behavioral apparatus. In most cases,
the monkey oriented its head either directly toward the speaker,
in which case the ears were at an equivalent distance, or toward
the food hopper, which would slightly favor one ear over the
other. Both hemispheres of OM 1 had similar areas of represen-
tations; however, there were significant differences between the
right and left hemispheres with respect to both QlOdB and
Q40dB. It is unclear at present how these binaural stimuli are
processed by the two hemispheres.
Comparisons with earlier studies. The caveats discussed above
do not detract from the findings that there is a significant im-
provement in behavioral performance at the frequency discrim-
ination task with training that was significantly correlated with
the spatial representation of the trained frequencies in primary
auditory cortex. Improvements in performance at a perceptual
or motor skill have been well documented for a variety of tasks
(see Volkmann, 1858; James, 1890; Gibson, 1953; Fitts, 1964)
including relatively simple sensory discrimination tasks in adult
primates (Sinnott et al., 1985; Prosen et al., 1990; Recanzone
et al., 1992a), and have been shown to continue for up to several
years of continuous practice. The initial, rapid component of
the improvement probably represents a “conceptual” learning
of the task and development of the most appropriate strategies
(see deJong, 1957; Crossman, 1959; Seibel, 1963; Recanzone et
al., 1992a). This is supported by the large variability in the
performance between these early sessions, and by the fact that
it was not seen when the task was changed to different frequen-
cies in well-trained monkeys (e.g., 2.5 kHz to 8 kHz in OM2).
The second stage of learning was characterized by an im-
provement in performance at each S2 frequency in a roughly
sequential fashion, and an increase in the slope of the psycho-
 100 Recanzone et al. * Auditory Cortex Plasticity with
metric function at threshold. This improvement has been hy-
pothesized to result from an enhancement of the central rep-
resentation of the appropriate stimulus parameters (Recanzone
et al., 1992a,d). The psychophysical results derived in these
monkeys are in agreement with this hypothesis. For example,
there was little transference of the improvement in performance
when untrained frequency ranges were tested in widely separated
sessions, consistent with studies on absolute auditory detection
thresholds (Zwislocki et al., 1958) or tactile frequency discrim-
ination thresholds (Recanzone et al., 1992a). Improvements were
only recorded when training was continuous for several days.
The representations of these untrained frequencies were com-
parable to those of normal monkeys. An interesting observation
was the apparent decrease in performance for decreasing S2
frequencies (-AF values) with training at increasing S2 fre-
quencies (+AF values). The thresholds for -AF values have
been reported to be slightly lower in humans but vary depending
on the species for Old-World monkeys (Sinnott et al., 1987). In
those experiments, the behavioral performances were reported
following training at both increasing and decreasing AF tasks.
In OM4, where several measuresof -AF threshold were made,
there were no cortical locations observed where the CF was
centeredin the rangeof thesedecreasingS2 frequencies.Finally,
our psychophysicalresultssuggestthat there is little, if any, effect
of stimulus level on discrimination performance. There was a
slight decreasein threshold with increasingstimulus level, but
this difference wasnot statistically significant. Increasinginten-
sity can increasefrequency discrimination ability in humans,
but decreasesperformance in macaquemonkeys (Sinnott et al.,
1987). Our results suggesteither that frequency discrimination
performance is normally independent of stimulus level in this
species,or that the performance at the trained stimulus level
improved to match that for lower stimulus levels. The effects
of stimulus amplitude on frequency discrimination in a naive
animal of this specieswould provide further information on this
issue.
The electrophysiological resultsare consistentwith a variety
of preparations in the somatosensorysystem (for reviews, see
Merzenich et al., 1988, 1990; Kaas, 1991) and with the hy-
pothesisthat the functional organization of the cortex changes
in parallel with perceptual changes(seeMerzenich et al., 1988,
1990; Recanzone et al., 1991b,d; Recanzone and Merzenich,
1992). The finding that the large cortical representation of the
behaviorally trained receptor surfacewasonly seenin monkeys
trained to discriminate those frequencies,and not in monkeys
trained at an unrelated task, is consistentwith resultsfrom the
somatosensorysystem (Jenkins et al., 1990; Recanzone et al.,
1992b-d). This representationalplasticity occurredeven though
the monkeys in this study were acoustically stimulated for ap-
proximately 15-20 min during the 2 hr sessioneach day, and
were allowed to hear other auditory stimuli, someof which were
presumably behaviorally significant, for the remaining 22 hr
each day. Thus, it is not necessaryto present the behaviorally
important stimuli to the exclusion of all other stimuli for these
effects to occur, and consistent with reports on the effects of
classicalconditioning on responsesof auditory cortical neurons,
behavioral relevance of the stimulus is necessary.
The demonstration that there is a larger CF representationin
trained monkeys is not altogether unexpected given the fre-
quency-specific changesin neural responsesof other auditory
cortical areasfollowing classicalconditioning paradigmsin the
cat (Diamond and Weinberger, 1986)and guineapig (Bakin and
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Training
Weinberger, 1990) and changesin the tonotopic representation
of AI after restricted cortical lesionsin the adult guinea pig and
cat (Robertson and Irvine, 1989; Irvine et al., 1991). In contrast
to the conditioning studies,there wasno obvious changein the
overall responsivenessof neurons representingthe frequencies
usedin behavioral training, and there wasno significant change
in the tuning of theseneuronsat intensitieswell above threshold.
Similarly, severalmonths after restricted lesionsin the cochlea,
the overall activity of neurons in the expanded zone of CF
representation in AI is similar to neurons in the adjacent, un-
affected region. These differencesmay reflect differencesin the
comparison between short-term effects as in the conditioning
studiesand the long-term effectsasin this and the lesionstudies,
in the anestheticstateof the animal, or in the differencesbetween
the conditioning and discrimination behavioral paradigms.
Several models advanced to account for cortical plasticity
have proposedthat synaptic efficaciesare altered consequentto
the temporal correlation of pre- and postsynaptic activity by a
Hebbian mechanism(e.g., seeEdelman, 1987; Merzenich et al.,
1988, 1990; von der Malsburg and Singer, 1988; Palm, 1990;
Singer, 1990; Weinberger et al., 1990). Thesemechanisms,al-
though presentedwith respect to the cerebral cortex, could be
equally effective for representationalchangesin subcortical ar-
eas. Computer models using these types of synaptic efficacy
changesresult in changesin topography similar to those ob-
served in the somatosensory(Pearsonet al., 1987; Grajski and
Merzenich, 1990) and visual cortices (Miller et al., 1989). The
present results are consistent with models of this class,as the
increasedarea of representationand the increasedsharpnessof
tuning could result from the strengtheningof the synapsesrep-
resentingthe behaviorally trained frequencies,and a weakening
of synapsesrepresenting other, neighboring frequencies. The
unaffected representationsin the passive stimulation monkeys
suggestthat attention to the stimulus enhancesthe response,
perhaps via neuromodulators such as ACh or norepinephrine
(seeSinger, 1990; Weinberger et al., 1990).
The inverse relationship between receptive field size and the
cortical area of representation describedin the visual and so-
matosensorysystems(Hubel and Wiesel, 1974; Sur et al., 1980)
is thought to ariseby competitive interactions of excitatory and
inhibitory inputs between cortical neurons comprising a hori-
zontally oriented network (seeEdelman, 1987; Merzenich et al.,
1990). Theselocal interactions between cortical neuronswould
also restrict the improvement to the behaviorally trained and
immediately surrounding frequencies,and thus only limited ef-
fects on either the cortical representationor behavioral perfor-
mance for other frequenciesare expected. The decreasingper-
formance at -AF frequenciesobserved in this study suggests
that these horizontal influencesdegrade the representation of
unused neighboring frequencies, perhaps by “borrowing” ter-
ritory from the representation of those frequencies.The local
interactions would also be expected to sharpen the temporal
cohesivenessof the responsesover a wider area, as has been
demonstrated in the somatosensorycortex (Recanzone et al.,
1992d), which would result in a stronger representationof the
stimulus in both spaceand time that could then be transmitted
to other areasfor further processing.
Technical considerations. Several technical considerations
must be rememberedwhen attempting to correlate data derived
by presentingsingletone bursts in a closedsound systemin an
anesthetizedmonkey with behavioral measuresperformed using
sequential tone bursts in the free field. The barbiturate anes-

thesia may have masked differential neuronal responses to be-
haviorally relevant stimuli. The effects of barbiturate anesthesia
on tuning properties of cells in AI of the monkey are not well
described; however, comparison of our data with that of others
derived in awake animals (i.e., Funkenstein and Winter, 1973;
Pelleg-Toiba and Wollberg, 1989) suggests little effect on mea-
sures of CF or temporal response properties of cortical neurons.
In this study, all of the comparisons of the neural responses
were made between monkeys anesthetized in a similar manner;
thus, all anesthetic effects are constant and do not differentially
affect one set of monkeys.
A second consideration is that the stimuli used in generating
the cortical maps were shorter in duration (50 msec vs. 150
msec) and were presented through a closed sound system. The
acoustic environment in the free field and the resonance prop-
erties of the external ear may have distorted the stimuli in the
behavioral condition. The definition of response properties of
AI neurons using a closed sound system has the advantage of
better control of the stimulus intensity within and between in-
dividual experiments. Direct comparisons can also be made
between these results and those of others. If the electrophysi-
ological experiments were performed in the free field, idiosyn-
cratic distortions of the stimulus could occur between individual
monkeys due to variations in head size and location and pinna
morphology. To match the acoustic environment of the behav-
ioral apparatus in the electrophysiology experiment, it would
have been necessary to perform both in the same apparatus,
which was not possible. Finally, the technique we have used is
appropriate to achieve the primary goal of the study, namely,
to define alterations in the tonotopic organization of AI con-
sequent to the behavioral training.
A third potential problem is in correlating the pooled data
from normal monkeys with the behavioral results from other
monkeys either early in training or when tested on untrained
frequencies. This method increased the number of data points
such that reasonable correlation analysis between physiological
and behavioral data could be performed, but is based on the
assumption that the physiological and behavioral measures are
representative for all monkeys. The estimates of “untrained”
behavioral performance are probably reasonable because the
thresholds for sessions 5-10 in monkey OMl were within 15%
of the initial five sessions for OM2 at the 8 kHz task, as were
the measures for sessions 5-l 0 for monkeys OM2 and OM3 in
the 2.5 kHz task. These differences were not significantly dif-
ferent, and all of our data for both auditory and tactile frequency
discrimination measures using this paradigm show low inter-
individual variability (see Recanzone et al., 199 1, 1992a). There
was also some variability in the cortical representations between
normal monkeys, as has been seen previously in AI (Merzenich
and Brugge, 1973; Imig et al., 1977; Aitken et al., 1986; see also
Merzenich et al., 1975) consistent with interanimal variability
described in primary somatosensory cortex (Merzenich et al.,
1987). This variability was most evident in the large represen-
tation of frequencies in the range of 10 kHz in monkey N2 and
for frequencies near 11 kHz in monkey CM 1. This interanimal
variability of frequency representations was reduced when con-
sidered from the point of view of the expanded representations
in the trained monkeys, where cortical representations were
invariably several times greater than the mean representations
from untrained hemispheres. These idiosyncrasies of primate
AI organization prompted us to utilize the strategy of using
different comparison frequencies for each of the limited number
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
The Journal of Neuroscience, January 1993, 73(l) 101
of owl monkeys trained in the behavioral task. Obviously, if
one had trained monkey N2 on a 10 kHz task, they may have
been misled by the preexisting large representation of those
frequencies. By the same logic, the probability of matching the
frequency in the behavioral paradigm with the idiosyncratically
overrepresented frequency in each of the studied individual
monkeys is remote.
One potential inadequacy of the data is that frequency re-
sponse areas, and hence the measures of QlOdB and Q40dB,
were based on the single presentation of each of the 675 stimuli.
Spontaneous activity occurring in the 50 msec time window of
data collection could artificially broaden the response area, and
only stimuli that evoke a response on several repeated trials
might be consistently considered within the response area. How-
ever, this method has been shown to be a reliable measure of
multiple-unit responses. The objective criteria used to define
QlOdB and Q40dB take into account the rate of spontaneous
activity as well as the consistency of the response latency, and
have been argued to accurately reflect the tuning and timing
properties of the neurons under study (see Schreiner and Men-
delson, 1990; Sutter and Schreiner, 199 1). The minimum la-
tency estimate was based on all 15 levels at three frequencies,
and therefore represents 45 separate measurements.
Concluding remarks. The demonstration that the extent of
the cortical area of representation of a specific frequency range
is correlated with the behaviorally measured discrimination per-
formance represents a first approximation for determining the
neural correlate of this perceptual judgement. The results of
these experiments are consistent with hypotheses that auditory
cortical representations are modifiable and that this plasticity
reflects the acquisition of skills and behaviors throughout life.
A key prediction from these hypotheses is that the changes in
cortical representation should occur in parallel with the behav-
ioral gains, and repeated measures of these representations should
correlate with the repeated measures ofbehavioral performance.
A second prediction is that inactivation of restricted areas of
the central representation of these stimuli should result in re-
stricted behavioral deficits. Finally, the effects of interrupting
the normal release of putative neuromodulators during the pe-
riod of acquisition of this behavioral improvement could begin
to reveal the pharmacology of this representational plasticity as
has been done in other systems (i.e, Kasamatsu and Pettigrew,
1979; Bear and Singer, 1986). Such studies would provide key
evidence in extending these hypotheses and models of cortical
function.
References
Aitkin LM, Merzenich MM, Irvine DRF, Clarey JC, Nelson JE (1986)
Frequency representation in auditory cortex of the common mar-
moset (Cullithrixjucchus jucchus). J Comp Neurol 252: 175-185.
Ashe JH, McKenna TS, Weinberger NM (1989) Cholinergic modu-
lation of frequency receptive fields in auditory cortex. II. Frequency-
specific effects of anticholinesterases provide evidence for a modu-
latory action of endogenous ACh. Synapse 444-54.
Bakin JS, Weinberger NM (1990) Classical conditioning induces CS-
specific receptive field plasticity in the auditory cortex of the guinea
pig. Brain Res 536:271-286.
Bear MF, Singer W (1986) Modulation of visual cortical plasticity by
acetylcholine and noradrenaline. Nature 320: 172-l 76. ^ . _
Beecher MD (1974) Hearing in the owl monkev Lotus trivirmtusl I.
Auditory sensitivity. J Camp Physiol Psycho1’86:898-901. ” ’
Butler RA, Diamond IT, Neff WD (1957) Role of auditory cortex in
discrimination of changes in frequency. J Neurophysiol20: 108-120.
Craig JC (1988) The role of experience in tactual pattern perception:
a preliminary report. Int J Rehabil Res 11:167-l 7 1.
 102 Recanzone et al. * Auditory Cortex Plasticity with Training
Crossman ERFW (1959) A theory of the acquisition of speed skill.
Ergonomics 2: 153-166.
deJong JR (1957) The effects of increasing skill on cycle time and its
consequences for time standards. Ergonomics 1:5 l-60.
Diamond DM, Weinberger NM (1986) Classical conditioning rapidly
induces specific changes in frequency receptive fields of single neurons
in secondary and ventral ectosylvian auditory cortical fields. Brain
Res 372~357-360.
Diamond IT, Neff WD (1957) Ablation of temporal cortex and dis-
crimination of auditory patterns. J Neurophysiol 20:300-3 15.
Disterhoft JF, Stuart DK (1976) Trial sequence ofchanged unit activity
in auditory system of alert rat during conditioned response acquisition
and extinction. J Neurophysiol 39:266-28 1.
Edelman GM (1987) Neuronal Darwinism: the theory of neuronal
group selection. New York: Basic.
Fitts PM (1964) Perceptual-motor skill learning. In: Categories of
human learning (Melton AW, ed), pp 243-285. New York: Academic.
Funkenstein HH, Winter P (1973) Responses to acoustic stimuli of
units in the auditory cortex of awake squirrel monkeys. Exp Brain
Res 18:464-488.
Gibson EJ (1953) Improvement in perceptual judgments as a function
of controlled practice or training. Psycho1 Bull 50:40 143 1.
Goldberg JM, Neff WD (196 1) Frequency discrimination after bilat-
eral ablation of cortical auditory areas. J Neurophysiol 24: 119-l 28.
Grajski KA, Merzenich MM (1990) Hebb-type dynamics is sufficient
to account for the inverse magnification rule in cortical somatotopy.
Neural Comput 2:7 l-84.
Green DM, Swets JA (1966) Signal detection theory. New York: Wiley.
Hubel DH, Wiesel TN (1974) Uniformity of monkey striate cortex:
a parallel relationship between field size, scatter and magnification
factor. J Comp Neurol 158:295-302.
Imig TJ, Rugero MA, Kitzes LM, Javel E, Brugge JF (1977) Orga-
nization of auditory cortex in the owl monkey (Lotus trivirgutus). J
Comp Neurol 17l:ll l-128.
Imig TJ, Irons A, Samson FR (1991) Single-unit selectivity to azi-
muthal direction and sound pressure level of noise bursts in cat high-
frequency primary auditory cortex. J Neurophysiol63:1448-1466.
Irvine DRF, Rajan R, Wize LZ, Heil P (1991) Reorganization in
auditory cortex of adult cats with unilateral restricted cochlear lesions.
Sot Neurosci Abstr 17: 1485.
James W (1890) The principles of psychology, Vol I. New York: Do-
ver.
Jenkins WM, Melzenich MM, Ochs MT, Allard T, Guic-Robles E (1990)
Functional reorganization of primary somatosensory cortex in adult
owl monkeys after behaviorally controlled tactile stimulation. J Neu-
rophysiol 63:82-104.
Kaas JH (199 1) Plasticity of sensory and motor maps in adult mam-
mals. Annu Rev Neurosci 14: 137-l 67.
Kasamatsu T, Pettigrew JD (1979) Preservation of binocularity after
monocular deprivation in the striate cortex of kittens treated with
6-hydroxydopamine. J Comp Neurol 185:139-162.
Kitzes LM. Farlev GR. Starr A (1978) Modulation of auditorv cortex
unit activity during the performance of a conditioned response. Exp
Neurol 62:678-697.
McKenna TM, Ashe JH, Weinberger NM (1989) Cholinergic mod-
ulation of frequency receptive fields in auditory cortex. I. Frequency-
specific effects of muscarinic agonists. Synapse 4:30-43.
Mendelson JR. Schreiner CE. Grasse K. Sutter ML (1988) Soatial
distribution ofresponses to FM sweeps in cat primary auditory cortex.
Abstr Assoc Res Otolaryngol 11:36.
Merzenich MM, Brugge JF (1973) Representation of the cochlear par-
tition on the superior temporal plane of the macaque monkey. Brain
Res 50:275-296. - -
Merzenich MM, Knight PL, Roth GL (1975) Representation of coch-
lea within primary auditory cortex in the cat. J Neurophysiol38:231-
249.
Merzenich MM, Nelson RJ, Kaas JH, Stryker MP, Jenkins WM, Zook
JM. Cvnader MS. Schouumann A (1987) Variabilitv in hand surface
representations in areas 3b and 1 in adult owl and squirrel monkeys.
J Comp Neurol 258:281-396.
Merzenich MM, Recanzone G, Jenkins WM, Allard TT, Nudo RJ (1988)
Cortical representational plasticity. In: Neurobiology of neocortex
(Rakic P, Singer W, eds), pp 41-67. New York: Wiley.
Merzenich MM, Recanzone GH, Jenkins WM, Nudo RJ (1990) How
the brain functionally rewires itself. In: Natural and artificial parallel
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
computations (Arbib M, Robinson JA, eds), pp 177-210. Cambridge,
MA: MIT Press.
Merzenich MM, Schreiner CE, Recanzone GH, Beitel RE, Sutter ML
(199 1) Topographic organization of cortical field AI in the owl mon-
key Aotus trivirgutus. Abstr Assoc Res Otolaryngol 14:44.
Metherate R, Weinberger NM (1990) Cholinergic modulation of re-
sponses to single tones produces tone-specific receptive field altera-
tions in cat auditory cortex. Synapse 6:133-145.
Meyer DR, Woolsey CN (1952) Effects of localized cortical destruction
on auditory discriminative conditioning in cat. J Neurophysiol 15:
149-162.
Miller KD, Keller JB, Stryker MP (1989) Ocular dominance column
development: analysis and simulation. Science 245:605-6 15.
Olds J, Disterhoft JF, Segal M, Komblith CL, Hirsh R (1972) Learning
centers of rat brain mapped by measuring latencies of conditioned
unit responses. J Neurophysiol35:202-219.
Palm G (1990) Cell assemblies as a auideline for brain research. Con-
cepts Neurosci 1: 133-l 47. -
Pearson JC, Finkel LH, Edelman GM (1987) Plasticity in the orga-
nization of adult cerebral cortical maps: a computer simulation based
on neuronal group selection. J Neurosci 7:4209-4223.
Pelleg-Toiba R, Wollberg Z (1989) Tuning properties of auditory cor-
tex cells in the awake squirrel monkey. Exp Brain Res 74:353-364.
Prosen CA, Moody DB, Sommers MS, Stebbins WC (1990) Frequency
discrimination in the monkey. J Acoust Sot Am 88:2 152-2 158.
Recanzone GH, Merzenich MM (1992) Alterations of the functional
organization of primary somatosensory cortex following intracortical
microstimulation or behavioral training. In: Memory: organization
and locus ofchange (Squire LM, Weinberger NM, Lynch G, McGaugh
JL, eds), pp 217-238. New York: Oxford UP.
Recanzone GH, Jenkins WM, Hradek GT, Merzenich MM (199 1) A
behavioral frequency discrimination paradigm for use in adult pri-
mates. Behav Res Methods Instrum Comput 23:357-369.
Recanzone GH, Jenkins WM, Hradek GT,^ Merzenich MM (1992a)
Progressive improvements in discriminative abilities in adult owl
monkeys performing a tactile frequency discrimination task. J Neu-
rophysiol 67:1015-1030.
Recanzone GH, Merzenich MM, Jenkins WM, Grajski KA, Dinse HR
(1992b) Topographic reorganization of the hand representation in
cortical area 3b of owl monkeys trained in a frequency discrimination
task. J Neurophysiol 67:1031-1056.
Recanzone GH. Merzenich MM. Jenkins WM (1992~) Freauencv dis-
I I _
crimination training engaging a restricted skin surface results in an
emergence of a cutaneous response zone in cortical area 3a. J Neu-
rophysiol 67:1057-1070.
Recanzone GH, Merzenich MM, Schreiner CE (1992d) Changes in
the distributed temporal response properties of SI cortical neurons
reflect improvements in performance on a temporally-based tactile
discrimination task. J Neurophysiol 67: 107 l-109 1.
Robertson D, Irvine DRF (1989) Plasticity of frequency organization
in auditory cortex of guinea pigs with partial unilateral deafness. J
Comp Neurol 282:456117 1.
Ryugo DK, Weinberger NM (1978) Differential plasticity of morpho-
logically distinct populations in the medial geniculate body of the cat
during classical conditioning. Behav Biol 22:275-301.
Schreiner CE, Mendelson JR (1990) Functional topography of cat
primary auditory cortex. Distribution of integrated excitation. J Neu-
rophysiol 6411442-1459.
Schreiner CE, Mendelson JR, Grasse K, Sutter ML (1988) Spatial
distribution of basic response properties in cat primary auditory cor-
tex. Abstr Assoc Res Otolaryngol 11:36.
Seibel R (1963) Discrimination reaction time for a 1,023-alternative
task. J Exp Psycho1 66:215-226.
Sineer W (1990) Search for coherence: a basic nrincinle of cortical
silf-organization. Concepts Neurosci 1: l-26. - -
Sinnott JM, Petersen MR, Hopp SL (1985) Frequency and intensity
discrimination in humans and monkeys. J Acoust Sot Am 78: 1977-
1985.
Sinnott JM, Owren MJ, Petersen MR (1987) Auditory frequency dis-
crimination in primates: species differences (Cercopithecus, A~UCUCU,
Homo). J Comp Psycho1 101:126-131.
Sur M, Merzenich MM, Kaas JH (1980) Magnification, receptive-field
area, and “hypercolumn” size in areas 3b and 1 of somatosensory
cortex in owl monkeys. J Neurophysiol44:295-3 11.
Sutter ML, Schreiner CE (199 1) Physiology and topography of neurons
 The Journal of Neuroscience, January 1993, 13(l) 103

with multipeaked tuning curves in cat primary auditory cortex. J
Neurophysiol 65: 1207-l 226.
Thompson RF (1960) Function of auditory cortex of cat in frequency
discrimination. J Neurophysiol 23:321-334.
Volkman AW (1858) Uber den Einfluss der Ubung. Leipzig Ber Math-
Phys Classe 10:38-69.
von der Malsburg C, Singer W (1988) Principles of cortical network
organization. In: Neurobiology of neocot-tex (Rakic P, Singer W, eds),
pp 69-99. New York: Wiley.
Weinberger NM, Ashe JH, Metherate R, McKenna TM, Diamond DM,
Bakin J (1990) Returning auditory cortex by learning: a preliminary
model of receptive field plasticity. Concepts Neurosci 1:9 l-l 32.
Wier CC, Jesteadt W, Green DM (1977) Frequency discrimination as
a function of frequency and sensation level. J Acoust Sot Am 61:
178-184.
Wong-Riley MTT, Merzenich MM, Leake PS (1978) Changes in en-
dogenous enzymatic reactivity to DAB induced by neuronal inactiv-
ity. Brain Res 141:185-192.
Zwislocki J, Maire F, Feldman AS, Rubin H (1958) On the effect of
practice and motivation on the threshold of audibility. J Acoust Sot
Am 30:254-262.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Current Biology 21, 2109–2114, December 20, 2011 ª2011 Elsevier Ltd All rights reserved
Acquiring ‘‘the Knowledge’’
of London’s Layout
Drives Structural Brain Changes
Katherine Woollett1 and Eleanor A. Maguire1,*
1Wellcome Trust Centre for Neuroimaging, Institute of
Neurology, University College London, 12 Queen Square,
London WC1N 3BG, UK
Summary
The last decade has seen a burgeoning of reports associ-
ating brain structure with specific skills and traits (e.g.,
[1–8]). Although these cross-sectional studies are informa-
tive, cause and effect are impossible to establish without
longitudinal investigation of the same individuals before
and after an intervention. Several longitudinal studies have
been conducted (e.g., [9–18]); some involved children or
young adults, potentially conflating brain development with
learning, most were restricted to the motor domain, and all
concerned relatively short timescales (weeks or months).
Here, by contrast, we utilized a unique opportunity to study
average-IQ adults operating in the real world as they learned,
over four years, the complex layout of London’s streets while
training to become licensed taxi drivers. In those who quali-
fied, acquisition of an internal spatial representation of
London was associated with a selective increase in gray
matter (GM) volume in their posterior hippocampi and con-
comitant changes to their memory profile. No structural brain
changes were observed in trainees who failed to qualify or
control participants. We conclude that specific, enduring,
structural brain changes in adult humans can be induced
by biologically relevant behaviors engaging higher cognitive
functions such as spatial memory, with significance for the
‘‘nature versus nurture’’ debate.
Results and Discussion
In order to qualify as a licensed London taxi driver, a trainee must
learn the complex and irregular layout of London’s w25,000
streets (Figure 1) within a 6-mile radius of Charing Cross train
station, along with the locations of thousands of places of
interest. This spatial learning is known as acquiring ‘‘the Knowl-
edge’’ and typically takes between 3 and 4 years, leading to
a stringent set of examinations, called ‘‘appearances,’’ which
must be passed in order to obtain an operating license from
the Public Carriage Office (PCO, the official London taxi-
licensing body). This comprehensive training and qualification
procedure is unique among taxi drivers anywhere in the world.
Previous cross-sectional studies of qualified London taxi
drivers documented more gray matter (GM) volume in their
posterior hippocampi and less in their anterior hippocampi
relative to non-taxi-driving matched control participants [2–4].
Moreover, correlation of hippocampal GM volume with years
of taxi driving suggested that structural differences may have
been acquired through the experience of navigating, to accom-
modate the internal representation of London, and were not
merely due to preexisting hippocampal GM volume patterns
*Correspondence: e.maguire@ucl.ac.uk
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
DOI 10.1016/j.cub.2011.11.018
Report
disposing individuals to being taxi drivers [2–4, 19, 20]. As well
as displaying a specific pattern of hippocampal GM volume,
qualified taxi drivers have been found to display better memory
for London-based information, but surprisingly poorer learning
and memory for certain types of new visual information (e.g.,
delayed recall of complex figures), compared with control
participants, suggesting there might be a price to pay for the
acquisition of their spatial knowledge, perhaps linked to their
reduced anterior hippocampal volume (see [3, 4, 20] for more
on this). Interestingly, the opposite pattern of hippocampal
GM volume and memory profile has been described in retired
taxi drivers, hinting that any changes acquired through learning
might be reversed or ‘‘normalized’’ when the call on stored
memory representations lessens [21].
London taxi drivers are therefore a useful model of memory,
illuminating the role of the hippocampus and intrahippocam-
pal functional differentiation and potentially informing about
whether hippocampal structure and memory capacity are
hardwired or amenable to change. Given current economic
imperatives and increasing longevity, the need to keep re-
training and learning throughout adulthood has never been
more acute. Direct evidence for hippocampal plasticity in
response to environmental stimulation could allow us to
understand the boundaries within which human memory op-
erates and the scope for improving or rehabilitating memory
in educational and clinical contexts. Moreover, given the
dearth of longitudinal magnetic resonance imaging (MRI)
structural association studies focusing on higher cognitive
functions in average adults engaged in truly naturalistic
behaviors, taxi drivers could contribute new information to
the wider debate about whether key aspects of cognition
are fixed or malleable.
The above aspirations are predicated upon hippocampal
plasticity in adult humans, evidence for which remains sparse
[10, 18]. We therefore conducted a longitudinal study exam-
ining 79 male trainee London taxi drivers at the start of their
training (T1, time 1) and then again 3–4 years later just after
qualification (T2, time 2), as well as 31 male control partici-
pants, with two aims. First, given that the PCO suggests that
typically 50%–60% of trainees fail to qualify, we anticipated
having three groups of participants: trainees who qualified
(Q), trainees who failed to qualify (F), and the controls (C).
With this design, we could retrospectively examine whether
MRI and/or neuropsychological findings at T1 could predict
who would eventually qualify 3–4 years later at T2. Second,
we sought to ascertain whether the pattern of hippocampal
GM volume and memory profile observed in previous cross-
sectional taxi driver studies would be induced and observable
within the same participants as a result of acquiring ‘‘the
Knowledge.’’
Of the 79 trainees, 39 went on to qualify as licensed London
taxi drivers, while 20 did not qualify (ceased training or failed
their appearances) but agreed to return for testing at T2. Of
the other 20 trainees who did not wish to return at T2, two
had qualified, and the remaining 18 decided to stop training
or made no appearances. Thus in our sample, 51.9% of
trainees qualified, in line with PCO figures. All 31 control
participants returned for testing at T2.
 Current Biology Vol 21 No 24
2110
Figure 1. Central London
Trainee taxi drivers must acquire ‘‘the Knowledge’’ of London’s layout within a 6-mile radius of Charing Cross train station. This map shows just a part of the
total area that must be learned. Map reproduced by permission of Geographers’ A-Z Map Co. Ltd. ª Crown copyright 2005. All rights reserved. License
number 100017302.
Time 1
We first focused on the data acquired at T1 and examined the
status before training occurred of those trainees who subse-
quently qualified or not, as well as the control participants.
There were no statistically significant differences between
the three groups on a range of background measures such
as age, handedness, education, and IQ [F(8,166) = 1.81; p =
0.07; see Table 1]. We next looked at their performance at
T1 on a battery of tests assessing working and long-term
memory, recognition and recall, and visual and verbal material
(see Supplemental Experimental Procedures and Table S1A
available online). There were no statistically significant differ-
ences between the three groups for any of the memory
measures [F(18,158) = 1.33; p = 0.18]. Finally, we examined
the structural MRI brain scans of participants acquired at T1,
before training. An automated whole-brain analysis method,
voxel-based morphometry (VBM; [22, 23]) implemented in
SPM8, was used to compare GM volume between the groups.
No significant differences were found, even with a liberal
statistical threshold (p < 0.005 uncorrected for multiple
comparisons).
Our data therefore showed that at T1, before acquisition of
a detailed internal spatial representation of London’s layout,
there was no general intellectual, mnemonic, or structural
brain difference that was associated with subsequent success
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
at qualification. In particular, hippocampal GM volume, both
anteriorly and posteriorly, was indistinguishable between
those who would qualify, those who would not, and the con-
trols. This means that the groups started out on equal terms
and that any changes that subsequently emerged would be
due to acquiring ‘‘the Knowledge.’’
Time 2
In the first instance, we compared the two trainee groups on
a range of training-related variables (see Table 1). There was
no significant difference between those who qualified and
those who did not in terms of the total time they had spent in
training [t(57) = 0.67; p = 0.5]. There was a difference, however,
in the number of hours per week the groups spent training, with
the trainees who qualified spending twice as many hours per
week training compared to the group who failed to qualify
[t(57) = 6.62; p = 0.001]. Unsurprisingly, the two groups differed
also in the number of appearances that they made, with the
failed candidates making hardly any [t(57) = 13.1; p = 0.001].
Finally, there was no significant difference in the time elapsed
between testing at T1 and testing at T2 for the Q, F, and C
groups [F(2, 86) = 1.27; p = 0.28].
We next examined performance at T2 on parallel versions
of the memory tests that had first been employed at T1 (see
Table S1B). Whereas at T1 there were no differences between
 Hippocampal Changes in Trainee London Taxi Drivers
2111

Table 1. Background Characteristics of the Participants scans, total GM volume, and participant age were modeled as
confounding variables. Given our a priori interest in the hippo-
Qualified Failed to Qualify Controls campus, the significance level was set at p < 0.001 corrected
Measure (n = 39) (n = 20) (n = 31)
for the volume of the hippocampus; otherwise the significance
Age (years) 37.97 6 7.96 40.50 6 5.27 35 6 8.99 level was set at p < 0.05 corrected for multiple comparisons
Age left school 16.66 6 1.32 16.75 6 1.40 16.77 6 1.30 across the whole brain. We first looked at the scans of the qual-
(years)
ified trainees and found that GM volume had increased in the
Estimated verbal IQ 97.72 6 6.29 98.66 6 3.49 100.79 6 3.79
Matrix reasoning 11.89 6 2.06 12.20 6 1.98 11.83 6 2.39
posterior hippocampi bilaterally (30, 242, 1, z = 5.44; 20, 237,
(scaled score) 10, z = 5.64; 24, 239, 7, z = 4.02; 229, 242, 3, z = 5.91) at T2 rela-
Handedness 87.97 6 24.09 88.10 6 21.50 71.74 6 38.69 tive to T1 (see Figure 2 and Figure 3). The opposite comparison
(laterality index) (T1 > T2) did not show any significant differences, nor were
Total training time 38.84 6 7.02 35.80 6 10.32 – there any significant differences in GM volume anywhere else
(months)
in the brain for either contrast. Similar analyses were performed
Training time per 34.56 6 12.40 16.70 6 8.21 –
week (hours)a
for the trainees who failed to qualify. No significant differences
Number of 15.64 6 3.66 2.60 6 3.45 – in GM volume were found anywhere in the brain, including the
appearancesa hippocampi, for T2 > T1 or T1 > T2. When the data from the
Time between T1 35.28 6 8.19 36.15 6 8.53 32.8 6 7.37 two time points of the control participants were compared,
and T2 testing again no significant differences in GM volume emerged.
(months)
As with the memory data, the structural MRI brain data
Measurements are given in means 6 SD. replicate and extend the cross-sectional taxi driver studies
a
Significantly more for trainees who qualified compared to trainees who [2–4] by showing that the increased posterior hippocampal
did not qualify.
GM volume previously observed most likely occurred as a result
of acquiring the detailed spatial representation of London’s
the groups, at T2 this was no longer the case: significant differ- layout. Importantly, the posterior hippocampal change in GM
ences between the three groups now emerged [F(18,158) = volume cannot be attributed to the qualified trainee taxi drivers’
2.35; p = 0.004], driven by two main effects. A significant dif- training procedures, methods, or general attempts to learn,
ference was found between the groups on the London land- as the trainees who failed to qualify were exposed to the
marks proximity judgments test [F(2,87) = 8.09 p = 0.001], same milieu. Although the posterior hippocampal increase
with trainees who qualified being significantly better at judging accords with previous findings, we did not observe a decrease
the spatial relations between London landmarks than the in anterior hippocampal GM volume. This is interesting,
control participants (p = 0.003), as were nonqualified trainees because it may provide an insight into the time frame of the
(p = 0.003). The groups also differed on the delayed recall of hippocampal structural changes. It could be that they occur
the Taylor complex figure [F(2,87) = 4.38; p = 0.015], with qual- serially, with the increase in posterior hippocampus happening
ified trainees being significantly worse at recalling the complex first and within 3–4 years. We speculate that the anterior hippo-
figure after 30 min delay than the control participants (p = 0.01). campal GM volume might then decrease subsequently and in
By contrast, the performance of the nonqualified trainees response to the posterior increase. In fact, the poor perfor-
was not significantly different from that of control participants mance of the qualified trainees on the delayed recall of the
(p = 0.14). complex figure at T2 may be an indication that changes are
The memory profile displayed by the now qualified train- already afoot in the anterior hippocampus but are not yet
ees mirrors exactly the pattern displayed in several pre- detectable with MRI.
vious cross-sectional studies of licensed London taxi drivers That acquiring ‘‘the Knowledge,’’ which encompasses
[3, 4, 20] (and that which normalized in the retired taxi drivers spatial learning and memory, can drive changes in posterior
[21]). In those studies also, the taxi drivers displayed more hippocampus illustrates the close relationship between this
knowledge of the spatial relationships between landmarks region and spatial navigation [25–27] and suggests that the
in London, unsurprisingly, given their increased exposure to hippocampus acts as a storage site for the spatial information
the city compared to control participants. By contrast, this acquired during ‘‘the Knowledge,’’ or as a processing hub for
enhanced spatial representation of the city was accompanied detailed navigational information. Our results underline the
by poorer performance on a complex figure test, a visuospatial existence of functional differentiation in the hippocampus,
task designed to assess the free recall of visual material after with anterior and posterior regions diverging in their response
30 min. Our findings therefore not only replicate those of to spatial memory [28, 29]. The hippocampal plasticity we have
previous cross-sectional studies but extend them by showing observed in vivo in adult humans parallels the effects reported
the change in memory profile within the same participants. in nonhumans where intraindividual hippocampal volume
That the only major difference between T1 and T2 was acquiring changes occur in response to demands placed on spatial
‘‘the Knowledge’’ strongly suggests that this is what induced memory [30, 31]. Having documented this hippocampal
the memory change. change, the question is what mechanism underpins this
We then turned to the structural MRI brain scans acquired at process.
T2 and compared them to the scans acquired in the same indi- Using standard structural MRI scanning in humans, it is not
viduals at T1 in order to assess whether acquiring ‘‘the Knowl- possible to address this question directly, but based on work
edge’’ had any impact on GM volume. This was accomplished in nonhumans, there are several candidates. Studies in ro-
by implementing high-dimensional warping (HDW) in SPM8. dents have demonstrated that when learning requires cogni-
HDW safeguards against nonspecific subtle differences that tive effort and where learning actually takes place (i.e., where
may arise between the first and second scans within subjects material is remembered after a delay), there is an effect on
in a longitudinal study (see Supplemental Experimental the rate of hippocampal neurogenesis [32]. Moreover, the
Procedures and also [24] for full details). The time between animals that learn best have more new neurons after training

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Current Biology Vol 21 No 24
2112

Figure 2. Gray Matter Volume Changes between T1 and T2 in Qualified Trainees

Gray matter (GM) volume increased in the most posterior part of the hippocampus bilaterally between T1 and T2 as a result of acquiring a detailed
representation of London’s layout. This change was only apparent in the trainees who qualified. The upper row shows axial views, and the middle and lower
rows show sagittal views through the right (R) and left (L) sides of the brain, respectively, that encompass the GM changes (shown in orange and yellow) in
the peak voxels detailed in the text.

than those who do not learn, or do not learn efficiently [33, 34].
If neurogenesis is what underpins the hippocampal volume
change in qualified taxi drivers, it may be related to recruitment
of new neurons following neurogenesis [35, 36] that are
pressed into the service of spatial memory. The development
of greater communication between neurons in the form of
increased synaptogenesis [37, 38] might also be involved,
and proliferation in dendritic arborization, augmenting con-
nectivity between neurons, could in turn increase memory
capacity and also lead to volumetric changes [39]. Glial cells,
which continue to be produced, albeit at a slow pace, through-
out adulthood [40], could also be implicated and have been
shown to increase in volume with the addition of synapses
following learning [41]. In the future, new approaches to human
brain scanning in vivo may eventually be able to provide more
direct insight into the key mechanisms supporting human
hippocampal plasticity [42].
To conclude, we have shown that there is a capacity for
memory improvement and concomitant structural changes
to occur in the human brain well into adulthood. That there
are many thousands of licensed London taxi drivers shows
that acquisition of ‘‘the Knowledge,’’ and presumably the brain
changes that arise from it, is not uncommon, offering encour-
agement for lifelong learning, and possibly a role in neuroreha-
bilitation in the clinical context. However, this needs to be
balanced by our finding that memory improvement in one
domain may come at the expense of memory performance
elsewhere. One final point to consider concerns the PCO fig-
ures and our data showing that only half of trainees who under-
take ‘‘the Knowledge’’ actually qualify. The reasons we were
given for ceasing training included the time commitment being
too great, financial imperatives, and family obligations. Very
few trainees reported ceasing because they found the spatial
memory demands to be too great, although it is possible, or

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Hippocampal Changes in Trainee London Taxi Drivers
2113
Figure 3. Plot of Gray Matter Intensities across Groups and Time
GM intensity values were extracted from the four peak hippocampal voxels
identified by the HDW procedure (see text for details). Here we show a plot
from one of those peaks (the others resulted in very similar plots) at 20, 237,
10 in the right posterior hippocampus. Only the qualified trainees experi-
enced a significant increase in posterior hippocampal gray matter between
T1 and T2; this change was evident in every qualified trainee. Data are
presented as means 6 two standard errors of the mean. *p < 0.05.
even likely, that the reasons given may have masked such diffi-
culties in some individuals. It could be that there are inherited
factors that feed into individual differences in spatial memory
and navigation ability. One source of influence may come
from genes. Genetic association studies have demonstrated
effects of specific gene polymorphisms on the volume of the
hippocampus [43, 44] and memory performance [45, 46].
Although our data show that environmental stimulation can
drive structural brain changes, it may be that this hippocampal
plasticity expresses itself only in certain individuals. The
trainees that qualified may have had a genetic predisposition
toward plasticity that the nonqualified individuals lacked.
Thus, the perennial question of ‘‘nature versus nurture’’ is still
open for future investigations that incorporate genetic and
other influences on individual differences, as well as cognitive
and structural brain dimensions.
Experimental Procedures
All participants gave informed written consent to participation in accordance
with the local research ethics committee. Full details of the participants,
cognitive tests, MRI scans, and data analysis procedures are provided in
the Supplemental Experimental Procedures.
Supplemental Information
Supplemental Information includes one table and Supplemental Experi-
mental Procedures and can be found with this article online at doi:10.
1016/j.cub.2011.11.018.
Acknowledgments
This work was supported by the Wellcome Trust. We thank all participants
for their time and continued interest throughout the study. We are very
grateful to the Public Carriage Office, in particular Tony Bishop and Alison
Said. Thanks also to the Knowledge Schools, the Physics and Imaging
Support teams at the Wellcome Trust Centre for Neuroimaging, Ralf
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Deichmann, Amanda Brennan, John Ashburner, Jenny Crinion, Alex Leff,
Ric Davis, and Chris Freemantle.
Received: September 8, 2011
Revised: October 26, 2011
Accepted: November 8, 2011
Published online: December 8, 2011
References
1. Gaser, C., and Schlaug, G. (2003). Gray matter differences between
musicians and nonmusicians. Ann. N Y Acad. Sci. 999, 514–517.
2. Maguire, E.A., Gadian, D.G., Johnsrude, I.S., Good, C.D., Ashburner, J.,
Frackowiak, R.S.J., and Frith, C.D. (2000). Navigation-related structural
change in the hippocampi of taxi drivers. Proc. Natl. Acad. Sci. USA 97,
4398–4403.
3. Maguire, E.A., Woollett, K., and Spiers, H.J. (2006). London taxi drivers
and bus drivers: a structural MRI and neuropsychological analysis.
Hippocampus 16, 1091–1101.
4. Woollett, K., and Maguire, E.A. (2009). Navigational expertise may
compromise anterograde associative memory. Neuropsychologia 47,
1088–1095.
5. Mechelli, A., Crinion, J.T., Noppeney, U., O’Doherty, J., Ashburner, J.,
Frackowiak, R.S., and Price, C.J. (2004). Neurolinguistics: structural
plasticity in the bilingual brain. Nature 431, 757.
6. Hüfner, K., Binetti, C., Hamilton, D.A., Stephan, T., Flanagin, V.L., Linn,
J., Labudda, K., Markowitsch, H., Glasauer, S., Jahn, K., et al. (2011).
Structural and functional plasticity of the hippocampal formation in
professional dancers and slackliners. Hippocampus 21, 855–865.
7. Martin, S.B., Covell, D.J., Joseph, J.E., Chebrolu, H., Smith, C.D., Kelly,
T.H., Jiang, Y., and Gold, B.T. (2007). Human experience seeking cor-
relates with hippocampus volume: convergent evidence from manual
tracing and voxel-based morphometry. Neuropsychologia 45, 2874–
2881.
8. Kanai, R., Feilden, T., Firth, C., and Rees, G. (2011). Political orientations
are correlated with brain structure in young adults. Curr. Biol. 21,
677–680.
9. Draganski, B., Gaser, C., Busch, V., Schuierer, G., Bogdahn, U., and
May, A. (2004). Neuroplasticity: changes in grey matter induced by
training. Nature 427, 311–312.
10. Draganski, B., Gaser, C., Kempermann, G., Kuhn, H.G., Winkler, J.,
Büchel, C., and May, A. (2006). Temporal and spatial dynamics of brain
structure changes during extensive learning. J. Neurosci. 26, 6314–
6317.
11. May, A., Hajak, G., Gänssbauer, S., Steffens, T., Langguth, B., Kleinjung,
T., and Eichhammer, P. (2007). Structural brain alterations following
5 days of intervention: dynamic aspects of neuroplasticity. Cereb.
Cortex 17, 205–210.
12. Ilg, R., Wohlschläger, A.M., Gaser, C., Liebau, Y., Dauner, R., Wöller, A.,
Zimmer, C., Zihl, J., and Mühlau, M. (2008). Gray matter increase
induced by practice correlates with task-specific activation: a combined
functional and morphometric magnetic resonance imaging study.
J. Neurosci. 28, 4210–4215.
13. Boyke, J., Driemeyer, J., Gaser, C., Büchel, C., and May, A. (2008).
Training-induced brain structure changes in the elderly. J. Neurosci.
28, 7031–7035.
14. Ceccarelli, A., Rocca, M.A., Pagani, E., Falini, A., Comi, G., and Filippi,
M. (2009). Cognitive learning is associated with gray matter changes
in healthy human individuals: a tensor-based morphometry study.
Neuroimage 48, 585–589.
15. Hyde, K.L., Lerch, J., Norton, A., Forgeard, M., Winner, E., Evans, A.C.,
and Schlaug, G. (2009). Musical training shapes structural brain devel-
opment. J. Neurosci. 29, 3019–3025.
16. Scholz, J., Klein, M.C., Behrens, T.E., and Johansen-Berg, H. (2009).
Training induces changes in white-matter architecture. Nat. Neurosci.
12, 1370–1371.
17. Bezzola, L., Mérillat, S., Gaser, C., and Jäncke, L. (2011). Training-
induced neural plasticity in golf novices. J. Neurosci. 31, 12444–12448.
18. Erickson, K.I., Voss, M.W., Prakash, R.S., Basak, C., Szabo, A.,
Chaddock, L., Kim, J.S., Heo, S., Alves, H., White, S.M., et al. (2011).
Exercise training increases size of hippocampus and improves memory.
Proc. Natl. Acad. Sci. USA 108, 3017–3022.
 Current Biology Vol 21 No 24
2114

19. Maguire, E.A., Spiers, H.J., Good, C.D., Hartley, T., Frackowiak, R.S.J., 45. Egan, M.F., Kojima, M., Callicott, J.H., Goldberg, T.E., Kolachana, B.S.,
and Burgess, N. (2003). Navigation expertise and the human hippo- Bertolino, A., Zaitsev, E., Gold, B., Goldman, D., Dean, M., et al. (2003).
campus: a structural brain imaging analysis. Hippocampus 13, 250–259. The BDNF val66met polymorphism affects activity-dependent secretion
20. Woollett, K., and Maguire, E.A. (2010). The effect of navigational ex- of BDNF and human memory and hippocampal function. Cell 112,
pertise on wayfinding in new environments. J. Environ. Psychol. 30, 257–269.
565–573. 46. Hariri, A.R., Goldberg, T.E., Mattay, V.S., Kolachana, B.S., Callicott, J.H.,
21. Woollett, K., Spiers, H.J., and Maguire, E.A. (2009). Talent in the taxi: Egan, M.F., and Weinberger, D.R. (2003). Brain-derived neurotrophic
a model system for exploring expertise. Philos. Trans. R. Soc. Lond. B factor val66met polymorphism affects human memory-related hippo-
Biol. Sci. 364, 1407–1416. campal activity and predicts memory performance. J. Neurosci. 23,
22. Ashburner, J., and Friston, K.J. (2000). Voxel-based morphometry—the 6690–6694.
methods. Neuroimage 11, 805–821.
23. Ashburner, J., and Friston, K.J. (2005). Unified segmentation.
Neuroimage 26, 839–851.
24. Kipps, C.M., Duggins, A.J., Mahant, N., Gomes, L., Ashburner, J., and
McCusker, E.A. (2005). Progression of structural neuropathology in
preclinical Huntington’s disease: a tensor based morphometry study.
J. Neurol. Neurosurg. Psychiatry 76, 650–655.
25. O’Keefe, J., and Dostrovsky, J. (1971). The hippocampus as a spatial
map. Preliminary evidence from unit activity in the freely-moving rat.
Brain Res. 34, 171–175.
26. O’Keefe, J., and Nadel, L. (1978). The Hippocampus as a Cognitive Map
(Oxford: Oxford University Press).
27. Burgess, N., Maguire, E.A., and O’Keefe, J. (2002). The human hippo-
campus and spatial and episodic memory. Neuron 35, 625–641.
28. Moser, M.B., and Moser, E.I. (1998). Functional differentiation in the
hippocampus. Hippocampus 8, 608–619.
29. Fanselow, M.S., and Dong, H.W. (2010). Are the dorsal and ventral
hippocampus functionally distinct structures? Neuron 65, 7–19.
30. Lee, D.W., Miyasato, L.E., and Clayton, N.S. (1998). Neurobiological
bases of spatial learning in the natural environment: neurogenesis and
growth in the avian and mammalian hippocampus. Neuroreport 9,
R15–R27.
31. Sherry, D.F., and Hoshooley, J.S. (2010). Seasonal hippocampal plas-
ticity in food-storing birds. Philos. Trans. R. Soc. Lond. B Biol. Sci.
365, 933–943.
32. Shors, T.J. (2009). Saving new brain cells. Sci. Am. 300, 46–52, 54.
33. Curlik, D.M., 2nd, and Shors, T.J. (2011). Learning increases the survival
of newborn neurons provided that learning is difficult to achieve and
successful. J. Cogn. Neurosci. 23, 2159–2170.
34. Dayer, A.G., Ford, A.A., Cleaver, K.M., Yassaee, M., and Cameron, H.A.
(2003). Short-term and long-term survival of new neurons in the rat
dentate gyrus. J. Comp. Neurol. 460, 563–572.
35. Doetsch, F., and Hen, R. (2005). Young and excitable: the function of
new neurons in the adult mammalian brain. Curr. Opin. Neurobiol. 15,
121–128.
36. Shors, T.J. (2008). From stem cells to grandmother cells: how neurogen-
esis relates to learning and memory. Cell Stem Cell 3, 253–258.
37. Kolb, B., Gorny, G., Söderpalm, A.H., and Robinson, T.E. (2003).
Environmental complexity has different effects on the structure of
neurons in the prefrontal cortex versus the parietal cortex or nucleus
accumbens. Synapse 48, 149–153.
38. Toni, N., Teng, E.M., Bushong, E.A., Aimone, J.B., Zhao, C., Consiglio,
A., van Praag, H., Martone, M.E., Ellisman, M.H., and Gage, F.H.
(2007). Synapse formation on neurons born in the adult hippocampus.
Nat. Neurosci. 10, 727–734.
39. McEwen, B.S. (1999). Stress and hippocampal plasticity. Annu. Rev.
Neurosci. 22, 105–122.
40. Rakic, P. (2002). Neurogenesis in adult primate neocortex: an evaluation
of the evidence. Nat. Rev. Neurosci. 3, 65–71.
41. Anderson, J.C., Douglas, R.J., Martin, K.A., and Nelson, J.C. (1994). Map
of the synapses formed with the dendrites of spiny stellate neurons of
cat visual cortex. J. Comp. Neurol. 341, 25–38.
42. Manganas, L.N., Zhang, X., Li, Y., Hazel, R.D., Smith, S.D., Wagshul,
M.E., Henn, F., Benveniste, H., Djuric, P.M., Enikolopov, G., and
Maletic-Savatic, M. (2007). Magnetic resonance spectroscopy identifies
neural progenitor cells in the live human brain. Science 318, 980–985.
43. Szeszko, P.R., Lipsky, R., Mentschel, C., Robinson, D., Gunduz-Bruce,
H., Sevy, S., Ashtari, M., Napolitano, B., Bilder, R.M., Kane, J.M., et al.
(2005). Brain-derived neurotrophic factor val66met polymorphism and
volume of the hippocampal formation. Mol. Psychiatry 10, 631–636.
44. Bueller, J.A., Aftab, M., Sen, S., Gomez-Hassan, D., Burmeister, M., and
Zubieta, J.K. (2006). BDNF Val66Met allele is associated with reduced
hippocampal volume in healthy subjects. Biol. Psychiatry 59, 812–815.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Report

Intermittent Stimulation of the Nucleus Basalis of

Meynert Improves Working Memory in Adult
Monkeys
Highlights Authors
d Intermittent stimulation of nucleus basalis (NB) improves Ruifeng Liu, Jonathan Crawford,
working memory performance Patrick M. Callahan, Alvin V. Terry, Jr.,
Christos Constantinidis, David T. Blake
d Continuous NB stimulation degrades performance in young
adult monkeys Correspondence
d Donepezil provided no greater improvement than intermittent dblake.mcg@gmail.com
stimulation
In Brief
d Effects were blocked by nicotinic and muscarinic receptor Liu et al. show that an intermittent train of
antagonists electrical stimulation in the nucleus
basalis improves working memory in
adult monkeys. The improvement
depends on cholinergic receptors. The
finding reveals a new mechanism to
improve cognitive performance in
conditions that compromise cholinergic
action, including aging and Alzheimer’s
disease.

Liu et al., 2017, Current Biology 27, 1–7

September 11, 2017 ª 2017 Elsevier Ltd.
http://dx.doi.org/10.1016/j.cub.2017.07.021

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021

Current Biology

Report

Intermittent Stimulation of the Nucleus Basalis

of Meynert Improves Working Memory
in Adult Monkeys
Ruifeng Liu,1,2 Jonathan Crawford,2 Patrick M. Callahan,3 Alvin V. Terry, Jr.,3 Christos Constantinidis,4
and David T. Blake2,5,*
1State Key Laboratory of Ophthalmology, Zhongshan Ophthalmic Center, Sun Yat-sen University, Guangzhou, China
2Brain and Behavior Discovery Institute, Department of Neurology, Medical College of Georgia, Augusta University, 1120 15th Street,
Augusta, GA 30912, USA
3Department of Pharmacology and Toxicology, Medical College of Georgia, Augusta University, 1120 15th Street, Augusta, GA 30912, USA
4Department of Neurobiology and Anatomy, Wake Forest School of Medicine, Winston-Salem, NC, USA
5Lead Contact

*Correspondence: dblake.mcg@gmail.com
http://dx.doi.org/10.1016/j.cub.2017.07.021

SUMMARY
Acetylcholine in the neocortex is critical for executive
function [1–3]. Degeneration of cholinergic neurons
in aging and Alzheimer’s dementia is commonly
treated with cholinesterase inhibitors [4–7]; however,
these are modestly effective and are associated with
side effects that preclude effective dosing in many
patients [8]. Electrical activation of the nucleus basa-
lis (NB) of Meynert, the source of neocortical acetyl-
choline [9, 10], provides a potential method of
improving cholinergic activation [11, 12]. Here we
tested whether NB stimulation would improve perfor-
mance of a working memory task in a nonhuman pri-
mate model. Unexpectedly, intermittent stimulation
proved to be most beneficial (60 pulses per second,
for 20 s every minute), whereas continuous stimula-
tion often impaired performance. Pharmacological
experiments confirmed that the effects depended
on cholinergic activation. Donepezil, a cholines-
terase inhibitor, restored performance in animals
impaired by continuous stimulation but did not
improve performance further during intermittent
stimulation. Intermittent stimulation was rendered
ineffective by either nicotinic or muscarinic receptor
antagonists. In the months after stimulation began,
performance also improved in sessions without stim-
ulation. Our results reveal that intermittent NB stimu-
lation can improve working memory, a finding that
has implications for restoring cognitive function in
aging and Alzheimer’s dementia.
RESULTS AND DISCUSSION
Five adult Rhesus monkeys were trained with a delayed match-
to-sample task, requiring them to remember a stimulus pre-
sented on a touch screen and, after a delay period, to select
the stimulus that was initially presented (Figure 1). Once the
animals were proficient with the task, they were implanted bilat-
erally with NB stimulation electrodes. Placement and anatomical
verification are described in STAR Methods. The zone of stimu-
lation was broad, on the order of a 4-mm-diameter sphere [15],
and the accuracy of placement of the electrode tip within the
nucleus was not critical in this experiment. Experiments in the
first two animals (K and S) primarily guided proper positioning
of stimulation leads in the next three (CH, DI, and PU). No histo-
logical confirmation of electrode placement occurred in these
three animals, and electrode positioning is presumptive based
on stereotaxic targeting.
Effects of Electrical Stimulation Parameters
We initially tested the hypothesis that NB stimulation improves
working memory performance by applying a continuous train
of stimulation pulses in blocks of 100 trials interleaved with
blocks of 100 trials without stimulation. Contrary to our expecta-
tions, we found that continuous stimulation always impaired per-
formance, and effects were larger at higher stimulation rates
(Figure 2A). Results for continuous stimulation at 80 Hz reached
statistical significance (binomial tests; animal CH: n = 800 trials,
p < 0.0001; animal DI: n = 1000, p < 0.001).
In an effort to determine whether stimulation during a partic-
ular interval of the task was disrupting working memory, the
task was altered to stimulate only during the inter-trial period
or to stimulate only during trials. Unexpectedly, either condition
resulted in supranormal performance. We established an inter-
mittent stimulation condition that provided 20 s of pulses at a
rate of 60 per second followed by 40 s without pulses. Results
from this condition are shown in Figure 2B. Trials with stimulation
resulted in better performance than trials with no stimulation in
the animals tested (binomial test, n = 1000, p < 0.01). Note that
a longer delay period was used in Figure 2B than in Figure 2A
to avoid ceiling effects. We conclude that intermittent stimulation
results in supranormal working memory performance.
We explored the parameters for optimal stimulation in two
ways. First, the number of pulses per minute was fixed, and
the duration and rate of stimulation were altered. The 1,200
pulses per minute were delivered in 10 s (i.e., at a rate of

Current Biology 27, 1–7, September 11, 2017 ª 2017 Elsevier Ltd. 1
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021
120 Hz), 20 s, 40 s, or 60 s. Results are shown in Figure 2C for
animals CH and DI. A two-way ANOVA found a significant effect
of these conditions on performance (test for animal number and
condition, F(4,4) = 12.92 for main effect of condition, p < 0.02;
F(1,4) = 1.48 for main effect of animal, p > 0.2). Post hoc compar-
isons showed that performance was significantly better at the
10 or 20 s period than the 60 s condition of continuous stimula-
tion (binomial statistic,
1,050 trials for each condition, Bonfer-
roni-corrected p < 0.005).
In a second experiment, the rate of stimulation was fixed at
80 pulses per second but the duration was varied to 10 s,
20 s, 40 s, or 60 s per minute (Figure 2D). A two-way ANOVA
found a significant effect of condition and animal (F(1,4) =
15.09 for animal, p < 0.02; F(4,4) = 8.48 for condition,
p < 0.04). The highest performance occurred for the 20 s stim-
ulation condition, which was significantly better than the 60 s
stimulation in post hoc tests (binomial statistic,
1,350 trials
each, Bonferroni-corrected p < 0.005). The significant ANOVA
finding on the animal variable occurred as animal CH averaged
higher performance in the 80 Hz stimulation conditions than
animal DI. These parametric tests revealed that the condition
of 1,200 pulses being delivered 15–20 s per minute was optimal
within the parameters tested.
The effect of intermittent stimulation across varying delay
periods can be seen in Figures 2E and 2F. The performance
change, expressed as change in percent correct, was larger
for longer delays, up to 5% improvement, than for shorter delays.
However, the variability of the measure is a function of the
percent correct, and if the percent correct is closer to 100, the
variance is lower. To enable a fair comparison of the perfor-
2 Current Biology 27, 1–7, September 11, 2017
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Figure 1. Task Paradigm and Electrical
Stimulation Position in the Brain
(A) Macaque monkeys were trained to interact with
a touchscreen. The first step in the task is touching
the cue.
(B) After the cue is touched, the screen blanks
during the delay period, followed by presentation
of two potential matches. Three colors were used
in training, and two were randomly selected as
cue/match and distractor on each trial.
(C) Stimulation of the nucleus basalis (NB) of
Meynert was used in experimental sessions. The
position of the nucleus is shown by the dotted
yellow rectangle, with the curved arrows approxi-
mating the pathways from the nucleus to neo-
cortices [13].
(D) An MRI coronal section of a Rhesus monkey
brain with the implantation target outlined in red.
The MRI was taken from the Macaque Scalable
Brain Atlas [14] (M.F. Dubach, and D.M. Bowden,
2009, Soc. Neurosci., abstract).
See also Figures S1 and S2.
mance change caused by stimulation as
a function of delay length, we converted
these percent change differences to
changes in signal size using the signal
detection theory metric d’ (see STAR
Methods), which measured how far
above the noise the signal size, or mem-
ory, rose. This metric inferred the signal size from the behavioral
responses, and it was normalized by the noise standard devia-
tion. Using either signal detectability metrics or percent correct,
no statistically significant changes in stimulation effect were
found as a function of delay length using ANOVAs (p > 0.1).
Stimulation improved performance across the entire working
memory delay curve, and the change in d’ ranged from 0.13 to
0.39, with an average at the longest delay of 0.16. Alternately,
if the animals were at zero delay and 96% correct, improvements
were 1.4%–1.7%, while at longer delays animals improved
3.6%–4% above 75%. Changes of this magnitude reached
statistical significance in two to four behavioral sessions.
Insight into the speed with which these effects occurred after
stimulation began was determined by analyzing the perfor-
mance shortly after a condition change. If effects appeared
slowly, then performance shortly after stimulation began should
be lower than performance later in the trial block. Similarly, if
the effects decreased slowly once stimulation stopped, the
no-stimulation blocks should have had lower performance later
in the block. Performance compared to trial position in the block
is shown in Figures S3A and S3B for continuous and intermit-
tent stimulation. No significant trends based on position within
the blocks were found, which indicated that effects occurred
in less than 3 min, which was the average time to complete
ten trials.
The Interactions of Stimulation Effects with Cholinergic
Pharmacology
To test the hypothesis that the improvement of working
memory performance after intermittent stimulation depended
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021

A 90 B 90 parisons relative to the control condition were significant (bino-

CONTINUOUS STIM INTERMITTENT STIM mial test, n > 3,000 for each condition, p < 0.005 each). The result
Performance (%)

Performance (%)
suggested that intermittent stimulation achieved a ceiling effect
on cholinergic pathways. It also suggested that continuous stim-
80 80 ulation degraded behavior by decreasing acetylcholine levels.
A further test of the cholinergic basis of effects was conducted
using acetylcholine receptor antagonists. Mecamylamine was
70 70 used to block nicotinic acetylcholine receptors in one set of
Ctrl StimCtrl Stim Ctrl Stim Ctrl Stim experiments, and scopolamine was used to block muscarinic
Monkey CH Monkey DI Monkey CH Monkey DI acetylcholine receptors in the second set of experiments. In
either case (Figure 3B), intermittent stimulation did not improve
C D
82 1200 TOTAL PULSES 82 80 PULSES/SEC performance (p > 0.1). The result suggested that both receptor
Performance (%)

Performance (%)

subtypes were necessary for the performance increase caused

78 78 by intermittent stimulation. As shown in Figure 3C, administra-
tion of mecamylamine and scopolamine alone reproduced
74 74 known effects [16, 17]. Scopolamine impaired working memory
(binomial test, n = 550, p < 0.001). Mecamylamine, at the dose
70 70 tested, improved working memory (binomial test, n = 2,700,
p < 0.001).
10 20 40 60 10 20 40 60
Stimulation period Stimulation period Donepezil is reported to cause side effects including nausea,
per minute (s) per minute (s) which limits its clinical efficacy in some patients [8, 18, 19]. We
plotted food consumption of the animals under control condi-
E F
Monkey CH Monkey DI tions, days in which intermittent stimulation was delivered in
100 100
blocks, and days in which Donepezil was given. The results (Fig-
Performance (%)

Performance (%)

90 Control 90 Control ure 3D) showed a significant effect of condition (unbalanced

Stim Stim
80 80
70 70
60 60
0 3 10 20 0 1 3 5
Delay time (s) Delay time (s)
Figure 2. Effects of NB Stimulation on Working Memory Behavior
(A) The impact of continuous stimulation on performance. Monkeys performed
the task with concurrent 80 Hz continuous stimulation or under a control
condition.
(B) The impact of intermittent stimulation on performance. The gray bars
indicate an intermittent stimulation condition distinct from the continuous
stimulation condition indicated by empty bars in (A).
(C) Both monkeys were tested with 1,200 stimulation pulses per minute,
delivered in 10 s (120 Hz), 20 s (60 Hz), 40 s (30 Hz), or 60 s (20 Hz). The
performance under the control (no stimulation) condition for this week is
indicated with the dashed line, and its standard error is 1.8%.
(D) Performance as both monkeys were tested with 80 pulses per second for
different fractions of a minute. The standard error in control condition is 1.7%.
(E) Delay performance curve with and without intermittent stimulation in
animal CH.
(F) Delay performance curve with and without intermittent stimulation in animal
DI. Up to 1,000 trials were used to determine each point.
Error bars indicate the standard error of the mean. See also Figures S3 and S4.
on cholinergic pathways, stimulation was delivered with and
without the cholinesterase inhibitor Donepezil. When the Done-
pezil dose was combined with intermittent stimulation, perfor-
mance was no different from Donepezil alone, or stimulation
alone, as shown in Figure 3A. The impact of combining Donepezil
with 80 pulses per second continuous stimulation is shown on
the right in Figure 3A. Continuous stimulation significantly
impaired performance, but Donepezil reversed this impairment
and resulted in supranormal performance. In Figure 3A, all com-
ANOVA; F(2,56) = 3.4 for condition, p = 0.04; F(1,56) = 1.85 for
animal, p = 0.179). Post hoc testing found that monkeys
consumed less food on days in which Donepezil was given rela-
tive to days in which stimulation was given (t test, t = 2.81,
df = 35, p < 0.009).
The reaction times of animals CH and DI to the cue and match
are shown in Figure S4. These reaction times were measured in
the same data shown in Figures 2E and 2F. Under no conditions
were there significant effects of stimulation on reaction times
(paired t tests, p > 0.1 for all), which argued against stimulation
causing a general effect on arousal or alertness.
Prior work stimulating NB in anesthetized animals has demon-
strated consistent spectrum shifts of cortical electroencephalo-
gram (EEG) recordings [20–23]. The low-frequency portions
of the EEG, under 12 Hz, tend to decrease in energy, or de-
synchronize, while the higher-frequency gamma range increases
under some experimental conditions. We recorded the subcor-
tical local field potential (LFP), via the stimulating electrode,
immediately before and after stimulation during periods of inter-
mittent and continuous stimulation in the awake animal. As
shown in Figures S2A–S2D, consistent reductions in LFP power
spectrum occurred as a result of intermittent stimulation. Similar
effects occurred after continuous stimulation at 80 Hz, shown in
Figures S2E and S2F. All effects were significant (t tests,
p < 0.001, frequency bands <4 Hz, 4–8 Hz, 8–12 Hz, 12–20 Hz,
20–40 Hz).
Long-Term Changes Caused by Stimulation
Unexpected improvements in working memory performance
were observed in the three animals that were tested with stimu-
lation of NB for months. In each animal, the delay interval was
adjusted to maintain performance between 75% and 80%. The
first indication of this change is shown in Figure 4A, in animal
CH. The working memory delay duration had been increased

Current Biology 27, 1–7, September 11, 2017 3

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021
A B 90
Performance (%)
80 80
70
70 60
er n th t n th Mec Mec+St Sc Sc+St
Int Do Bo Con Do Bo
C D
1.25
Food consumed
Performance (%)
90
80 1
70
Figure 3. Pharmacological Interactions with Stimulation, and
Effects on the Animals’ Appetites
(A) On the left, the experimental conditions of intermittent stimulation (Inter) are
compared to control (dashed line), Donepezil (Don), and both stimulation and
Donepezil (Both). On the right, in a separate experiment, the same compari-
sons are applied with continuous stimulation (Cont). The gray bars indicate
intermittent stimulation involved. Standard errors for control conditions
(dashed lines) on the left and right are 1% and 1.1%, respectively.
(B) Effects of stimulation and acetylcholine receptor blockers. Mecamylamine
(Mec) was given to animals prior to behavior, and working memory perfor-
mance was assessed during intermittent stimulation and without stimulation.
Scopolamine (Sc) was similarly tested. Average of the two animals tested in
each condition is displayed. Dashed line shows the baseline performance in
control condition, and standard error is 1.6%.
(C) Effects of Mecamylamine and Scopolamine. Animals performed behavior
in control condition (left Ctrl), then were given Mecamylamine (Mec), or
Scopolamine (Sc). Then animals performed washout behavior (right Ctrl, 72 hr
after drug application).
(D) Effects of stimulation and Donepezil on appetite. Individual animal con-
sumption under all conditions was normalized by mean control consumption
to allow pooling across animals for this plot. Food consumption data were
collected from animals CH and DI.
Error bars indicate the standard error of the mean.
from 6 s to 30 s over 9 weeks. Animal CH was improving shortly
before stimulation was introduced, so further improvement could
have been caused by a long-term stimulation effect or a contin-
uation of an improvement trend.
Figures 2E and 2F give a chance to convert stimulation effects
into equivalent changes in delay length. If the working memory
delay chosen caused an animal to perform close to 75% correct,
and the delay were halved, performance would have increased
5%–6%. This increase is slightly larger than the impact of inter-
mittent stimulation shown in Figure 2B. A 5-fold decrease in the
length of the delay period corresponds to a 12%–14% increase
in percent correct. In signal-detection terms, a 14% increase in
performance, from 75% to 89%, corresponds to a d’ change of
0.78, which is 4.9 times greater than the average acute d’ change
caused by stimulation at long delays in Figures 2E and 2F.
Animal DI was allowed to reach asymptotic performance in the
absence of stimulation for a longer period, and performance was
stable over 11 weeks with a working memory delay of 2–3 s. Over
4 Current Biology 27, 1–7, September 11, 2017
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
the ensuing 5 months, the working memory delay increased from
3 s to 10 s, which is shown in Figure 4B. Under the null hypoth-
esis that working memory ability was stable, this change
was highly significant (binomial model, see STAR Methods,
p < 105). Animal PU (Figure 4C) performed the task for 17 weeks
prior to initiating stimulation. Its performance was stable in this
pre-stimulation period. In the following 7 weeks, the 1.5 s delay
increased to 8 s, more than 5-fold, which again was significant
(p < 105). The results from animals DI and PU showed that stim-
ulation increased working memory duration. All of the perfor-
mance data used in this analysis were collected under control
conditions, i.e., without the acute boost in performance shown
in Figure 2B. For all data shown in Figure 4, the stimulation deliv-
ered in the several-month-long period of stimulation was mixed
while the experiments of Figures 2, 3, and 4 were conducted.
DISCUSSION
The delayed match-to-sample task was used because it is a
well-studied test of executive function in the nonhuman primate
[24]. It embodies several mnemonic processes that are critical to
human cognition, including attention, stimulus discrimination,
encoding, and memory for recent events [25]. The task is sensi-
tive to cholinergic modulation, as indicated by its enhancement
by cholinesterase inhibitors [26, 27], and to cholinergic receptor
agonists [28, 29], and it is sensitive to impairments of perfor-
mance by cholinergic antagonists [17, 30]. Intermittent stimula-
tion of NB led to short-term improvement in the task, whereas
continuous stimulation suppressed performance. Pharmacolog-
ical administration of cholinesterase inhibitors and acetylcholine
receptor blockers demonstrated that the effects occur through
modulation of acetylcholine levels and require activation of
both classes of acetylcholine receptors. Unlike cholinesterase
inhibitors, appetite is not suppressed by cholinergic modulation
through deep brain stimulation. Even though stimulation was
delivered in varied amounts each week, each of three animals
increased their working memory delay duration 3- to 5-fold in
the months after stimulation began. Two of those animals had
extended periods of stable task performance before stimulation
began. Future work may address the specificity of these effects,
because working memory only comprises one component of
executive function.
A major unexpected finding in this study is the form of stimu-
lation that produced optimal results: 60 pulses per second deliv-
ered for 20 s and followed by 40 s without stimulation. Deep brain
stimulation’s mechanisms have been a subject of debate [31].
Locally, a depolarization block and possibly activation of local
inhibitory circuits dominates to prevent somatic activation,
whereas recording and imaging evidence suggests that axonal
efferent outflow is activated by each stimulation pulse. Accord-
ingly, the NB axons should be activated each time a pulse is
delivered. Why the intermittent pattern results in more cortical
acetylcholine, as inferred from pharmacological experiments,
than continuous stimulation is an obvious question for subse-
quent studies. We speculate that intermittent, not continuous,
neuronal stimulation results in higher levels of acetylcholine
synthesis and release. The rate-limiting step in acetylcholine
synthesis is the uptake of choline from the synaptic cleft via
the high-affinity choline transporter hCHT [32]. Before release,
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021

A B Figure 4. Changes in Working Memory over

Monkey CH Monkey DI Time
50 100 15 100
(A) The working memory delay and performance

Performance (%)
Performance (%)
40 80 12 80 for animal CH. Delay and performance are plotted

Delay time (s)

Delay time (s)

against time relative to initiating intermittent

30 60 9 60 stimulation. The delay was chosen to maintain
performance range. The vertical line before week 6
20 40 6 40 indicates when intermittent stimulation was intro-
duced. Left axis illustrates the delay, which is
10 20 3 20 marked by circle symbols, while the right axis
illustrates the performance, plotted as triangles. All
0 0 0 0 data in this figure are defined under no-stimulation
0 5 10 15 0 5 10 15 20 25 30 35
control conditions.
Time(week) Time(week)
(B) Data for animal DI.
C
15 Monkey PU 100 (C) Data for animal PU.
Performance (%)

12 80
Delay time (s)

9 60
6 Delay Time 40
Performance (%)
3 20
0 0
0 1 10 15 20 25
Time(week)
the proton-exchanging acetylcholine transporter VAChT brings
more acetylcholine into the vesicle, and the high-affinity choline
transporter, which would transport choline to the cytosol, is
turned off by the low vesicular pH [33, 34]. After release, the
vesicular membrane becomes part of the cell membrane, and
the decreased proton concentrations turn the acetylcholine
transporter off and the choline transporter on. If frequent vesic-
ular release from high-frequency stimulation alters post-release
activation of the high-affinity choline transporter by changing
local pH, overall acetylcholine synthesis and release could be
reduced.
A similar curiosity is that the stimulation does not have any
temporal relation to the behavioral task or to known cholinergic
kinetics other than the recycle time [35, 36]. Donepezil’s effects
on behavioral performance are arguably boosting the efficacy
after normal cholinergic release, which would preserve the
normal temporal function of acetylcholine [26]. However, some
experiments with subtype-selective agonists also improve work-
ing memory function [37–39], and these would similarly lack any
temporal specificity.
Improvement in cognition from cholinergic system activation
could be caused by cholinergic modulation of neurons or by
cholinergic-induced increases in blood flow, and these mecha-
nisms are not mutually exclusive. In the neocortex, cholinergic
activation of neurons leads to increased thalamic input into layer
IV while suppressing other cortical inputs [40, 41]. This, at least
partly, explains how the activation of the cholinergic system
desynchronizes the EEG and local field potential (Figure S2),
which was also reported by previous studies [21, 22, 42, 43].
The coupling of desynchronization to behavioral improvements
suggests that direct cholinergic modulation of neurons improves
the behavior.
At the same time, we cannot exclude the possibility that indi-
rect mechanisms such as increases of blood flow played a role in
the improved cognitive performance. In
the brain, acetylcholine is acting on both
neurons and glia adjacent to blood ves-
sels [44, 45]. Cholinesterase inhibitors in
humans cause an increase of 10%–15%
in cerebral blood flow [46], and this blood
flow is typically reduced in Alzheimer’s
patients [47–49]. Reductions in cerebral
blood flow likely contribute to the cogni-
tive impairments [50]. Future work may shed light on the relative
contributions of neural modulation and blood flow regulation in
these effects, as well as on the question of their independence.
The long-term effects observed indicate that working memory
duration is increased 3- to 5-fold in each of three animals tested
over periods of up to 5 months. The changes even occurred after
long stable performance periods prior to stimulation. These re-
sults suggest that cholinergic modulation combined with execu-
tive function behavior causes brain plasticity to support behavioral
improvements. Prior work has associated phasic stimulation of
NB with neural plasticity in the sensory cortex [20, 22, 51].
Several clinical trials have been conducted on deep brain stim-
ulation of cholinergic pathways [52, 53]. None have found
consistent cognitive improvement using continuous stimulation.
Similarly, animal work with continuous stimulation finds weak to
no effects on learning and memory [11]. We measured analogous
effects using continuous stimulation, which also suggest that
high-frequency continuous stimulation suppresses acetylcho-
line levels. Our data show that making the stimulation intermit-
tent is critical to turning cognitive suppression into cognitive
enhancement. Animals had stronger appetites while receiving
stimulation than while receiving Donepezil. The use of the inter-
mittent stimulation parameters outlined in this work reach effi-
cacies as high as those of high doses of cholinesterase inhibitors
without peripheral side effects and point toward new candidates
for therapeutic treatment of Alzheimer’s dementia.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING

Current Biology 27, 1–7, September 11, 2017 5

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Animal Model
d METHOD DETAILS
B Behavioral Task
B Surgery and Deep Brain Stimulation
B Pharmacology and Stimulation Studies Administration
d QUANTIFICATION AND STATISTICAL ANALYSIS
B d’ Calculation
B Binomial Statistical Model
d DATA AND SOFTWARE AVAILABILITY
SUPPLEMENTAL INFORMATION
Supplemental Information includes four figures and can be found with this
article online at http://dx.doi.org/10.1016/j.cub.2017.07.021.
AUTHOR CONTRIBUTIONS
Conceptualization and Funding Acquisition: D.T.B and C.C.; Project Adminis-
tration: D.T.B.; Pharmacology Methodology: A.V.T. and P.M.C.; Investigation
and Analysis: R.L., J.C., and D.T.B. performed experiments; Software: R.L.
and D.T.B.; Writing: D.T.B., C.C., and R.L.; Review and Editing: all authors.
ACKNOWLEDGMENTS
Veterinary support provided by N. Rodriguez and P. Otovic. Artwork in Figure 1
was done by D. Bliss. This work funded by NIH grant 5R01MH097695.
Received: April 18, 2017
Revised: June 7, 2017
Accepted: July 11, 2017
Published: August 17, 2017
REFERENCES
1. Bartus, R.T. (2000). On neurodegenerative diseases, models, and treat-
ment strategies: lessons learned and lessons forgotten a generation
following the cholinergic hypothesis. Exp. Neurol. 163, 495–529.
2. Terry, A.V., Jr., and Buccafusco, J.J. (2003). The cholinergic hypothesis of
age and Alzheimer’s disease-related cognitive deficits: recent challenges
and their implications for novel drug development. J. Pharmacol. Exp.
Ther. 306, 821–827.
3. Sarter, M., and Bruno, J.P. (1997). Cognitive functions of cortical acetyl-
choline: toward a unifying hypothesis. Brain Res. Brain Res. Rev. 23,
28–46.
4. Birks, J.S. (2006). Cholinesterase inhibitors for Alzheimer’s disease. In
Cochrane Database of Systematic Reviews, J.S. Birks, ed. (Chichester,
UK: John Wiley & Sons, Ltd).
5. Hogan, D.B., Bailey, P., Carswell, A., Clarke, B., Cohen, C., Forbes, D.,
Man-Son-Hing, M., Lanctôt, K., Morgan, D., and Thorpe, L. (2007).
Management of mild to moderate Alzheimer’s disease and dementia.
Alzheimers Dement. 3, 355–384.
6. Qaseem, A., Snow, V., Cross, J.T., Jr., Forciea, M.A., Hopkins, R., Jr.,
Shekelle, P., Adelman, A., Mehr, D., Schellhase, K., Campos-Outcalt, D.,
et al.; American College of Physicians/American Academy of Family
Physicians Panel on Dementia (2008). Current pharmacologic treatment
of dementia: a clinical practice guideline from the American College of
Physicians and the American Academy of Family Physicians. Ann.
Intern. Med. 148, 370–378.
7. Sabbagh, M., and Cummings, J. (2011). Progressive cholinergic decline in
Alzheimer’s Disease: consideration for treatment with donepezil 23 mg in
patients with moderate to severe symptomatology. BMC Neurol. 11, 21.
8. Farlow, M.R., Salloway, S., Tariot, P.N., Yardley, J., Moline, M.L., Wang,
Q., Brand-Schieber, E., Zou, H., Hsu, T., and Satlin, A. (2010).
Effectiveness and tolerability of high-dose (23 mg/d) versus standard-
6 Current Biology 27, 1–7, September 11, 2017
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
dose (10 mg/d) donepezil in moderate to severe Alzheimer’s disease:
A 24-week, randomized, double-blind study. Clin. Ther. 32, 1234–1251.
9. Hendry, S.H.C., Jones, E.G., Killackey, H.P., and Chalupa, L.M. (1987).
Choline acetyltransferase-immunoreactive neurons in fetal monkey cere-
bral cortex. Brain Res. 465, 313–317.
10. Mesulam, M.M., Mufson, E.J., Levey, A.I., and Wainer, B.H. (1983).
Cholinergic innervation of cortex by the basal forebrain: cytochemistry
and cortical connections of the septal area, diagonal band nuclei, nucleus
basalis (substantia innominata), and hypothalamus in the rhesus monkey.
J. Comp. Neurol. 214, 170–197.
11. Lee, J.E., Jeong, D.U., Lee, J., Chang, W.S., and Chang, J.W. (2016). The
effect of nucleus basalis magnocellularis deep brain stimulation on mem-
ory function in a rat model of dementia. BMC Neurol. 16, 6.
12. Pinto, L., Goard, M.J., Estandian, D., Xu, M., Kwan, A.C., Lee, S.-H.,
Harrison, T.C., Feng, G., and Dan, Y. (2013). Fast modulation of visual
perception by basal forebrain cholinergic neurons. Nat. Neurosci. 16,
1857–1863.
13. Selden, N.R., Gitelman, D.R., Salamon-Murayama, N., Parrish, T.B., and
Mesulam, M.-M. (1998). Trajectories of cholinergic pathways within the
cerebral hemispheres of the human brain. Brain 121, 2249–2257.
14. Rohlfing, T., Kroenke, C.D., Sullivan, E.V., Dubach, M.F., Bowden, D.M.,
Grant, K.A., and Pfefferbaum, A. (2012). The INIA19 template and
NeuroMaps atlas for primate brain image parcellation and spatial normal-
ization. Front. Neuroinform. 6, 27.
15. McIntyre, C.C., Grill, W.M., Sherman, D.L., and Thakor, N.V. (2004).
Cellular effects of deep brain stimulation: model-based analysis of activa-
tion and inhibition. J. Neurophysiol. 91, 1457–1469.
16. Terry, A.V., Buccafusco, J.J., and Prendergast, M.A. (1999). Dose-specific
improvements in memory-related task performance by rats and aged
monkeys administered the nicotinic-cholinergic antagonist mecamyl-
amine. Drug Dev. Res. 47, 127–136.
17. Buccafusco, J.J., Terry, A.V., Jr., Webster, S.J., Martin, D., Hohnadel, E.J.,
Bouchard, K.A., and Warner, S.E. (2008). The scopolamine-reversal para-
digm in rats and monkeys: the importance of computer-assisted operant-
conditioning memory tasks for screening drug candidates.
Psychopharmacology (Berl.) 199, 481–494.
18. Yoshida, T., Ha-Kawa, S., Yoshimura, M., Nobuhara, K., Kinoshita, T., and
Sawada, S. (2007). Effectiveness of treatment with donepezil hydrochlo-
ride and changes in regional cerebral blood flow in patients with
Alzheimer’s disease. Ann. Nucl. Med. 21, 257–265.
19. Han, S.-H., Lee, J.-H., Kim, S.Y., Park, K.W., Chen, C., Tripathi, M., Dash,
A., and Kubota, N. (2017). Donepezil 23 mg in Asian patients with moder-
ate-to-severe Alzheimer’s disease. Acta Neurol. Scand. 135, 252–256.
20. Bakin, J.S., and Weinberger, N.M. (1996). Induction of a physiological
memory in the cerebral cortex by stimulation of the nucleus basalis.
Proc. Natl. Acad. Sci. USA 93, 11219–11224.
21. Bjordahl, T.S., Dimyan, M.A., and Weinberger, N.M. (1998). Induction of
long-term receptive field plasticity in the auditory cortex of the waking
guinea pig by stimulation of the nucleus basalis. Behav. Neurosci. 112,
467–479.
22. Kilgard, M.P., and Merzenich, M.M. (1998). Cortical map reorganization
enabled by nucleus basalis activity. Science 279, 1714–1718.
23. McLin, D.E., 3rd, Miasnikov, A.A., and Weinberger, N.M. (2002). Induction
of behavioral associative memory by stimulation of the nucleus basalis.
Proc. Natl. Acad. Sci. USA 99, 4002–4007.
24. Riley, M.R., and Constantinidis, C. (2016). Role of prefrontal persistent
activity in working memory. Front. Syst. Neurosci. 9, 181.
25. Rodriguez, J.S., and Paule, M.G. (2009). Working memory delayed
response tasks in monkeys. In Methods Behav. Anal. Neurosci.,
Second Edition, J.J. Buccafusco, ed. (Boca Raton: CRC Press/Taylor &
Francis).
26. Buccafusco, J.J., and Terry, A.V. (2004). Donepezil-induced improvement
in delayed matching accuracy by young and old rhesus monkeys. J. Mol.
Neurosci. 24, 85–91.
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021
27. Callahan, P.M., Hutchings, E.J., Kille, N.J., Chapman, J.M., and Terry,
A.V., Jr. (2013). Positive allosteric modulator of a7 nicotinic-acetylcholine
receptors, PNU-120596 augments the effects of donepezil on learning and
memory in aged rodents and non-human primates. Neuropharmacology
67, 201–212.
28. Terry, A.V., Jr., Plagenhoef, M., and Callahan, P.M. (2016). Effects of the
nicotinic agonist varenicline on the performance of tasks of cognition in
aged and middle-aged rhesus and pigtail monkeys. Psychopharmacology
(Berl.) 233, 761–771.
29. Terry, A.V., Jr., Risbrough, V.B., Buccafusco, J.J., and Menzaghi, F.
(2002). Effects of (+/-)-4-[[2-(1-methyl-2-pyrrolidinyl)ethyl]thio]phenol
hydrochloride (SIB-1553A), a selective ligand for nicotinic acetylcholine
receptors, in tests of visual attention and distractibility in rats and mon-
keys. J. Pharmacol. Exp. Ther. 301, 284–292.
30. Zhou, X., Qi, X.-L., Douglas, K., Palaninathan, K., Kang, H.S., Buccafusco,
J.J., Blake, D.T., and Constantinidis, C. (2011). Cholinergic modulation of
working memory activity in primate prefrontal cortex. J. Neurophysiol. 106,
2180–2188.
31. Vitek, J.L. (2002). Mechanisms of deep brain stimulation: excitation or
inhibition. Mov. Disord. 17 (Suppl 3 ), S69–S72.
32. Okuda, T., Haga, T., Kanai, Y., Endou, H., Ishihara, T., and Katsura, I.
(2000). Identification and characterization of the high-affinity choline trans-
porter. Nat. Neurosci. 3, 120–125.
33. Iwamoto, H., Blakely, R.D., and De Felice, L.J. (2006). Na+, Cl-, and pH
dependence of the human choline transporter (hCHT) in Xenopus oocytes:
the proton inactivation hypothesis of hCHT in synaptic vesicles.
J. Neurosci. 26, 9851–9859.
34. Nguyen, M.L., Cox, G.D., and Parsons, S.M. (1998). Kinetic parameters for
the vesicular acetylcholine transporter: two protons are exchanged for one
acetylcholine. Biochemistry 37, 13400–13410.
35. Trabucchi, M., Cheney, D.L., Hanin, I., and Costa, E. (1975). Application of
principles of steady-state kinetics to the estimation of brain acetylcholine
turnover rate: effects of oxotremorine and physostigmine. J. Pharmacol.
Exp. Ther. 194, 57–64.
36. Cheney, D.L., Trabucchi, M., Racagni, G., Wang, C., and Costa, E. (1974).
Effects of acute and chronic morphine on regional rat brain acetylcholine
turnover rate. Life Sci. 15, 1977–1990.
37. Terry, A.V., Jr., Buccafusco, J.J., Borsini, F., and Leusch, A. (2002).
Memory-related task performance by aged rhesus monkeys administered
the muscarinic M(1)-preferring agonist, talsaclidine. Psychopharmacology
(Berl.) 162, 292–300.
38. Castner, S.A., Smagin, G.N., Piser, T.M., Wang, Y., Smith, J.S., Christian,
E.P., Mrzljak, L., and Williams, G.V. (2011). Immediate and sustained im-
provements in working memory after selective stimulation of a7 nicotinic
acetylcholine receptors. Biol. Psychiatry 69, 12–18.
39. Buccafusco, J.J., and Jackson, W.J. (1991). Beneficial effects of nicotine
administered prior to a delayed matching-to-sample task in young and
aged monkeys. Neurobiol. Aging 12, 233–238.
40. Metherate, R., and Ashe, J.H. (1993). Nucleus basalis stimulation facili-
tates thalamocortical synaptic transmission in the rat auditory cortex.
Synapse 14, 132–143.
41. Aramakis, V.B., and Metherate, R. (1998). Nicotine selectively enhances
NMDA receptor-mediated synaptic transmission during postnatal devel-
opment in sensory neocortex. J. Neurosci. 18, 8485–8495.
42. Bakin, J.S., South, D.A., and Weinberger, N.M. (1996). Induction of recep-
tive field plasticity in the auditory cortex of the guinea pig during instru-
mental avoidance conditioning. Behav. Neurosci. 110, 905–913.
43. McLin, D.E., 3rd, Miasnikov, A.A., and Weinberger, N.M. (2002). The
effects of electrical stimulation of the nucleus basalis on the electroen-
cephalogram, heart rate, and respiration. Behav. Neurosci. 116, 795–806.
44. Vaucher, E., Borredon, J., Seylaz, J., and Lacombe, P. (1995).
Autoradiographic distribution of cerebral blood flow increases elicited by
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
stimulation of the nucleus basalis magnocellularis in the unanesthetized
rat. Brain Res. 691, 57–68.
45. Vaucher, E., Linville, D., and Hamel, E. (1997). Cholinergic basal forebrain
projections to nitric oxide synthase-containing neurons in the rat cerebral
cortex. Neuroscience 79, 827–836.
46. Lojkowska, W., Ryglewicz, D., Jedrzejczak, T., Minc, S., Jakubowska, T.,
Jarosz, H., and Bochynska, A. (2003). The effect of cholinesterase inhibi-
tors on the regional blood flow in patients with Alzheimer’s disease and
vascular dementia. J. Neurol. Sci. 216, 119–126.
47. Sharp, P., Gemmell, H., Cherryman, G., Besson, J., Crawford, J., and
Smith, F. (1986). Application of iodine-123-labeled isopropylamphetamine
imaging to the study of dementia. J. Nucl. Med. 27, 761–768.
48. Jagust, W.J., Budinger, T.F., and Reed, B.R. (1987). The diagnosis of de-
mentia with single photon emission computed tomography. Arch. Neurol.
44, 258–262.
49. Schuff, N., Matsumoto, S., Kmiecik, J., Studholme, C., Du, A., Ezekiel, F.,
Miller, B.L., Kramer, J.H., Jagust, W.J., Chui, H.C., and Weiner, M.W.
(2009). Cerebral blood flow in ischemic vascular dementia and
Alzheimer’s disease, measured by arterial spin-labeling magnetic reso-
nance imaging. Alzheimers Dement. 5, 454–462.
50. Ruitenberg, A., den Heijer, T., Bakker, S.L.M., van Swieten, J.C.,
Koudstaal, P.J., Hofman, A., and Breteler, M.M.B. (2005). Cerebral hypo-
perfusion and clinical onset of dementia: the Rotterdam Study. Ann.
Neurol. 57, 789–794.
51. Bao, S., Chang, E.F., Davis, J.D., Gobeske, K.T., and Merzenich, M.M.
(2003). Progressive degradation and subsequent refinement of acoustic
representations in the adult auditory cortex. J. Neurosci. 23, 10765–
10775.
52. Laxton, A.W., Tang-Wai, D.F., McAndrews, M.P., Zumsteg, D., Wennberg,
R., Keren, R., Wherrett, J., Naglie, G., Hamani, C., Smith, G.S., and
Lozano, A.M. (2010). A phase I trial of deep brain stimulation of memory
circuits in Alzheimer’s disease. Ann. Neurol. 68, 521–534.
53. Kuhn, J., Hardenacke, K., Lenartz, D., Gruendler, T., Ullsperger, M.,
Bartsch, C., Mai, J.K., Zilles, K., Bauer, A., Matusch, A., et al. (2015).
Deep brain stimulation of the nucleus basalis of Meynert in Alzheimer’s de-
mentia. Mol. Psychiatry 20, 353–360.
54. Mesulam, M.-M., Mufson, E.J., Levey, A.I., and Wainer, B.H. (1984). Atlas
of cholinergic neurons in the forebrain and upper brainstem of the ma-
caque based on monoclonal choline acetyltransferase immunohisto-
chemistry and acetylcholinesterase histochemistry. Neuroscience 12,
669–686.
55. Gielow, M.R., and Zaborszky, L. (2017). The input-output relationship of
the cholinergic basal forebrain. Cell Rep. 18, 1817–1830.
56. Paxinos, G., and Huang, X.F.A.W.T. (2000). The Rhesus Monkey Brain in
Stereotaxic Coordinates (San Diego: Academic Press).
57. McCairn, K.W., and Turner, R.S. (2015). Pallidal stimulation suppresses
pathological dysrhythmia in the parkinsonian motor cortex.
J. Neurophysiol. 113, 2537–2548.
58. McCairn, K.W., and Turner, R.S. (2009). Deep brain stimulation of the
globus pallidus internus in the parkinsonian primate: local entrainment
and suppression of low-frequency oscillations. J. Neurophysiol. 101,
1941–1960.
59. Tsukada, H., Nishiyama, S., Fukumoto, D., Ohba, H., Sato, K., and
Kakiuchi, T. (2004). Effects of acute acetylcholinesterase inhibition on
the cerebral cholinergic neuronal system and cognitive function:
Functional imaging of the conscious monkey brain using animal PET in
combination with microdialysis. Synapse 52, 1–10.
60. Green, D.M., and Swets, J.A. (1966). Signal Detection Theory and
Psychophysics (New York: Wiley).
61. Smith, D.G., and Duncan, M.J.J. (2004). Testing theories of recognition
memory by predicting performance across paradigms. J. Exp. Psychol.
Learn. Mem. Cogn. 30, 615–625.
Current Biology 27, 1–7, September 11, 2017 7
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021

STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Chemicals, Peptides, and Recombinant Proteins
Donepezil Hydrochloride Memory Pharmaceuticals, N/A
Montvale, NJ
Scopolamine Hydrobromide Sigma Aldrich CAS # 114-49-8
Mecamylamine Hydrochloride Sigma Aldrich CAS # 826-39-1
Deposited Data
http://dx.doi.org/10.6084/m9.figshare.5039153 N/A N/A
Experimental Models: Organisms/Strains
Macaca mulatta Mannheimer Foundation N/A
Software and Algorithms
MATLAB R2016a MathWorks N/A
DTS_Gracewood PBTLI http://www.pbtli.com
d’ calculation Smith and Duncan 2004 N/A
Binomial model This paper http://dx.doi.org/10.6084/m9.figshare.5039153

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to, and will be fulfilled by, the Lead Contact, David
T Blake (dblake.mcg@gmail.com).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animal Model
Five rhesus monkeys (Macaca mulatta) were used in this study. Two initial animals (K and S) were used to optimize the placing of the
stimulation electrodes. These were a 7 years old male and a 12 years old female weighing 7 and 6 kg. The three subsequent animals
(animals CH, DI, and PU), used for data in Figures 2–4, were male, 6 years old and weighed 8 11 kg. The monkeys were pair-housed
in 12-h light/dark cycle rooms. All monkeys were task naive at the start of the experiments. All animals were chaired during task
performance, with one arm restrained to both establish consistency in the arm used to perform the task, and to prevent spinning
in the primate chair. Animals were not head fixed during task performance.
All animal studies comply with the Guide for the Care and Use of Animals, 8th Edition, and were approved by the IACUC at Augusta
University.

METHOD DETAILS

Behavioral Task
All behavioral software was based on prior work [28], and available from PBTLI Prime Behavior Testing Laboratories, Inc, http://www.
pbtli.com/, Augusta, GA. The behavioral task, shown in Figure 1, initiated with a colored square cue in the center-top part of the
touchscreen. After the animal touched the square, the cue disappeared, and the screen remained blank for the delay period. In
the match phase, two colored squares were presented in the lower right, and lower left, of the touchscreen. One of the two squares,
randomized in location, had the same color as the cue square. Correctly touching the target square resulted in delivery of food slurry
reward, while incorrect responses resulted in doubling the inter trial period as a brief timeout. The animal reaction times were not
strongly constrained. Animals had 10 s to make a response. Percent correct and reaction times were recorded. Animals required
several months of daily training to achieve stable performance at the task. Three colors (red, blue and yellow) were used. In each trial,
two were randomly chosen as cue/match, and as the distractor.
Dates of data collection for each animal for each figure.

e1 Current Biology 27, 1–7.e1–e4, September 11, 2017

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021

Experiment Animal CH Animal DI Animal PU

Figures 2A, 2B, and S3 8/15-10/15 11/15-1/16
Figures 2C and 2D 6/16-7/16 6/16-6/16
Figures 2E, 2F, and S4 10/15-11/15 11/15-12/15
Figure 3 2/16-8/16 6/16-8/16 8/2016
Figure 4 7/15-11/15 3/15-11/15 12/15-8/16

Surgery and Deep Brain Stimulation

Surgery was performed in a sterile surgical suite under isoflurane anesthesia monitored by a clinical veterinarian. Stereotaxic
measurements were made to target implantation. In our first two animals, targeting was 8mm lateral, 12 and 14 mm anterior interaural
respectively, and 29 mm below the cortical surface in a vertical penetration. The final three animals were implanted at 16 mm anterior
interaural with the same lateral and depth coordinates. The final lateral and anterior coordinates, and depth, were chosen to corre-
spond to the center of the anterior portion of the Nucleus Basalis of Meynert, which would contain the highest density of projections to
the prefrontal cortex [10, 54, 55]. At the targeted brain position, the stimulation region, which is estimated to be a 4 mm diameter
sphere [31], should include the Nucleus Basalis of Meynert as well as portions of the anterior amygdala including the central nucleus,
the anterior commissure, and the inferior internal globus pallidus [56]. A skin flap was performed, and two holes were drilled above the
location of each implant. A small cylindrical titanium chamber (5-mm inner diameter and 7-mm outer) was mounted on the cranium
and the skin was sutured into place leaving the chamber exposed. A 26 ga. hypodermic guide tube was lowered and the tip advanced
5 mm below the dura mater. The electrode was inserted into the guide tube, and a stylus was used to push it to the appropriate depth.
The use of the guide tube and stylus followed published methods [57, 58], which are analogous to methods used in human neuro-
surgical deep brain electrode implantation. The guide tube was then raised while the stylus depth maintained. The chamber was
evacuated of fluid, flushed with ceftriaxone, and thereafter fluid evacuated a second time. Silicone was poured into the chamber
to seal the fenestrations in the skull and the inside of the chamber. The electrode was fixed in depth with a drop of cyanoacrylate,
and its rear wire was stripped and soldered to a connector that was fixed on the chamber outer wall. One week after the surgery,
the animals returned to behavioral studies.
The stimulation pulses were created by a multiple functional I/O device (USB-6211, National Instruments, Austin, TX), which was
controlled by a custom programed software. Impedances of electrodes were checked daily prior to each behavioral session so that
stimulation voltages could be tailored to deliver the appropriate currents (200 mA). Continuous stimulation was performed from
20 pulses per second to 120. Intermittent stimulation was initially applied at 60 pulses per second only in-between trials, or only
during trials, which were roughly 3 and 5 s periods. Thereafter, experiments standardized on intermittent stimulation using a one-min-
ute cycle length, and stimulation for 20 s at 60 pulses per second, followed by 40 s with no stimulation. In these one-minute cycle
stimulation experiments, no attempt was made to synchronize the stimulation with the behavioral trials.
Electrodes were custom manufactured in our laboratory based on published specifications [57, 58]. Conductors were 50 mm Pt/Ir,
Teflon-insulated wire (A-M systems, Seattle, WA) embedded within a 30 ga. hypodermic tube, which was encased in a 28 ga. poly-
imide sheath. The wire extended from the end of the sheath into the brain tissue by roughly 1 mm, and the last 0.7 mm of insulation
was stripped to achieve impedances of 5-10 kOhm at 1 kHz. The far end of the electrode was soldered to an extracranial connector
fixed on the chamber outer wall. Preliminary experiments on electrode placement in the two pilot animals tested the effects of short
periods of stimulation on EEG desynchronization. Stimulation was delivered with biphasic, negative first, unipolar 200 mA pulses with
100 mS per phase, and 10 pulses were delivered in 100 msec. Under anesthesia, this resulted in EEG desynchronization when aver-
aged over 10 trials when the electrode was at a depth corresponding to the atlas position of Nucleus Basalis. EEG was recorded with
3 1000 gain from two bone screws at the same M-L position, with one near vertex, and one lateralized. An electrode movement verti-
cally in either direction of more than 1 mm was adequate to make desynchronization not possible using the same protocol. After our
second pilot animal, we determined the depth for our A-P and M-L coordinates was 29 mm below the cortical surface. We further
recovered our electrode track from our second pilot animal, shown in Figure S1. It was implanted 2 mm more caudally than the elec-
trodes from the three monkeys presented in the results. The tip of the electrode was centered 1 mm medial to Nucleus Basalis at
14 mm anterior interaural. After our pilot animals, we implanted to the same depth in all animals, and 2 mm more rostrally, for the
current study, and measured LFP desynchronization in the awake animals using the stimulation electrode as the positive lead
(Figure S2).
In these first two pilot animals, we frequently moved electrodes vertically to new positions. At this point in the study, we understood
that the electrode depth at which EEG desynchronization occurred after brief phasic stimulation precisely matched that at which
80 Hz continuous stimulation resulted in significant performance decrements. The rear ends of the electrodes were always visible,
protruding from the implanted titanium cylinder, which gave an absolute depth measurement.

Pharmacology and Stimulation Studies Administration

We used donepezil, mecamylamine and scopolamine (Sigma-Aldrich, St. Louis, MO) in this study. The drugs were dissolved in a
0.5 cc saline vehicle to an amount specified by the experiments (donepezil: 200 mg/kg, mecamylamine: 300 mg/kg and scopolamine:

Current Biology 27, 1–7.e1–e4, September 11, 2017 e2

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021

6.2 mg/kg). Donepezil was given via I.M. administration 15 min before behavior testing on the mornings in which data was collected.
The rationale for the doses selected for Donepezil was based on previous behavioral and functional brain imaging data (dose range
50-250 mg/kg) in rhesus monkeys [27, 59]. Data from days without donepezil were at least 72 hr after the last administration. Doses
higher than 200 mg/kg could not be used in this study, as the animals administered these doses refused to engage in behavioral trials
for food reward. Nausea is a well-known, clinical side effect of the drug in humans. Each week, control performance was evaluated.
Then, performances under the intermittent stimulation alone, donepezil alone, and donepezil+stimulation were evaluated on succes-
sive days. Similar weekly plans were used for continuous stimulation, and mecamylamine (300 mg/kg) and scopolamine (6.2 mg/kg)
testing. This weekly testing schedule was chosen over an intraday comparison because an intraday design would compare drug
versus no-drug conditions at different times, and with different appetitive motivations. The recovery test sessions from drug were
recorded 72 hr after drug application. The doses of mecamylamine and scopolamine used in this study refer to previous publications
[16, 30].
Initial deep brain stimulation studies, shown in Figures 2A and 2B, provided stimulation in blocks. For continuous stimulation, 100
trials without stimulation were interleaved with 100 trials with stimulation. For intermittent stimulation, blocks were 50 trials. For Fig-
ures 2C and 2D, all stimulation conditions were executed randomly block by block and interleaved with each other. Each block also
contained 50 trials. For Figures 3A–3C, we collected animals’ performance data in control conditions for 1-2 days first, then collected
performance data after administering pharmacological agents. Drug and Stim+Drug conditions were interleaved with each other in
blocks of 50 trials. The recording procedures for the pharmacology experiments were 2-3 times to improve statistical power. For Fig-
ures 2E, 2F, and S4, which includes performance at different delays with and without stimulation, we tested the animals’ performance
with interleaved blocks in control and stimulation conditions. In each block, variable delay times were pseudo-randomly chosen and
arranged in the test sequence. Software was administered directly from PBTLI software while a second computer was used to
generate stimulation pulses. Data for Figure 4 were selected from all available data in which no drug or stimulation condition
occurred.
Delay times used in each experiment.

Experiment Animal CH Animal DI Animal PU

Figures 2A & S3A 10 s 3s
Figures 2B & S3B 13 s 3s
Figures 2C and 2D 6s 7s
Figures 2E, 2F, and S4 0 s, 3 s, 10 s, 20 s 0 s, 1 s, 3 s, 5 s
Figure 3A left 6s 5s
Figure 3A right 6s 10 s
Figure 3B 8s 8s
Figure 3C 10 s 3s

QUANTIFICATION AND STATISTICAL ANALYSIS

d’ Calculation
The signal detection theoretic d’ is the salience of the target relative to the distractor as inferred from the animals’ behavioral choices.
In our two alternative forced
pchoice
ffiffiffi framework, d’ is calculated using only the percent correct with the MATLAB (Mathworks, Natick,
MA) norminv function, as 2  norminvðHit RateÞ [60, 61]. This metric measures the amount of the memory that is retained, on
average. It assumes the memory can be measured in terms of a mean and a noise standard deviation. As time passes, that
mean, or memory, regresses toward zero. The metric measures the magnitude, normalized by the noise standard deviation, of
the mean at the end of the delay period. The following table details how d’ relates to percent correct in the task.

d’ Percent Correct
0.25 57
0.5 64
0.75 70
1.0 76
1.25 81
1.5 85
1.75 89
2.0 92

e3 Current Biology 27, 1–7.e1–e4, September 11, 2017

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 Please cite this article in press as: Liu et al., Intermittent Stimulation of the Nucleus Basalis of Meynert Improves Working Memory in Adult Monkeys,
Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.07.021

Binomial Statistical Model

We relied on a binomial statistic model to test if Nucleus Basalis stimulation, over weeks and months, produced long-term improve-
ments in working memory that could be measured which resulted in ability to perform the task at high performance with a longer delay
interval (Figure 4). Prior to stimulation, animals performed 3000 trials weekly at a performance level averaging 78% correct, with a
delay interval ranging between 1.5-6 s for the three animals. For each animal, we asked what performance improvement would occur
if the delay, at which performance level is 78% correct, were doubled? In each animal’s delay curves, shown in Figures 2E and 2F,
halving the delay would result in changes in percent correct of 5.4 and 5.1 percent. In other words, if we time scaled the graphs in
Figures 2E and 2F to the same percentage correct at twice the delays, animal performance at the original delay would be predicted,
based on the delay curves in Figures 2E and 2F, to increase by just over 5%. We conservatively rounded this percentage to a
5 percent change. Next, the two-tailed probabilities were calculated that a binomial process with a fraction correct of 0.78 over
3000 trials could result in more correct trials than a process with a fraction correct 5 percent higher or lower. This probability is
less than 105. The implication of the binomial model is that if an animal performs at 78% correct for duration 2X, then it would
be expected to perform at 83%, or higher, at duration X, and that this performance would be significantly better than its pre-stimu-
lation performance of 78% at duration X.

DATA AND SOFTWARE AVAILABILITY

Data used to generate these figures are archived at http://dx.doi.org/10.6084/m9.figshare.5039153.

Current Biology 27, 1–7.e1–e4, September 11, 2017 e4

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 14442 • The Journal of Neuroscience, December 26, 2007 • 27(52):14442–14447

Behavioral/Systems/Cognitive

Focusing Effect of Acetylcholine on Neuroplasticity in the

Human Motor Cortex
Min-Fang Kuo,1 Jan Grosch,1 Felipe Fregni,2 Walter Paulus,1 and Michael A. Nitsche1
1Department of Clinical Neurophysiology, Georg-August-University Göttingen, 37075 Göttingen, Germany, and 2Harvard Center for Noninvasive Brain
Stimulation, Boston, Massachusetts 02215

Cholinergic neuromodulation is pivotal for arousal, attention, and cognitive processes. Loss or dysregulation of cholinergic inputs leads
to cognitive impairments like those manifested in Alzheimer’s disease. Such dysfunction can be at least partially restored by an increase
of acetylcholine (ACh). In animal studies, ACh selectively facilitates long-term excitability changes induced by feed-forward afferent
input. Consequently, it has been hypothesized that ACh enhances the signal-to-noise ratio of input processing. However, the neurophys-
iological foundation for its ability to enhance cognition in humans is not well documented. In this study we explore the effects of
rivastigmine, a cholinesterase inhibitor, on global and synapse-specific forms of cortical plasticity induced by transcranial direct current
stimulation (tDCS) and paired associative stimulation (PAS) on 10 –12 healthy subjects, respectively. Rivastigmine essentially blocked
the induction of the global excitability enhancement elicited by anodal tDCS and revealed a tendency to first reduce and then stabilize
cathodal tDCS-induced inhibitory aftereffects. However, ACh enhanced the synapse-specific excitability enhancement produced by
facilitatory PAS and consolidated the inhibitory PAS-induced excitability diminution. These findings are in line with a cholinergic
focusing effect that optimizes the detection of relevant signals during information processing in humans.
Key words: neuroplasticity; acetylcholine; transcranial direct current stimulation; paired associative stimulation; motor cortex; human

Introduction
Extensive evidence concerning cholinergic modulation of several
cognitive functions supports an important role of acetylcholine
(ACh) in arousal, attention, learning, and memory formation
(Gold, 2003; Sarter et al., 2003). In Alzheimer’s disease, enhanc-
ing cerebral ACh level has been shown to improve impaired
learning and memory functions caused by cholinergic dysfunc-
tion. With regard to its specific functional properties, neurophys-
iological data from animal studies reveal dual neuromodulatory
effects of ACh on cortical excitability and synaptic plasticity (Ras-
musson, 2000; Gu, 2002). Cholinergic blockade has been shown
to reduce long-term potentiation (LTP), whereas cholinergic
agonists enhance LTP in the hippocampus, piriform cortex, and
neocortex (Blitzer et al., 1990; Brocher et al., 1992; Hasselmo and
Barkai, 1995). In humans, use-dependent plasticity of the motor
cortex is facilitated by an acetylcholinesterase inhibitor and
blocked by a cholinergic antagonist (Sawaki et al., 2002;
Meintzschel and Ziemann, 2006). However, it is also reported
that ACh enhanced long-term depression (LTD) induced with
paired-pulse stimulation in the rat visual cortex (Kirkwood et al.,
1999). Furthermore, it suppressed excitatory glutamatergic syn-
aptic transmission via presynaptic inhibition at intrinsic, recur-
Received May 17, 2007; revised Oct. 18, 2007; accepted Nov. 1, 2007.
This work was supported by the Deutsche Forschungsgemeinschaft (DFG) Program “Nicotine (SPP 1226).” M.-
F.K. was supported by European Graduiertenkolleg 632, “Neuroplasticity: From Molecules to Systems,” which was
funded by DFG.
Correspondence should be addressed to Michael A. Nitsche, Department of Clinical Neurophysiology, Georg-
August-University Göttingen, Robert-Koch-Strasse 40, 37075 Göttingen, Germany. E-mail: mnitsch1@gwdg.de.
DOI:10.1523/JNEUROSCI.4104-07.2007
Copyright © 2007 Society for Neuroscience 0270-6474/07/2714442-06$15.00/0
rent synapses but not afferent fiber synapses (Hasselmo and
Bower, 1992; Hasselmo et al., 1995; Vogt and Regehr, 2001). This
suggests a differential, activity-dependent cholinergic modifica-
tion of neural networks in which ACh facilitates the detection of
incoming afferent inputs, whereas it decreases intrinsic feedback
excitability, thereby focusing the encoding of relevant, associated
information processing (Blokland et al., 1992; Winters and Bus-
sey, 2005).
To test the focusing action of ACh on neuroplasticity in hu-
mans, two protocols of brain stimulation were introduced in the
present study. In the paired associative stimulation (PAS) proto-
col, repetitive peripheral nerve stimulation is paired with trans-
cranial magnetic stimulation (TMS) of the human motor cortex
(Stefan et al., 2000). It is postulated that PAS-induced excitability
changes share the features of associative synaptic LTP and LTD,
depending on the sequence of the near-synchronous pair of stim-
uli from different stimulation modalities in the motor cortex
(Stefan et al., 2000; Wolters et al., 2003), which parallels the spike-
timing-dependent rule for Hebbian LTP and LTD induction in
animal studies (Dan and Poo, 2004). Facilitatory PAS (PAS25)
with peripheral nerve stimulation applied 25 ms earlier than TMS
pulse in M1 results in synchronous activation of motor cortical
neurons by the afferent somatosensory stimulus and motor cor-
tex TMS and thus enhances cortical excitability. However, inhib-
itory PAS (PAS10) diminishes cortical excitability with the inter-
stimuli interval 10 ms, with which the somatosensory input
reaches the motor cortex relevantly later than the TMS pulse,
thereby inducing asynchronous stimulation on motor cortical
neurons. PAS should specifically induce neuroplasticity in
somatosensory-motor cortical synapses. In contrast, transcranial

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Kuo et al. • Acetylcholine Focuses Neuroplasticity
direct current stimulation (tDCS) encompasses the global mod-
ulation of cortical network plasticity by application of weak direct
currents through the surface of the scalp. Anodal tDCS enhances
cortical excitability, whereas cathodal tDCS diminishes it for up
to 1 h after the end of stimulation (Nitsche and Paulus, 2000,
2001; Nitsche et al., 2003a). The primary mechanism is a modu-
lation of the resting membrane potential, and the resulting
polarity-specific excitability changes subsequently induce
changes of synaptic strength, which are, however, not restricted
to specific synaptic connections (Bindman et al., 1964; Purpura
and McMurtry, 1965). Both plasticity-inducing protocols ac-
complish long-lasting, NMDA receptor-dependent excitability
changes (Liebetanz et al., 2002; Stefan et al., 2002; Nitsche et al.,
2003b). The main difference lies in the synapse-specific focal ef-
fects of PAS: whereas the plasticity induced by tDCS is relatively
nonfocal and not synapse-specific because it is thought to change
cortical excitability under the whole area covered by the relatively
large stimulation electrode, the plasticity induced by PAS is re-
stricted to the intercortical connections between the somatosen-
sory and motor cortex. According to the focusing hypothesis of
ACh, it should selectively enhance and consolidate specific syn-
aptic modifications induced by PAS, while depressing global ones
accomplished by tDCS, to sharpen the signal-to-noise ratio in
human cortical networks.
Materials and Methods
Subjects. Ten to twelve healthy subjects [tDCS experiment: six men, six
women, aged (mean ⫾ SD) 24 ⫾ 3 years; PAS25 experiment: four women
and six men, aged 28 ⫾ 4 years; PAS10 experiment: 5 women and 5 men,
aged 27 ⫾ 4 years], without receiving acute or chronic medication, par-
ticipated in each experiment. The groups did not differ significantly with
regard to age and gender. Both studies were approved by the ethics com-
mittee of the University of Goettingen, and we conform to the Declara-
tion of Helsinki. All subjects had given written informed consent. Be-
cause we were interested primarily in the physiological effects of ACh in
this study, we recruited relatively young subjects to guarantee the com-
parability of the results with former pharmacological studies on neuro-
plasticity and to avoid including subjects with subclinical (e.g., microvas-
cular) brain lesions, which might have influenced our results
unintentionally.
Transcranial direct current stimulation. tDCS was performed with a
pair of saline-soaked surface sponge electrodes (35 cm 2) with one of the
electrodes placed over the representational area of the right abductor
digiti minimi muscle (ADM), as determined by TMS, and the other
electrode above the right orbit as reference. The currents ran continu-
ously for 13 min (anodal tDCS) or 9 min (cathodal tDCS) with an inten-
sity of 1 mA. In previous studies, these stimulation durations have been
shown to induce aftereffects of tDCS lasting for ⬃1 h (Nitsche and Pau-
lus, 2001; Nitsche et al., 2003a).
Paired associative stimulation. Peripheral nerve stimulation was ap-
plied on the right ulnar nerve at the level of the wrist with a Digitimer
D185 stimulator (Digitimer, Welwyn Garden City, UK). Single-pulse
TMS was delivered over the representing area of the right ADM. Each
TMS pulse, at an intensity eliciting a muscle evoked potential of ⬃1 mV
peak-to-peak amplitude, was preceded by an ulnar nerve stimulus with
an interval of 25 ms (PAS25) or 10 ms (PAS10) using a standard stimu-
lation block (cathodal proximal) at a stimulation width of 200 ␮s and
stimulation intensity of 300% of the perceptual threshold, defined as the
lowest intensity of the stimuli that is perceivable by the subject. Ninety
pairs were applied at 0.05 Hz over 30 min, which has been shown to
induce long-lasting excitability changes in the motor cortex (Stefan et al.,
2000).
Pharmacological interventions. Rivastigmine (3 mg) or equivalent pla-
cebo drugs were taken by the subjects 2 h before the start of the interven-
tion (Kennedy et al., 1999). This dose was chosen to minimize drug-
induced side effects, but to enhance the cholinergic level of the CNS
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
J. Neurosci., December 26, 2007 • 27(52):14442–14447 • 14443
effectively. The experimental sessions were performed in a randomized
order and were separated by at least 1 week to avoid cumulative drug or
stimulation effects.
Measurement of motor cortical excitability. TMS-elicited muscle-
evoked potentials (MEPs) were recorded to measure excitability changes
of the representational motor cortical area of the right ADM. Single-
pulse TMS was conducted by a Magstim 200 magnetic stimulator (Mag-
stim Company, Whiteland, Dyfed, UK) with a figure-eight-shaped mag-
netic coil (diameter of one winding, 70 mm; peak magnetic field, 2.2
tesla). The coil was held tangentially to the skull, with the handle pointing
backward and laterally at an angle of 45° from midline. The optimal
position was defined as the site at which stimulation resulted consistently
in the largest MEPs. Surface EMG was recorded from the right ADM with
Ag-AgCl electrodes in a belly-tendon montage. The signals were ampli-
fied and filtered with a time constant of 10 ms and a low-pass filter of 2.5
kHz and then digitized at an analog-to-digital rate of 5 kHz and further
relayed into a laboratory computer using the Signal software and CED
1401 hardware (Cambridge Electronic Design, Cambridge, UK). The
intensity was adjusted to elicit baseline MEPs of, on average, 1 mV peak-
to-peak amplitude and was kept constant for the poststimulation assess-
ment unless adjusted (see below).
Experimental procedures. The experiments were conducted in a re-
peated measurement design. For the tDCS experiment, a complete cross-
over design was chosen. For the PAS experiments, separate subject
groups participated in the PAS10 and PAS25 experiments. Subjects were
seated comfortably in a reclining chair. First the optimal position of the
magnetic coil for eliciting MEPs in the resting ADM was assessed over the
left motor cortex, and 20 MEPs were recorded for the first baseline. Two
hours after intake of the medication, a second baseline was determined to
control for a possible influence of the drug on cortical excitability and
adjusted if necessary. Subjects were not aware about and could not dis-
tinguish between the specific stimulation protocols used, nor were they
informed about the medication administered in a specific session.
In experiment 1 with tDCS, one of the DC electrodes, to which in the
following the terms cathodal or anodal tDCS refer, was fixed at the cor-
tical representational area of ADM as defined during the first baseline
recording, and the other one was fixed at the contralateral forehead area
above the right orbit. Direct currents were applied on 12 subjects for 9
min (cathodal) or 13 min (anodal). After cessation of tDCS, 20 MEPs
were recorded at 0.25 Hz every 5 min for half an hour and then every 30
min until 2 h after the end of DC stimulation, because tDCS-induced
aftereffects without medication will not last longer than this period of
time (Nitsche and Paulus, 2000; Nitsche et al., 2003a). Because of the
relatively diurnal stability of corticospinal excitability, we did not expect
additional excitability changes after MEPs returned to baseline level
again. Only for the rivastigmine conditions, TMS recordings were per-
formed at four additional time points: same day evening, next morning,
next noon, and next evening.
In experiment 2 with PAS, the interventional PAS protocol as de-
scribed above was used on 10 subjects. TMS recording procedures were
the same as described above.
Data analysis and statistics. MEP amplitude means were calculated first
individually and then interindividually for each time bin including both
baseline values. The postintervention MEPs were normalized and are
given as ratios of the baseline determined immediately before interven-
tion (tDCS/PAS).
In the tDCS experiment, a repeated-measure ANOVA for the time
bins up to 120 min after tDCS was calculated with the within-subject
factors time course, current stimulation (anodal and cathodal tDCS),
drug condition (rivastigmine vs placebo), and the dependent variable
MEP amplitude. For PAS experiment, we performed a repeated-measure
ANOVA with PAS (PAS25 and PAS10) as between-subject factor and
within-subject factors drug condition (rivastigmine vs placebo) and time
course for MEPs up to 120 min after intervention. If appropriate, post hoc
Student’s t tests (paired samples, two-tailed, p ⬍ 0.05, not adjusted) were
performed to determine whether the MEP amplitudes before and after
the interventional brain stimulations differed in each intervention con-
dition and whether those differences depended on the drug conditions.
 14444 • J. Neurosci., December 26, 2007 • 27(52):14442–14447 Kuo et al. • Acetylcholine Focuses Neuroplasticity

Table 1. Results of the ANOVAs

Parameters df F value p value
Experiment 1 (tDCS) TDCS 1 40.585 ⬍ 0.001*
Drug 1 3.650 0.082
Time course 10 1.700 0.099
tDCS ⫻ drug 1 3.417 0.092
tDCS ⫻ time course 10 1.868 0.066
Drug ⫻ time course 10 0.290 0.976
tDCS ⫻ drug ⫻ time course 10 4.206 ⬍ 0.001*
Experiment 2 (PAS) PAS 1 73.006 ⬍ 0.001*
Drug 1 0.140 0.712
Time course 10 1.556 0.123
Drug ⫻ PAS 1 23.886 ⬍ 0.001*
Time course ⫻ PAS 10 9.934 ⬍ 0.001*
Drug ⫻ time course 10 2.207 0.019*
Drug ⫻ time course ⫻ PAS 10 3.208 0.001* Figure 1. Cholinergic modulation of global cortical plasticity induced by tDCS. Rivastigmine
In both experiments, the ANOVAs encompass the time course up to 120 min after tDCS or PAS, because the remaining abolished the induction of anodal tDCS-elicited excitability increases, as recorded by TMS-
time points were only measured for the rivastigmine conditions. ⬙Drug⬙ represents rivastigmine and placebo, evoked MEP amplitudes. Additionally, rivastigmine initially diminished the excitability reduc-
whereas ⬙tDCS⬙ indicates anodal and cathodal polarity, and ⬙PAS⬙ represents PAS25 and PAS10. p ⱕ 0.05. tion induced by cathodal tDCS under rivastigmine. However, the respective excitability diminu-
tion was later consolidated. Filled symbols indicate significant deviations from baseline with
Additional post hoc tests (Student’s t tests, p ⬍ 0.05) were performed to regard to each drug condition. Hash symbols indicate significant differences in anodal tDCS-
explore whether rivastigmine modified baseline MEPs. induced excitability changes between placebo and rivastigmine conditions; asterisks represent
significant differences in inhibition caused by cathodal stimulation between the placebo and
Results rivastigmine medication conditions (Student’s t test, two-tailed, repeated measures; #p ⬍
Three subjects experienced prominent side effects under rivastig- 0.05). a, Anodal; c, cathodal; plc, placebo; riva, rivastigmine; se, same evening; nm, next morn-
ing; nn, next noon; ne, next evening. Error bars indicate SEM.
mine (nausea, vomiting, and dizziness) and were excluded from
the experiment. Two of the participating subjects complained
about slight nausea after taking rivastigmine, but the symptoms
subsided before the neuroplasticity-inducing intervention. Base-
line MEP amplitudes did not differ significantly before and after
drug intake in all conditions. Absolute baseline MEP amplitudes
were not different in all medication and stimulation subgroups
(Student’s t tests, paired, two-tailed, p ⬎ 0.05).

Effects of rivastigmine on tDCS-induced motor cortex

excitability shifts (experiment 1)
The ANOVA revealed a significant main effect of tDCS (F ⫽
40.585; p ⬍ 0.001) and tDCS ⫻ drug ⫻ time course (F ⫽ 4.206;
p ⬍ 0.001) (Table 1). In the PLC conditions, the anodal tDCS-
induced excitability increase stayed significant until 30 min after
tDCS, and the cathodal tDCS-induced inhibition lasted until 60
min after DC stimulation. As revealed by the post hoc t tests
(paired, two-tailed, p ⬍ 0.05), rivastigmine initially abolished the
induction of both the anodal tDCS-elicited excitability enhance-
ment and the cathodal tDCS-elicited excitability diminution.
However, a delayed, consolidated inhibition induced by cathodal
tDCS was observed in the rivastigmine condition, compared with
placebo conditions. The decrease of excitability generated by
cathodal tDCS under rivastigmine remained significant for 2 h
after tDCS, whereas it returned to baseline level after 60 min in
the PLC condition (Fig. 1).
Effects of rivastigmine on PAS-induced motor cortex
excitability shifts (experiment 2)
The ANOVA displayed a significant main effect of PAS (F ⫽
73.006; p ⬍ 0.001). The interactions between drug ⫻ PAS (F ⫽
23.886; p ⬍ 0.001), time course ⫻ PAS (F ⫽ 9.934; p ⬍ 0.001),
drug ⫻ time course (F ⫽ 2.207; p 0.019), and drug ⫻ time
course ⫻ PAS (F ⫽ 3.208; p ⬍ 0.001) were also significant (Table
1). In the PAS25 experiment, the excitatory shift of MEP ampli-
tudes returned to baseline 25 min after PAS in the placebo med-
ication condition, as revealed by Student’s t tests (paired, two-
tailed, p ⬍ 0.05), whereas rivastigmine enhanced and prolonged
the excitatory effects of PAS25 (Fig. 2). In the PAS10 experiment,
Figure 2. The PAS25-induced synapse-specific excitability enhancement is facilitated by
ACh. The PAS-induced excitability enhancement was increased and prolonged under rivastig-
mine until 30 min after PAS, whereas MEP amplitudes returned to baseline 25 min after PAS in
the placebo condition. Hash symbols represent significant differences between placebo and
rivastigmine conditions; filled symbols indicate significant deviations from baseline with regard
to each drug condition (Student’s t test, two-tailed, repeated measures; #p ⬍ 0.05). plc, Pla-
cebo; riva, rivastigmine; se, same evening; nm, next morning; nn, next noon; ne, next evening.
Error bars indicate SEM.
the inhibitory effect without medication lasted for half an hour.
Rivastigmine further enhanced and prolonged the inhibition un-
til the same day evening after intervention (Fig. 3).
Discussion
The present study demonstrates that ACh enhances the synapse-
specific cortical excitability increase induced by PAS25 and con-
solidates the PAS10-induced reduction of motor cortical excit-
ability, whereas it prevents global excitatory aftereffects produced
by anodal tDCS. ACh also delayed the induction of the cathodal
tDCS-elicited excitability decrease and slightly prolonged its
overall duration. Because MEP amplitudes were not modified by
rivastigmine in the dosage applied alone, we have no evidence for
a direct cholinergic influence of the drug on corticospinal excit-

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Kuo et al. • Acetylcholine Focuses Neuroplasticity
Figure 3. The PAS10-induced synapse-specific excitability diminution is consolidated by
ACh. In the placebo condition, the inhibitory effect of PAS10 lasted for approximately half an
hour, whereas rivastigmine enhanced the inhibition and further prolonged it until same day
evening of the intervention. Significant differences between placebo and rivastigmine are
shown with asterisks. Filled symbols indicate significant deviations from baseline with regard to
each drug condition (Student’s t test, two-tailed, repeated measures, p ⬍ 0.05). plc, Placebo;
riva, rivastigmine; se, same evening; nm, next morning; nn, next noon; ne, next evening. Error
bars indicate SEM.
ability. The results support the hypothesis of a focusing effect of
ACh on neuroplasticity of cortical networks. ACh not only in-
creased selectively the efficacy of synapse-specific excitability-
enhancing neuroplasticity, but also prolonged the excitability
diminution in case of asynchronous synapse-specific inputs and
suppressed global excitability enhancements. By these mecha-
nisms, ACh is well suited to improve the signal-to-noise ratio and
to refine information processing in neural networks.
ACh diminishes tDCS-driven neuroplasticity
At first glance, the inhibitory effect of rivastigmine on facilitatory
neuroplasticity induced by anodal tDCS seems contradictory to
the results obtained from animal studies in which LTP was facil-
itated by cholinergic stimulation (Brocher et al., 1992; Abe et al.,
1994; Hasselmo and Barkai, 1995; Patil et al., 1998). The major
conceptual difference between these studies and our tDCS exper-
iment is the manipulation applied for neuroplasticity induction.
In slice preparation, the plastic changes induced with either
paired or high-frequency suprathreshold electrical stimulation
are caused by synchronous activation of synaptic connections. In
contrast, tDCS-elicited neuroplasticity is a consequence of in-
creasing general, asynchronous network activity, possibly added
by modification of the postsynaptic resting membrane potential
(Bindman et al., 1964; Purpura and McMurtry, 1965). The dim-
inution of the anodal tDCS-induced excitability enhancement
might thus be because of an ACh-induced decrease of general
excitation within global neuronal networks, probably modulated
via the cholinergic presynaptic inhibition of excitatory feedback
potentials or excitatory transmission at recurrent connections.
Moreover, evidence reveals that the suppression of synaptic
transmission is selective for recently modified synapses (Linster
et al., 2003) but does not apply to silent synapses (Fernandez de
Sevilla et al., 2002). Thus it is probable that synapses that are
globally modified by tDCS in the present study are more suscep-
tible to cholinergic suppression of synaptic transmission during
plasticity induction. A similar effect is also demonstrated by a
recent study using a dynamic clamp system to mimic in-vivo-like
background activities in motor cortical slices, in which cholin-
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
J. Neurosci., December 26, 2007 • 27(52):14442–14447 • 14445
ergic facilitation of LTP was attenuated in the presence of random
background noise (Desai and Walcott, 2006).
However, ACh revealed a tendency to consolidate excitability-
diminishing aftereffects generated by cathodal tDCS, although it
abolished its initial induction phase. The biphasic effect of riv-
astigmine on cathodal tDCS-induced plasticity can also be ex-
plained by a cholinergic regulation of different neurons within
different temporary profiles. The initial blockade could be be-
cause of the fast negative modulation of inhibitory neurons (e.g.,
inhibitory interneurons) known to be induced by ACh (Ji and
Dani, 2000), whereas the late-onset and prolonged excitability
diminution can be explained by the inhibitory modification of
excitatory neurons with the combination of ACh and cathodal
tDCS. Further studies are required to test this hypothesis.
Cholinergic consolidation of PAS-induced cortical plasticity
The PAS experiment demonstrates a positive cholinergic modu-
lation of PAS-elicited synaptic-specific plasticity. ACh seems to
enhance specifically the ability of synchronous motor cortical
input to enhance excitability, whereas the excitability diminution
accomplished by asynchronous input is prolonged. It is suggested
that PAS25/PAS10 relates to associative LTP/LTD in the human
motor cortex (Stefan et al., 2000, 2002; Wolters et al., 2003; Stefan
et al., 2006). Because ACh has been shown to facilitate cortical
sensory plasticity by enhancing sensory input processing (Ras-
musson and Dykes, 1988; Tremblay et al., 1990; Patil et al., 1998)
and suppressing irrelevant input (Hasselmo and Barkai, 1995),
one might further speculate that rivastigmine specifically im-
proved the efficacy of PAS by (1) enhancing the signal-to-noise
ratio, thereby facilitating meaningful information processing as
represented by synchronous input; and (2) suppressing non-
meaningful input as represented by asynchronous stimulation
within neural networks.
Thus, the results of our study are in accordance with the re-
spective animal experiments (Blitzer et al., 1990; Hasselmo and
Barkai, 1995; Kirkwood et al., 1999). Moreover, they are concor-
dant with the results of recent behavioral studies in the human
motor cortex exploring the effect of ACh modulation on use-
dependent plasticity, which was blocked by the ACh antagonist
scopolamine (Sawaki et al., 2002) and enhanced by an acetylcho-
linesterase inhibitor (Meintzschel and Ziemann, 2006). There-
fore, our results offer a neurophysiological mechanism of how
ACh might improve behavioral plasticity.
It is also notable that the duration of cholinergic effects on
cortical plasticity induced by PAS25 and PAS10 is asymmetric.
ACh was more effective in stabilizing the PAS10-generated excit-
ability reduction than the PAS25-induced excitability enhance-
ment. This might be because of the generally less efficient PAS25
protocol, as revealed in the placebo condition in the present
study, which might have limited the ability of ACh to stabilize
neuroplasticity. Alternatively, it was suggested that the direction
of cholinergic modulation on synaptic plasticity could be deter-
mined by ACh concentration and subtype receptors activation
(Kuczewski et al., 2005), which could also explain an asymmetry
of ACh effects on different PAS protocols.
Summary of cholinergic modulation in human
cortical plasticity
The results of the present study suggest that ACh has fairly spe-
cific effects on cortical plasticity. Rivastigmine selectively en-
hanced the efficacy of PAS25, a synchronous associative stimula-
tion protocol, to increase excitability, whereas it shifted the
effects of the remaining plasticity-inducing protocols in an inhib-
 14446 • J. Neurosci., December 26, 2007 • 27(52):14442–14447
itory direction. This indicates a general inhibitory effect of riv-
astigmine on network excitability, with the exception of PAS25-
induced plasticity, and further explains how rivastigmine works
as a cognitive enhancer via increasing the signal-to-noise ratio of
cortical activity. Associative plasticity is suggested as a neuro-
physiological correlate of learning and memory formation, and,
indeed, excitability-enhancing PAS has been shown to be tightly
connected to learning processes (Ziemann et al., 2004; Stefan et
al., 2006), thus strengthening the efficacy of synchronous stimuli
to enhance excitability might improve learning. However, riv-
astigmine can reduce “noisy” synaptic modification, as demon-
strated in our study via abolishing the excitability-enhancing
properties of anodal tDCS, which increases general and most
probably not synchronous cortical activity (Bindman et al., 1964;
Purpura and McMurtry, 1965), and via strengthening the inhib-
itory effects of the asynchronous PAS10 protocol. Moreover, ri-
vastigmine enhanced the efficacy of cathodal tDCS to diminish
cortical excitability. Cathodal tDCS applied synchronously with
PAS25 has been shown to enhance the efficacy of PAS25 to in-
crease cortical excitability (Nitsche et al., 2007), which can be
explained by a noise-reducing function of cathodal tDCS-
induced inhibition. Thus the enhancing effect of rivastigmine on
cathodal tDCS-driven inhibition should also increase the efficacy
of meaningful information processing via increasing the signal-
to-noise ratio.
A recently conducted study revealed similar effects of L-dopa
(L-3,4-dihydroxyphenylalanine) on neuroplasticity (Kuo et al.,
2007). Here, dopamine reversed the anodal tDCS-induced excit-
ability enhancement into inhibition and prolonged the excitabil-
ity diminution caused by cathodal tDCS. However, dopamine
enhanced the PAS25-induced synapse-specific increase of corti-
cal excitability. The similarity of the dopamine and ACh effects
could be explained by the fact that dopaminergic and cholinergic
neurons serve similar functions and are tightly interconnected in
neuronal networks (Sarter et al., 1999; Zahm, 2006). They also
provide, at least, a theoretical option for the application of dopa-
mine in neurological disorders associated with cognitive deficits.
Indeed, dopaminergic medication has been shown to improve
learning in healthy subjects and patients after stroke (Floel et al.,
2005, 2007).
Together, the results of the present study are in line with cor-
tical cholinergic functions enhancing the contrast of relevant
stimuli against background noise or distractors, thereby improv-
ing signal processing during information encoding. Therefore, it
might strengthen the rational basis for application of cholinest-
erase inhibitors to improve cognitive functions in patients with
Alzheimer’s disease or vascular dementia. However, it has to be
kept in mind that the results were obtained with healthy young
subjects. Further studies are needed to elucidate whether the ef-
fect of ACh on neuroplasticity is identical in elderly healthy and
demented patients and whether the impact of ACh on neuroplas-
ticity is correlated with clinical outcome.
References
Abe K, Nakata A, Mizutani A, Saito H (1994) Facilitatory but nonessential
role of the muscarinic cholinergic system in the generation of long-term
potentiation of population spikes in the dentate gyrus in vivo. Neuro-
pharmacology 33:847– 852.
Bindman LJ, Lippold OCJ, Redfearn JWT (1964) The action of brief polar-
izing currents on the cerebral cortex of the rat (1) during current flow and
(2) in the production of long-lasting after-effects. J Physiol (Lond)
172:369 –382.
Blitzer RD, Gil O, Landau EM (1990) Cholinergic stimulation enhances
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Kuo et al. • Acetylcholine Focuses Neuroplasticity
long-term potentiation in the CA1 region of rat hippocampus. Neurosci
Lett 119:207–210.
Blokland A, Honig W, Raaijmakers WG (1992) Effects of intra-
hippocampal scopolamine injections in a repeated spatial acquisition task
in the rat. Psychopharmacology (Berl) 109:373–376.
Brocher S, Artola A, Singer W (1992) Agonists of cholinergic and noradren-
ergic receptors facilitate synergistically the induction of long-term poten-
tiation in slices of rat visual cortex. Brain Res 573:27–36.
Dan Y, Poo MM (2004) Spike timing-dependent plasticity of neural cir-
cuits. Neuron 44:23–30.
Desai NS, Walcott EC (2006) Synaptic bombardment modulates musca-
rinic effects in forelimb motor cortex. J Neurosci 26:2215–2226.
Fernandez de Sevilla D, Cabezas C, de Prada AN, Sanchez-Jimenez A, Buno W
(2002) Selective muscarinic regulation of functional glutamatergic
Schaffer collateral synapses in rat CA1 pyramidal neurons. J Physiol
(Lond) 545:51– 63.
Floel A, Hummel F, Breitenstein C, Knecht S, Cohen LG (2005) Dopami-
nergic effects on encoding of a motor memory in chronic stroke. Neurol-
ogy 65:472– 474.
Floel A, Garraux G, Xu B, Breitenstein C, Knecht S, Herscovitch P, Cohen LG
(2007) Levodopa increases memory encoding and dopamine release in
the striatum in the elderly. Neurobiol Aging, in press.
Gold PE (2003) Acetylcholine modulation of neural systems involved in
learning and memory. Neurobiol Learn Mem 80:194 –210.
Gu Q (2002) Neuromodulatory transmitter systems in the cortex and their
role in cortical plasticity. Neuroscience 111:815– 835.
Hasselmo ME, Barkai E (1995) Cholinergic modulation of activity-
dependent synaptic plasticity in the piriform cortex and associative mem-
ory function in a network biophysical simulation. J Neurosci
15:6592– 6604.
Hasselmo ME, Bower JM (1992) Cholinergic suppression specific to intrin-
sic not afferent fiber synapses in rat piriform (olfactory) cortex. J Neuro-
physiol 67:1222–1229.
Hasselmo ME, Schnell E, Barkai E (1995) Dynamics of learning and recall at
excitatory recurrent synapses and cholinergic modulation in rat hip-
pocampal region CA3. J Neurosci 15:5249 –5262.
Ji D, Dani JA (2000) Inhibition and disinhibition of pyramidal neurons by
activation of nicotinic receptors on hippocampal interneurons. J Neuro-
physiol 83:2682–2690.
Kennedy JS, Polinsky RJ, Johnson B, Loosen P, Enz A, Laplanche R, Schmidt
D, Mancione LC, Parris WC, Ebert MH (1999) Preferential cerebrospi-
nal fluid acetylcholinesterase inhibition by rivastigmine in humans. J Clin
Psychopharmacol 19:513–521.
Kirkwood A, Rozas C, Kirkwood J, Perez F, Bear MF (1999) Modulation of
long-term synaptic depression in visual cortex by acetylcholine and nor-
epinephrine. J Neurosci 19:1599 –1609.
Kuczewski N, Aztiria E, Gautam D, Wess J, Domenici L (2005) Acetylcho-
line modulates cortical synaptic transmission via different muscarinic
receptors, as studied with receptor knockout mice. J Physiol (Lond)
566:907–919.
Kuo MF, Paulus W, Nitsche MA (2007) Boosting focally-induced brain
plasticity by dopamine. Cereb Cortex, in press.
Liebetanz D, Nitsche MA, Tergau F, Paulus W (2002) Pharmacological ap-
proach to the mechanisms of transcranial DC-stimulation-induced after-
effects of human motor cortex excitability. Brain 125:2238 –2247.
Linster C, Maloney M, Patil M, Hasselmo ME (2003) Enhanced cholinergic
suppression of previously strengthened synapses enables the formation of
self-organized representations in olfactory cortex. Neurobiol Learn Mem
80:302–314.
Meintzschel F, Ziemann U (2006) Modification of practice-dependent plas-
ticity in human motor cortex by neuromodulators. Cereb Cortex
16:1106 –1115.
Nitsche MA, Paulus W (2000) Excitability changes induced in the human
motor cortex by weak transcranial direct current stimulation. J Physiol
527 Pt 3:633– 639.
Nitsche MA, Paulus W (2001) Sustained excitability elevations induced by
transcranial DC motor cortex stimulation in humans. Neurology
57:1899 –1901.
Nitsche MA, Nitsche MS, Klein CC, Tergau F, Rothwell JC, Paulus W
(2003a) Level of action of cathodal DC polarisation induced inhibition of
the human motor cortex. Clin Neurophysiol 114:600 – 604.
Nitsche MA, Fricke K, Henschke U, Schlitterlau A, Liebetanz D, Lang N,
 Kuo et al. • Acetylcholine Focuses Neuroplasticity
Henning S, Tergau F, Paulus W (2003b) Pharmacological modulation
of cortical excitability shifts induced by transcranial direct current stim-
ulation in humans. J Physiol (Lond) 553:293–301.
Nitsche MA, Roth A, Kuo MF, Fischer AK, Liebetanz D, Lang N, Tergau F,
Paulus W (2007) Timing-dependent modulation of associative plastic-
ity by general network excitability in the human motor cortex. J Neurosci
27:3807–3812.
Patil MM, Linster C, Lubenov E, Hasselmo ME (1998) Cholinergic agonist
carbachol enables associative long-term potentiation in piriform cortex
slices. J Neurophysiol 80:2467–2474.
Purpura DP, McMurtry JG (1965) Intracellular activities and evoked poten-
tial changes during polarization of motor cortex. J Neurophysiol
28:166 –185.
Rasmusson DD (2000) The role of acetylcholine in cortical synaptic plastic-
ity. Behav Brain Res 115:205–218.
Rasmusson DD, Dykes RW (1988) Long-term enhancement of evoked po-
tentials in cat somatosensory cortex produced by co-activation of the
basal forebrain and cutaneous receptors. Exp Brain Res 70:276 –286.
Sarter M, Bruno JP, Turchi J (1999) Basal forebrain afferent projections
modulating cortical acetylcholine, attention, and implications for neuro-
psychiatric disorders. Ann NY Acad Sci 877:368 –382.
Sarter M, Bruno JP, Givens B (2003) Attentional functions of cortical cho-
linergic inputs: what does it mean for learning and memory? Neurobiol
Learn Mem 80:245–256.
Sawaki L, Boroojerdi B, Kaelin-Lang A, Burstein AH, Butefisch CM, Kopylev
L, Davis B, Cohen LG (2002) Cholinergic influences on use-dependent
plasticity. J Neurophysiol 87:166 –171.
Stefan K, Kunesch E, Cohen LG, Benecke R, Classen J (2000) Induction of
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
J. Neurosci., December 26, 2007 • 27(52):14442–14447 • 14447
plasticity in the human motor cortex by paired associative stimulation.
Brain 123 Pt 3:572–584.
Stefan K, Kunesch E, Benecke R, Cohen LG, Classen J (2002) Mechanisms of
enhancement of human motor cortex excitability induced by interven-
tional paired associative stimulation. J Physiol (Lond) 543:699 –708.
Stefan K, Wycislo M, Gentner R, Schramm A, Naumann M, Reiners K, Clas-
sen J (2006) Temporary occlusion of associative motor cortical plasticity
by prior dynamic motor training. Cerebral Cortex 16:376 –385.
Tremblay N, Warren RA, Dykes RW (1990) Electrophysiological studies of
acetylcholine and the role of the basal forebrain in the somatosensory
cortex of the cat. II. Cortical neurons excited by somatic stimuli. J Neu-
rophysiol 64:1212–1222.
Vogt KE, Regehr WG (2001) Cholinergic modulation of excitatory synaptic
transmission in the CA3 area of the hippocampus. J Neurosci 21:75– 83.
Winters BD, Bussey TJ (2005) Removal of cholinergic input to perirhinal
cortex disrupts object recognition but not spatial working memory in the
rat. Eur J Neurosci 21:2263–2270.
Wolters A, Sandbrink F, Schlottmann A, Kunesch E, Stefan K, Cohen LG,
Benecke R, Classen J (2003) A temporally asymmetric Hebbian rule
governing plasticity in the human motor cortex. J Neurophysiol
89:2339 –2345.
Zahm DS (2006) The evolving theory of basal forebrain functional-
anatomical “macrosystems.” Neurosci Biobehav Rev 30:148 –172.
Ziemann U, Iliac TV, Pauli C, Meintzschel F, Ruge D (2004) Learning mod-
ifies subsequent induction of long-term potentiation-like and long-term
depression-like plasticity in human motor cortex. J Neurosci 24:1666 –
1672.
 European Journal of Neuroscience, Vol. 41, pp. 1358–1371, 2015 doi:10.1111/ejn.12897

COGNITIVE NEUROSCIENCE

Differentiating neural systems mediating the

acquisition vs. expression of goal-directed and
habitual behavioral control

Mimi Liljeholm,1,2 Simon Dunne1,3 and John P. O’Doherty1,3

1
Trinity College Institute of Neuroscience, Trinity College Dublin, Dublin, Ireland
2
Department of Cognitive Sciences, University of California, 2312 Social & Behavioral Science Gateway Building, Irvine,
CA 92697, USA
3
Division of the Humanities and Social Sciences and Computation and Neural Systems Program, California Institute of
Technology, Pasadena, CA, USA

Keywords: action–outcome, functional magnetic resonance imaging, habit formation, human, stimulus–response

Abstract
Considerable behavioral data indicate that operant actions can become habitual, as demonstrated by insensitivity to changes in
the action–outcome contingency and in subjective outcome values. Notably, although several studies have investigated the neural
substrates of habits, none has clearly differentiated the areas of the human brain that support habit formation from those that
implement habitual control. We scanned participants with functional magnetic resonance imaging as they learned and performed
an operant task in which the conditional structure of the environment encouraged either goal-directed encoding of the conse-
quences of actions, or a habit-like mapping of actions to antecedent cues. Participants were also scanned during a subsequent
assessment of insensitivity to outcome devaluation. We identified dissociable roles of the cerebellum and ventral striatum, across
learning and test performance, in behavioral insensitivity to outcome devaluation. We also showed that the inferior parietal lobule
(an area previously implicated in several aspects of goal-directed action selection, including the attribution of intent and aware-
ness of agency) predicted sensitivity to outcome devaluation. Finally, we revealed a potential functional homology between the
human subgenual cortex and rodent infralimbic cortex in the implementation of habitual control. In summary, our findings sug-
gested a broad systems division, at the cortical and subcortical levels, between brain areas mediating the encoding and expres-
sion of action–outcome and stimulus–response associations.

Introduction
Habitual action selection is defined by insensitivity to changes in
the causal efficacy with which actions produce rewards and to the
current subjective value of those rewards (Balleine & Dickinson,
1998). Neuroscientific research on humans and rodents has demon-
strated that the brain areas mediating habitual performance are disso-
ciable from those supporting more deliberate, goal-directed, action
selection (Balleine & Dickinson, 1998; Yin et al., 2004, 2005; Val-
entin et al., 2007; Tricomi et al., 2009). Intriguingly, work in
rodents also suggests that distinct neural substrates make specialized
contributions to the development vs. deployment of habits (Killcross
& Coutureau, 2003). In contrast, in humans there has been no clear
differentiation between brain areas that support habit formation and
those that implement habitual control. Although a couple of neuroi-
maging studies have demonstrated increases in neural activity con-
comitant with the development of habits in a posterior area of the
lateral striatum (Tricomi et al., 2009; Wunderlich et al., 2012), the
Correspondence: Mimi Liljeholm, 2Department of Cognitive Sciences, as above.
E-mail: m.liljeholm@uci.edu
Received 3 October 2014, revised 6 March 2015, accepted 12 March 2015
use of overtraining to induce habitual responding in these studies
confounds well-trained performance with habitual control. In the
current study, in order to discriminate between neural substrates sup-
porting the acquisition vs. expression of habits, we scanned human
participants with functional magnetic resonance imaging as they per-
formed a novel instrumental task (see Task section in Materials and
methods), designed to rapidly induce habitual responding without
the potentially confounding process of overtraining.
Pharmacological disruptions and electrophysiological recordings
of the rodent brain have strongly implicated the dorsolateral striatum
in the acquisition of habits. Pre-training lesions of the dorsolateral
striatum abolish habit formation (Yin et al., 2004), and distinct
changes in neuronal activity patterns (Jog et al., 1999), including
substantial decreases in firing rates (Carelli et al., 1997; Tang et al.,
2007), have been demonstrated in this area across the development
of habitual responding. In contrast, the infralimbic region of the pre-
frontal cortex has been suggested to support an executive control
system that facilitates the expression of habits. Post-training musci-
mol-induced inactivation of the infralimbic cortex disrupts habitual
performance (Coutureau & Killcross, 2003; Haddon & Killcross,
2011), and changes in neuronal ensemble activity patterns in this

© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159


area occur very late in training and closely track the behavioral
manifestation of habitual control (Smith & Graybiel, 2013). Further
evidence for the involvement of the infralimbic cortex in the expres-
sion of habits comes from studies in which optogenetic perturbation
of this area disrupts well-ingrained habitual behavior (Smith et al.,
2012). Based on these findings in rodents, we hypothesized that
human homologs of the dorsolateral striatum and infralimbic cortex
would be involved in the formation and expression of habits respec-
tively, such that behavioral insensitivity to outcome devaluation
would correlate with neural activity during learning in the dorsal pu-
tamen (Carelli & West, 1991; Draganski et al., 2008), but with
activity during actual test performance in the subgenual cortex (On-
gur & Price, 2000; Ongur et al., 2003).
Materials and methods
Participants
Twenty volunteers (mean age 21.4  2.63 years, range 19–
28 years; 11 males) participated in the experiment. Due to technical
problems (loss of power to the stimulus computer), one of the sub-
jects was excluded, yielding a total of 19 participants. All partici-
pants were healthy and were recruited locally from the city of
Dublin, Ireland. The study was approved by the Trinity College
Dublin School of Psychology research ethics committee, and all par-
ticipants gave informed consent. The study conformed to the guide-
lines set out in the 2013 World Medical Association Declaration of
Helsinki.
Task
Goal-directed actions, defined by their sensitivity to changes in both
action–outcome contingency and outcome value, have been pro-
posed to depend on an internal model of the world that explicitly
relates alternative actions to future environmental states (Doya
et al., 2002; Daw et al., 2005). Consistent with this theoretical
framework, data from rodent studies suggest that reliance on a goal-
directed vs. habitual strategy might depend on the ease with which
alternative actions can be associated with distinct outcomes; goal-
directed performance appears to dominate, in spite of overtraining,
when alternative actions yield distinct sensory-specific outcomes
(e.g. grain vs. sucrose pellets) (Colwill & Rescorla, 1985; Holland,
2004), as well as when the rate of outcome delivery depends on the
rate of responding, rather than on a particular time interval passing
between successive reinforced responses (i.e. ratio vs. interval
schedules of reinforcement) (Dickinson et al., 1983). The current
task was structured on these potential bases of behavioral control,
in that external contingencies either facilitated or impeded a reliable
mapping of alternative actions to distinct sensory-specific outcome
states.
Participants were required to maintain the balance of a system of
fluid-filled beakers (see Fig. 1A for details). As long as all beakers
had sufficient fluid, system balance was maintained and randomly
occurring balance checks yielded monetary reward. However, on
each trial, one of the beakers would be emptied causing ‘system
imbalance’, with balance checks resulting in monetary loss until the
participant refilled the beaker by performing a particular instrumen-
tal action. The emptying of a beaker was always accompanied by
the onset of one of four abstract cues. In the Stimulus–Response (S-
R) condition, the identity of the presented cue determined which
instrumental action would refill the emptied beaker, regardless of
which of the beakers had lost its fluid. Consequently, across trials,
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Acquisition and expression of habitual control 1359
each rewarded action was paired with a specific antecedent cue, but
was decorrelated from the refilling of any particular beaker. Con-
versely, in a Response–Outcome (R-O) condition, each instrumental
action refilled a particular beaker, regardless of which abstract cue
was presented, such that identification of the relevant subgoal (e.g.
refilling beaker 1), combined with knowledge about specific action–
outcome contingencies (i.e. action 1 refills beaker 1), indicated
which action would restore system balance. To ensure that discrimi-
natory neural activity was not due to differences in the visual pro-
cessing of abstract cues vs. beakers, a matching task was interleaved
with the instrumental task (see Fig. 1B).
The persistent execution of an action after its outcome has been
devalued is a defining feature of habitual performance (Adams &
Dickinson, 1981; Adams, 1982). In the current study, following
acquisition of the instrumental task, we devalued one of the four
beakers by degrading its relationship to monetary gain, such that
system balance was maintained, and continued to yield points
even when the liquid in this beaker dropped below threshold.
Because there was a small cost for regulating the system, attempts
to refill this beaker now resulted in a net loss (see Fig. 1A and
Experimental Procedure section for details). Based on the notion
that failure to associate alternative actions with distinct outcome
states obstructs goal-directed encoding, we predicted that the S-R
condition would bolster habit formation during acquisition and
bias participants towards habitual action selection, defined as
responding to refill the devalued beaker, in subsequent test perfor-
mance. Note that, in both conditions, distinct features of the stim-
ulus environment can enter into S-R associations. Whereas, in the
S-R condition, each rewarded action was reliably preceded by a
particular arbitrary cue, rewarded actions in the R-O condition
were reliably preceded by the emptying of a particular beaker.
However, critically, it was only in the R-O condition that alterna-
tive actions could be associated with distinct outcome states.
Consequently, we expected performance in this condition to be
goal-directed.
Experimental procedure
We scanned participants as they acquired and performed the instru-
mental task, as well as during a subsequent devaluation test phase.
Each subject participated in both the R-O and S-R condition, with
conditions being run in separate, immediately consecutive, sessions
(order counterbalanced across subjects) and with a novel set of
four instrumental actions being used in each condition. Each condi-
tion included a response pre-training phase, learning phase, devalu-
ation phase, and final test phase as described below. The response-
training phase of the first condition was conducted outside the
scanner (in a separate testing room), before participants were trans-
ferred to the scanner in which they remained throughout all subse-
quent stages of the experiment. Before being transferred to the
scanner, participants were presented with a cover story describing
the beaker system and task. They were told that they would be in
one of two possible conditions (one in which each instrumental
action refilled a particular beaker regardless of which cue was pre-
sented, and another in which the identity of the cue determined
which of the four actions was required to refill an emptied beaker,
regardless of the identity of that beaker) and that part of their task
was to determine which of the two conditions they were in. The
entire experiment lasted for approximately 2 h, with 1.5 h being
spent in the scanner, and with approximately 60 min of active
scanning during the learning phases and devaluation tests in each
condition.
 1360 M. Liljeholm et al.

Fig. 1. Illustrations of instrumental learning and matching control tasks. (A) Instrumental learning. Participants are required to maintain the balance of a system
of fluid-filled beakers using a set of four instrumental actions (each a three-press sequence, see Materials and methods). During the inter-trial interval, the liquid
in the beakers continually fluctuates but remains high, and ‘balance checks’, occurring at brief random intervals, yield points for system balance (1). At the trial
onset, one of four abstract cues appears, the liquid in one of the beakers drops to its bottom, and balance checks begin to indicate a loss of points due to system
imbalance (2). Points are continually lost until the participant successfully refills the emptied beaker using one of the four actions. Following completion of the
correct action (3), the abstract cue disappears, the beaker is refilled, a small fee is charged for regulating the system, and balance checks again yield points for
system balance. If the correct action is not performed within 7 s, the beaker is automatically refilled, in which case there is no charge for system regulation. (B)
Matching task. The inter-trial interval (1) and trial onset (2) were as in the instrumental task but were followed, at 1500 ms after trial onset, by a blank screen
for 700 ms (3). The subsequent, final screen (4) showed a matching/non-matching stimulus together with a query about the match. In the S-R condition, the
final screen always showed one of the abstract cues (top); conversely, in the R-O condition, the stimulus to be matched was always a set of beakers (bottom).

Response pre-training at least five times without any prompts, again with feedback given
at the completion of each sequence. Participants were allowed to
Prior to the instrumental learning phases, participants received pre-
repeat these two phases as many times as they wanted to, knowing
training on the four instrumental actions (each being a three-press
that they would have to use the actions to earn monetary reward in
sequence). During this training, key-press sequences were illustrated
a subsequent phase.
by a white dot moving across three gray squares, horizontally
aligned at the center of the screen. Initially, participants viewed and
then immediately attempted to replicate each sequence, with feed-
Instrumental learning phase
back (i.e. correct/incorrect) given on each trial. After a total of five
correct replications of each response sequence, they proceeded to a The instrumental task was as illustrated in Fig. 1A. Note that, in
retrieval phase, in which they had to generate each unique sequence addition to the increase or decrease in monetary points based on

© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159


system balance, there was a small cost for regulating the system.
This response cost was included to ensure that, during test, partici-
pants would not respond simply based on any reinforcement intrin-
sic to executing the correct response. The response cost message
(screen 3 in Fig. 1A) also served to inform participants that their
action had successfully regulated the system, rather than the system
having self-regulated (in which case no response cost was charged)
due to failure to perform the correct action within 7 s of trial onset.
Participants were allowed to perform as many key-presses as they
wanted during system imbalance (i.e. during the 7 s following trial
onset), with the correct sequence of three consecutive key-presses
immediately restoring system balance and terminating the trial.
There was no constraint on the temporal spacing between key-
presses, as long as all three presses were performed consecutively
within the trial window.
Critically, the stimulus materials presented in Fig. 1A were identi-
cal across the two conditions; our manipulation consisted entirely of
differences in the contingencies between cues, actions and beaker
outcomes. To rule out visual processes involved in selectively
attending to the abstract cues vs. the beakers as a source of any
imaging effects, a matching task was block-interleaved with the
instrumental task during instrumental learning (see Fig. 1B). Briefly,
in matching blocks, the inter-trial intervals and trial onsets were
exactly as in the system balance task, except that the words ‘Match-
ing trial’ were displayed center screen. Without this indication,
matching trials would be identical to instrumental trials during the
relevant trial period, and thus could not have served as controls.
Following the appearance of the abstract cue and emptying of the
relevant beaker, a white masking screen was displayed, followed by
a depiction of either an abstract cue (S-R condition) or a set of
beakers (R-O condition), together with a query about whether the
currently shown cue/beaker set matched that on the previous screen.
In each condition, the instrumental learning phase consisted of four
blocks of trials, with each block being further divided into one sub-
block of 24 instrumental trials, followed by a sub-block of eight
matching trials, and with the order of trials randomized within each
sub-block.
At the end of the instrumental learning phase, participants were
asked whether they felt confident that they had learned how to regu-
late the system or whether they wanted to receive an additional set
of 20 training trials (five with each action). Five participants (two in
the R-O and three in the S-R condition) requested and received
additional training. No scanning was conducted during additional
training, nor were the added trials included in assessments of accu-
racy and response times during acquisition.
Devaluation phase
Following instrumental learning, participants were instructed that
the system had changed such that one of the beakers was no
longer relevant for system balance, which would be maintained,
and continued to yield points even when the liquid in this beaker
dropped below threshold. They then observed as the system regu-
lated itself (i.e. no actions were performed) across 16 trials (four
with each beaker) in order to discover the identity of the devalued
beaker. Participants were told that they would not lose or gain
any of the displayed points during this phase. In the imaging
experiment, two participants failed to correctly identify the deva-
lued beaker after this devaluation procedure, which was therefore
repeated once for these participants. All other participants success-
fully identified the devalued beaker at the end of the devaluation
procedure.
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Acquisition and expression of habitual control 1361
Test phase
Having correctly identified the devalued beaker, participants were
again given the opportunity to regulate the system for personal mon-
etary reward. During this phase, all text messages indicating gains
or losses were covered up, in order to prevent additional learning
(i.e. simulating extinction). Participants were instructed that, in spite
of these gray strips, they should assume that all was exactly as they
had learned before, i.e. they would still lose points whenever the
system was not balanced, there was still a cost for regulating the
system, and the previously identified irrelevant beaker was still irrel-
evant for system balance. Importantly, because there was a small
charge for each instrumental regulation of the system, refilling the
now irrelevant beaker resulted in a net monetary loss. The test phase
consisted of a single instrumental block with 44 randomly ordered
trials, i.e. 11 trials with each beaker, including the devalued (i.e.
irrelevant) one. We operationally defined devaluation insensitivity,
our assay of habitual performance, as the proportion of the 11 deva-
lued trials on which participants initiated a response to refill the bea-
ker. Thus, if a participant initiated a response on two of those 11
trials, their devaluation insensitivity score would be 0.18.
Pay-off structure
Throughout the instrumental task, system balance checks occurred
on average every 3 s. During the inter-trial interval (mean 5 s), the
system was always balanced, with each check yielding a reward of
0.5 points. At each trial onset, the system became imbalanced, with
each balance check resulting in a loss of 0.5 points, and would
remain so for 7 s unless regulated by the participant. Thus, if no
action were taken to balance the system, each trial would entail an
average loss of 1.17 points. If the system was balanced immediately,
no points were lost due to system imbalance, but there was a
response cost of 0.1 points (this cost was only applied to correct
responses resulting in system regulation). Thus, the average gain of
responding to balance the system was 1.07 points. After beaker
devaluation, during the test phase, this was still the case for regula-
tory responses refilling non-devalued beakers; however, responding
to refill the devalued beaker now resulted in a net loss of 0.1 points
(the response cost).
Imaging acquisition and analyses
A 3 Tesla scanner (MAGNETOM Trio, Siemens) was used to
acquire structural T1-weighted images and T2*-weighted echoplanar
images (repetition time, 2.65 s; echo time, 30 ms; flip angle, 90°;
45 transverse slices; matrix, 64 9 64; field of view, 192 mm; thick-
ness, 3 mm; slice gap, 0 mm) with blood oxygenation level-depen-
dent contrast. To recover signal loss from dropout in the medial
orbitofrontal cortex (O’Doherty et al., 2002), each horizontal section
was acquired at 30° to the anterior commissure–posterior commis-
sure axis. Image processing and statistical analyses were performed
using statistical parametric mapping (SPM)8 (http://www.fil.ion.u-
cl.ac.uk/spm). The first four volumes of images were discarded to
avoid T1 equilibrium effects. All remaining volumes were corrected
for differences in slice acquisition, realigned to the first volume, spa-
tially normalized to the Montreal Neurological Institute echoplanar
imaging template, and spatially smoothed with a Gaussian kernel
(8 mm, full width at half-maximum). We used a high-pass filter
with a cutoff of 128 s.
As previously noted, instrumental conditions were run in separate
consecutive sessions, with the acquisition phase of each condition
 1362 M. Liljeholm et al.
consisting of interleaved instrumental and matching block, and with
active scanning occurring only during the acquisition and test phase
of each condition (i.e. the scanner was off during response pre-train-
ing and devaluation phases), for a total of four active scanning peri-
ods per subject. For each subject, we constructed a design matrix
that included the acquisition and test phases of both experimental
conditions. For each of the two instrumental acquisition phases (i.e.
R-O and S-R), stick functions modeled the trial onsets in each block
of instrumental learning and in matching blocks. In addition, three
regressors, respectively modeling the onsets of error trials, the times
of all individual key-presses and the times of all point displays (i.e.
gains and losses), were entered as regressors of no interest, together
with six regressors accounting for the residual effects of head
motion. For each of the two test phases, two stick functions respec-
tively modeled non-devalued and devalued trials. Again, for each
instrumental condition, additional regressors modeling the onsets of
error trials and the times of all key-presses were added together with
six regressors accounting for the residual effects of head motion as
regressors of no interest. All regressors were convolved with a
canonical hemodynamic response function.
Group-level random-effects statistics were generated by entering
contrasts of parameter estimates for the different regressors into
between-subjects analyses of variance (ANOVAs). Specifically, to
delineate neural substrates engaged during early acquisition vs.
during expression, contrasts of parameter estimates for the first
two blocks of instrumental learning, and for the devaluation test
phase, were entered for each instrumental condition into a 2 9 2
(condition by experimental phase) ANOVA. Contrasts of parameter
estimates for all learning blocks and for matching blocks were
entered into a separate 2 9 2 (stimulus by task) ANOVA to compare
differences between instrumental conditions during learning with
those between matching control conditions. An analogous ANOVA
was performed using contrasts of parameter estimates for devalua-
tion test trials and matching trials. Finally, a 4 9 2 (acquisition
block by instrumental condition) ANOVA assessed training-related
changes in neural activity, with contrasts of parameter estimates
for each of the four blocks of instrumental learning, for each
instrumental condition. Following estimation of the second-level
model, F-tests were specified by adding linear weights to each
block of learning (e.g. [ 1.5 0.5 0.5 1.5] for increases across
blocks and [1.5 0.5 0.5 1.5] for decreases) in each instrumen-
tal condition.
To assess whether neural discrimination between instrumental
conditions correlated with the degree of devaluation insensitivity, a
simple t-test was performed on first-level interaction contrasts [S-
R > R-O 9 Instrumental > Matching] of parameter estimates from
the learning phase, with the degree of devaluation insensitivity
entered as a covariate. An analogous t-test using interaction con-
trasts of parameter estimates from the devaluation test phase was
also performed, and exclusive functional masks were used to assess
the specificity of neural effects, such that all voxels that reached sig-
nificance at a threshold of P < 0.1 when assessing correlates of
devaluation insensitivity during the test phase were removed from
the effects observed during learning, and vice versa. Exclusive func-
tional masks were also used to assess the directions of simple effects
underlying the Instrumental Condition by Experimental Phase inter-
action. Finally, we used exclusive functional masks to selectively
assess training-related increases vs. decreases in neural activity, such
that, for example, when testing for increases in neural activity across
blocks of training in the R-O condition, voxels were removed
that reached significance at P < 0.1 for the same test in the S-R
condition.
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Small volume corrections (SVCs) were performed on several a-
priori regions of interest using a 10 mm sphere, with center coordi-
nates obtained from highly relevant studies. Specifically, in an anal-
ogous study on observational learning (Liljeholm et al., 2012), we
found selective recruitment of the extrastriate cortex (45, 72, 9),
the tail of the caudate nucleus/thalamus (21, 30, 9) and the lin-
gual gyrus (12, 69, 0) by the S-R condition during acquisition.
Conversely, areas selectively recruited by the R-O condition in that
same observational learning study, and also identified in our other
work on goal-directed performance (Liljeholm et al., 2011), include
the inferior parietal lobule (IPL) (51, 52, 33) and anterior cau-
date nucleus (16, 8, 19). Finally, several human neuroimaging
studies have implicated a posterior region of the putamen (33,
24, 0) in habit formation (Tricomi et al., 2009; de Wit et al.,
2012; Wunderlich et al., 2012; Lee et al., 2014), and the ventrome-
dial prefrontal cortex (4, 55, 7) in goal-directed processes
(Hampton et al., 2006; Tanaka et al., 2008; Liljeholm et al., 2011).
It should be noted that, although specified a priori based on a clo-
sely related literature, effects in several of these regions survived
whole-brain cluster-size thresholding (CST) correction, as well as
SVC, in the current study.
As noted, the putamen, considered a human homolog of the
rodent dorsolateral striatum based on its afferent and efferent projec-
tions (Carelli & West, 1991; Draganski et al., 2008), has been
implicated in habitual performance in several human imaging studies
(Tricomi et al., 2009; de Wit et al., 2012; Wunderlich et al., 2012;
Lee et al., 2014). In contrast, the role of the infralimbic cortex in
instrumental performance has been exclusively demonstrated in
rodents (Killcross & Coutureau, 2003; Haddon & Killcross, 2011;
Smith et al., 2012; Smith & Graybiel, 2013). Consequently, and
based on anatomical and functional evidence for a homology
between the rodent infralimbic cortex and the human subgenual cor-
tex (Ongur & Price, 2000; Ongur et al., 2003; Drevets et al., 2008),
predictions regarding this area were tested using an anatomical mask
of the subgenual cortex (Brodmann area 25), defined using the
Wake Forest University Pickatlas (Maldjian et al., 2003). All other
effects were reported at P < 0.05, using CST of SPM t-maps to
adjust for multiple comparisons (Forman et al., 1995). AlphaSim
(Ward, 2000), a Monte Carlo simulation, was used to determine
cluster size and significance. Using an individual voxel probability
threshold of P = 0.005 indicated that using a minimum cluster size
of 111 Montreal Neurological Institute transformed voxels resulted
in an overall significance of P < 0.05.
To eliminate non-independence bias for plots of contrast values, a
leave-one-subject-out (Esterman et al., 2010) approach was used, in
which 19 general linear models were run with one subject left out
in each, and with each general linear model defining the voxel clus-
ter for the left-out subject. Using rfxplot (Glascher, 2009), mean
contrast values were extracted from spheres (10 mm) centered on
these leave-one-subject-out peaks (identified within regions of inter-
est for SVCs) and were averaged across subjects to plot overall
effect sizes.
Results
Behavioral results
The results from the test phase indicated that our manipulation did
indeed produce differences in devaluation sensitivity. Having cor-
rectly identified the devalued beaker, participants initiated a response
on a significantly greater proportion of devalued trials in the S-R
condition (mean proportion 0.43, SEM 0.09) than in the R-O
European Journal of Neuroscience, 41, 1358–1371

condition (mean proportion 0.19, SEM 0.09) [t18 = 2.3, P < 0.031].
Indeed, whereas only five of 19 participants initiated a response on
any devalued trials in the R-O condition, 15 of 19 participants initi-
ated a response on at least one devalued trial in the S-R condition
(v2 = 10.56, P < 0.005).
Note that devaluation insensitivity was defined such that a
response to obtain the devalued outcome (i.e. to refill the devalued
beaker) was counted even if the entire three-press sequence was not
completed on that trial. Indeed, in the S-R condition, when a partici-
pant responded to fill up the devalued beaker, they often did so
without completing the entire three-press sequence, consistent with
evidence from the rodent literature that the response most proximal
to reward remains sensitive to devaluation long after more distal
responses have become habitual (Killcross & Coutureau, 2003). In
contrast, those few participants that responded on devalued trials in
the R-O condition did so on most or all devalued trials and com-
pleted all three key-presses whenever initiating a response, suggest-
ing a more deliberate decision to respond.
Importantly, during the test phase, whereas in the R-O condition
participants could determine both the accuracy and value of a partic-
ular action solely by identifying the emptied beaker, in the S-R con-
dition, determining the value and accuracy of a given response
required identification of both the abstract cue and the emptied bea-
ker. It is possible therefore that the differences in devaluation perfor-
mance were due to this additional aspect of the S-R condition. The
overall pattern of behavioral results, however, strongly suggests that
this was not the case, and generally rules out task difficulty as the
source of our effects. First, there were no significant differences
between conditions in the percent of incorrect responses, during
either acquisition [R-O: mean 14%, SEM 3; S-R: mean 11%, SEM
2; t18 = 1.1, P = 0.30] or test performance [R-O: mean 4%, SEM 1;
S-R: mean 4%, SEM 2; t18 = 0, P = 1.0]. Likewise, response times
did not differ between conditions during either acquisition [R-O:
mean 1482 ms, SEM 62; S-R: mean 1393 ms, SEM 78; t18 = 1.95,
P = 0.21] or test performance [R-O: mean 1258 ms, SEM 54; S-R:
mean 1361 ms, SEM 68; t18 = 1.52, P = 0.15]. The fact that there
were no significant differences between conditions in either accuracy
or reaction times during test makes it highly unlikely that differ-
ences in devaluation sensitivity reflected differences in task diffi-
culty. Perhaps most pertinently, differences in devaluation
insensitivity were not correlated with differences in accuracy
(P = 0.5) or reaction times (P = 0.8), nor did individual differences
in accuracy or reaction times predict neural effects in any of the
areas identified by our imaging analyses.
To assess the influence of counterbalancing order, we performed
order-by-condition analyses of variance on response times and accu-
racy scores during the learning and test phases, as well as on the
devaluation insensitivity scores. There was no significant interaction
between order and condition for devaluation insensitivity
(F1,17 = 0.55, P = 0.47), or for either response times (F1,17 = 0.89,
P = 0.36) or accuracy scores (F1,17 = 0.07, P = 0.79) during the
test phase. In contrast, during training, there was an anticipated
interaction for both response times (F1,17 = 5.96, P < 0.03) and
accuracy scores (F1,17 = 23.01, P < 0.001), such that response times
were longer and the percent incorrect was greater for whichever
condition came first. Thus, whereas the difference in response times
between instrumental conditions differed across the two orders
(mean difference for R-O first 0.19; mean difference for S-R first
0.32), there was no difference between instrumental conditions
when compared for a given order (i.e. for R-O vs. S-R with both
first, P = 0.84 and for R-O vs. S-R with both second, P = 0.71). A
similar pattern was observed for percent incorrect scores (mean
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Acquisition and expression of habitual control 1363
difference for R-O first 0.08; mean difference for S-R first 0.01;
significance test for R-O vs. S-R with both first, P = 0.26 and for
R-O vs. S-R with both second, P = 0.90). Consequently, we can
rule out counterbalancing order as a source of our behavioral and
imaging effects of interest.
There were no differences between instrumental conditions in the
number of requested repetitions of response pre-training phases [R-
O: mean 1.98, SEM 0.29; S-R: mean 1.75, SEM 0.18; t18 = 0.72,
P = 0.48], or in the total number of key-presses executed during
instrumental learning [R-O: mean 470.63, SEM 32.14; S-R: mean
503.74, SEM 31.63; t18 = 0.85, P = 0.40] or on non-devalued tri-
als during test [R-O: mean 121.84, SEM 4.57; S-R mean: 125.53,
SEM 4.53; t18 = 1.34, P = 0.57]. Finally, there was no difference
between instrumental conditions in the total number of points earned
during instrumental learning [R-O: mean 33.6, SEM 3.61; S-R:
mean 33.4, SEM 3.33; t18 = 0.071, P = 0.94]. This allowed us to
rule out the number of motor responses as well as total earnings as
sources of the difference in devaluation performance.
Neuroimaging results
Our primary objective was to investigate whether neural discrimina-
tion between the two instrumental conditions differed across the
acquisition and expression of behavioral control, particularly with
respect to blood oxygenation level-dependent activity predictive of
individual differences in devaluation insensitivity (see Table 1). To
rule out the possibility that differences between our instrumental
conditions were due to differences in the processing of relevant
visual features (i.e. of abstract cues in the S-R condition and of the
set of beakers in the R-O condition), additional analyses formally
contrasted such differences with those emerging between matching
control conditions (see Table 2). Unless otherwise noted, all effects
reported below survived the subtraction of matching control condi-
tions. All of the results described below survived correction for mul-
tiple comparisons at P < 0.05 using either whole-brain CST or
SVCs based on coordinates from relevant previous studies. Notably,
many of the areas specified a priori as targets of SVC, including
the lingual gyrus, extrastriate cortex and IPL, also survived correc-
tion using whole-brain CST, as indicated in the fourth columns of
Tables 1 and 2.
Discrimination between instrumental conditions during
acquisition vs. performance
To delineate the neural substrates engaged during early acquisition
vs. during expression of operant behavior, we entered the imaging
data from the first two blocks of instrumental learning together with
that from the subsequent test phase, for each instrumental condition,
into a 2 9 2 (condition by experimental phase) ANOVA (see Table 1).
Although processes supporting acquisition should certainly dominate
during the first two blocks of instrumental learning, we cannot com-
pletely rule out the possibility that some element of expression was
also present during this phase. Inclusion of only the very earliest
learning trials would not eliminate this possibility, but would intro-
duce dramatically different error levels and severely reduce the statis-
tical power. We note, however, that the presence of processes related
to expression during the early stages of instrumental learning should
reduce, rather than enhance, discrimination between the levels of the
‘experimental phase’ variable. Thus, in areas where we did observe a
significant difference in discriminatory activity across the different
phases of the experiment, such differences were probably attributable
to the differential effects of acquisition vs. expression.
 1364 M. Liljeholm et al.

Table 1. Imaging results from a 2 9 2 ANOVA contrasting instrumental conditions and training phases

Test Area x, y, z T at peak level Correction* Cluster size at P < 0.005*

Main effects
S-R > R-O Inferior occipital 36, 85, 8 6.33 CST 859
Middle occipital 36, 85, 7 6.82
Superior occipital 27, 67, 28 5.16
SPL 18, 61, 52 4.22
R-O > S-R DMPFC 18, 56, 25 3.62 CST 843
VMPFC 3, 41, 1 3.25 CST
IPL 60, 55, 28 3.99 CST 187
Posterior cing. 3, 34, 43 3.13 CST 190
Insula 39, 14, 41 3.89 CST 157
Acq. > Perf. VMPFC 3, 62, 4 6.12 CST 2181
DMPFC 3, 47, 37 5.65
VS 6, 5, 8 4.90
aCN 9, 20, 7 3.53
VLPFC 45, 32, 11 5.46 CST 354
IPL 51, 64, 34 4.76 CST 200
Perf. > Acq. Cerebellum 27, 61, 26 9.17 CST 9773
SMA 0, 8, 46 5.15
Insula/putamen 33, 8, 4 5.69
Post-central S. 42, 37, 58 6.33
Pre-central S. 30, 4, 52 5.97
Calcarine 24, 58, 10 4.94
SPL 15, 64, 49 5.31
Interactions
S-R > R-O (Perf.>Acq.) Occipital 33, 79, 19 5.10 CST 859
SPL 24, 52, 46 3.58
R-O > S-R (Perf.>Acq.) DMPFC 9, 41, 40 4.26 CST 168

*The absence of an entry in the ‘correction’ and ‘cluster size’ columns indicates that the relevant cluster is continuous (at the P < 0.005 threshold) with that
listed above it. Acq., Acquisition; Perf., Performance; SPL, superior parietal lobule; DMPFC, dorsomedial prefrontal cortex; aCN, anterior caudate nucleus;
cing., cingulate; VLPFC, ventrolateral prefrontal cortex; SMA, supplementary motor area; S., post-/pre-central sulcus.

Table 2. Imaging results from 2 9 2 ANOVAs contrasting instrumental and matching tasks, and from tests of neural correlates of devaluation insensitivity

Contrast Area x, y, z T at peak level Correction* Cluster size at P < 0.005*

Learning phase
S-R > R-O (instr. > ctrl) EBA 39, 79, 10 4.09 SVC 45
R-O > S-R (instr. > ctrl) IPL 57, 58, 31 4.41 CST 275
Insula 33, 14, 4 3.57 CST 421
Precuneus 6, 70, 37 4.28 CST 395
Post. cing. 9, 31, 43 3.36
aCN 12, 2, 16 3.40 SVC 18
Test phase
S-R > R-O (instr. > ctrl) Occipital 30, 79, 19 5.30 CST 1153
SPL 24, 62, 49 4.47
R-O > S-R (instr. > ctrl) IPL 57, 46, 37 3.09 SVC 18
Insula 42, 1, 5 4.06 CST 383
DMPFC 6, 41, 40 3.87 CST 382
Matching effects
S-R control > R-O control Lateral LG 24, 91, 8 5.20 CST 198
R-O control > S-R control Medial LG 9, 76, 8 5.15 CST 1541
Cuneus 9, 91, 19 5.01
Correlates of devaluation insensitivity
Positive correlation CRBL/LG 6, 52, 8 5.06 CST 125
tCN/th 12, 28, 10 4.19 SVC 21
Subgenual 15, 23, 14 5.24 CST 224
VS 9, 5, 14 8.47
Negative correlation IPL 45, 43, 34 4.62 SVC 54

*The absence of an entry in the ‘correction’ and ‘cluster size’ columns indicates that the relevant cluster is continuous (at the P < 0.005 threshold) with that
listed above it. Instr., instrumental; ctrl, control; Post. cing., posterior cingulate; aCN, anterior caudate nucleus; SPL, superior parietal lobule; DMPFC, dorsome-
dial prefrontal cortex; tCN/th, tail of the caudate nucleus/thalamus; CRBL, cerebellum; LG, lingual gyrus.

Stimulus–Response-related activity
superior parietal lobule. Interaction tests and (exclusively masked)
Activity that was greater in the S-R than the R-O condition (Fig. 2) tests of simple effects revealed that activity in the superior parietal
emerged in the middle and superior occipital cortex, and in the lobule and superior occipital cortex was significantly greater in the

© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159


Fig. 2. Areas selectively involved in the S-R condition. Statistical maps show main effects of the S-R > R-O contrast, in the middle occipital cortex (top row),
and an interaction effect {i.e. [S-R > R-O (performance > acquisition)]} in the superior occipital cortex, extending into the superior parietal lobule (middle and
bottom rows). Bar plots show contrast values estimated at leave-one-subject-out coordinates for each instrumental condition, during both acquisition and perfor-
mance, and for matching controls. Error bars = SEM. a.u., arbitrary units.
S-R than the R-O condition only during test performance, but not
during early acquisition (Table 1). Moreover, during early acquisi-
tion, only a small area of the middle occipital cortex, the extrastriate
body area (EBA; SVC), survived subtraction of corresponding
matching controls (Table 2). In contrast, S-R selective activity
throughout the occipital cortex, including the EBA, survived sub-
traction of matching controls during the test phase.
Response–Outcome-related activity
Activity significantly greater in the R-O condition than the S-R condi-
tion (Fig. 3) was observed throughout a fronto-parietal network,
including the dorsal and ventral medial prefrontal cortex (VMPFC),
IPLs, posterior cingulate and bilateral insula. Interaction tests and
(exclusively masked) tests of simple effects revealed that, in the dorso-
medial prefrontal cortex, discrimination between instrumental condi-
tions occurred during test performance only (Table 1). Analyses
assessing differences between conditions relative to matching controls
also revealed R-O selective effects in the dorsal anterior caudate, but
did not yield significant effects in the VMFPC (Table 2).
Discrimination between conditions across blocks of acquisition
To further assess differential activity related to acquisition processes,
we tested for differences between conditions in training-dependent
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Acquisition and expression of habitual control 1365
changes in neural activity, by adding linear weights to blocks of
instrumental learning. An interaction test assessing neural activity
that increased in the R-O condition and decreased in the S-R condi-
tion yielded significant effects throughout the right putamen and glo-
bus pallidus (CST, 33, 7, 5) and in the right IPL (CST, 63,
31, 37). The reverse interaction contrast, assessing neural activity
that decreased in the R-O condition and increased in the S-R condi-
tion, did not reveal any significant effects. Specific assessments of
increases vs. decreases revealed decreases across training blocks in
the S-R condition, but not in the (exclusively masked) R-O condi-
tion, throughout the right putamen and globus pallidus (Fig. 4A). In
contrast, significant increases in activity across blocks in the R-O
condition, but not in the (exclusively masked) S-R condition,
emerged in the right IPL (Fig. 4B).
Neural correlates of devaluation insensitivity
To relate the neuroimaging data to our behavioral effects, we tested
whether neural discrimination between instrumental conditions cor-
related with differences between conditions in the degree of devalua-
tion insensitivity. This was indeed the case. Participants with
stronger activation of the tail of the caudate nucleus/thalamus and of
the cerebellum, extending into the lingual gyrus, in the S-R relative
to the R-O condition during the first two blocks of instrumental
learning responded on a greater proportion of devalued trials in the
 1366 M. Liljeholm et al.

Fig. 3. Areas selectively involved in the R-O condition. Statistical maps show main effects of the R-O > S-R contrast in the posterior ventral IPL (top) and
insula (middle), as well as an interaction effect {i.e. [R-O > S-R (performance > acquisition)]} in the dorsomedial prefrontal cortex (bottom). Bar plots show
contrast values estimated at leave-one-subject-out coordinates for each instrumental condition, during both acquisition and performance, and for matching con-
trols. Error bars = SEM. a.u., arbitrary units.

S-R condition relative to the R-O condition during the subsequent
test phase (Fig. 5A). These effects all survived exclusive masking
by an identical contrast applied, with a threshold of 0.1, to the imag-
ing data from the test phase, suggesting that, in these areas, discrim-
inatory activity during early acquisition, but not during test
performance, predicted differences in devaluation insensitivity.
Conversely, during the test phase, differences in neural activity
between the S-R and R-O conditions in the subgenual cortex (region
of interest) and ventral striatum (VS) were positively correlated with
the difference between conditions in devaluation insensitivity
(Fig. 5B), i.e. greater subgenual and VS activity in the S-R relative
to the R-O condition predicted greater devaluation insensitivity in
the S-R than R-O condition. We also found a negative correlation
with test performance in the right IPL (SVC), such that lesser IPL
activity in the S-R relative to the R-O condition predicted greater
devaluation insensitivity in the S-R than R-O condition. Again, the
specificity of these results was confirmed using exclusive masking
by an identical contrast, at a threshold of 0.1, applied to the imaging
data from the training phase.
Discussion
In this study we explored the neural substrates of goal-directed and
habitual action selection, with a focus on how the recruitment of rel-
evant brain areas might differ across acquisition and implementation
of behavioral control. We scanned human participants with
functional magnetic resonance imaging as they learned and per-
formed a novel task designed to encourage either goal-directed
encoding of the specific outcomes of instrumental responses (R-O
condition), or a habitual mapping of responses to antecedent cues
(S-R condition). In a subsequent test phase, participants were more
likely to respond for a devalued outcome, indicative of habits, in the
S-R condition. We found that neural activity in striatal and cortical
areas (i) discriminated between our two instrumental conditions, (ii)
predicted individual differences in devaluation insensitivity and (iii)
did so differentially across acquisition and test performance.
Our results identified several areas in which the degree of neural
discrimination between instrumental conditions predicted differences
in devaluation insensitivity. Specifically, S-R selective activity in the
tail of the caudate nucleus and cerebellum during learning, but not
during test performance, predicted greater devaluation insensitivity
in the S-R, relative to the R-O, condition. The cerebellum has been
strongly implicated in response automatization (Doyon et al., 1998,
2002; Lang & Bastian, 2002; Balsters & Ramnani, 2011), a resis-
tance to dual-task interference postulated to be closely related to,
and sometimes treated as synonymous with, habitual performance.
Indeed, in the rodent literature, the cerebellum has been directly
implicated in habit formation, such that cerebellar lesions abolish
devaluation insensitivity in overtrained rats (Callu et al., 2007).
However, to our knowledge, no previous study has directly linked
the cerebellum to outcome devaluation insensitivity in humans.
As with the cerebellum, the tail of the caudate nucleus has been

© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159


Fig. 4. Discrimination between conditions across blocks of acquisition. (A) Statistical map showing results from disjunction tests assessing decreases across
blocks of acquisition in the S-R but not the R-O condition with effects emerging in the right putamen/globus pallidus. (B) Results from a disjunction test of
increases across blocks in the R-O but not the S-R condition yielded effects in the IPL. Bar plots show contrast values estimated at leave-one-subject-out coordi-
nates for each instrumental condition in each training block, as well as for matching control conditions. Error bars = SEM. a.u., arbitrary units.
implicated in skill learning (Poldrack & Gabrieli, 2001; Yamamoto
et al., 2013). Moreover, single neuron recordings in the monkey tail
of the caudate nucleus have revealed that, relative to the body and
head of the caudate, this area is specifically involved in encoding
stable, non-flexible, values of visual cues (Kim & Hikosaka, 2013;
Yamamoto et al., 2013; Hikosaka et al., 2014; Kim et al., 2014),
consistent with its role in devaluation insensitivity demonstrated
here and elsewhere (Valentin et al., 2007; Liljeholm et al., 2012).
In the subgenual cingulate (Brodmann area 25), S-R selective
activity during test performance, but not during learning, correlated
with behavioral devaluation insensitivity. Based on its cytoarchitec-
tonic subdivisions and connections, this area has been identified as
homologous to the rodent infralimbic cortex. The rodent infralimbic
and human subgenual areas both project heavily to the shell of the
nucleus accumbens, and both are agranular, relatively poorly lami-
nated areas located ventrally on the medial wall (Gabbott et al.,
1997; Ongur & Price, 2000; Ongur et al., 2003). In rodents, the in-
fralimbic cortex has been shown to make specialized contributions
to executive control processes facilitating the deployment of habits
(Coutureau & Killcross, 2003; Haddon & Killcross, 2011; Smith
et al., 2012; Smith & Graybiel, 2013). Consistent with such find-
ings, our results suggest that a putative human homolog of this area
does indeed play selective roles in the expression of habits. It should
be noted, however, that the rodent infralimbic cortex is primarily
known for its involvement in the extinction of Pavlovian responses
(Milad & Quirk, 2002; Rhodes & Killcross, 2004, 2007; Santini
et al., 2008), whereas the human subgenual cortex has predomi-
nantly featured in studies on depression (Greicius et al., 2007;
Drevets et al., 2008; Johansen-Berg et al., 2008; Matthews et al.,
2009; Keedwell et al., 2010). Future work is needed to reconcile
these apparently divergent functions and their potential homology
across species.
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Acquisition and expression of habitual control 1367
As with the subgenual cortex, activity in the VS during the test
phase, but not during acquisition, predicted devaluation insensitivity.
The VS has been frequently shown to support both the acquisition
and expression of Pavlovian (stimulus–outcome) associations (Day
et al., 2007; Blaiss & Janak, 2009), and to mediate the general
motivational influence of such associations on instrumental perfor-
mance, a phenomenon referred to as Pavlovian-instrumental transfer
(Corbit & Balleine, 2011). Of particular importance for interpreting
the current findings is the fact that, in rodents, Pavlovian-instrumen-
tal transfer appears to selectively influence habitual, rather than
goal-directed, responding, i.e. the greater the degree of insensitivity
to outcome devaluation, the greater the general motivational influ-
ence of Pavlovian cues on instrumental performance (Holland, 2004;
Balleine & Ostlund, 2007). We interpret the currently observed
activity in the VS as reflecting a greater influence of Pavlovian cues
on instrumental responding in the S-R than the R-O condition, and
conjecture that this selective engagement of Pavlovian processes
supported our behavioral effect.
Contrary to our predictions, we did not find a correlation between
activity in the dorsal putamen during early learning and subsequent
devaluation insensitivity. We did, however, find S-R selective
decreases in right putamen activity across blocks of training, an
effect that is consistent with a substantial literature demonstrating a
decreased dependence on the putamen with extended (Ungerleider
et al., 2002; Poldrack et al., 2005; Turner et al., 2005; Ashby et al.,
2010), as well as intermediate (Brovelli et al., 2011), levels of train-
ing. Taken together, these results suggest that the contributions of
the putamen to automatic and habitual behavioral control may take
place early in acquisition, with long-term storage and mediation of
well-trained performance occurring elsewhere (Orban et al., 2010).
However, some neuroimaging studies have found increases in puta-
men activity with extended training (Floyer-Lea & Matthews, 2004;
 1368 M. Liljeholm et al.
A habitual learning paradigm allowed us to do so. It is also worth not-
B not be discernable in analyses that categorically compare conditions.
Fig. 5. Correlations between contrast values for the [S-R > R-O (instrumen-
tal > control)] contrast and behavioral differences between the S-R and R-O
conditions in the proportion of devaluation insensitive responses. Contrast
values are extracted from 8 mm spheres centered on the peak voxel in each
area. (A) Correlation between contrast values from the first two blocks of
instrumental learning (x-axis) and differences between instrumental condi-
tions in the proportion of devaluation insensitive responses (y-axis), showing
effects in the cerebellum. (B) Correlation between contrast values from the
test phase (x-axis) and concurrent differences between instrumental condi-
tions in the proportion of devaluation insensitive responses (y-axis), showing
a positive correlation in the subgenual cortex (top) and a negative correlation
in the right IPL (bottom).
Tricomi et al., 2009; Wunderlich et al., 2012). For instance, Tri-
comi et al. (2009) found an increase in activity with overtraining in
a far posterior region of the right putamen, concomitant with the
development of habitual performance. One key difference between
the study of Tricomi et al. (2009) and the present one is that the S-
R condition in the present study is optimized to generate the rapid
acquisition and expression of habitual behavior, whereas in the
study of Tricomi et al. (2009), behavioral expression of habits (and
perhaps also acquisition) emerged much more slowly, becoming evi-
dent in behavior only after several days of training. Thus, if the
involvement of the putamen in S-R learning dissipates after a period
of stable habitual performance, Tricomi et al. (2009) may not have
sampled behavior beyond that stable period, whereas our accelerated
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
ing, that, whereas Tricomi et al. (2009) reported effects in a small
area in the very far posterior putamen, our effects extend throughout
the right putamen and globus pallidus.
One feature of the present results is that areas in which discrimi-
natory neural activity was correlated with between-subject variation
in devaluation insensitivity did not also show differential main
effects in a comparison of S-R and R-O conditions. This may be
due in part to the fact that our efforts to differentially encourage S-
R and R-O learning resulted only in relative differences in devalua-
tion insensitivity, with the majority of participants showing some
degree of sensitivity even in the S-R condition. Thus, neural pro-
cesses directly related to the degree of habitual responding might
However, not all aspects of encoding S-R associations necessarily
contribute to the dominance of such associations over performance.
For example, the S-R selective decreases in neural activity across
blocks of acquisition, found throughout the right putamen and glo-
bus pallidus, may reflect a gradual disengagement of processes that
provide critical support during the early development of S-R associ-
ations (such as the maintenance of relevant representations, the
encoding of discrepancies between attempted and accurate map-
pings, or the retrieval of response alternatives) but that are sup-
planted during subsequent performance by processes mediating
behavioral control.
We also found selective recruitment by the S-R condition of a lat-
eral middle occipital area that overlaps closely with what has been
termed the ‘EBA’ (Downing et al., 2001). As the name implies, this
region is known for its responses to images of human body parts,
and has also been frequently implicated in action observation as
well as execution (Astafiev et al., 2004; Kuhn et al., 2011; Lilje-
holm et al., 2012). Notably, studies assessing the role of the EBA
in action execution have employed visually guided actions, such that
an arbitrary visual stimulus indicates either the location towards
which a movement should be directed (Astafiev et al., 2004), or
which one of alternative actions should be performed (Kuhn et al.,
2011). It is possible therefore that this area is specifically involved
in limb movements that are largely stimulus-driven. In the current
study, the S-R condition was designed to encourage a mapping of
responses to arbitrary antecedent visual cues by decorrelating actions
from sensory-specific outcome features (and indeed from any visual
features that were intrinsically related to the goal of maintaining
beaker fluids). Thus, one possible explanation for the selective
recruitment of EBA by the S-R condition is that this area mediates
the detection and mapping of arbitrary stimuli to behavioral
responses.
The VMPFC has been implicated in goal-directed performance in
numerous human neuroimaging studies, and is a proposed homolog
of the rodent prelimbic cortex (Balleine & O’Doherty, 2010), lesions
of which disrupt the acquisition, but not the expression, of goal-
directed performance (Ostlund & Balleine, 2005). Although we did
find selective recruitment of the VMPFC by the R-O condition dur-
ing learning, this effect did not survive the subtraction of matching
control conditions. It is possible, of course, that the process sup-
ported by the VMPFC, for example assigning values to subgoals,
was elicited by the relevant stimuli in the matching task as well as
the instrumental task. Another possibility, given our robust effect of
experimental phase in the VMPFC, such that activity was greater
during early acquisition than during test in both conditions, is that
the goal-directed function supported by the VMPFC was present in
both instrumental conditions during early acquisition. Further work
is needed to arbitrate between these possibilities.
European Journal of Neuroscience, 41, 1358–1371

Further dissociating the contributions of medial prefrontal areas,
we found that activity in a more dorsal medial prefrontal region was
greater in the R-O than the S-R condition during test performance,
but not during early acquisition. This region, along with the adjacent
dorsal anterior cingulate, has been implicated in tasks in which the
attainment of goals and rewards requires high levels of cognitive
control, such as when monitoring for unfavorable outcomes, and
during response conflict and decision uncertainty (Ridderinkhof
et al., 2004a,b). The currently observed pattern of activity in this
area may reflect a decrease in performance monitoring and outcome
evaluation during habitual relative to goal-directed control.
In our novel task the stimulus materials were identical across S-R
and R-O conditions, but the instrumental contingencies encouraged
participants to attend to different visual features (cues vs. beakers).
Although we used a simple match-to-sample task to rule out visual
processes involved in selectively attending to such features as a
source of any imaging effects, there may be aspects of visual atten-
tion that are intrinsically related to instrumental responding. For
example, we found greater activity in the R-O than the S-R condi-
tion in a posterior ventral region of the IPL during both the learning
and test phase; during test, discriminatory activity in this area pre-
dicted greater sensitivity to devaluation. The posterior ventral region
of the IPL has been proposed to function as a ‘circuit break’ that
redirects attention towards behaviorally relevant information that is
either exogenously presented (Corbetta & Shulman, 2002; Corbetta
et al., 2008; Cabeza et al., 2012) or retrieved into working memory
(Cabeza et al., 2012). Importantly, substantial evidence also indi-
cates that the posterior ventral region of the IPL is deactivated when
the new information is not relevant to the current task (Shulman
et al., 2003). As can be seen in Fig. 3 (top), activity in the posterior
ventral region of the IPL appears to be deactivated in both instru-
mental conditions relevant to the matching control, but markedly
more so in the S-R condition. A possible explanation for this pattern
of results is that greater suppression was required in the S-R condi-
tion because the sensory features of the beaker subgoal, although
ultimately irrelevant for response selection, was intimately related to
the trial outcome. Indeed, this augmented suppression of sensory-
specific outcome features may be a general property of S-R learning,
particularly as the retrieval of such features into working memory is
considered central to goal-directed encoding (Balleine & Ostlund,
2007).
Consistent with evidence from the rodent literature that goal-
directed performance persists in tasks in which alternative actions
produce distinct rewards (Colwill & Rescorla, 1985; Holland, 2004),
we found that human action selection was more goal-directed in a
condition in which instrumental actions obtained unique subgoals.
Formally, the degree to which alternative actions yield distinct out-
come states can be quantified as the divergence of their outcome
probability distributions. In a previous study (Liljeholm et al.,
2013), we found that activity in the right anterior IPL (the anterior
dorsal supramarginal gyrus) increased with increasing outcome
divergence. Likewise, in the current study, training-dependent
increases in this area were found in the R-O, but not the S-R, condi-
tion, potentially reflecting the incremental increase in divergence as
actions became associated with their respective outcome states. The
selective recruitment of the IPL by the R-O condition, and the corre-
lation between such discriminatory neural activity and behavioral
sensitivity to outcome devaluation, is also consistent with a large
body of research implicating this area in various goal-directed pro-
cesses, including the computation of instrumental contingencies (Seo
et al., 2009; Liljeholm et al., 2011), attribution of intent (den Ouden
et al., 2005), awareness of agency (Farrer et al., 2008), and
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Acquisition and expression of habitual control 1369
mediation of executive function (Friedman & Goldman-Rakic,
1994). Further determination of the exact contributions of the IPL to
goal-directed action selection, its role in corollary attentional pro-
cesses, and in representations of outcome divergence, is an impor-
tant avenue for future work.
In summary, our results are novel in several critical ways. To our
knowledge, we are the first to dissociate the roles of the tail of the
caudate nucleus and VS, across learning and test performance, in
behavioral insensitivity to outcome devaluation. We are also the
first to demonstrate that activity in the IPL, an area that has been
previously implicated in several processes closely linked to goal-
directed action selection, including the attribution of intent and
awareness of agency, predicts sensitivity to outcome devaluation.
Finally, we reveal a potential functional homology between the
human subgenual and rodent infralimbic cortex in the implementa-
tion of habitual control. Taken together, our findings suggest a
broad systems division, at the cortical and subcortical levels,
between brain areas mediating the acquisition and expression of
action–outcome and S-R associations. Notably, a fundamental issue
in the search for treatments of behavioral disorders is how to both
facilitate the automatization of actions that lead to healthful conse-
quences and abolish well-established deleterious habits. An
improved understanding of how distinct neural substrates in the
human brain mediate the acquisition vs. expression of habitual con-
trol, the aim of the present study, is therefore of significant clinical
interest.
Acknowledgements
This work was supported by a grant from the Wellcome Trust and by an
NIH grant (DA033077-01) to J.P.O’D. There are no conflicts of interest.
Abbreviations
CST, cluster-size thresholding; EBA, extrastriate body area; IPL, inferior
parietal lobule; R-O, Response–Outcome; S-R, Stimulus–Response; SVC,
small volume correction; VMPFC, ventromedial prefrontal cortex; VS, ven-
tral striatum.
References
Adams, C.D. (1982) Variations in the sensitivity of instrumental responding
to refinforcer devaluation. Q. J. Exp. Psychol., 34, 77–98.
Adams, C.D. & Dickinson, A. (1981) Instrumental responding following
reinforcer devaluation. Q. J. Exp. Psychol., 33, 109–121.
Ashby, F.G., Turner, B.O. & Horvitz, J.C. (2010) Cortical and basal ganglia
contributions to habit learning and automaticity. Trends Cogn. Sci., 14,
208–215.
Astafiev, S.V., Stanley, C.M., Shulman, G.L. & Corbetta, M. (2004) Extras-
triate body area in human occipital cortex responds to the performance of
motor actions. Nat. Neurosci., 7, 542–548.
Balleine, B.W. & Dickinson, A. (1998) Goal-directed instrumental action:
contingency and incentive learning and their cortical substrates. Neuro-
pharmacology, 37, 407–419.
Balleine, B.W. & O’Doherty, J.P. (2010) Human and rodent homologies in
action control: corticostriatal determinants of goal-directed and habitual
action. Neuropsychopharmacology, 35, 48–69.
Balleine, B.W. & Ostlund, S.B. (2007) Still at the choice-point: action selec-
tion and initiation in instrumental conditioning. Ann. NY Acad. Sci., 1104,
147–171.
Balsters, J.H. & Ramnani, N. (2011) Cerebellar plasticity and the automation
of first-order rules. J. Neurosci., 31, 2305–2312.
Blaiss, C.A. & Janak, P.H. (2009) The nucleus accumbens core and shell are
critical for the expression, but not the consolidation, of Pavlovian condi-
tioned approach. Behav. Brain Res., 200, 22–32.
Brovelli, A., Nazarian, B., Meunier, M. & Boussaoud, D. (2011) Differential
roles of caudate nucleus and putamen during instrumental learning. Neuro-
Image, 57, 1580–1590.
 1370 M. Liljeholm et al.
Cabeza, R., Ciaramelli, E. & Moscovitch, M. (2012) Cognitive contributions
of the ventral parietal cortex: an integrative theoretical account. Trends
Cogn. Sci., 16, 338–352.
Callu, D., Puget, S., Faure, A., Guegan, M. & El Massioui, N. (2007) Habit
learning dissociation in rats with lesions to the vermis and the interpositus
of the cerebellum. Neurobiol. Dis., 27, 228–237.
Carelli, R.M. & West, M.O. (1991) Representation of the body by single
neurons in the dorsolateral striatum of the awake, unrestrained rat. J.
Comp. Neurol., 309, 231–249.
Carelli, R.M., Wolske, M. & West, M.O. (1997) Loss of lever press-related
firing of rat striatal forelimb neurons after repeated sessions in a lever
pressing task. J. Neurosci., 17, 1804–1814.
Colwill, R.M. & Rescorla, R.A. (1985) Instrumental responding remains
insensitive to reinfrocer devaluation after extensive training. J. Exp. Psy-
chol. Anim. B., 11, 520–536.
Corbetta, M. & Shulman, G.L. (2002) Control of goal-directed and stimulus-
driven attention in the brain. Nat. Rev. Neurosci., 3, 201–215.
Corbetta, M., Patel, G. & Shulman, G.L. (2008) The reorienting system of the
human brain: from environment to theory of mind. Neuron, 58, 306–324.
Corbit, L.H. & Balleine, B.W. (2011) The general and outcome-specific
forms of Pavlovian-instrumental transfer are differentially mediated by the
nucleus accumbens core and shell. J. Neurosci., 31, 11786–11794.
Coutureau, E. & Killcross, S. (2003) Inactivation of the infralimbic prefrontal
cortex reinstates goal-directed responding in overtrained rats. Behav. Brain
Res., 146, 167–174.
Daw, N.D., Niv, Y. & Dayan, P. (2005) Uncertainty-based competition
between prefrontal and dorsolateral striatal systems for behavioral control.
Nat. Neurosci., 8, 1704–1711.
Day, J.J., Roitman, M.F., Wightman, R.M. & Carelli, R.M. (2007) Associa-
tive learning mediates dynamic shifts in dopamine signaling in the nucleus
accumbens. Nat. Neurosci., 10, 1020–1028.
Dickinson, A., Nicholas, D.J. & Adams, C.D. (1983) The effect of the instru-
mental training contingency on susceptibility to reinforcer devaluation. Q.
J. Exp. Psychol.-B., 35, 35–51.
Downing, P.E., Jiang, Y., Shuman, M. & Kanwisher, N. (2001) A cortical
area selective for visual processing of the human body. Science, 293,
2470–2473.
Doya, K., Samejima, K., Katagiri, K. & Kawato, M. (2002) Multiple model-
based reinforcement learning. Neural Comput., 14, 1347–1369.
Doyon, J., Laforce, R. Jr., Bouchard, G., Gaudreau, D., Roy, J., Poirier, M.,
Bedard, P.J., Bedard, F. & Bouchard, J.P. (1998) Role of the striatum, cer-
ebellum and frontal lobes in the automatization of a repeated visuomotor
sequence of movements. Neuropsychologia, 36, 625–641.
Doyon, J., Song, A.W., Karni, A., Lalonde, F., Adams, M.M. & Ungerleider,
L.G. (2002) Experience-dependent changes in cerebellar contributions to
motor sequence learning. Proc. Natl. Acad. Sci. USA, 99, 1017–1022.
Draganski, B., Kherif, F., Kloppel, S., Cook, P.A., Alexander, D.C., Parker,
G.J., Deichmann, R., Ashburner, J. & Frackowiak, R.S. (2008) Evidence
for segregated and integrative connectivity patterns in the human Basal
Ganglia. J. Neurosci., 28, 7143–7152.
Drevets, W.C., Savitz, J. & Trimble, M. (2008) The subgenual anterior cin-
gulate cortex in mood disorders. CNS Spectr., 13, 663–681.
Esterman, M., Tamber-Rosenau, B.J., Chiu, Y.C. & Yantis, S. (2010) Avoid-
ing non-independence in fMRI data analysis: leave one subject out. Neuro-
Image, 50, 572–576.
Farrer, C., Frey, S.H., Van Horn, J.D., Tunik, E., Turk, D., Inati, S. & Graf-
ton, S.T. (2008) The angular gyrus computes action awareness representa-
tions. Cereb. Cortex, 18, 254–261.
Floyer-Lea, A. & Matthews, P.M. (2004) Changing brain networks for visuo-
motor control with increased movement automaticity. J. Neurophysiol., 92,
2405–2412.
Forman, S.D., Cohen, J.D., Fitzgerald, M., Eddy, W.F., Mintun, M.A. &
Noll, D.C. (1995) Improved assessment of significant activation in func-
tional magnetic resonance imaging (fMRI): use of a cluster-size threshold.
Magn. Reson. Med., 33, 636–647.
Friedman, H.R. & Goldman-Rakic, P.S. (1994) Coactivation of prefrontal
cortex and inferior parietal cortex in working memory tasks revealed by
2DG functional mapping in the rhesus monkey. J. Neurosci., 14, 2775–
2788.
Gabbott, P.L., Dickie, B.G., Vaid, R.R., Headlam, A.J. & Bacon, S.J. (1997)
Local-circuit neurones in the medial prefrontal cortex (areas 25, 32 and
24b) in the rat: morphology and quantitative distribution. J. Comp. Neu-
rol., 377, 465–499.
Glascher, J. (2009) Visualization of group inference data in functional neu-
roimaging. Neuroinformatics, 7, 73–82.
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Greicius, M.D., Flores, B.H., Menon, V., Glover, G.H., Solvason, H.B., Ken-
na, H., Reiss, A.L. & Schatzberg, A.F. (2007) Resting-state functional
connectivity in major depression: abnormally increased contributions from
subgenual cingulate cortex and thalamus. Biol. Psychiat., 62, 429–437.
Haddon, J.E. & Killcross, S. (2011) Inactivation of the infralimbic prefrontal
cortex in rats reduces the influence of inappropriate habitual responding in
a response-conflict task. Neuroscience, 199, 205–212.
Hampton, A.N., Bossaerts, P. & O’Doherty, J.P. (2006) The role of the ven-
tromedial prefrontal cortex in abstract state-based inference during decision
making in humans. J. Neurosci., 26, 8360–8367.
Hikosaka, O., Kim, H.F., Yasuda, M. & Yamamoto, S. (2014) Basal ganglia
circuits for reward value-guided behavior. Annu. Rev. Neurosci., 37, 289–
306.
Holland, P.C. (2004) Relations between Pavlovian-instrumental transfer and
reinforcer devaluation. J. Exp. Psychol. Anim. B., 30, 104–117.
Jog, M.S., Kubota, Y., Connolly, C.I., Hillegaart, V. & Graybiel,
A.M. (1999) Building neural representations of habits. Science, 286,
1745–1749.
Johansen-Berg, H., Gutman, D.A., Behrens, T.E., Matthews, P.M.,
Rushworth, M.F., Katz, E., Lozano, A.M. & Mayberg, H.S. (2008) Ana-
tomical connectivity of the subgenual cingulate region targeted with deep
brain stimulation for treatment-resistant depression. Cereb. Cortex, 18,
1374–1383.
Keedwell, P.A., Drapier, D., Surguladze, S., Giampietro, V., Brammer, M. &
Phillips, M. (2010) Subgenual cingulate and visual cortex responses to sad
faces predict clinical outcome during antidepressant treatment for depres-
sion. J. Affect. Disorders, 120, 120–125.
Killcross, S. & Coutureau, E. (2003) Coordination of actions and habits in
the medial prefrontal cortex of rats. Cereb. Cortex, 13, 400–408.
Kim, H.F. & Hikosaka, O. (2013) Distinct basal ganglia circuits controlling
behaviors guided by flexible and stable values. Neuron, 79, 1001–1010.
Kim, H.F., Ghazizadeh, A. & Hikosaka, O. (2014) Separate groups of dopa-
mine neurons innervate caudate head and tail encoding flexible and stable
value memories. Front. Neuroanatomy, 8, 120.
Kuhn, S., Keizer, A., Rombouts, S.A. & Hommel, B. (2011) The functional
and neural mechanism of action preparation: roles of EBA and FFA in
voluntary action control. J. Cognitive Neurosci., 23, 214–220.
Lang, C.E. & Bastian, A.J. (2002) Cerebellar damage impairs automaticity of
a recently practiced movement. J. Neurophysiol., 87, 1336–1347.
Lee, S.W., Shimojo, S. & O’Doherty, J.P. (2014) Neural computations
underlying arbitration between model-based and model-free learning. Neu-
ron, 81, 687–699.
Liljeholm, M., Tricomi, E., O’Doherty, J.P. & Balleine, B.W. (2011) Neural
correlates of instrumental contingency learning: differential effects of
action-reward conjunction and disjunction. J. Neurosci., 31, 2474–2480.
Liljeholm, M., Molloy, C.J. & O’Doherty, J.P. (2012) Dissociable brain sys-
tems mediate vicarious learning of stimulus-response and action-outcome
contingencies. J. Neurosci., 32, 9878–9886.
Liljeholm, M., Wang, S., Zhang, J. & O’Doherty, J.P. (2013) Neural corre-
lates of the divergence of instrumental probability distributions. J. Neuro-
sci., 33, 12519–12527.
Maldjian, J.A., Laurienti, P.J., Kraft, R.A. & Burdette, J.H. (2003) An auto-
mated method for neuroanatomic and cytoarchitectonic atlas-based interro-
gation of fMRI data sets. NeuroImage, 19, 1233–1239.
Matthews, S., Simmons, A., Strigo, I., Gianaros, P., Yang, T. & Paulus, M.
(2009) Inhibition-related activity in subgenual cingulate is associated with
symptom severity in major depression. Psychiat. Res., 172, 1–6.
Milad, M.R. & Quirk, G.J. (2002) Neurons in medial prefrontal cortex signal
memory for fear extinction. Nature, 420, 70–74.
O’Doherty, J.P., Deichmann, R., Critchley, H.D. & Dolan, R.J. (2002) Neu-
ral responses during anticipation of a primary taste reward. Neuron, 33,
815–826.
Ongur, D. & Price, J.L. (2000) The organization of networks within the orbi-
tal and medial prefrontal cortex of rats, monkeys and humans. Cereb. Cor-
tex, 10, 206–219.
Ongur, D., Ferry, A.T. & Price, J.L. (2003) Architectonic subdivision of the
human orbital and medial prefrontal cortex. J. Comp. Neurol., 460, 425–
449.
Orban, P., Peigneux, P., Lungu, O., Albouy, G., Breton, E., Laberenne, F.,
Benali, H., Maquet, P. & Doyon, J. (2010) The multifaceted nature of the
relationship between performance and brain activity in motor sequence
learning. NeuroImage, 49, 694–702.
Ostlund, S.B. & Balleine, B.W. (2005) Lesions of medial prefrontal cortex
disrupt the acquisition but not the expression of goal-directed learning. J.
Neurosci., 25, 7763–7770.
European Journal of Neuroscience, 41, 1358–1371

den Ouden, H.M., Frith, U. & Frith, C.S.J.B. (2005) Thinking about inten-
tions. NeuroImage, 28, 787–796.
Poldrack, R.A. & Gabrieli, J.D. (2001) Characterizing the neural mechanisms
of skill learning and repetition priming: evidence from mirror reading.
Brain, 124, 67–82.
Poldrack, R.A., Sabb, F.W., Foerde, K., Tom, S.M., Asarnow, R.F., Book-
heimer, S.Y. & Knowlton, B.J. (2005) The neural correlates of motor skill
automaticity. J. Neurosci., 25, 5356–5364.
Rhodes, S.E. & Killcross, S. (2004) Lesions of rat infralimbic cortex enhance
recovery and reinstatement of an appetitive Pavlovian response. Learn.
Memory, 11, 611–616.
Rhodes, S.E. & Killcross, A.S. (2007) Lesions of rat infralimbic cortex
enhance renewal of extinguished appetitive Pavlovian responding. Eur. J.
Neurosci., 25, 2498–2503.
Ridderinkhof, K.R., van den Wildenberg, W.P., Segalowitz, S.J. & Carter,
C.S. (2004a) Neurocognitive mechanisms of cognitive control: the
role of prefrontal cortex in action selection, response inhibition, perfor-
mance monitoring, and reward-based learning. Brain Cognition, 56,
129–140.
Ridderinkhof, K.R., Ullsperger, M., Crone, E.A. & Nieuwenhuis, S. (2004b)
The role of the medial frontal cortex in cognitive control. Science, 306,
443–447.
Santini, E., Quirk, G.J. & Porter, J.T. (2008) Fear conditioning and extinc-
tion differentially modify the intrinsic excitability of infralimbic neurons.
J. Neurosci., 28, 4028–4036.
Seo, H., Barraclough, D.J. & Lee, D. (2009) Lateral intraparietal cortex and
reinforcement learning during a mixed-strategy game. J. Neurosci., 29,
7278–7289.
Shulman, G.L., McAvoy, M.P., Cowan, M.C., Astafiev, S.V., Tansy, A.P.,
d’Avossa, G. & Corbetta, M. (2003) Quantitative analysis of attention
and detection signals during visual search. J. Neurophysiol., 90, 3384–
3397.
Smith, K.S. & Graybiel, A.M. (2013) A dual operator view of
habitual behavior reflecting cortical and striatal dynamics. Neuron, 79,
361–374.
Smith, K.S., Virkud, A., Deisseroth, K. & Graybiel, A.M. (2012) Reversible
online control of habitual behavior by optogenetic perturbation of medial
prefrontal cortex. Proc. Natl. Acad. Sci. USA, 109, 18932–18937.
© 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd
European Journal of Neuroscience, 41, 1358–1371
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Acquisition and expression of habitual control 1371
Tanaka, S.C., Balleine, B.W. & O’Doherty, J.P. (2008) Calculating conse-
quences: brain systems that encode the causal effects of actions. J. Neuro-
sci., 28, 6750–6755.
Tang, C., Pawlak, A.P., Prokopenko, V. & West, M.O. (2007) Changes in
activity of the striatum during formation of a motor habit. Eur. J. Neuro-
sci., 25, 1212–1227.
Tricomi, E., Balleine, B.W. & O’Doherty, J.P. (2009) A specific role for pos-
terior dorsolateral striatum in human habit learning. Eur. J. Neurosci., 29,
2225–2232.
Turner, R.S., McCairn, K., Simmons, D. & Bar-Gad, I. (2005) Sequential
motor behavior and the basal ganglia. In Bolam, J.P., Ingham, C.A. &
Magill, P.J. (Eds), The Basal Ganglia VIII (Advances in Behavioral Biol-
ogy). Springer, New York, pp. 563–574.
Ungerleider, L.G., Doyon, J. & Karni, A. (2002) Imaging brain plasticity
during motor skill learning. Neurobiol. Learn. Mem., 78, 553–564.
Valentin, V.V., Dickinson, A. & O’Doherty, J.P. (2007) Determining the
neural substrates of goal-directed learning in the human brain. J. Neuro-
sci., 27, 4019–4026.
Ward, B.D. (2000) Simultaneous inference for fMRI data [Computer soft-
ware manual]. AFNI 3dDeconvolve Documentation, Medical College of
Wisconsin. [Internet] Available http://afni.nimh.nih.gov/afni/doc/manual/
AlphaSim.
de Wit, S., Watson, P., Harsay, H.A., Cohen, M.X., van de Vijver, I. & Rid-
derinkhof, K.R. (2012) Corticostriatal connectivity underlies individual dif-
ferences in the balance between habitual and goal-directed action control.
J. Neurosci., 32, 12066–12075.
Wunderlich, K., Dayan, P. & Dolan, R.J. (2012) Mapping value based plan-
ning and extensively trained choice in the human brain. Nat. Neurosci.,
15, 786–791.
Yamamoto, S., Kim, H.F. & Hikosaka, O. (2013) Reward value-contingent
changes of visual responses in the primate caudate tail associated with a
visuomotor skill. J. Neurosci., 33, 11227–11238.
Yin, H.H., Knowlton, B.J. & Balleine, B.W. (2004) Lesions of dorsolateral
striatum preserve outcome expectancy but disrupt habit formation in
instrumental learning. Eur. J. Neurosci., 19, 181–189.
Yin, H.H., Knowlton, B.J. & Balleine, B.W. (2005) Blockade of NMDA
receptors in the dorsomedial striatum prevents action-outcome learning in
instrumental conditioning. Eur. J. Neurosci., 22, 505–512.
 Copyright of European Journal of Neuroscience is the property of Wiley-Blackwell and its
content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder's express written permission. However, users may print, download, or email
articles for individual use.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 PS67CH11-Wood ARI 1 September 2015 12:57

Review in Advance first posted online

V I E W
E on September 10, 2015. (Changes may
R

still occur before final publication

S
online and in print.)

C E
I N

A
D V A

Psychology of Habit
Wendy Wood and Dennis Rünger
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Department of Psychology, University of Southern California, Los Angeles,

California 90089-1061; email: wendy.wood@usc.edu, dennis.ruenger@gmail.com
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

Annu. Rev. Psychol. 2016. 67:11.1–11.26 Keywords

The Annual Review of Psychology is online at
psych.annualreviews.org
This article’s doi:
10.1146/annurev-psych-122414-033417
Copyright  c 2016 by Annual Reviews.
All rights reserved
automaticity, dual-process models, goals, behavior change, model-free
learning
Abstract
As the proverbial creatures of habit, people tend to repeat the same behaviors
in recurring contexts. This review characterizes habits in terms of their cog-
nitive, motivational, and neurobiological properties. In so doing, we identify
three ways that habits interface with deliberate goal pursuit: First, habits
form as people pursue goals by repeating the same responses in a given con-
text. Second, as outlined in computational models, habits and deliberate goal
pursuit guide actions synergistically, although habits are the efficient, default
mode of response. Third, people tend to infer from the frequency of habit
performance that the behavior must have been intended. We conclude by
applying insights from habit research to understand stress and addiction as
well as the design of effective interventions to change health and consumer
behaviors.

11.1

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

Contents
INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2
HABIT AUTOMATICITY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4
Automatic Cuing of Habits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4
Habit Automaticity and Deliberation About Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.5
HABIT FORMATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.6
Associative and Reward Mechanisms in Habit Learning. . . . . . . . . . . . . . . . . . . . . . . . . . . 11.6
Measures of Habit Strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.7
COMPUTATIONAL MODELS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.8
NEUROBIOLOGY OF HABITS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11.10
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Neural Systems Associated with Habits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11.10

Neural Systems that Integrate Habit and Goal-Directed Action Control . . . . . . . . . . .11.12
FACTORS THAT SHIFT THE BALANCE BETWEEN HABITS
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

AND DELIBERATE GOAL PURSUIT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11.12

Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11.13
Addiction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11.14
INFERENCES ABOUT THE CAUSES OF HABIT PERFORMANCE. . . . . . . . . . . .11.15
CHANGING HABITS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11.16
Interventions to Impede Unwanted Habit Performance . . . . . . . . . . . . . . . . . . . . . . . . . . .11.17
Interventions to Promote Formation of Desired Habits . . . . . . . . . . . . . . . . . . . . . . . . . . .11.17
CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11.18

INTRODUCTION
The yin and yang of the history of habits is closely tied to broader trends in the history of
psychology. William James’s (1890) view that “habit covers a very large part of life,” necessitated
that we “define clearly just what its limits are” (p. 104). Psychology complied, and the habit
construct acquired specific meanings in the behaviorist traditions of Thorndike’s (1898) law of
effect, Hull’s (1943) formalized drive theory, and Skinner’s (1938) operant conditioning. However,
these reinforcement-based models of habit were soon supplanted as the field embraced more
purposive and cognitive perspectives. To Tolman (1948), repeated behaviors reflected learning
of internal representations and maps, and Miller et al. (1960) argued that habits be replaced
with information-processing mechanisms of goal pursuit. Accordingly, cognitive psychology and
decision-making research in the 1960s and 70s developed largely separately from research on habit.
More recent theories have captured the complexity of action control and enabled integration
of these opposing conceptualizations. The concept of automaticity (Shiffrin & Schneider 1977)
and theories of dual information processing (Wason & Evans 1975) provided frameworks
inclusive of habits and thoughtful decision making. Through procedural memory, habits could be
cognitively represented as distinct from other types of implicit processes as well as from explicit,
declarative memories (Squire & Zola-Morgan 1991). Reinforcement learning (RL) research with
animals identified a behavioral criterion for detecting habit performance, involving insensitivity
to changes in rewarding outcomes (Dickinson 1985), and neuroscience began to identify the brain
regions and circuits involved in habitual behavior (Graybiel 1998). Social-cognitive approaches
outlined a variety of ways in which habits interface with goals (Verplanken & Aarts 1999, Ouellette
& Wood 1998), and computational models included goal pursuit and prospective planning as

11.2 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

well as habit-like mechanisms of stimulus-driven or model-free action control (Cooper et al.

2014, Daw et al. 2005). Empirical fuel for the study of habit was provided by evidence of the high
levels of repetition in daily activities (Khare & Inman 2006, Wood et al. 2002). Further framing
these developments is the reading public’s interest in understanding their own habits (e.g.,
Rubin 2015).
In this article, we take stock of the fast-growing research on habits. Our review is necessar-
ily wide reaching, covering a variety of behavioral domains, cognitive tasks, and neuroscientific
findings. We also consider animal learning research when relevant, given that habit learning
mechanisms are largely conserved across mammalian species and do not appear to be degraded
in humans or replaced by higher cortical functions (Bayley et al. 2005). Across these diverse re-
search paradigms, the common elements are habit learning through repeated responding so as
to form context-response associations in memory, and automated habit performance that is rel-
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

atively insensitive to changes in the value or contingency of response outcomes. As we explain,

these definitions of habit learning and performance capture the cognitive and neural mechanisms
involved in habit memory and are reflected in characteristic patterns in habitual responding.
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

The schematic in Figure 1 provides a framework for this review by depicting three ways in
which habits interface with goals to guide behavior. Goals energize and direct action by defining a
desired end state. In our three-pronged model, habits and goals interact through habit formation,
habit performance, and inferences about the causes of behavior. First, goals influence habit for-
mation by initially motivating people to repeat actions and to expose themselves to performance
contexts (see Habit Formation section). This is illustrated by the arrows from goal system to con-
text cues and habitual response in Figure 1. Once habits form, context cues come to automatically
activate the habit representation in memory (see Habit Automaticity section). Second, people act
on the habit in mind as well as on their prevailing goals by tailoring their behavior to the current
circumstances (see Computational Models section). External factors such as stress and distraction
influence the impact of these two processes by reducing the motivation or ability to deliberately
pursue goals and increasing reliance on habits (see section Factors that Shift the Balance Between
Habits and Goals section). As outlined in dual-process frameworks, habits provide a sort of de-
fault response unless people are sufficiently motivated and able to tailor their behavior to current
circumstances. Finally, people make inferences about their goals based on observing their own
frequent behavior, as reflected by the double-headed arrow in Figure 1 between the habitual
response and the goal system (see Inferences about the Causes of Habit Performance section).

Goal
system Inference

Activation or
Exposure inhibition

Memory
Context representation of Habitual Outcome
cues habitual response response

Figure 1
Schematic of three ways in which habits interface with deliberate goal pursuit: through initial repetition and
exposure to contexts during habit formation (illustrated by the arrows from goal system to context cues and
habitual response), through activation or inhibition of the habitual response, and through inferences about
the probable causes of habit responding (reflected by the double-headed arrow between habitual response
and goal system).

www.annualreviews.org • Habit 11.3

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

To conclude the review, we apply insights from basic research on habit to understand stress and
addiction and to design successful behavior-change interventions (see Changing Habits section).

HABIT AUTOMATICITY
The terms habit and automaticity are sometimes used interchangeably. Like other automatic re-
sponses, habits are activated in memory in an autonomous fashion without requiring executive
control (Evans & Stanovich 2013). Habits, however, are not synonymous with automaticity but
are best understood as learned automatic responses with specific features (Wood et al. 2014).
Two defining features of habit automaticity are (a) activation by recurring context cues and
(b) insensitivity to short-term changes in goals (a.k.a., not goal dependent), including changes
in the value of response outcomes and the response-outcome contingency. Additional features
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

that apply to most habitual responses include speed and efficiency, limited thought, rigidity, and
integration of sequences of responses that can be executed as a unit (Seger & Spiering 2011,
Smith & Graybiel 2013). However, each of these additional features may not be assessed in all
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

habit research paradigms. Instrumental learning studies, for example, often do not evaluate re-
sponse speed, whereas learning of motor sequences in cognitive-experimental research is assessed
primarily via changes in response latency.
Habits differ from other automatic, implicit processes including priming, classical conditioning,
and nonassociative learning (Evans & Stanovich 2013, Squire & Zola-Morgan 1991). For example,
the priming of goals, attitudes, or concepts can activate a range of responses, not only the repetition
of a particular well-learned response (see Wood et al. 2014). Even strongly desired goals that stably
characterize people’s motives do not necessarily yield stability in the particular means of goal
pursuit. In contrast, habit automaticity applies to a specific response. Furthermore, unlike habits,
automated goals (e.g., implementation intentions) influence behavior primarily to the extent that
they are consistent with people’s explicit motivations (Sheeran et al. 2005).

Automatic Cuing of Habits

A variety of cues might trigger habit performance, including aspects of physical environments,
other people, and preceding actions in a sequence. Once habits form, perception of the relevant
context cues automatically activates the mental representation of the habitual response. Exposure
to cues might be deliberate, as when sitting at a computer in order to activate thoughts of work.
Or exposure can be inadvertent, as when a chance sighting of a fast-food outlet activates thoughts
of eating. We assume that the memory representation of a habit response is cognitively richer
than a mere motor program that controls response execution. Given that human cognition is
based on integrated sensorimotor units (Hommel 2009), a habitual response will be represented
in terms of response features as well as perceptual features. Specifically, the sensory feedback while
making a response, which gives rise to the experience of performing the action, is included in the
mental representation. As a consequence, a habit cue not only triggers a motor program, it also
activates a multimodal representation, or thought, of the habitual response. Consistent with this
view, Neal et al. (2012) found that runners with strong running habits automatically brought to
mind thoughts of running and jogging when exposed to words designating the physical locations
in which they typically ran.
Once habitual responses are activated, people can act on the response in mind without making
a decision to do so. That is, habit performance follows relatively directly from the perception of
context cues and thoughts about the behavior, reflecting the tight linkage between an internal
action representation and the action itself ( James 1890). For example, when students who

11.4 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

frequently went to the sports stadium on campus were incidentally exposed to an image of the
stadium, they raised their voices as they would habitually in that context, despite no change in
their motivation to speak loudly (Neal et al. 2012). Also, in a study conducted in a local cinema,
participants with stronger habits to eat popcorn at the movies consumed more than those with
weak habits, even when they disliked the popcorn because it was stale and unpalatable (Neal et al.
2011). However, when in a campus meeting room watching music videos, participants with strong
cinema-popcorn eating habits were guided by their preferences and ate little stale popcorn.
Habit performance is typified by the insensitivity to outcomes apparent in speaking loudly
in a quiet laboratory setting and eating popcorn despite disliking it. This insensitivity has been
demonstrated directly in instrumental learning experiments in which participants were first trained
extensively to choose a reward to a certain image cue (e.g., Tricomi et al. 2009). Participants then
ate as much of the reward as they desired, so that they did not want any more of that specific
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

food. Nonetheless, when tested again, extensively trained participants in this paradigm continued
to make the habitual but unwanted choice to the associated image (see also Hogarth et al. 2012b,
Schwabe & Wolf 2010).
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

Triandis’s (1977) early work on behavior prediction sparked social psychologists’ interest in
the idea that repeated behavior becomes more habitual and less dependent on goal pursuit. In
prediction studies, behavioral intentions and habit strength (usually operationalized as frequency
of past performance) are used to predict future performance. In a meta-analysis of 64 such studies,
Ouellette & Wood (1998) found that intentions were stronger predictors of actions that were
performed only occasionally (e.g., getting flu shots) than actions that could be repeated more
regularly (e.g., wearing seat belts). Actions performed regularly apparently became habitual and
persisted with little guidance from intentions (see also Gardner et al. 2011). In addition, habits and
intentions interact in guiding daily variations in behavior. For example, on days when participants’
intentions to engage in physical activity were weaker than usual, they fell back on their exercise
habits and worked out only to the extent that exercise was habitual (Rebar et al. 2014). Also, in
a longitudinal test, habit strength to donate blood determined the relation between intentions
and actual donations (P. Sheeran, G. Godin, M. Conner, and M. Germain, unpublished data).
That is, for participants with weak donation habits, increasingly favorable intentions predicted a
greater number of future donations. However, as habit strength increased, the predictive power
of intentions diminished, and participants with the strongest habits simply repeated their past
donations without input from intentions.
When acting out of habit, the ready response in mind reduces deliberation and narrows focus
even when some explicit decision making is required. In a multiattribute choice task involving a
series of travel mode decisions, participants with stronger habits to ride a bike or drive their car
conducted less extensive information searches, considered fewer action alternatives, and biased the
searches toward their habitual choice (Aarts et al. 1997; see also Betsch et al. 2001). Experimentally
enhancing attention to the decision process temporarily increased alternative choices, but the
habitual choice reemerged with continued decisions (Verplanken et al. 1997). These results may
in part reflect that the repeated activation of one response in a context reduces the cognitive
accessibility of alternatives (Danner et al. 2007). Essentially, people with strong habits process
information in ways that reduce the likelihood that they will consider acting otherwise.

Habit Automaticity and Deliberation About Action

Dual-process theories of decision making and judgment outline the mechanisms that lead people
to respond automatically or to engage in deliberate information processing that draws on the
limited-capacity resource of working memory (Evans & Stanovich 2013). From this view, people

www.annualreviews.org • Habit 11.5

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

often act habitually in the interests of efficiency (Wood et al. 2014). When motivated and able to
engage in deliberate goal pursuit, however, they might identify desired outcomes, set and initiate
behavioral intentions, end actions, and evaluate outcomes (Gollwitzer & Brandstätter 1997).
Dual-process models in psychology often specify that automatic and deliberate systems in-
teract through a default-interventionist architecture (Evans & Stanovich 2013) so that responses
are largely habitual unless the deliberative system intervenes to impose an alternative. In con-
trast, reinforcement-learning computational models of routine behavior and decision making (see
Computational Models section) often invoke the less psychologically plausible parallel-competitive
form of dual-process architecture, in which planning proceeds in parallel with habitual control.
As Evans & Stanovich (2013) point out, parallel processing assumes that a costly central executive
is almost continuously ready to be engaged in planning and deliberation. More recently, some
computational models have adapted a more default interventionist approach in which planning
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

intervenes to alter habits only when necessary (e.g., Pezzulo et al. 2013). It is useful to note that, al-
though we invoke the dual-process framework in this review, action control is influenced by more
than two processes, including automated goal pursuit (Sheeran et al. 2005) as well as Pavlovian
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

conditioning of incentive motivation (Balleine & O’Doherty 2010).

In summary, perception of habit cues automatically activates habit representations, and people
typically carry out the habit in mind. Thus, behavior prediction studies reveal that people often
act out of habit, even when it is in conflict with their intentions. In dual-process models depicting
the ways that habits integrate with goals in guiding behavior, habits are a ready default unless
people are motivated and able to intervene and engage in more deliberate goal pursuit.

HABIT FORMATION
Given that everyday habits develop as people go about pursuing life goals, habit formation is closely
intertwined with goal pursuit. Nonetheless, an implication of the basic context-response mecha-
nism underlying habits is that behavior becomes less responsive to current goals and planning as
habit associations strengthen.
Habits develop through instrumental learning and build on the fundamental principle that
rewarded responses are repeated (Thorndike 1898). When repeatedly pursuing a goal such as
making coffee, people experience covariations between context cues (e.g., coffee filter) and ac-
tions (e.g., measure grounds) that lead to goal attainment. Daily life is full of such repetition. In
experience-sampling research in which people recorded once per hour what they were thinking,
feeling, and doing, about 43% of actions were performed almost daily and usually in the same
context (Wood et al. 2002; see also Khare & Inman 2006). Particular actions, such as types of
food eaten, also tend to be performed in particular physical locations (Liu et al. 2015). Typically,
the learning of context-response associations is an unintended consequence of this repetition.
Suggestive of this automaticity, participants in Wood et al.’s (2002) study often reported that they
did not think about repeated behaviors during performance.

Associative and Reward Mechanisms in Habit Learning

Habits strengthen through associative and reward-learning mechanisms that capture the slow,
incremental nature of habit formation. With each repetition, small changes occur in the cognitive
and neural mechanisms associated with procedural memory. Through Hebbian learning of re-
peated connections, cognitive associations between context cues and a response are strengthened
gradually so that people are prepared to repeat performance when the context cues are encountered
again (called direct cuing by Wood & Neal 2007).

11.6 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

The strength of context-response associations is further modulated by the reward following

the response. At a neural level, midbrain dopamine systems support this reinforcement process. By
signaling reward prediction errors, or the discrepancy between an anticipated and actual reward,
a phasic dopamine response acts as a teaching signal for habit learning in the striatum (Balleine &
O’Doherty 2010). Specifically, the dopaminergic signal that is triggered by an unexpected change
in reward magnitude works retroactively to stamp in associations between the still-active memory
traces of the response and the cues in the performance context (Wise 2004). Thus, dopamine
signals promote habit learning as people initially repeat responses to a reward, but the signals
become less active with repetition, as the reward recurs.
In computational RL models, habit formation is conceptualized as the learning of value signals
that represent the expected future rewards for the different response options in a given context and
provide the basis for selecting an action (see Computational Models section). Value-based selection
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

of habitual responses can be can be regarded as a form of motivated cuing (Wood & Neal 2007).
Another way in which habitual responding continues to be influenced by motivational processes
is through context cues that have become associated with the reward that follows an action. As
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

learned predictors of reward, such Pavlovian context cues can cause the habitual response to be
performed with increased vigor (Balleine & O’Doherty 2010). Of note, the motivational effect of
such Pavlovian predictors of rewarding outcomes is distinct from the motivational value of the
outcome itself. Holland (2004), for example, found that responses extensively trained into habits
were insensitive to changes in outcome value but continued to be influenced by reward-related cues.
A standard finding from animal learning is that habits develop most readily when rewards are
provided on interval schedules, meaning that responses are rewarded only after a time has elapsed.
Such rewards mimic natural resources that are replenished over time. With these schedules,
changes in response rate during the time interval do not change the amount of reward deliv-
ered, reducing the experience of instrumental contingency between the response and the reward
(Dickinson 1985). Thus, interval rewards likely promote habit learning because context-response
associations can form without including a representation of the goal or outcome of the action.
As a caveat to the principle that habits form from repetition, habits do not always emerge with
complex tasks in which different response choices lead to different rewards. In animal research,
even after extensive training at a task involving a lever press (yielding sucrose) or chain pull (yielding
a food pellet), rats failed to form habits—they continued to be sensitive to reward value and ceased
responding for one of these rewards after it had been paired with a toxin (Colwill & Rescorla 1985).
Perhaps in an analogous fashion, decision making in humans impedes habit formation. In a repeated
sequential choice task, participants failed to form habits to the extent that they spontaneously used
a planning strategy and based their choices on the value and probability of response outcomes
(Gillan et al. 2015). Similarly, habit formation was hindered when an instrumental task promoted
planning compared with mapping of responses to cues (Liljeholm et al. 2015). It appears, then,
that deliberative decision making is protective against habit formation even when people respond
repeatedly to particular cues.
In summary, habits are likely to form from responses repeated contiguously with context
cues, especially when responses are rewarded on an interval schedule. Through the activation of
dopamine systems, habits form that are insensitive to current shifts in reward value and structure.
However, planning and making deliberate choices during responding can hinder habit formation.

Measures of Habit Strength

Habitization is a process, with no clear demarcation point when strong habits have formed.
Nonetheless, instrumental learning tasks provide a clear behavioral criterion. That is, habits have

www.annualreviews.org • Habit 11.7

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

formed when participants continue to repeat a well-practiced response after the reward has been
reduced in value (e.g., for food rewards, after consuming to satiety) or is no longer tied to the
response (Dickinson 1985). In these paradigms, habitual responding is evaluated during a subse-
quent extinction phase (rewards withheld) in order to preclude additional learning based on the
changed reward values.
Behavioral indicators of habit strength are captured in a variety of experimental paradigms
beyond the simple motor responses often thought emblematic of habits. Habit strength has been
manipulated in experimental paradigms involving word-association tasks (Hay & Jacoby 1996,
Quinn et al. 2010), choice tasks involving pictorial and other judgment stimuli (de Wit et al.
2009, Gillan et al. 2015), two-stage decision-making tasks (Daw et al. 2011), and problem-solving
tasks such as tower building (Patsenko & Altmann 2010). Strong habits also have been formed
as humans repeatedly navigate through virtual mazes (Marchette et al. 2011) and rats through
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

actual mazes (Packard & Goodman 2013). Echoing insensitivity to reward, strong habits in these
paradigms often emerge as errors, reflecting persistent responding despite task changes in the
correct, rewarding outcome. Habits also emerge as chunked responses into a unit (Dezfouli &
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

Balleine 2013, Graybiel 1998), which is related to the performance gains (reduced response time,
increased accuracy) in many sequential learning tasks (e.g., Lungu et al. 2014). In addition, eye-
tracking measures have been used to capture visual attention to task structures triggering habitual
responses (Patsenko & Altmann 2010).
The strength of everyday habits is typically assessed from people’s self-reports. One method
is to combine self-reported ratings of behavioral frequency with ratings of the stability of the
performance context, reflecting the logic that habits represent the repeated pairing of responses
and recurring context cues (Galla & Duckworth 2015, Wood & Neal 2009). Using an alternative
approach, the Self-Report Habit Index estimates habit strength from a questionnaire measure of
the experience of automaticity and frequency of past performance (Verplanken & Orbell 2003),
which has further been streamlined to measure only automaticity (Gardner 2014, Gardner et al.
2012). However, as Labrecque & Wood (2015) noted, experienced automaticity measures often
fail to assess triggering contexts and may capture automation more generally, as opposed to habit
automaticity that guides performance with limited input from current intentions. However, these
measures can successfully capture habit strength when triggering cues are present (P. Lin, W.
Wood, and J. Monterosso, unpublished data). Perhaps the most valid assessments of everyday
habit strength involve reaction time measures of the accessibility of the habitual response given
exposure to associated context cues (e.g., Neal et al. 2012).
In summary, assessments of habit formation rest most importantly on evidence that responses
are insensitive to changes in rewarding outcomes. Habit formation has been documented in a
variety of laboratory tasks using a variety of behavioral assays. For everyday habits, self-report as-
sessments can capture habit strength, although direct assessment of context-response associations
is probably most effective.

COMPUTATIONAL MODELS
Computational models offer detailed accounts of the cognitive processes that support habit
learning and performance. We selectively focus on models that incorporate habit-like control
systems as well as deliberate goal pursuit. Instead of making the simplified assumption that a
behavior is either goal directed or habitual, these models explore how adaptive behavior can
emerge from the interplay of different models of action control. Competing ideas have been
proposed about how the different action controllers work together to produce a response.
In Cooper et al.’s (2014) goal circuit (GC) model, goals structure the learning of habits and
control their expression. The GC model is an artificial neural network composed of two interlinked

11.8 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

habit and goal subnetworks. The habit system, which was originally proposed by Botvinick & Plaut
(2004), selects actions in a bottom-up manner based on the current stimulus environment and on
internal feedback about the network’s previous state. This response selection process is biased by
input from the goal system. A habit develops gradually as the network repeats the same sequence
of responses while learning to attain goals in a particular environment. Eventually, the habit
system becomes capable of performing a sequence autonomously without goal input. In addition
to guiding learning in the habit system, the goal network enables top-down control over habitual
action sequences, as when a person deliberately overrides a habitual response.
Taatgen et al. (2008) developed a model within the ACT-R (adaptive control of thought–
rational) cognitive architecture that shows how behavioral control shifts from an internal, declar-
ative task representation to environmental cues when acquiring a new action routine. Initially,
explicit task knowledge is used to control behavior in a goal-directed manner. With practice, ex-
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

plicit memory retrieval is gradually transformed into a process by which perceptual cues trigger the
relevant action directly. This proceduralization of explicit knowledge accounts for performance
improvements during skill learning. Additional learning is possible by combining stimulus-cued
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

productions of sequences into a single new production. The creation of new, specialized knowl-
edge structures or chunks that can be used more efficiently is a central element in cognitive theories
of skill acquisition (Newell 1990).
In cognitive neuroscience, the prevalent theoretical perspective is that goal-directed actions and
habits can be described by different classes of RL (Daw et al. 2005, Dolan & Dayan 2013). Goal-
directed, or model-based, learning is a computationally demanding process of mental simulation
and planning. Using this approach, an agent computes on the fly which action maximizes long-
run cumulative reward. Habit formation, in contrast, relies on model-free RL involving trial-
and-error learning to estimate and store the long-run values of actions that are available in the
different states or contexts. Actions are then chosen based on the stored or cached action values,
reflecting predictions about future reward. Model-free control lacks the flexibility of model-based
learning because short-term changes in the reward value of an action outcome have only limited
effect on the cached value. Thus, model-free RL theoretically captures a key property of habits—
insensitivity to changes in reward. Both types of learning are driven by prediction errors, with
model-based learning capturing the discrepancy between the current state and the expected one
and model-free learning capturing the difference between predicted and experienced reward.
In an initial dual-system RL model, Daw et al. (2005) assumed a competitive, winner-take-all
mechanism in which the habit system or the goal system gained control over action, depending on
which system provided the more reliable estimates of action values. However, subsequent investi-
gations favor a more dynamic integration in which both systems contribute to the computation of
action values according to their relative reliabilities (Lee et al. 2014). A related proposal involves
Bayesian model averaging that takes into account prediction accuracy and model complexity. In
this view, with experience, goal-directed actions are replaced by habits because the habit system
becomes increasingly reliable and favored over the computationally more complex goal-directed
system (FitzGerald et al. 2014). Furthermore, is it likely that model-based judgments are imple-
mented only selectively, given the psychological costs of planning (see Habit and Deliberation
section)? In recognition, a number of contingent RL systems engage model-based processing con-
ditional on trade-offs between accuracy and efficiency (Keramati et al. 2011, Pezzulo et al. 2013).
Beyond integrating independently computed action values, some RL models assume a direct
influence of goal values on habit learning. For example, in Gershman et al.’s (2014) two-step de-
cision task, participants changed their choice preferences in the first task step after independently
learning about second-step reward contingencies. To account for this finding, Gershman and col-
leagues argued that the model-based system simulated a complete two-step decision process, and

www.annualreviews.org • Habit 11.9

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

the model-free system learned from the simulations. Similarly, Pezzulo et al. (2013) proposed that
model-based planning can update and improve the value estimates of the model-free system. By
enabling goals to influence the action values represented in habit learning, these models promote
the formation of habits compatible with goals.
The RL approach can account not only for the insensitivity of habits to changes in reward
value but also for the chunking feature of habit automaticity. Theories of hierarchical RL show
that it can be advantageous to concatenate individual actions and treat the sequence as a single
response unit or chunk (Botvinick & Weinstein 2014) because the faster responding enabled by
chunking can lead to greater average reward (Keramati et al. 2011). The chunked units could be
deployed in a goal-directed (Dezfouli & Balleine 2012) or model-free manner on the basis of the
reward history (Botvinick et al. 2009).
Questions remain, however, about the appropriateness of equating model-free learning with
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

habit processes. Dezfouli & Balleine’s (2012) proposal that habits are action chunks that are ac-
quired and controlled through model-based processes marks a radical departure from the common
RL assumption that habits are the result of model-free learning. Even more significant, individual-
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

difference studies have reported that the strength of model-free learning was unrelated to habit
formation and insensitivity to the value of the task outcome (Friedel et al. 2014, Gillan et al. 2015).
Instead, outcome insensitivity was greater among participants evidencing little model-based learn-
ing. It is possible that the standard two-stage RL decision task does not capture the process that
produces outcome-insensitive habits, perhaps because participants have to choose between multi-
ple outcomes with varying reward rates (see discussion in the Associative and Reward Mechanisms
in Habit Learning section). Thus, model-free learning in these tasks might reflect other stimulus-
driven strategies such as simple decision heuristics instead of habits. Future research on model-free
processes may need to develop new experimental tasks to better capture habits.
In summary, habits form as a product of repeated behaviors in the service of goal pursuit.
Learning in the habit system may proceed independently or be guided by the goal system. Recent
theories suggest that rather than being independent action controllers, habit and goal systems
integrate in ways that reflect the reliability of each system and the costs of planning. In some
contingent RL models, the habit system serves as an efficient default, and people plan only when
motivated and able.

NEUROBIOLOGY OF HABITS
From a dual-system perspective, a fundamental objective is to identify brain regions whose activity
is uniquely associated with habitual and goal-directed behavior, respectively. Current neuroscien-
tific research is guided, to an increasing extent, by the computational RL models that we discussed
in the previous section. Thanks to the rapid advancement of functional neuroimaging with hu-
man subjects, it is now possible to relate computational dual-process models to brain functioning
at increasingly fine-grained levels of analysis. The emerging picture is one of a neurocognitive
system that integrates the computations of partially overlapping neural systems of habitual and
goal-directed control.

Neural Systems Associated with Habits

Research conducted with rodents, monkeys, and humans has provided converging evidence that
habitual and goal-directed behaviors are mediated by neural circuits that link cortical brain ar-
eas and the basal ganglia (BG), a collection of subcortical nuclei. These circuits are organized
as anatomically separate reentrant loops, two of which are essential for deliberate and habitual
responding (Burton et al. 2015, Yin & Knowlton 2006). The associative cortico-BG loop supports

11.10 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

working memory functions and goal-directed actions and links the prefrontal cortex (PFC) with
two striatal BG regions, the caudate nucleus and the anterior putamen. The sensorimotor loop
underlies automatic, habitual behaviors and connects the somatosensory and motor cortex with
the medial and posterior putamen. Though anatomically separate, the two loops can interact, for
example, through spiraling dopaminergic connectivity (Haber et al. 2000).
Animal learning studies have demonstrated the importance of the sensorimotor loop for habit-
ual responding. Rats do not acquire a lever-pressing habit when their dorsolateral striatum (DLS),
the equivalent of the primate putamen, is lesioned prior to lever-press training for sucrose. Even
after extended training, these rats continued to be goal directed and pressed the lever less fre-
quently when sucrose was devalued (Yin et al. 2004). Furthermore, when the DLS was inactivated
pharmacologically after a lever-pressing habit was acquired, outcome sensitivity was reinstated
(Yin et al. 2006).
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Suggesting that the DLS is involved in the chunking of individual actions into a sequence,
electrical recordings from neurons in the DLS of rats exhibited a task-bracketing pattern of
activity during habitual runs through a maze—high neuronal activity at the beginning and end of
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

a run, with lower activity in-between. Task bracketing emerged when the learned behavior was
still goal directed (Smith & Graybiel 2013), which indicates that habits develop in parallel with
goal-directed learning but do not influence overt behavior early in training. By contrast, when
goal-directed control is abolished by lesioning associated brain regions such as the rodent posterior
dorsomedial striatum (DMS, corresponding to the caudate nucleus in primates) or the prelimbic
medial PFC, behavior immediately comes under habitual control (Killcross & Coutureau 2003,
Yin et al. 2005).
Neuroimaging research with human participants implicates similar networks of brain regions.
Practicing sequences of finger movements for days or weeks decreased brain activation in areas
associated with goal-directed control [e.g., premotor and prefrontal cortical areas, anterior cin-
gulate cortex (ACC), and associative BG territories] and increased activation in the sensorimotor
network, including the putamen (e.g., Lehéricy et al. 2005, Steele & Penhune 2010). Participants
who developed a lever-pressing habit for potato chips and candy over three days of training showed
similar increases in activity in the sensorimotor striatum (posterior putamen) both within prac-
tice days as well as across days (Tricomi et al. 2009). Neuroimaging studies of motor sequence
learning further confirmed the role of the sensorimotor striatum in chunk formation, along with
a frontoparietal network and the mediotemporal lobes (Lungu et al. 2014).
The sensorimotor loop is critical for habit learning, and an extended network of brain regions
modulates its activity. Motivational influences on habit acquisition are mediated by ascending
dopamine projections from the substantia nigra pars compacta to the dorsal striatum that modulate
habit plasticity at corticostriatal synapses (Balleine & O’Doherty 2010). Lesioning this nigrostriatal
pathway in rats disrupted habit formation (Faure et al. 2005). Habit acquisition also was impeded
by lesions of the amygdala central nucleus, most likely due to its effect on substantia nigra pars
compacta function (Lingawi & Balleine 2012). Finally, rodents’ infralimbic cortex, a medial PFC
region, directly participates in the formation of a habit and is required for its expression (Killcross
& Coutureau 2003, Smith & Graybiel 2013).
Whether the sensorimotor loop is responsible for long-term habit storage remains unclear. Af-
ter six months of practicing sequences of joystick movements, monkeys that had their sensorimotor
loop disrupted pharmacologically were not impaired in the expression of sequence knowledge and
still executed overlearned sequences faster and more accurately than random control sequences
(Desmurget & Turner 2010). One explanation is that, with extensive practice, habit learning is
consolidated in cortical brain areas (Atallah et al. 2007). This possibility fits Penhune & Steele’s
(2012) conclusion that long-term representations of learned skills are encoded in a network of

www.annualreviews.org • Habit 11.11

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

motor cortical regions. In their research, delayed recall of a motor sequence engaged not the BG
but rather cortical regions (primary motor and premotor cortices and parietal lobe). This idea is
supported further by the finding that participants who practiced motor sequences for more than
six weeks showed, late in training, neural specialization in cortical motor areas [primary motor
and premotor cortices and the supplementary motor area (SMA)] (Wymbs & Grafton 2014).
In summary, converging evidence implicates the sensorimotor cortico-BG loop as the core
neural substrate of habit learning and performance. Whether the BG is also required for the
long-term retention of habits is a matter of current debate.

Neural Systems that Integrate Habit and Goal-Directed Action Control

Based on computational RL theory, a stream of neuroimaging research has evaluated performance
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

in sequential decision tasks that elicit both habit-based (model-free) learning and prospective
planning about outcomes (model based). The results are broadly consistent with the mapping of
habitual and goal-directed control onto the associative and sensorimotor cortico-BG loops. For
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

example, Wunderlich et al. (2012) related activity in the posterior putamen to cached action values
(habits) that were acquired through extensive training, and activity in the anterior caudate nucleus
was related to values used in model-based planning. Similarly, Lee et al. (2014) reported that the
posterior putamen, SMA, dorsomedial PFC, and dorsolateral PFC encoded model-free action
values, whereas model-based values were associated with activity in orbitofrontal cortex (OFC)
and medial PFC as well as the ACC (see also de Wit et al. 2009, Valentin et al. 2007).
Recent dual-system RL models propose that response selection is based on action values rep-
resented in ventromedial prefrontal cortex (vmPFC) that are jointly determined by a model-free
(habit) and prospective planning controllers (e.g., Daw et al. 2011). This integration of model-
based and model-free value signals is thought to be conducted by an arbitrator associated with
activity in the inferior lateral PFC, frontopolar cortex, and ACC (Lee et al. 2014). How exactly
such an arbitrator regulates the contribution of each system is still largely unknown. According
to one analysis, shifts in response strategy are achieved primarily by strengthening or inhibiting
the influence of the model-free habit system (Lee et al. 2014). Other findings suggest dynamic
changes in both the habit system and the goal system (Gremel & Costa 2013).
Consistent with the idea that habits develop as people pursue goals, recent evidence suggests
that multiple brain regions participate in both goal-directed and habitual control. For exam-
ple, Lee et al. (2014) found with the multi-step decision task that two regions, the SMA and
dmPFC, represented both model-free and model-based values. Similarly, Gremel & Costa (2013)
trained rats to lever press for sucrose using either a habitual or a goal-directed strategy. They
reported that a large proportion of neurons in DLS, DMS, and OFC participated in both ha-
bitual and goal-directed responding, and that the relative engagement of neurons in these areas
depended on the current response strategy.
In summary, research guided by RL theory has identified the neural substrates of model-based
(goal-directed) and model-free (habitual) control. These neural systems are partially overlapping,
and their computations are integrated by brain regions that regulate the relative influence of the
two modes of behavioral control.

FACTORS THAT SHIFT THE BALANCE BETWEEN HABITS

AND DELIBERATE GOAL PURSUIT
Given the cognitive and neural features that differentiate habitual responding from more deliber-
ate pursuit of goals, action control proceeds by balancing these and other processes (e.g., Pavlovian

11.12 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

HABIT SLIPS

Habit slips, or errors of inadvertent habit performance, occur primarily when an intended action and a habit share a
performance context or some action component (Norman 1981). In daily life, people appear to slip up by performing
unwanted habits about six times a week, especially when their attention is diverted from the task at hand (Reason
1979). For example, habit slips underlie errors in responding to innocent-appearing email phishing attacks (Vish-
wanath 2015). In laboratory tests, habits were most likely to be performed inadvertently when goal-directed control
was taxed by, for example, advanced age or performing a secondary task (e.g., de Wit et al. 2014, Ruh et al. 2010).
Moving beyond the truism that people make errors when not attending to what they are doing, research has
identified several sources of habit slips. Slips arising from failures in planning reflect limits in motivation or knowl-
edge about completion of task goals, and they are most likely at decision points such as the end-of-task subroutines
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

(Norman 1981, Reason 1979). However, habit slips also emerge from failures in automated, habitual guides to
performance, as when degraded or forgotten representations of task context occur within a sequence of well-
learned actions (Botvinick & Bylsma 2005). In still another analysis, habit slips arise from normal processing,
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

especially open-loop action control in which habits are executed ballistically once they are launched, even when
they are not the optimal response (Dezfouli et al. 2014, Orbell & Verplanken 2010). Thus, habit slips reflect
failures to select the correct action through top-down control or bottom-up activation as well as ballistic habit
performance.

conditioning; de Wit & Dickinson 2009). Tipping the balance between habits and goal pursuit are
factors such as distraction (see Habit Slips sidebar), time pressure, limited task ability, or limited
willpower (for a review, see Wood et al. 2014). These factors drive action control by pitting effi-
ciency in processing against more effortful reliable processing (Evans & Stanovich 2013). That is,
people act on strong habits when they lack the ability and motivation to engage the central executive
in deliberation or, in RL terminology, when pressured by the time costs of model-based planning.
Considerable research reveals increased habit performance with impaired executive function-
ing. For example, when willpower was reduced by previously performing a taxing decision-making
task, participants did not tailor their responses to their current circumstances but instead fell back
on strongly habitual choices (Neal et al. 2013, Vohs et al. 2005). Also, when distracted by per-
forming a demanding task, participants completing a categorization task increased the use of
stimulus-response strategies over rule-based ones (Foerde et al. 2006), and participants complet-
ing a multistep decision task increased model-free responding (Otto et al. 2013a). Furthermore, in
individual-difference paradigms, older adults and those with lower cognitive-control abilities were
less able to leverage higher-order goal representations for model-based responding in order to
overcome habitual, model-free solutions to a variety of tasks (de Wit et al. 2014, Otto et al. 2015). In
like manner, participants possessing low spatial perspective-taking ability, after practicing navigat-
ing a maze, used more habitual and less goal-directed navigation strategies (Marchette et al. 2011).
Stress and drug addiction are of particular interest because of the multiple routes by which they
tip the balance toward habits and away from deliberate decision making. As we explain in the next
sections, these factors not only impede executive processes but also perhaps promote habit learning.

Stress
Acute as well as chronic stress can increase people’s reliance on habits (Schwabe & Wolf 2013). For
example, participants exposed to a combination of physical and psychosocial stressors (immersing a

www.annualreviews.org • Habit 11.13

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

hand into ice water while being monitored by a stranger and videotaped) after instrumental learning
acted more habitually and were less sensitive to changes in the value of task rewards (Schwabe &
Wolf 2010). This stress-induced shift toward habits is due in part to stress impeding deliberate
action control. At the neural level, the shift toward habitual behavior was accompanied by decreased
activity in OFC and medial PFC, brain regions associated with goal-directed learning (Schwabe
et al. 2012). In sequential decision-making tasks, acute stress selectively attenuated deliberate,
model-based control and promoted habit performance in vulnerable participants—those with low
working-memory capacity (Otto et al. 2013b) or high levels of chronic stress (Radenbach et al.
2015). Similarly, in a study of visual classification learning, stressed participants were biased toward
relying on a habit-linked procedural learning strategy at the expense of explicit learning (Schwabe
& Wolf 2012). These results could reflect simply the breakdown of higher-order decision-making
functions under stress, or stress could lead to a shift in the allocation of cognitive resources so that
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

people fall back on habits and other strategies to prevent unreliable performance. From an RL
perspective, stress shifts the balance toward habits over lengthy planning by increasing experienced
time pressure (Doll et al. 2012).
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

Along with impeding deliberate thought, stress also might promote habit acquisition. Research
with rodents suggests that stress can, under specific conditions, facilitate habit learning through
mechanisms associated with dorsolateral striatal function (Dias-Ferreira et al. 2009). In humans,
however, stress does not clearly affect habit formation. To isolate stress effects on habit learning,
stress is induced before training, and learning is assessed after acute stress effects have dissipated so
that these do not affect performance. Administering stress hormones led to improved learning in a
simple stimulus-response task but had no effect on habit learning in a virtual radial maze (Guenzel
et al. 2014a). Furthermore, a combination of pretraining physical and psychosocial stressors actu-
ally impaired habitual performance at maze navigation, albeit only in male participants (Guenzel
et al. 2014b).
In summary, neurophysiological responses to stress increase habitual responding by impeding
deliberate action control and, potentially, by promoting habit formation. More generally, stress
research highlights the benefits of habits for rescuing performance. In support of this functional
role, stressed participants’ task performance was impaired to the extent that they attempted to
engage goal-directed neural systems (Schwabe & Wolf 2013). Given the ready acquisition and
performance of habits, they provide a useful default when threat and pressure derail more thought-
ful responding.

Addiction
From a habit perspective, the path to drug addiction involves not a pathological motivation for
drugs but rather a shift from goal-directed to habitual drug seeking and consumption (Everitt
2014, Hogarth et al. 2013). Initial drug seeking is voluntary and reflects the hedonic value of the
drug. Through instrumental learning with drug rewards, context cues rapidly become associated
with drug use. In addition, Pavlovian mechanisms contribute to the want for drugs and for
cue-evoked cravings (Berridge 2007). These various learning mechanisms are involved in the
cues that come to trigger drug seeking and consumption independently of the drug outcome,
much as people repeat habits with limited sensitivity to goals and outcomes (Zapata et al. 2010).
Phenomenologically, the addict no longer likes the drug yet uses it compulsively, often despite
intentions to quit. Drug-outcome insensitivity is promoted further as repeated exposures build
tolerance to rewarding drug effects.
Drug use promotes habit formation in part by impairing goal-directed control. In illustra-
tion, study participants who had consumed alcohol responded habitually and continued to choose

11.14 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

chocolate in a repeated food-choice task despite having just eaten three chocolate bars (Hogarth
et al. 2012a). Comparable findings have emerged with chronic addicts abstinent at test. For ex-
ample, participants who were obese, obsessive-compulsive, or dependent on methamphetamine
(abstinent at test) showed compromised goal-directed learning at a decision-making task, along
with a maladaptive reliance on habits (Voon et al. 2015). Furthermore, these responses in chronic
addicts were associated with neural markers of lower gray matter volumes in the caudate, medial
OFC, and lateral PFC. Also, alcohol-dependent participants responded habitually after rewards
had been devalued in an associative learning task, and greater reliance on habits was associated
with reduced engagement of brain areas implicated in goal-directed action (vmPFC) and increased
engagement of areas implicated in habit learning (posterior putamen; Sjoerds et al. 2013). Even
when not under drug influence, simply being in the presence of cues repeatedly associated with
drug exposure can disrupt goal-directed responding (Ostlund et al. 2010). In general, goal-directed
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

impairments increasingly narrow addicts’ behavioral repertoires onto drug habits by restricting
their capacity for intentionally selecting alternative actions.
Drug use also promotes habit responding through neurobiological processes that sensitize users
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

to the incentive properties of drugs. Drug rewards appear to engage habits more rapidly than other
reinforcers (see review in Everitt 2014). Stimulants in particular accelerate and consolidate the
development of drug use habits, speeding the neural shifts from associative to sensorimotor areas
typically found with habit formation. The accelerated formation of habits hastens the transition
from initial or occasional user to addict.
In summary, drug exposure hijacks the habit learning system by exerting a continuous pressure
in favor of habitual, context-driven behavior and away from the evaluation of the outcomes of
action. As a result of these habitual and deliberative processes, drug use escalates so that people
ultimately seek drugs compulsively (Redish et al. 2008).

INFERENCES ABOUT THE CAUSES OF HABIT PERFORMANCE

People often are aware of their habitual responses although they are largely unaware of the cuing
mechanism that activates habits. Given this limited introspective access, people’s explanations
for their habitual responses are largely post hoc accounts. According to classic social psychology
theories, when internal cues to action are weak, ambiguous, or uninterpretable, people infer
what their motivations must be from observing their behavior and the context in which it occurs
(Bem 1972).
The simple frequency of habit performance plausibly implies strong, consistent underlying
motives. Such inferences could be correct in an historical sense, because people might accurately
remember the goals that initially guided habit formation. However, they will not be correct
concerning current habit performance. In support of such an inference, participants with stronger
habits were more certain about their behavioral intentions and perceived the behavior as guided
more by their goals than were participants with weaker habits, when in fact the opposite was true—
intentions and goals were particularly poor predictors of strongly habitual behaviors ( Ji & Wood
2007, Neal et al. 2012). People infer that goals underlie a range of habitual behaviors, even habits
of compulsive drug seeking and use. Everitt & Robbins (2005) suggested that addicts’ experience
of wanting a drug is not a precursor of consumption but rather a post hoc rationalization of
compulsive drug-use habits. Similarly, obsessive-compulsive disorders may originate in excessive
habit formation, with irrational threat beliefs then inferred to explain the compulsively repeated
behaviors (Gillan & Robbins 2014). Although people infer intentionality for a wide variety of
automatically cued responses, when internal cues are strong and unambiguous, causal inferences

www.annualreviews.org • Habit 11.15

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

are unnecessary. Then, inconsistencies between habits and goals might just be labeled as such
(e.g., “I can’t help it, it’s a habit”).
The inference of motives behind habit performance is represented in some computational
models as non-goal-mediated routine responding giving rise to goal representations. For example,
in Sun et al.’s (2001) CLARION (Connectionist Learning with Adaptive Rule Induction Online)
model, habitual responses that are controlled through bottom-up procedural knowledge can, over
time, come to be represented in top-down rules via a rule extraction-refinement algorithm. In
Cooper et al.’s (2014) goal circuit model, stimulus-driven habitual responses can activate goal
representations that subsequently provide input to habits. In this way, active goals may be a
consequence, rather than a precursor, of habitual action. From the perspective of RL models, this
inference might occur through the model-based system adopting the values expressed in habit
learning, perhaps to reduce computational overhead (Doll et al. 2012).
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Goal inferences are sparked not only by the simple frequency of habit performance but also by
the positive affect associated with many habits. Habits are likely to be favored due to the ease with
which they can be performed compared with alternatives. Thus, consumers value using existing
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

products and services over new ones because of the difficulty of learning new usage behaviors (e.g.,
computer program updates; Murray & Häubl 2007). Habits also are likely to be viewed positively
due to the fluency or speed and ease of processing associated with frequently performed behaviors.
High fluency is experienced as positive in part because it signals familiarity over uncertainty and
success at processing and understanding, and this positive affect generalizes to evaluation of the
activity (Reber et al. 2004). Habit inferences thus exploit a psychological calculus that favors what
feels easy because it is well practiced over what feels more difficult because it is new. Being favorably
disposed toward habits, people may infer that they intended to perform the response.
Although the inferences that follow habit performance may not be accurate descriptions of
the mechanisms actually generating action, such inferences sometimes contribute to well-being.
Repeated behaviors, such as students’ choices of the same seat in a classroom, heighten feelings
of comfort, confidence, and control despite that these choices might initially be random (Avni-
Babad 2011). Furthermore, habit performance may promote coherence or comprehensibility of
experiences and thus enhance meaning in life (S. Heintzelman and L. King, unpublished obser-
vation). However, inferences about habits are not always beneficial, and the transparency of habit
knowledge to introspection can lead people to underestimate its usefulness. For example, when
highly motivated to perform well, participants with good procedural knowledge at a task overrode
their habits and responded thoughtfully, despite that this impaired task performance (L. Carden,
W. Wood, D. Neal, and A. Pascoe, unpublished observation).
In summary, people may explain habit performance, even addictive habits, by inferring relevant
goals and intentions. Despite being largely erroneous, the inference that habits were intended may
seem intuitively plausible given response frequency. Also relevant, switching costs can discourage
deviating from habits, and experienced fluency can increase liking for them.

CHANGING HABITS
Unwanted habits are at the root of many failed attempts at behavior change. Evidence comes from
Webb & Sheeran’s (2006) meta-analysis of 47 studies that successfully used persuasive appeals
and other interventions to change participants’ intentions. The changed intentions, however, only
yielded change in behaviors that participants performed sporadically (e.g., course enrollment) and
not in behaviors that could be repeated into habits (e.g., seat belt use). Even the largely effective
implementation intentions, or if-then plans to act on intentions at particular times and places (Goll-
witzer & Sheeran 2006), are not successful at controlling many strong habits (Webb et al. 2009).

11.16 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

From a habit perspective, difficulties in changing established behavior patterns do not reflect
people’s continuing desire to perform the old behavior or a failure of willpower. The central
challenge is that old habits continue to be activated automatically by recurring environmental
cues ( J. Labrecque and W. Wood, unpublished observation; Walker et al. 2014). Even after new
habits have been learned, old memory traces are not necessarily replaced (Bouton et al. 2011).
Relapse can occur when old habit memories are activated by prior routines and other context cues.
Addressing the role of habit learning is a central challenge for the next generation of behavior
change interventions (Marteau et al. 2012, Rothman et al. 2015). In response to this challenge,
interventions can be designed to (a) impede the automated cueing of old, unwanted habits as well
as (b) promote the repetition of a new, desired behavior into a habit.
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Interventions to Impede Unwanted Habit Performance

To reduce interference from old habits, behavior change interventions can incorporate mecha-
nisms of inhibition. In research investigating the spontaneous inhibition of unwanted habits in
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

daily life, the most successful strategy involved thinking, “Don’t do it,” and being mindful of slip-
ups (Quinn et al. 2010). This strategy also successfully controlled habit errors when participants
were instructed to use it in an experimental task. As expected, it worked by enhancing cognitive
control, not by decreasing habit strength (Quinn et al. 2010). Also effective is tying inhibitory
plans to the cues that activate unwanted habits (e.g., “After dinner, I’ll skip dessert as usual and
substitute fruit”; Adriaanse et al. 2010).
Interference from old habits also can be reduced by changing cues in performance environ-
ments. Animal research suggests that habit performance is especially impaired when contexts
shift, with goal-directed responding transferring more successfully across contexts (Thrailkill
& Bouton 2015). One way to change habit cues is through managing exposure. For example,
unhealthy eating habits can be curbed by increasing the salience or accessibility of healthy foods
(Sobal & Wansink 2007). A study illustrated how people do this in daily life: At all-you-can-eat
Chinese buffets, patrons with lower body mass index used chopsticks, chose small plates, put
napkins on their laps, and sat with their sides or backs to the buffet (Wansink & Payne 2012).
Another way that habit cues change is through naturally occurring life transitions, such as when
people switch jobs or move house. Habit discontinuity interventions capitalize on this reduced
exposure to cues that trigger old habits (Thøgersen 2012, Verplanken et al. 2008, Walker et al.
2014). Life transitions that alter habit cues can provide a window of opportunity to act on new
intentions without competition from old habits (Wood et al. 2005).

Interventions to Promote Formation of Desired Habits

By encouraging the formation of new habits, behavior change interventions can be designed
to habitize a new behavior so that it is maintained despite short-term desires and temptations.
In an experiment illustrating how healthy habits can arm people against succumbing to food
temptations (P.-Y. Lin, W. Wood, and J. Monterosso, unpublished data), participants learned
in a computerized task to avoid chocolate (study 1) or to choose carrots (study 2). When later
given options of eating unhealthy treats, participants continued to make habitual choices, at least
when context cues automatically triggered the healthy habit. Also, echoing the factors that balance
habits and goals, participants were most likely to act on their healthy habits when they lacked the
willpower to deliberate about food choices.
Despite the promise of habit formation in maintaining desired behaviors, only a few interven-
tions with nonclinical populations have built on the three pillars of habit formation–repetition in

www.annualreviews.org • Habit 11.17

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

stable contexts with appropriate reward schedules (see Lally & Gardner 2013, Rünger & Wood
2015). The importance of frequent responding was illustrated in Lally et al.’s (2010) field exper-
iment in which repetitions of a simple health behavior (e.g., walking after dinner) required from
18 days to as many as 254 days in the same context to become habitual and performed without
thinking. For exercise, Kaushal & Rhodes (2015) estimated that going to the gym became
automatic within six weeks, assuming visits of four times per week. Unfortunately, few health
or other behavioral interventions have adapted interval reward schedules to facilitate habit
formation (Burns et al. 2012). However, the importance of stable cues was demonstrated in
interventions using tooth brushing to cue dental flossing and form flossing habits (e.g., Judah
et al. 2013, Orbell & Verplanken 2010). The few interventions built on the three components
of habit formation have yielded promising results for weight loss (Carels et al. 2014, Lally et al.
2008) and consumption of healthy food in families (Gardner et al. 2014).
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Interestingly, in a habit formation intervention, electronic reminders to perform a desired

behavior increased repetition but also impeded automaticity and learning of context-response
associations (Stawarz et al. 2015). Perhaps reminders engaged deliberate decision making that
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

impaired learning context–response associations (see Associative and Reward Mechanisms in Habit
Learning section). More passive reminders in the form of physical signs, although helpful in
prompting initial repetition and habit formation, ultimately lost potency over time (Tobias 2009).
Additional evidence of the utility of habit formation comes from interventions that did not
specifically target habits. For example, forming gym-going habits enabled new members of a
health club to sustain working out (Armitage 2005), and forming nonsmoking habits enabled
former smokers to remain abstinent a year after the end of a smoking cessation program (Baldwin
et al. 2006). In a study of regular exercisers, approximately 90% had a location or time cue to
exercise, and exercising was more automatic for those who exercised in a routine way and were
cued by a particular location (Tappe et al. 2013).
Research on self-control also suggests the usefulness of habits for maintaining desired be-
haviors. People with high trait self-control do not appear to attain goals through inhibition of
problematic desires but instead through forming habits that allow them to achieve goals without
experiencing unwanted temptations (Galla & Duckworth 2015, Hofmann et al. 2012). Trait self-
control generally fosters proficiency at performing tasks that require automation (de Ridder et al.
2012). Nonetheless, everyone tends to fall back on performing habits—both good and bad—when
they lack the capacity or motivation to make decisions to act in nonhabitual ways (Neal et al. 2013).
In summary, through combating unwanted habits and ensuring that desired behaviors are
repeated in ways that promote habit formation, interventions can promote adoption of behaviors
that endure over time. These interventions adapt the habit strategies that people with effective
self-control use in their daily lives to ensure successful goal attainment.

CONCLUSION
The current state of the science on habits has provided the definition that James (1890) requested,
overturned behaviorists’ conceptions of simple stimulus-response associations, and placed habits
within broader models of goal-directed action. Habits reflect associative learning and the formation
of context-response associations in procedural memory. Once habits form, perception of the
context automatically brings the response to mind, and people often carry out that response. As
habits strengthen, they gradually become independent of the incentive value of their consequences,
and neural activation shifts from associative toward sensorimotor cortico-striatal brain regions.
When repeated in a sequence, habitual responses also may be chunked together and activated as a
unit. In short, our review provides a framework for understanding, predicting, and changing that

11.18 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

common component of everyday life in which behavioral control has been outsourced onto the
context cues contiguous with past performance.
Although habits are largely insensitive to changes in goal structure and value, they interact in
three different ways with deliberate goal pursuit. First, habits form in daily life as people pursue
goals by repeating actions in particular performance contexts. Initially, goals and declarative task
knowledge structure behavior. With repetition, responses and associated context cues are captured
in procedural learning systems. Goals also may contribute to habit formation through heightening
attention to certain stimuli and identifying the value of action outcomes. Given the profusion of
direct and indirect connections between neural circuits underlying goal-directed and habitual
(model-free) behaviors, goals can have a biasing influence on habit formation (Doll et al. 2012).
Cross talk between habit systems and more deliberative action control, especially during habit
formation, is consistent with an evolutionary history in which neural systems supporting more
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

sophisticated planning capacities evolved on top of neural mechanisms associated with habits.
A second interface between goals and habits emerges after habits form. That is, habits provide
an efficient baseline response that likely integrates with more effortful goal pursuit only when
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

necessary, as when habits prove unreliable in a given context or when people are especially mo-
tivated and able to tailor responses to particular circumstances. Various factors impede people’s
ability to deliberate and thus tip the balance toward relying on habits, including time pressure,
distraction, stress, and addiction. Addictive substances may in addition promote habit responding
by accelerating habit learning.
A third way in which goals integrate with habits is through the explanations that people gen-
erate for their habits. Because habit automaticity is inaccessible to subjective experience, people
must infer the reasons for such responses. A plausible inference for repeated behaviors is strong,
consistent underlying motivations and goals.
The research we reviewed highlights a number of advantages to acting habitually. For example,
habit knowledge is protected from short-term whims and occasional happenings, given that habits
form through incremental experience and do not shift readily with changes in people’s goals and
plans. Also, by outsourcing action control to environmental cues, people have a ready response
when distraction, time pressure, lowered willpower, and stress reduce the capacity to deliberate
about action and tailor responses to current environments. Furthermore, habit systems are smart
in the sense that they enable people to efficiently capitalize on environmental regularities.
As we noted at the end of the present review, understanding habits is important from the applied
perspective of human health and welfare. Drug addictions and other compulsions appear to co-
opt habit processes and reduce people’s capacity to purposively guide their behavior. Lifestyle
habits of poor diet and limited exercise are major contributors to chronic diseases. By building on
an understanding of habit mechanisms, addiction treatments as well as interventions to change
lifestyle behaviors may successfully disrupt these unwanted habits and help people to form more
effective habits that meet their goals for healthy, productive lives.

DISCLOSURE STATEMENT
The authors are not aware of any affiliations, memberships, funding, or financial holdings that
might be perceived as affecting the objectivity of this review.

ACKNOWLEDGMENTS
The authors thank Peter Dayan, Sanne de Wit, Benjamin Gardner, Barbara Knowlton, David
T. Neal, Yael Niv, Sheina Orbell, A. Ross Otto, Carol Seger, Kyle Smith, and Bas Verplanken

www.annualreviews.org • Habit 11.19

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

for their thoughtful comments on an earlier version of the article. This article was made possible
through the support of a grant from the John Templeton Foundation. The opinions expressed
are those of the authors and do not necessarily reflect the Foundation’s views.

LITERATURE CITED
Aarts H, Verplanken B, Van Knippenberg A. 1997. Habit and information use in travel mode choices. Acta
Psychol. 96:1–14
Adriaanse MA, Oettingen G, Gollwitzer PM, Hennes EP, de Ridder DTD, de Wit JBF. 2010. When planning
is not enough: fighting unhealthy snacking habits by mental contrasting with implementation intentions
(MCII). Eur. J. Soc. Psychol. 40:1277–93
Armitage CJ. 2005. Can the theory of planned behavior predict the maintenance of physical activity? Health
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Psychol. 24:235–45
Atallah HE, Lopez-Paniagua D, Rudy JW, O’Reilly RC. 2007. Separate neural substrates for skill learning
and performance in the ventral and dorsal striatum. Nat. Neurosci. 10:126–31
Avni-Babad D. 2011. Routine and feelings of safety, confidence, and well-being. Br. J. Psychol. 102:223–44
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

Baldwin AS, Rothman AJ, Hertel AW, Linde JA, Jeffery RW, et al. 2006. Specifying the determinants of the
initiation and maintenance of behavior change: an examination of self-efficacy, satisfaction, and smoking
cessation. Health Psychol. 25:626–34
Balleine BW, O’Doherty JP. 2010. Human and rodent homologies in action control: corticostriatal determi-
nants of goal-directed and habitual action. Neuropsychopharmacology 35:48–69
Bayley PJ, Frascino JC, Squire LR. 2005. Robust habit learning in the absence of awareness and independent
of the medial temporal lobe. Nature 436:550–53
Bem DJ. 1972. Constructing cross-situational consistencies in behavior: some thoughts on Alker’s critique of
Mischel. J. Personal. 40:17–26
Berridge KC. 2007. The debate over dopamine’s role in reward: the case for incentive salience. Psychopharma-
cology 191:391–431
Betsch T, Haberstroh S, Glöckner A, Haar T, Fiedler K. 2001. The effects of routine strength on adaptation
and information search in recurrent decision making. Organ. Behav. Hum. Decis. Process. 84:23–53
Botvinick MM, Bylsma LM. 2005. Distraction and action slips in an everyday task: evidence for a dynamic
representation of task context. Psychol. Bull. Rev. 12:1011–17
Botvinick MM, Niv Y, Barto AC. 2009. Hierarchically organized behavior and its neural foundations: a
reinforcement learning perspective. Cognition 113:262–80
Botvinick MM, Plaut DC. 2004. Doing without schema hierarchies: a recurrent connectionist approach to
normal and impaired routine sequential action. Psychol. Rev. 111:395–429
Botvinick MM, Weinstein A. 2014. Model-based hierarchical reinforcement learning and human action con-
trol. Philos. Trans. R. Soc. B 369:20130480
Bouton ME, Todd TP, Vurbic D, Winterbauer NE. 2011. Renewal after the extinction of free operant
behavior. Learn. Behav. 39:57–67
Burns RJ, Donovan AS, Ackermann RT, Finch EA, Rothman AJ, Jeffery RW. 2012. A theoretically grounded
systematic review of material incentives for weight loss: implications for interventions. Ann. Behav. Med.
44:375–88
Burton AC, Nakamura K, Roesch MR. 2015. From ventral-medial to dorsal-lateral striatum: neural correlates
of reward-guided decision-making. Neurobiol. Learn. Mem. 117:51–59
Carels RA, Burmeister JM, Koball AM, Oehlhof MW, Hinman N, et al. 2014. A randomized trial comparing
two approaches to weight loss: differences in weight loss maintenance. J. Health Psychol. 19:296–311
Colwill RM, Rescorla RA. 1985. Postconditioning devaluation of a reinforcer affects instrumental responding.
J. Exp. Psychol.: Anim. Behav. Process. 11:120–32
Cooper RP, Ruh N, Mareschal D. 2014. The goal circuit model: a hierarchical multi-route model of the
acquisition and control of routine sequential action in humans. Cogn. Sci. 38:244–74
Danner UN, Vries NK, Aarts H. 2007. Habit formation and multiple means to goal attainment: Repeated
retrieval of target means causes inhibited access to competitors. Personal. Soc. Psychol. Bull. 33:1367–79

11.20 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

Daw ND, Gershman SJ, Seymour B, Dayan P, Dolan RJ. 2011. Model-based influences on humans’ choices
and striatal prediction errors. Neuron 69:1204–15
Daw ND, Niv Y, Dayan P. 2005. Uncertainty-based competition between prefrontal and dorsolateral striatal
systems for behavioral control. Nat. Neurosci. 8:1704–11
de Ridder DTD, Lensvelt-Mulders G, Finkenauer C, Stok FM, Baumeister RF. 2012. Taking stock of self-
control: a meta-analysis of how trait self-control relates to a wide range of behaviors. Personal. Soc. Psychol.
Rev. 16:76–99
de Wit S, Corlett PR, Fletcher PC, Dickinson A, Aitken MR. 2009. Differential engagement of the ventrome-
dial prefrontal cortex by goal-directed and habitual behavior toward food pictures in humans. J. Neurosci.
29:11330–38
de Wit S, Dickinson A. 2009. Associative theories of goal-directed behaviour: a case for animal-human trans-
lational models. Psychol. Res. 73:463–76
de Wit S, van de Vijver I, Ridderinkhof KR. 2014. Impaired acquisition of goal-directed action in healthy
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

aging. Cogn. Affect. Behav. Neurosci. 14:647–58

Desmurget M, Turner RS. 2010. Motor sequences and the basal ganglia: kinematics, not habits. J. Neurosci.
30:7685–90
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

Dezfouli A, Balleine BW. 2012. Habits, action sequences and reinforcement learning. Eur. J. Neurosci. 35:1036–
51
Dezfouli A, Balleine BW. 2013. Actions, action sequences and habits: evidence that goal-directed and habitual
action control are hierarchically organized. PLOS Comp. Biol. 9:e1003364
Dezfouli A, Lingawi NW, Balleine BW. 2014. Habits as action sequences: hierarchical action control and
changes in outcome value. Philos. Trans. R. Soc. B 369:20130482
Dias-Ferreira E, Sousa JC, Melo I, Morgado P, Cerqueira JJ. 2009. Chronic stress causes frontostriatal reor-
ganization and affects decision-making. Science 325:621–25
Dickinson A. 1985. Actions and habits: the development of behavioural autonomy. Philos. Trans. R. Soc. B
308:67–78
Dolan RJ, Dayan P. 2013. Goals and habits in the brain. Neuron 80:312–25
Doll BB, Simon DA, Daw ND. 2012. The ubiquity of model-based reinforcement learning. Curr. Opin.
Neurobiol. 22:1075–81
Evans J, Stanovich KE. 2013. Dual-process theories of higher cognition advancing the debate. Perspect. Psychol.
Sci. 8:223–41
Everitt BJ. 2014. Neural and psychological mechanisms underlying compulsive drug seeking habits and drug
memories—indications for novel treatments of addiction. Eur. J. Neurosci. 40:2163–82
Everitt BJ, Robbins TW. 2005. Neural systems of reinforcement for drug addiction: from actions to habits to
compulsion. Nat. Neurosci. 8:1481–89
Faure A, Haberland U, Condé F, El Massioui N. 2005. Lesion to the nigrostriatal dopamine system disrupts
stimulus-response habit formation. J. Neurosci. 25:2771–80
FitzGerald THB, Dolan RJ, Friston KJ. 2014. Model averaging, optimal inference, and habit formation. Front.
Hum. Neurosci. 8:457
Foerde K, Knowlton BJ, Poldrack RA. 2006. Modulation of competing memory systems by distraction. PNAS
103:11778–83
Friedel E, Koch SP, Wendt J, Heinz A, Deserno L, Schlagenhauf F. 2014. Devaluation and sequential
decisions: linking goal-directed and model-based behavior. Front. Hum. Neurosci. 8:587
Galla BM, Duckworth AL. 2015. More than resisting temptation: Beneficial habits mediate the relationship
between self-control and positive life outcomes. J. Personal. Soc. Psychol. 109:508–25
Gardner B. 2014. A review and analysis of the use of “habit” in understanding, predicting and influencing
health-related behaviour. Health Psychol. Rev. doi: 10.1080/17437199.2013.876238
Gardner B, Abraham C, Lally P, de Bruijn GJ. 2012. Towards parsimony in habit measurement: testing the
convergent and predictive validity of an automaticity subscale of the Self-Report Habit Index. Int. J.
Behav. Nutr. Phys. Act. 9:102
Gardner B, de Bruijn GJ, Lally P. 2011. A systematic review and meta-analysis of applications of the Self-Report
Habit Index to nutrition and physical activity behaviours. Ann. Behav. Med. 42:174–87

www.annualreviews.org • Habit 11.21

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

Gardner B, Sheals K, Wardle J, McGowan L. 2014. Putting habit into practice, and practice into habit: a process
evaluation and exploration of the acceptability of a habit-based dietary behaviour change intervention.
Int. J. Behav. Nutr. Phys. Act. 11:135
Gershman SJ, Markman AB, Otto AR. 2014. Retrospective revaluation in sequential decision making: a tale
of two systems. J. Exp. Psychol.: Gen. 143:182–94
Gillan CM, Otto AR, Phelps EA, Daw ND. 2015. Model-based learning protects against forming habits. Cogn.
Affect. Behav. Neurosci. 15:523–36
Gillan CM, Robbins TW. 2014. Goal-directed learning and obsessive-compulsive disorder. Philos. Trans. R.
Soc. B 369:20130475
Gollwitzer PM, Brandstätter V. 1997. Implementation intentions and effective goal pursuit. J. Personal. Soc.
Psychol. 73:186–99
Gollwitzer PM, Sheeran P. 2006. Implementation intentions and goal achievement: a meta-analysis of effects
and processes. In Advances in Experimental Social Psychology, Vol. 38, ed. MP Zanna, pp. 69–119. San
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Diego, CA: Elsevier

Graybiel AM. 1998. The basal ganglia and chunking of action repertoires. Neurobiol. Learn. Mem. 70:119–36
Gremel CM, Costa RM. 2013. Orbitofrontal and striatal circuits dynamically encode the shift between goal-
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

directed and habitual actions. Nat. Commun. 4:2264

Guenzel FM, Wolf OT, Schwabe L. 2014a. Glucocorticoids boost stimulus-response memory formation in
humans. Psychoneuroendocrinology 45:21–30
Guenzel FM, Wolf OT, Schwabe L. 2014b. Sex differences in stress effects on response and spatial memory
formation. Neurobiol. Learn. Mem. 109:46–55
Haber SN, Fudge JL, McFarland NR. 2000. Striatonigrostriatal pathways in primates form an ascending spiral
from the shell to the dorsolateral striatum. J. Neurosci. 20:2369–82
Hay JF, Jacoby LL. 1996. Separating habit and recollection: memory slips, process dissociations, and proba-
bility matching. J. Exp. Psychol.: Learn. Mem. Cogn. 22:1323–35
Hofmann W, Baumeister RF, Förster G, Vohs KD. 2012. Everyday temptations: an experience sampling study
of desire, conflict, and self-control. J. Personal. Soc. Psychol. 102:1318–35
Hogarth L, Attwood AS, Bate HA, Munafò MR. 2012a. Acute alcohol impairs human goal-directed action.
Biol. Psychol. 90:154–60
Hogarth L, Balleine BW, Corbit LH, Killcross S. 2013. Associative learning mechanisms underpinning the
transition from recreational drug use to addiction. Ann. NY Acad. Sci. 1282:12–24
Hogarth L, Chase HW, Baess K. 2012b. Impaired goal-directed behavioural control in human impulsivity.
Q. J. Exp. Psychol. 65:305–16
Holland PC. 2004. Relations between Pavlovian-instrumental transfer and reinforcer devaluation. J. Exp.
Psychol.: Anim. Behav. Proc. 30:104–17
Hommel B. 2009. Action control according to TEC (theory of event coding). Psychol. Res. 73:512–26
Hull CL. 1943. Principles of Behavior: An Introduction to Behavior Theory. New York: Appleton-Century
James W. 1890. The Principles of Psychology. New York: H. Holt
Ji MF, Wood W. 2007. Purchase and consumption habits: not necessarily what you intend. J. Consum. Psychol.
17:261–76
Judah G, Gardner B, Aunger R. 2013. Forming a flossing habit: an exploratory study of the psychological
determinants of habit formation. Br. J. Health Psychol. 18:338–53
Kaushal N, Rhodes RE. 2015. Exercise habit formation in new gym members: a longitudinal study. J. Behav.
Med. 38:652–63
Keramati M, Dezfouli A, Piray P. 2011. Speed/accuracy trade-off between the habitual and the goal-directed
processes. PLOS Comp. Biol. 7:e1002055
Khare A, Inman JJ. 2006. Habitual behavior in American eating patterns: the role of meal occasions. J. Consum.
Res. 32:567–75
Killcross S, Coutureau E. 2003. Coordination of actions and habits in the medial prefrontal cortex of rats.
Cereb. Cortex 13:400–8
Labrecque JS, Wood W. 2015. What measures of habit strength to use? Comment on Gardner 2014. Health
Psychol. Rev. doi: 0.1080/17437199.2014.992030

11.22 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

Lally P, Chipperfield A, Wardle J. 2008. Healthy habits: efficacy of simple advice on weight control based on
a habit-formation model. Int. J. Obes. 32:700–7
Lally P, Gardner B. 2013. Promoting habit formation. Health Psychol. Rev. 7:S137–58
Lally P, Van Jaarsveld CHM, Potts HWW, Wardle J. 2010. How are habits formed: modelling habit formation
in the real world. Eur. J. Neurosci. 40:998–1009
Lee SW, Shimojo S, O’Doherty JP. 2014. Neural computations underlying arbitration between model-based
and model-free learning. Neuron 81:687–99
Lehéricy S, Benali H, Van de Moortele PF, Pélégrini-Issac M, Waechter T, et al. 2005. Distinct basal ganglia
territories are engaged in early and advanced motor sequence learning. PNAS 102:12566–71
Liljeholm M, Dunne S, O’Doherty JP. 2015. Differentiating neural systems mediating the acquisition versus
expression of goal-directed and habitual behavioral control. Eur. J. Neurosci. 41:1358–71
Lingawi NW, Balleine BW. 2012. Amygdala central nucleus interacts with dorsolateral striatum to regulate
the acquisition of habits. J. Neurosci. 32:1073–81
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Liu JL, Han B, Cohen DA. 2015. Associations between eating occasions and places of consumption among
adults. Appetite 87:199–204
Lungu OL, Monchi O, Albouy G, Jubault T, Ballarin E, et al. 2014. Striatal and hippocampal involvement in
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

motor sequence chunking depends on the learning strategy. PLOS ONE 9:e103885
Maher JP, Conroy DE. 2015. Habit strength moderates the effects of daily action planning prompts on physical
activity but not sedentary behavior. J. Sport Exerc. Psychol. 37:97–107
Marchette SA, Bakker A, Shelton AL. 2011. Cognitive mappers to creatures of habit: differential engagement of
place and response learning mechanisms predicts human navigational behavior. J. Neurosci. 31:15264–68
Marteau TM, Hollands GJ, Fletcher PC. 2012. Changing human behavior to prevent disease: the importance
of targeting automatic processes. Science 337:1492–95
Miller GA, Galanter E, Pribram KH. 1960. Plans and the Structure of Behavior. New York: Holt, Rinehart, &
Winston
Murray KB, Häubl G. 2007. Explaining cognitive lock-in: the role of skill-based habits of use in consumer
choice. J. Consum. Res. 34:77–88
Neal DT, Wood W, Drolet A. 2013. How do people adhere to goals when willpower is low? The profits (and
pitfalls) of strong habits. J. Personal. Soc. Psychol. 104:959–75
Neal DT, Wood W, Labrecque JS, Lally P. 2012. How do habits guide behavior? Perceived and actual triggers
of habits in daily life. J. Exp. Soc. Psychol. 48:492–98
Neal DT, Wood W, Wu M, Kurlander D. 2011. The pull of the past: When do habits persist despite conflict
with motives? Personal. Soc. Psychol. Bull. 37:1428–37
Newell A. 1990. Unified Theories of Cognition. Cambridge, MA: Harvard Univ. Press
Norman DA. 1981. Categorization of action slips. Psychol. Rev. 88:1–15
Orbell S, Verplanken B. 2010. The automatic component of habit in health behavior: habit as cue-contingent
automaticity. Health Psychol. 29:374–83
Ostlund SB, Maidment NT, Balleine BW. 2010. Alcohol-paired contextual cues produce an immediate and
selective loss of goal-directed action in rats. Front. Integr. Neurosci. 4:19
Otto AR, Gershman SJ, Markman AB, Daw ND. 2013a. The curse of planning: dissecting multiple
reinforcement-learning systems by taxing the central executive. Psychol. Sci. 24:751–61
Otto AR, Raio CM, Chiang A. 2013b. Working-memory capacity protects model-based learning from stress.
PNAS 52:20941–46
Otto AR, Skatova A, Madlon-Kay S, Daw ND. 2015. Cognitive control predicts use of model-based reinforce-
ment learning. J. Cogn. Neurosci. 27:319–33
Ouellette JA, Wood W. 1998. Habit and intention in everyday life: the multiple processes by which past
behavior predicts future behavior. Psychol. Bull. 124:54–74
Packard MG, Goodman J. 2013. Factors that influence the relative use of multiple memory systems. Hip-
pocampus 23:1044–52
Patsenko EG, Altmann EM. 2010. How planful is routine behavior? A selective-attention model of perfor-
mance in the Tower of Hanoi. J. Exp. Psychol.: Gen. 139:95–116
Penhune VB, Steele CJ. 2012. Parallel contributions of cerebellar, striatal and M1 mechanisms to motor
sequence learning. Behav. Brain Res. 226:579–91

www.annualreviews.org • Habit 11.23

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

Pezzulo G, Rigoli F, Chersi F. 2013. The mixed instrumental controller: using value of information to combine
habitual choice and mental simulation. Front. Psychol. 4:92
Quinn JM, Pascoe A, Wood W, Neal DT. 2010. Can’t control yourself? Monitor those bad habits. Personal.
Soc. Psychol. Bull. 36:499–511
Radenbach C, Reiter AMF, Engert V, Sjoerds Z, Villringer A, et al. 2015. The interaction of acute and chronic
stress impairs model-based behavioral control. Psychoneuroendocrinology 53:268–80
Reason JT. 1979. Actions not as planned: the price of automatization. In Aspects of Consciousness, ed. G
Underwood, R Stevens, pp. 67–89. London: Academic
Rebar AL, Elavsky S, Maher JP, Doerksen SE, Conroy DE. 2014. Habits predict physical activity on days
when intentions are weak. J. Sport Exerc. Psychol. 36:157–65
Reber R, Schwarz N, Winkielman P. 2004. Processing fluency and aesthetic pleasure: Is beauty in the per-
ceiver’s processing experience? Personal. Soc. Psychol. Rev. 8:364–82
Redish AD, Jensen S, Johnson A. 2008. A unified framework for addiction: vulnerabilities in the decision
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

process. Behav. Brain Sci. 31:415–37

Rothman AJ, Gollwitzer PM, Grant AM, Neal DT, Sheeran P, Wood W. 2015. Hale and hearty policies:
how psychological science can create and maintain healthy habits. Perspect. Psychol. Sci. In press
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

Rubin G. 2015. Better than Before. New York: Random House

Ruh N, Cooper RP, Mareschal D. 2010. Action selection in complex routinized sequential behaviors. J. Exp.
Psychol.: Hum. Percept. Perform. 36:955–75
Rünger D, Wood W. 2015. Maintenance of healthy behaviors: forming and changing habits. In Behavioral
Economics and Public Health, ed. C Roberto, I Kawachi. Oxford, UK: Oxford Univ. Press. In press
Schwabe L, Tegenthoff M, Höffken O, Wolf OT. 2012. Simultaneous glucocorticoid and noradrenergic
activity disrupts the neural basis of goal-directed action in the human brain. J. Neurosci. 32:10146–55
Schwabe L, Wolf OT. 2010. Socially evaluated cold pressor stress after instrumental learning favors habits
over goal-directed action. Psychoneuroendocrinology 35:977–86
Schwabe L, Wolf OT. 2012. Stress modulates the engagement of multiple memory systems in classification
learning. J. Neurosci. 32:11042–49
Schwabe L, Wolf OT. 2013. Stress and multiple memory systems: from “thinking” to “doing.” Trends Cogn.
Sci. 17:60–68
Seger CA, Spiering BJ. 2011. A critical review of habit learning and the basal ganglia. Front. Syst. Neurosci.
5:1–9
Sheeran P, Webb TL, Gollwitzer PM. 2005. The interplay between goal intentions and implementation
intentions. Personal. Soc. Psychol. Bull. 31:87–98
Shiffrin RM, Schneider W. 1977. Controlled and automatic human information processing: II. Perceptual
learning, automatic attending and a general theory. Psychol. Rev. 84:127–90
Sjoerds Z, de Wit S, van den Brink W, Robbins TW, Beekman ATF, et al. 2013. Behavioral and neuroimaging
evidence for overreliance on habit learning in alcohol-dependent patients. Transl. Psychiatry 3:e337
Skinner BF. 1938. The Behavior of Organisms. New York: Appleton-Century-Crofts
Smith KS, Graybiel AM. 2013. A dual operator view of habitual behavior reflecting cortical and striatal
dynamics. Neuron 79:361–74
Sobal J, Wansink B. 2007. Kitchenscapes, tablescapes, platescapes, and foodscapes: influences of microscale
built environments on food intake. Environ. Behav. 39:124–42
Squire LR, Zola-Morgan S. 1991. The medial temporal lobe memory system. Science 253:1380–86
Stawarz K, Cox AL, Blandford A. 2015. Beyond self-tracking and reminders: designing smartphone apps that
support habit formation. In Proc. 33rd annu. ACM Conf. Hum. Factors Comput., pp. 2653–62. New York:
ACM
Steele CJ, Penhune VB. 2010. Specific increases within global decreases: a functional magnetic resonance
imaging investigation of five days of motor sequence learning. J. Neurosci. 30:8332–41
Sun R, Merrill E, Peterson T. 2001. From implicit skills to explicit knowledge: a bottom-up model of skill
learning. Cogn. Sci. 25:203–44
Taatgen NA, Huss D, Dickison D, Anderson JR. 2008. The acquisition of robust and flexible cognitive skills.
J. Exp. Psychol.: Gen. 137:548–65

11.24 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

Tappe K, Tarves E, Oltarzewski J, Frum D. 2013. Habit formation among regular exercisers at fitness centers:
an exploratory study. J. Phys. Act. Health 10:607–13
Thøgersen J. 2012. The importance of timing for breaking commuters’ car driving habits. Collegium 12:130–40
Thorndike EL. 1898. Animal intelligence: an experimental study of the associative processes in animals. Psychol.
Monogr. Gen. Appl. 2:1–109
Thrailkill EA, Bouton ME. 2015. Contextual control of instrumental actions and habits. J. Exp. Psychol.: Anim.
Learn. Cogn. 41:69–80
Tobias R. 2009. Changing behavior by memory aids: a social psychological model of prospective memory and
habit development tested with dynamic field data. Psychol. Rev. 116:408–38
Tolman EC. 1948. Cognitive maps in rats and men. Psychol. Rev. 55:189–208
Triandis HC. 1977. Interpersonal Behavior. Monterey, CA: Brooks/Cole Publ.
Tricomi E, Balleine BW, O’Doherty JP. 2009. A specific role for posterior dorsolateral striatum in human
habit learning. Eur. J. Neurosci. 29:2225–32
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.

Valentin VV, Dickinson A, O’Doherty JP. 2007. Determining the neural substrates of goal-directed learning
in the human brain. J. Neurosci. 27:4019–26
Verplanken B, Aarts H. 1999. Habit, attitude, and planned behaviour: Is habit an empty construct or an
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

interesting case of goal-directed automaticity? Eur. Rev. Soc. Psychol. 10:101–34

Verplanken B, Aarts H, Van Knippenberg A. 1997. Habit, information acquisition, and the process of making
travel mode choices. Eur. J. Soc. Psychol. 27:539–60
Verplanken B, Orbell S. 2003. Reflections on past behavior: a self-report index of habit strength. J. Appl. Soc.
Psychol. 33:1313–30
Verplanken B, Walker I, Davis A, Jurasek M. 2008. Context change and travel mode choice: combining the
habit discontinuity and self-activation hypotheses. J. Environ. Psychol. 28:121–27
Vishwanath A. 2015. Examining the distinct antecedents of e-mail habits and its influence on the outcomes of
a phishing attack. J. Comput. -Mediat. Commun. doi: 10.1111/jcc4.12126
Vohs KD, Baumeister RF, Ciarocco NJ. 2005. Self-regulation and self-presentation: regulatory resource
depletion impairs impression management and effortful self-presentation depletes regulatory resources.
J. Personal. Soc. Psychol. 88:632–57
Voon V, Derbyshire K, Rück C, Irvine MA, Worbe Y, et al. 2015. Disorders of compulsivity: a common bias
towards learning habits. Mol. Psychiatry 20:345–52
Walker I, Thomas GO, Verplanken B. 2014. Old habits die hard: travel habit formation and decay during an
office relocation. Environ. Behav. doi: 10.1177/0013916514549619
Wansink B, Payne CR. 2012. Eating behavior and obesity at Chinese buffets. Obesity 16:1957–60
Wason PC, Evans J. 1975. Dual processes in reasoning? Cognition 3:141–54
Webb TL, Sheeran P. 2006. Does changing behavioral intentions engender behavior change? A meta-analysis
of the experimental evidence. Psychol. Bull. 132:249–68
Webb TL, Sheeran P, Luszczynska A. 2009. Planning to break unwanted habits: Habit strength moderates
implementation intention effects on behaviour change. Br. J. Soc. Psychol. 48:507–23
Wise RA. 2004. Dopamine, learning and motivation. Nat. Rev. Neurosci. 5:483–94
Wood W, Labrecque JS, Lin PY, Rünger D. 2014. Habits in dual process models. In Dual Process Theories of
the Social Mind, ed. JW Sherman, B Gawronski, Y Trope, pp. 371–85. New York: Guilford
Wood W, Neal DT. 2007. A new look at habits and the habit-goal interface. Psychol. Rev. 114:843–63
Wood W, Neal DT. 2009. The habitual consumer. J. Consum. Psychol. 19:579–92
Wood W, Quinn JM, Kashy DA. 2002. Habits in everyday life: thought, emotion, and action. J. Personal. Soc.
Psychol. 83:1281–87
Wood W, Tam L, Witt MG. 2005. Changing circumstances, disrupting habits. J. Personal. Soc. Psychol.
88:918–33
Wunderlich K, Dayan P, Dolan RJ. 2012. Mapping value based planning and extensively trained choice in the
human brain. Nat. Neurosci. 15:786–91
Wymbs NF, Grafton ST. 2014. The human motor system supports sequence-specific representations over
multiple training-dependent timescales. Cereb. Cortex. doi: 10.1093/cercor/bhu144
Yin HH, Knowlton BJ. 2006. The role of the basal ganglia in habit formation. Nat. Rev. Neurosci. 7:464–76

www.annualreviews.org • Habit 11.25

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 PS67CH11-Wood ARI 1 September 2015 12:57

Yin HH, Knowlton BJ, Balleine BW. 2004. Lesions of dorsolateral striatum preserve outcome expectancy but
disrupt habit formation in instrumental learning. Eur. J. Neurosci. 19:181–89
Yin HH, Knowlton BJ, Balleine BW. 2006. Inactivation of dorsolateral striatum enhances sensitivity to changes
in the action-outcome contingency in instrumental conditioning. Behav. Brain Res. 166:189–96
Yin HH, Ostlund SB, Knowlton BJ, Balleine BW. 2005. The role of the dorsomedial striatum in instrumental
conditioning. Eur. J. Neurosci. 22:513–23
Zapata A, Minney VL, Shippenberg TS. 2010. Shift from goal-directed to habitual cocaine seeking after
prolonged experience in rats. J. Neurosci. 30:15457–63
Access provided by University of California - Santa Barbara on 09/13/15. For personal use only.
Annu. Rev. Psychol. 2016.67. Downloaded from www.annualreviews.org

11.26 Wood · Rünger

Changes may still occur before final publication online and in print
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 7510 • The Journal of Neuroscience, September 23, 2020 • 40(39):7510–7522

Systems/Circuits

Prefrontal Cortex-Driven Dopamine Signals in the Striatum

Show Unique Spatial and Pharmacological Properties
Martín F. Adrover,1* Jung Hoon Shin,1* Cesar Quiroz,2 Sergi Ferré,2 Julia C. Lemos,1 and
Veronica A. Alvarez1,3
1
Laboratory on Neurobiology of Compulsive Behavior, Intramural Research Program, National Institute on Alcohol Abuse and Alcoholism, National Institutes
of Health, Bethesda, Maryland 20892, 2Integrative Neurobiology Section, National Institute on Drug Abuse, Intramural Research Program, National
Institutes of Health, Baltimore, Maryland 21224, and 3Center on Compulsive Behaviors, National Institutes of Health, Bethesda, Maryland 20892

Dopamine (DA) signals in the striatum are critical for a variety of vital processes, including motivation, motor learning, and reinforce-
ment learning. Striatal DA signals can be evoked by direct activation of inputs from midbrain DA neurons (DANs) as well as cortical
and thalamic inputs to the striatum. In this study, we show that in vivo optogenetic stimulation of prelimbic (PrL) and infralimbic
(IL) cortical afferents to the striatum triggers an increase in extracellular DA concentration, which coincides with elevation of striatal
acetylcholine (ACh) levels. This increase is blocked by a nicotinic ACh receptor (nAChR) antagonist. Using single or dual optogenetic
stimulation in brain slices from male and female mice, we compared the properties of these PrL/IL-evoked DA signals with those
evoked by stimulation from midbrain DAN axonal projections. PrL/IL-evoked DA signals are undistinguishable from DAN evoked DA
signals in their amplitudes and electrochemical properties. However, PrL/IL-evoked DA signals are spatially restricted and preferen-
tially recorded in the dorsomedial striatum. PrL/IL-evoked DA signals also differ in their pharmacological properties, requiring activa-
tion of glutamate and nicotinic ACh receptors. Thus, both in vivo and in vitro results indicate that cortical evoked DA signals rely on
recruitment of cholinergic interneurons, which renders DA signals less able to summate during trains of stimulation and more sensi-
tive to both cholinergic drugs and temperature. In conclusion, cortical and midbrain inputs to the striatum evoke DA signals with
unique spatial and pharmacological properties that likely shape their functional roles and behavioral relevance.
Key words: DA release; dorsomedial striatum; fast-scan cyclic voltammetry; optogenetics; PFC

Significance Statement
Dopamine signals in the striatum play a critical role in basal ganglia function, such as reinforcement and motor learning.
Different afferents to the striatum can trigger dopamine signals, but their release properties are not well understood. Further,
these input-specific dopamine signals have only been studied in separate animals. Here we show that optogenetic stimulation
of cortical glutamatergic afferents to the striatum triggers dopamine signals both in vivo and in vitro. These afferents engage
cholinergic interneurons, which drive dopamine release from dopamine neuron axons by activation of nicotinic acetylcholine
receptors. We also show that cortically evoked dopamine signals have other unique properties, including spatial restriction
and sensitivity to temperature changes than dopamine signals evoked by stimulation of midbrain dopamine neuron axons.

Received May 27, 2020; revised July 22, 2020; accepted Aug. 17, 2020.
Author contributions: M.F.A., J.H.S., and V.A.A. designed research; M.F.A., J.H.S., C.Q., S.F., and J.C.L. performed
Introduction
research; M.F.A., J.H.S., C.Q., S.F., and J.C.L. analyzed data; M.F.A., J.H.S., and V.A.A. wrote the first draft of the paper; The striatum receives dense innervation from midbrain DA neu-
M.F.A., J.H.S., C.Q., S.F., J.C.L., and V.A.A. edited the paper; M.F.A., J.H.S., and V.A.A. wrote the paper. rons (DANs), which are the main source of DA in the striatum.
*M.F.A. and J.H.S. contributed equally to this work. DA plays a critical neuromodulatory role in regulating striatal
This work was supported by the Intramural Programs of National Institute on Alcohol Abuse and
Alcoholism, National Institute of Neurological Disorders and Stroke (ZIA-AA000421), and National Institute on
circuitry and function (Surmeier et al., 2007; Gerfen and
Drug Abuse. We thank Roland Bock (National Institute on Alcohol Abuse and Alcoholism, National Institutes of Surmeier, 2011; Burke et al., 2017). Disruptions in striatal DA
Health) for development of the VIGOR acquisition and analysis software; Drs. Karl Deisseroth (Stanford levels are associated with many neurologic and psychiatric disor-
University) and Ed Boyden for providing the channelrhodopsin-2 and ChrimsonR constructs, respectively; and
ders, such as Parkinson’s disease and substance abuse disorder
members of the A.A.V. laboratory for valuable comments on the manuscript.
M.F. Adrover’s present address: Instituto de Investigaciones en Ingeniería Genética y Biología Molecular, (Gerfen, 2000; Luscher et al., 2020). DA is released from a frac-
CONICET, Buenos Aires, C1428ADN, Argentina. tion of the varicosities distributed along DAN axons, which con-
J.C. Lemos’ present address: Department of Neuroscience, University of Minnesota, Minneapolis, Minnesota 55455. tain specialized active zone-like sites (Sulzer et al., 2016; Liu et
The authors declare no competing financial interests.
al., 2018; Liu and Kaeser, 2019). The axonal arborizations of
Correspondence should be addressed to Veronica A. Alvarez at alvarezva@mail.nih.gov.
https://doi.org/10.1523/JNEUROSCI.1327-20.2020 DANs ramify extensively within the striatum, and a single dopa-
Copyright © 2020 the authors mine axon can spread over a significant area (;3% in average)

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals
of the striatum (Matsuda et al., 2009). DA released from axonal
varicosities generates a rapid increase in local extracellular DA
concentration. This extracellular increase leads to the activation
of multiple types of DA receptors, which are localized in den-
drites, somas, and presynaptic terminals. Ultimately, this impacts
the network activity of the striatum and drives behavior output
(Tritsch and Sabatini, 2012; Chuhma et al., 2014; Mamaligas et
al., 2016; Burke et al., 2017; Shin et al., 2017; Lahiri and Bevan,
2020). This striatal DA signal is known to be triggered by two
mechanisms. One mechanism involves action potential firing
initiated at midbrain DAN somas, which propagates through
dense axonal arborizations to reach active zone-like release sites
in the striatum (Matsuda et al., 2009; Liu et al., 2018). In the
other mechanism, DAN axons in the striatum are locally acti-
vated, independent of the action potential firing at DAN somas.
This mode of DA release requires the activation of nAChRs
expressed on DAN axons and synchronized activation of cholin-
ergic interneurons (CINs), which is thought to give rise to local
release of ACh (Cachope et al., 2012; Threlfell et al., 2012; Wang
et al., 2014; Shin et al., 2015, 2017).
This local intrastriatal trigger of DA release has only recently
been demonstrated and is gaining attention as it represents a
newly discovered means for striatal DA modulation/transmission.
However, little is known about the differential/unique biophysical
and pharmacological properties of locally evoked versus DAN-
evoked DA transmission. Recent in vitro studies using optogenetic
stimulation and fast-scan cyclic voltammetry (FSCV) show that
local DA signals can be triggered in the striatum by stimulation of
glutamatergic inputs from thalamus (Threlfell et al., 2012; Kosillo
et al., 2016; Johnson et al., 2017; Cover et al., 2019), motor cortex
(Kosillo et al., 2016), or PFC (Mateo et al., 2017). These findings
agree with previous literature showing in vivo DA signals in the
striatum evoked by stimulation of PFC (Taber and Fibiger, 1993;
Quiroz et al., 2016; Hill et al., 2018), hippocampus (Tritschler et
al., 2018), or amygdala (Floresco et al., 1998). Furthermore, in vivo
intrastriatal administration of glutamate and ACh was also shown
to cause increases in extracellular DA concentration in the rat
striatum (Giorguieff et al., 1976, 1977; Leviel et al., 1990; Shimizu
et al., 1990). Together, this evidence supports an intrastriatal
mechanism for DA signal generation that can be initiated by exci-
tatory inputs to the striatum that require glutamate and ACh.
While cortically evoked DA signals in the striatum have been
reported, identifying the unique pharmacological and basic prop-
erties of these input-specific signals will allow us to selectively
manipulate and target them, leading to a better understanding of
their functional significance. In this study, we tackle this gap in
knowledge using both in vivo microdialysis and in vitro FSCV
with optogenetic stimulation. Particularly, we set up a novel
approach using dual optogenetic stimulation in the same brain sli-
ces to input-specific evoke DA release and compare the properties
of these DA signals in the striatum.
Materials and Methods
Animals. All animals used in the study were maintained in accord-
ance with the guidelines of the National Institutes of Health Animal
Care, and the animal research procedures were approved by the
National Institute on Alcohol Abuse and Alcoholism Animal Care and
Use Committee for mice and by the National Institute on Drug Abuse
Intramural Research Program Animal Care and Use Committee for rats.
Except for the in vivo microdialysis experiments, all experiments were
conducted using male and female mice of C57BL/6 background.
Heterozygote B6.SJL-Slc6a3tm1.1(cre)Bkmn/J mice (Backman et al., 2006)
(The Jackson Laboratory, 006660) referred to as DATIRES-Cre1 mice and
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
J. Neurosci., September 23, 2020 • 40(39):7510–7522 • 7511
negative DATIRES-Cre littermates considered as WT mice were used. To
fluorescently label CINs, homozygote B6N.129S6(B6)-Chattm2(cre)Lowl/J
mice (Rossi et al., 2011) (The Jackson Laboratory, 018957) were crossed
with homozygote B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J mice
(Madisen et al., 2010) (The Jackson Laboratory, 007914), and the prog-
eny is referred to as CIN-tdTomato mice. For the microdialysis experi-
ments, male Sprague Dawley albino rats (Charles River Laboratories)
were used. Rats were housed individually for the first week after intracra-
nial injection, until suture removal, after which the rats were housed 2
per cage. All animals were housed on a 12 h light/dark cycle (06:30 to
18:30 light) with food and water ad libitum.
Surgery and stereotaxic injection of AAV-ChR2 vectors. Injections
were conducted as described previously (Adrover et al., 2014). Briefly,
mice (5-6 weeks old) were anesthetized by inhalation of isoflurane-oxygen
mixture and placed in a stereotaxic frame (David Kopf). Adeno-associated
virus (AAV) vectors with Cre recombinase-dependent expression of
Channelrhodopsin2 (ChR2) protein, AAV5-EF1a-DIO-hChR2(H134R)-
EYFP (4  1012 IU/ml), were bilaterally injected into the VTA/SNc (AP:
3.30, ML: 60.60, DV: 4.50) of DATIRES-Cre1 mice. The AAV vectors
AAV5-CaMKII-hChR2(H134R)-EYFP (4  1012 IU/ml) or AAV5-Syn-
ChrimsonR-tdTomato (1.7  1013 IU/ml), were injected in the PFC (PrL
cortex, AP: 2.10, ML: 60.35, DV: 2.30; IL cortex, AP: 1.90, ML: 60.30,
DV: 3.20) of negative littermates or DATIRES-Cre1 mice. All stereotaxic
coordinates were from bregma (in mm) according to the mouse atlas by
Franklin and Paxinos (2007); 300-500 nl for VTA/SNc and 100-200 nl for
PFC were injected at a flow rate of 100 nl/min. Recordings were made after
a minimum of 4 weeks of incubation. For mice injected in both VTA/SNc
(DIO-ChR2-EYFP) and PFC (ChrimsonR-tdTomato), recordings were per-
formed after a minimum of 8 weeks of incubation. Viral vectors were pur-
chased from Gene Therapy Center Vector Core at University of North
Carolina and Penn Vector Core at University of Pennsylvania.
Slice preparation. Mice were anesthetized and rapidly decapitated.
Brains were quickly removed, mounted, and sliced using a vibratome
(VT-1200S, Leica Microsystems) in an ice-cold cutting solution contain-
ing the following (in mM): 225 sucrose, 13.9 NaCl, 26.2 NaHCO3, 1
NaH2PO4, 1.25 glucose, 2.5 KCl, 0.1 CaCl2, 4.9 MgCl2, and 3 kynurenic
acid. The sagittal slices (240 mm) were incubated for 20 min at 33°C in
ACSF containing the following (in mM): 124 NaCl, 1 NaH2PO4, 2.5 KCl,
1.3 MgCl2, 2.5 CaCl2, 20 glucose, 26.2 NaHCO3, and 0.4 ascorbic acid,
and kept in the dark at room temperature before use. Recording cham-
ber was perfused at 2 ml/min with ACSF heated at 32°C using an inline
heater (Harvard Apparatus).
FSCV and amperometry. FSCV was performed in the dorsal stria-
tum. Carbon-fiber electrodes (CFEs) were prepared with a cylindrical
carbon-fiber (7 mm diameter, ;150 mm of exposed fiber) inserted into a
glass pipette. Before use, the CFEs were conditioned with an 8-ms-long
triangular voltage ramp ( 0.4 to 1.2 and back to 0.4 V vs Ag/AgCl ref-
erence at 400 V/s) delivered every 15 ms. CFEs showing current .1.8 mA
or ,1.0 mA in response to the voltage ramp at ;0.6 V were discarded.
During the recording, the CFEs were held at 0.4 V versus Ag/AgCl,
and the same triangular voltage ramp was delivered every 100 ms. DA
transients were evoked by electrical or optical single pulse, or 5 pulses at
20 Hz stimulations. Using the same CFE and location, DA signals were
evoked by alternating electrical and optical stimulations in some experi-
ments. These data were combined with results from other experiments
where either electrical or optical stimulation was used to evoke DA sig-
nals. For electrical stimulation, a glass pipette filled with ACSF was
placed near the tip of the carbon fiber and a rectangular pulse (0.2 ms,
100 mA) was applied every 2 min. For optogenetic stimulation, a fiber-
optic (200 mm diameter, 0.22 NA, ThorLabs) connected to a blue LED
(470 nm, 1.8 mW of maximal output power measured at the tip of the
fiber-optics, ThorLabs) was placed over the carbon fiber and light pulses
(0.6 ms) were delivered every 2 min. For input-output curves, the widths
of light pulse were 0.1, 0.2, 0.5, 1, 2, and 5 ms. For dual optogenetic
recordings, two fiber-optics connected to a purple LED (420 nm, 3.0
mW, ThorLabs) and an orange LED (590 nm, 0.7 mW, ThorLabs),
respectively, were placed over the carbon fiber. Light pulses (0.6 ms for
420 nm and 0.6-2 ms for 590 nm) were delivered in an alternating pat-
tern every 2 min. Data were collected with a retrofit headstage (CB-7B/
 7512 • J. Neurosci., September 23, 2020 • 40(39):7510–7522
EC with 5 MX resistor) using a Multiclamp 700B amplifier (Molecular
Devices) after low-pass filter at 3 kHz and digitized at 100 kHz using a
DA board (NI USB-6229 BNC, National Instruments). Data acquisition
and analysis were performed using a custom-written software, VIGOR,
in Igor Pro (Wavemetrics) using mafPC (courtesy of MA Xu-Friedman).
The current peak amplitudes of the evoked DA transients were con-
verted to DA concentration according to the postexperimental calibra-
tion using 1-3 mM DA. Amperometric recordings were performed using
the same carbon-fiber electrodes held at 0.4 V versus Ag/AgCl and a
30-s-long step to 0.6 V was applied. Single pulse or 5 pulses at 20 Hz
stimulation were delivered at 20 s after switching to 0.6 V. Since we did
not find major differences between PrL and IL in evoking DA transients,
we combined the data obtained from two groups to represent DA transi-
ents evoked by PFC inputs (PFC-oDA).
Cell-attached recordings. Striatal CINs from CIN-TdTomato mice
injected with ChR2-EYFP in the PFC were identified by fluorescence
and confirmed by their characteristic spontaneous firing pattern. Cell-
attached recordings were performed from CINs in the striatum using
glass pipette electrodes with a resistance of ;3-4 MV, filled with an inter-
nal solution containing the following (in mM): 120 cesium methanesulfo-
nate, 20 CsCl, 10 HEPES, 0.2 EGTA, 10 sodium phosphocreatine, 4
Na2-ATP, and 0.4 Na-GTP, pH, 7.25 (290-310 mOsm), at a holding
potential of 0 mV. To estimate the strength of connectivity to CINs, the
lowest light intensity needed to reliably evoke action potentials in CINs
was determined, ranging from 0.4 mW for “high” connectivity to 2.1
mW for “low” connectivity. The data were collected using a Multiclamp
700B amplifier after low-pass filter at 1 kHz and digitized at 5 kHz, using
pClamp10 software (Molecular Devices). Spike fidelity was calculated as
percentages of the number of AP evoked from five trials for both single
pulse and trains stimulation.
In vivo microdialysis and optogenetic stimulation of the cortical
inputs. Rats (80-90 g) received unilateral intracranial injection of
AAV-CaMKIIa-hChR2(H134R)-EYFP (titer: 1012 IU/ml; Gene Therapy
Center Vector Core at University of North Carolina) in the PrL and IL
cortex (AP: 3.0 mm, ML: 0.5 mm, DV: 3.5 and 5 mm with respect to
bregma). Two injection sites per hemisphere were used, and 300 nl of vi-
ral vector solution was delivered per site via a 105-mm-thick silica tubing
injector coupled directly to a 1 ml syringe driven by an infusion pump
(rate = 50 nl/min) during 10 min. The injector was left in place for an
additional 10 min following each injection to allow for diffusion of the
suspension. Ten weeks after viral vector injection, rats (350-400 g)
underwent surgery for probe implantation according to previously pub-
lished procedures (Quiroz et al., 2016). Briefly, a modified microdialysis
probe with an embedded light-guiding optic fiber was implanted into
the NAc shell (AP: 1.2 mm, ML: 0.5 mm, DV: 8.0 mm with respect to
bregma). The probe was fixed to the skull with a stainless-steel screw
and glass-ionomer dental cement. All surgical procedures were per-
formed under anesthesia with 3 ml/kg of Equithesin (4.44  g of chloral-
hydrate, 0.972  g of Na pentobarbital, 2.124  g of MgSO4, 44.4 ml of
propylene glycol, 12 ml of ethanol, and distilled H2O up to 100 ml of
final solution; National Institute on Drug Abuse Pharmacy). To build
the microdialysis probe with embedded optic fiber, the tip of an optic
fiber (105 mm diameter core, 0.22 NA) was sculpted into a conical shape
using a Flaming-Brown pipette puller fitted with a custom platinum
heating filament (Sutter Instruments) to allow for a larger area of stimu-
lation. The conical optic fiber tip was embedded inside the microdialysis
probe and implanted. Microdialysis experiments were performed in
freely moving rats 24 h after probe implantation. Optical fiber was
coupled to a 473 nm solid-state laser module and light stimulation was
driven by a Grass S88 stimulator. Light stimulation was delivered for 20
min as trains of light pulses (2 ms pulse duration; at 100 Hz for 160 ms;
trains repeats once per second; intensity = 5-8 mW at probe tip). ACSF
containing (in mM) 144 NaCl, 4.8 KCl, 1.7 CaCl2, and 1.2 MgCl2 was
pumped through the optogenetic-microdialysis probe (rate = 1.25 ml/
min). After a washout period of 90 min, dialysate samples were collected
at 20 min intervals. After 80 min of baseline sampling, optogenetic stim-
ulation was applied for 20 min, and samples were collected for 80 addi-
tional minutes after the end of the stimulation. Samples were split and
analyzed separately for glutamate, DA, and ACh content. ACh and
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals
glutamate contents were measured by HPLC coupled to an ACh oxidase
and glutamate oxidase enzyme reactors, respectively, and electrochemi-
cal detection (Eicom). DA was measured by HPLC coupled with a coulo-
metric detector (5200a Coulochem III, ESA). At the end of the
microdialysis experiment, animals were deeply anesthetized with
Equithesin and perfused transcardially with 0.1 M PBS, followed by 4%
formaldehyde in 0.1 M PBS, pH 7.4. Brains were postfixed in the same
fixative for 2 h and immersed in 20% sucrose/0.1 M PBS, pH 7.4, solution
for 48 h at 4°C. Forty-micron-thick coronal sections were cut in a Leica
Microsystems CM3050S cryostat at 20°C, collected in PBS, and stored
in antifreeze-buffered solution (20% ethylene glycol, 10% glycerol, and
10% sucrose in PBS) at 80°C until processing. Sections were then eval-
uated for localization of implanted probes and ChR2-EYFP expression.
Wide-field images were acquired with a Typhoon laser scanner (GE
Healthcare). Confocal fluorescence microscopy images were acquired
with an Examiner Z1 microscope (Carl Zeiss) fitted with a confocal laser
module (LSM-710, Carl Zeiss).
Drugs. Dihydro- b -erythroidine hydrobromide (DH b E) was pur-
chased from Tocris. Kynurenic acid (sodium salt), NBQX, and 3-((R)-2-
carboxypiperazin-4-yl)-propyl-1-phosphonic acid (CPP) were purchased
from Abcam. All other chemicals were purchased from Sigma Millipore.
Statistical analysis. Statistical analysis was performed with Prism
(GraphPad). One-sample t test, two-tailed paired t test, repeated-
measures one-way ANOVA with or without mixed-effects models, or
two-way ANOVA was used as specified. Tukey’s or Dunnett’s multi-
ple comparisons test was used for post hoc analysis as specified. The
number of experiments, n, was expressed as the number of slices or
cells/the number of mice or number of rats for the in vivo microdialy-
sis experiments.
Results
In vivo stimulation of PrL/IL inputs increases striatal ACh
and DA concentration, and the increase in DA requires
nAChR activation
Published work in vivo shows that optogenetic stimulation of
cortical inputs to the striatum can evoke DA release (Quiroz et
al., 2016). However, this study provides no evidence that activa-
tion of these cortical inputs recruits striatal CINs or that the
cortically evoked DA release in vivo requires activation of striatal
nAChRs. To assess these matters, we used in vivo microdialysis
combined with optogenetic stimulation, as previously described
(Quiroz et al., 2016), and measured extracellular levels of ACh
and DA in response to optogenetic stimulation of cortical inputs.
A modified microdialysis probe with an embedded light-guiding
optic fiber was implanted into the striatum of rats expressing
ChR2-EYFP in the PrL/IL cortex (Fig. 1A,B). On the experiment
day, dialysate samples were collected at baseline for 80 min
before stimulation started. Optogenetic stimulation of cortical
axons within the striatum was delivered around the microdialysis
probe as trains of light pulses (16 pulses at 100 Hz every 1 s for
20 min). Samples were collected for an additional 80 min after
stimulation and were split and analyzed separately for DA and
ACh content or DA and glutamate. This long train of in vivo
optogenetic stimulation of PrL/IL cortical axon fibers within the
striatum produced an increase in extracellular concentration of
DA in the striatal dialysate (185.1 6 44.0% of baseline, n = 7;
F = 5.27, p = 0.0002, repeated-measures one-way ANOVA; p =
0.0002 for 60 min vs 80 min; Dunnett’s multiple comparisons
test; Fig. 1C), which coincided with an increase of extracellular
ACh concentration (196.7 6 36.6% of baseline, n = 7; F = 3.00,
p = 0.0002, repeated-measures one-way ANOVA; p = 0.004 for
60 min vs 80 min; Dunnett’s multiple comparisons test; Fig. 1D).
These in vivo microdialysis findings support the hypothesis that
cortical stimulation recruits CINs. In addition, when the in vivo
optogenetic stimulation was performed in constant perfusion of
 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals J. Neurosci., September 23, 2020 • 40(39):7510–7522 • 7513

A dialysate samples B a d
c

LASER

PrL/IL
cortex
b

C D
250 * 250
***
200 200
% of baseline

% of baseline
150 150

100 100

50 50
DA ACh
0 0
0 40 80 120 0 40 80 120
Time (min) Time (min)
E F
250 250
**
200 200
% of baseline

% of baseline

150 150

100 100

50 50
DA Glu
+ DHβE + DHβE
0 0
0 40 80 120 0 40 80 120
Time (min) Time (min)

Figure 1. Effect of the nAChR antagonist DHb E on the extracellular DA level evoked by local optogenetic stimulation of PFC fibers in vivo. A, Diagram showing the injection of ChR2-EYFP in
the PrL and IL cortices (green circle) and projection to the striatum with the microdialysis-optogenetics probe in a sagittal brain view. B, Left, Optogenetic-microdialysis probe schematics: a, liq-
uid inlet; b, dialysis membrane; c, liquid outlet; d, sculpted optic fiber. Middle, Probe tip detail. Scale bar, 0.2 mm. Right, Picture of the optogenetic-microdialysis probe showing light spread
pattern at the probe tip. Scale bar, 1 mm. C, D, Time course of extracellular concentrations of (C) DA (red) and (D) ACh (black) in the NAc. E, F, Time course of extracellular concentrations of
(E) DA (red) and (F) glutamate (Glu, green) in the NAc with constant perfusion (reverse dialysis) of the nAChR antagonist DHb E (10 mM). Time 0-60 represents the values of samples before
stimulation. Blue vertical lines indicate the period of optogenetic stimulation (20 min). Results are expressed as mean 6 SEM of percentage of the average of three values before stimulation.
*p , 0.05, **p , 0.01, ***p , 0.001.

the nAChR antagonist DH b E (10 mM) delivered locally via
reverse dialysis, the same stimulation protocol did not cause any
change in extracellular DA concentration in the striatal dialysates
(96.5 6 11.6% of baseline, n = 9; F = 0.64, p = 0.70, repeated-
measures one-way ANOVA; Fig. 1E), indicating that activation
of nAChRs in the striatum is required for the cortically evoked
DA release in vivo. As a control, we measured glutamate concen-
tration in these same dialysates, which showed increased gluta-
mate levels and confirmed the delivery of cortical stimulation in
the presence of the nAChR antagonist (184.3 6 27.3% of base-
line, n = 9; F(6,45) = 4.01, p = 0.003, repeated-measures one-way
ANOVA with mixed-effects model; p = 0.0006 for 60 min vs
80 min; Dunnett’s multiple comparisons test; Fig. 1F).
Stimulation of PrL/IL inputs to striatum is sufficient to
evoke local DA signals in brain slices
In order to precisely access the specific inputs and apply pharma-
cological agents, we used transgenic mice with combination of
optogenetic stimulation. DA signals were recorded in the stria-
tum using FSCV in mouse brain slice preparations. Release of
DA from axonal projections within the striatum was evoked by
electrical stimulation and compared with DA signals evoked by

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 7514 • J. Neurosci., September 23, 2020 • 40(39):7510–7522 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals

A B

SNc/VTA PrL/IL
cortex
DATires/cre+ mice C57Bl6/J mice
ChR2-EYFP ChR2-EYFP
PrL
IL
LV

Th PBP

LV LV
AcSh VP VTA
Tu 1 mm

DS DS

SNC
ac ac SNC
PBP
ac
ac SNR
AcC AcC
VP VP 1.0 mm
AcSh SNR 1.0 mm
AcSh

C eDA DAN-oDA PFC-oDA D PrL cortex

15 15 15 IL cortex
Current (nA)

Current (nA)

Current (nA)
10 10 10
5 5 5

-5 1.2 -5 1.2 1.2

Voltage (V) Voltage (V) -5 Voltage (V)

-0.4 -0.4 -0.4

15
Voltage (V)

10
1.2 1.2 1.2
5
0
-0.4 -0.4 -0.4 nA
0 2 4 0 2 4 0 2 4
Time (s) Time (s) Time (s) Lateral 1.20 mm

E oxidation
reduction
F DA concentration G Decay time constant
0.8 n. s. 0.5
*** **
** **
DA amplitude (μM)

0.6 1.0 0.4

Decay tau (s)
Voltage (V)

0.4
0.3
0.2 n. s.
0.5 0.2
0.0
-0.2 0.1

-0.4 0.0 0.0

E-stim DAN PFC E-stim DAN PFC E-stim DAN PFC

Figure 2. Striatal DA signals evoked by midbrain and cortical inputs. A, Example of the fluorescence pattern (bottom) observed in a sagittal brain slice from a DATIRES-Cre1 mouse injected
with DIO-ChR2-EYFP in the midbrain (top). B, Example of the fluorescence pattern (bottom) observed in a sagittal brain slice from a C57Bl6/J (DATIRES-Cre–) mouse injected with ChR2-EYFP in
the PrL or IL cortex (top). Inset, A more medial slice with the site of injection. C, Representative DA transients, current–voltage (CV) plots, and color voltammograms when evoked by electrical
stimulation (eDA, left), optogenetic stimulation of DAN fibers (DAN-oDA, middle), or optogenetic stimulation of PFC inputs (PFC-oDA, right). All three CV plots show the electrochemical profile
of DA oxidation. Scale bar, 100 nM. D, A sagittal brain section modified from the Mouse Brain Atlas (Franklin and Paxinos, 2007) showing the FSCV recording sites from mice injected with
ChR2-EYFP in the PrL (filled) and IL (empty) cortex. E, The oxidation (top) and reduction (bottom) voltages from the CV plots of eDA (left), DAN-oDA (middle), and PFC-oDA (right) were plotted
as average with SEM. Open and filled circles represent individual values. F, DA peak concentrations and G, decay time constants of eDA (left), DAN-oDA (middle), and PFC-oDA (right) were plot-
ted as average with SEM. Open circles represent individual values. For PFC-oDAs, the same color code was applied as in D for PrL and IL according to injection location. E–G, Data points include
experiments where electrical and optogenetic stimulation was delivered alternatingly in the same slice or in different slices (see Materials and Methods). **p , 0.01, ***p , 0.001. LV, Lateral
ventricle; DS, dorsal striatum; ac, anterior commissure; AcC, accumbens core; AcSh, accumbens shell; VP, ventral pallidum; PBP, parabrachial pigmented nucleus; SNR, substantia nigra reticulata;
Th, thalamus; Tu, olfactory tubercle. n.s., not significant.

input-specific optogenetic stimulation. To achieve input-specific As expected, single brief pulses of electrical stimulation deliv-
stimulation, ChR2-EYFP was expressed in either midbrain ered via an electrode placed within the striatal tissue reliably
DANs and their axonal projections to the striatum using Cre-de- evoked DA (electrically evoked DA transient [eDA]) signals in
pendent viral vector injection in DATIRES-Cre1 mice (Fig. 2A) or striatal brain slices (Fig. 2C). Also, direct optogenetic stimulation
ChR2-EYFP in cortical neurons of the PrL/IL cortex and their of midbrain DAN axonal projections within the striatum also
projections to the striatum using viral injection in Cre-negative evoked DA signals (referred here as DAN-oDA) on delivery of a
littermate mice (Fig. 2B). single brief pulse of blue light (0.6 ms), in agreement with

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals
previous published studies (Adrover et al., 2014; Melchior et al.,
2015). Last, similar to the in vivo findings in Figure 1, selective
optogenetic stimulation of axon projections from PrL/IL cortex
into the striatum was also sufficient to trigger DA signals
(referred as PFC-oDA) in the striatum of in slice preparation
(Fig. 2C). These PFC-oDA signals were also reliably evoked by
brief single pulses of blue light (0.6 ms) in the ventral portion of
the dorsomedial striatum in sagittal brain slices (Fig. 2D). The vi-
ral expression around injection site was often large enough to
include both PrL and IL regions of the cortex. For this reason, we
combined data from the two regions. It is worth noting that these
brain sections mainly contain the axon projections from PrL/IL
neurons, but not their cell bodies, which are localized to more
medial sagittal sections of the brain (Fig. 2B, inset).
The current–voltage plots of DA signals evoked via the three
different types of stimulation (electrical, optogenetic stimulation
of DANs, and optogenetic stimulation of PrL/IL) were indistin-
guishable and displayed current peaks at the expected oxidation
(0.56 6 0.01 V for eDA, n = 15/n = 8; 0.58 6 0.01 V for DAN-
oDA, n = 14/n = 6; 0.56 6 0.01 V for PFC-oDA, n = 22/n = 13;
F = 0.49, p = 0.62, repeated-measures one-way ANOVA with
mixed-effects model; Fig. 2E) and reduction voltages ( 0.17 6
0.01 V for eDA, 0.14 6 0.01 V for DAN-oDA, 0.16 6 0.02 V
for PFC-oDA; F = 0.76, p = 0.49, repeated-measures one-way
ANOVA with mixed-effects model; Fig. 2E) for DA. The peak
amplitude of eDA was significantly higher than DAN-oDA or
PFC-oDA (556 6 42 nM for eDA; 367 6 58 nM for DAN-oDA;
325 6 23 nM for PFC-oDA; F(2,12) = 13.430.76, p . 0.05,
repeated-measures one-way ANOVA with mixed-effects model;
p = 0.01 for eDA vs DAN-oDA, p = 0.0005 for eDA vs PFC-oDA,
and p = 0.99 for DAN-oDA vs PFC-oDA; Tukey’s multiple com-
parisons test; Fig. 2F), and the decay time constants (tau) were
similarly longer with eDA (0.25 6 0.01 s for eDA, 0.22 6 0.01 s
for DAN-oDA, 0.23 6 0.01 s for PFC-oDA; F(2,12) = 13.43,
p = 0.0009, repeated-measures one-way ANOVA with mixed-
effects model; p = 0.004 for eDA vs DAN-oDA, p = 0.009 for eDA
vs PFC-oDA, and p = 0.83 for DAN-oDA vs PFC-oDA; Tukey’s
multiple comparisons test; Fig. 2G). Thus, we were able to reli-
ably evoke steady DA signals by selective optogenetic stimulation
of axon projections from PrL/IL cortex in the ventral portion of
the dorsomedial striatum. The DA signals recorded using FSCV
showed indistinguishable chemical and temporal characteristics,
even when evoked by different inputs.
Unique physical and pharmacological properties across
input-specific DA signals
We first explored the threshold for triggering the input-specific
optogenetically evoked DA signals. The threshold was deter-
mined by constructing input-output curves and varying the light
pulse durations from 0.1 to 5 ms in brain slices from mice
expressing ChR2 in midbrain DANs or the PrL/IL region (Fig.
3A,B). The relative amplitudes of DA transients were overlap-
ping for pulses of 0.5-5 ms duration, and both DAN-oDA and
PFC-oDA signals showed maximal amplitude with 1-5 ms light
pulse durations. However, very short light pulses (0.1 ms) evoked
measurable DA signals only in slices in which ChR2 was
expressed in midbrain DAN projections to the striatum, indicat-
ing a lower threshold for evoking DA signals with direct stimula-
tion of DA neuron axons than with stimulation of cortical inputs
(duration  input, F(5,90) = 11.58, p , 0.0001, repeated-measures
two-way ANOVA; post hoc for the duration of 0.1 ms: 33 6 4%
for DAN-oDA, n = 12/n = 8, and 1.9 6 0.7% for PFC-oDA, n = 8/
n = 5, p , 0.0001; Fig. 3B). A higher threshold for the cortically
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
J. Neurosci., September 23, 2020 • 40(39):7510–7522 • 7515
evoked DA signals could reflect technical differences, such as
opsin expression levels, or other biological factors, such as the
density of innervation of cortical versus DAN axonal projections,
difference in the release probability of the DA varicosities that
respond to direct stimulation of DAN axons from those respond-
ing to cortical inputs, and/or the indirect and polysynaptic na-
ture of the cortical-evoked DA signals.
To determine whether PrL/IL activation elicits striatal DA
release and engages the CIN-dependent mechanism as shown in
in vivo microdialysis experiment (Fig. 1), we then studied the ba-
sic pharmacological characteristic of DA signals evoked by stim-
ulation of PFC inputs and compared them with those evoked by
electrical and DAN input stimulation. In striatal brain slices,
PFC-oDA signals were completely abolished by a mixture of
AMPA and NMDA receptor antagonists (NBQX/CPP), whereas
eDA and DAN-oDA signals were unaffected (amplitude after the
antagonists = 0.3 6 1.6% of baseline for PFC-oDA, n = 6/n = 5;
107.3 6 3.0% for eDA, n = 6/n = 4; 99.8 6 1.4% for DAN-oDA,
n = 6/n = 4; F(2,3) = 694.8, p , 0.0001, repeated-measures one-
way ANOVA with mixed-effects model; p = 0.14 for eDA vs
DAN-oDA, p values , 0.0001 for PFC-oDA vs eDA or PFC-
oDA vs DAN-oDA; Tukey’s multiple comparisons test; Fig. 3C,
D). These findings indicate a requirement for ionotropic gluta-
mate transmission for PrL/IL inputs to evoke local DA signals,
which is in agreement with previous reports on the requirement
of glutamate transmission for DA signals evoked by motor cortex
and thalamic inputs to the striatum (Threlfell et al., 2012; Kosillo
et al., 2016; Cover et al., 2019). Further, similar to our in vivo
data (Fig. 1D), DH b E also abolished PFC-oDA signals, and sig-
nificantly depressed eDA signals, while having no effect on
DAN-oDA signals (1.9 6 1.4% of baseline for PFC-oDA, n = 11/
n = 7; 21 6 3% for eDA, n = 10/n = 5; 99 6 3% for DAN-oDA,
n = 5/n = 2; F(2,7) = 395.2, p , 0.0001, repeated-measures one-
way ANOVA with mixed-effects model; p , 0.0001 for eDA vs
DAN-oDA, p = 0.0006 for eDA vs PFC-oDA, and p , 0.0001 for
DAN-oDA vs PFC-oDA; Tukey’s multiple comparisons test; Fig.
3E,F). Together, the pharmacology results support the hypothesis
that PFC-oDA signals are mediated by activation of glutamate
synapses from PrL/IL cortex to striatal CINs, which excites them
to fire and trigger ACh release, thereby driving DA release via
activation of nAChRs on DAN axon fibers within the striatum.
ACh is rapidly cleared by enzymatic degradation by AChE
(Quinn, 1987). The extremely dense presence of AChE in the
striatum (Zhou et al., 2001) assures the termination of choliner-
gic action. We also have previously shown the influence of the
AChE on the DA transmission in the striatum (Shin et al., 2015,
2017). As AChE activity shows temperature dependence (Vidal
et al., 1987), we then tested the temperature dependence of the
input-specific DA transients by varying the temperature of the
bath solution. The amplitude of both DAN-oDA and PFC-oDA
signals increased linearly as the temperature was lowered from
the standard 32°C to 25°C, and both showed a tight negative
regression with similar slopes significantly different from zero
( 10.8 6 0.6% per °C, n = 8/n = 3, for DAN-oDA, and 10.8 6
2.0% per °C, n = 6/n = 3, for PFC-oDA; p values , 0.0001; Fig.
3G,H). The temperature dependence of the DA signal amplitude
likely reflects the change in rate of DA transporter activity with
temperature. DA transporter activity was reported to have a tem-
perature coefficient Q10 ranging from 1.39 to 2.95 (Zhu and
Hexum, 1992), from which we estimate a Q10 of ;2.3 for the DA
transporter for 24°C–37°C range. Our estimate predicts that
large changes in the rate of DA reuptake by the transporter
would be around the tested and physiological temperatures.
 7516 • J. Neurosci., September 23, 2020 • 40(39):7510–7522 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals

A DAN-oDA PFC-oDA B Input-output

120

oDA amplitude
0.5

(% of max)
80
0.2 0.5

40
***
0.2 DAN
0.1 PFC
0.1
0
2 4 6 8 2 4
0.1 1 Duration (ms)

C NBQX/CPP D n.s.

DA peak after NBQX/CPP

pre post pre post pre post
100

(%baseline)
50

***
0
E-stim DAN PFC
eDA DAN-oDA PFC-oDA

E DHβE F
pre post pre post pre post

DA peak after DHβE

100

(%baseline)
50
*** ###
***
0
E-stim DAN PFC
eDA DAN-oDA PFC-oDA

G Temperature H oDA peak I oDA tau

25°C 25°C 200 DAN 0.6 DAN
Decay tau (sec)

PFC PFC
oDA amplitude

150
(% of 32°C)

0.4
32°C 32°C
100
35°C
50
* 0.2
35°C

0 0.0

DAN-oDA PFC-oDA 25 30 35 25 30 35
Temperature (°C) Temperature (°C)

Figure 3. DAN-oDA and PFC-oDA show different physical and pharmacological properties. A, Representative DAN-oDA (left) and PFC-oDA (right) transients evoked with different light pulse
duration (in ms). Scale bar, 100 nM, 0.5 s. B, oDA amplitudes normalized to their maximum response were averaged and plotted as a function of the stimulus duration. C, Representative traces
of eDA, DAN-oDA, and PFC-oDA transients before and after bath application of the glutamate receptor antagonists, NBQX and CPP (both 5 mM). Dotted line (top) indicates the amplitude of DA
transients before the drugs. Scale bar, 200 nM, 2 s. D, Averages with SEM of DA amplitude after NBQX and CPP were plotted. Open circles represent individual value. E, Representative traces of
eDA, DAN-oDA, and PFC-oDA transients before and after bath application of the b 2-contatining nAChR antagonist, DH b E (1 mM). Scale bar, 200 nM, 2 s. F, Averages with SEM of DA ampli-
tude after DHb E were plotted. Open circles represent individual value. ***Versus DAN-oDA. ###Versus eDA. G, Representative traces of DAN-oDA (left) and PFC-oDA (right) transients at 25°C,
32°C, and 35°C. Dotted line (top) indicates the oDA amplitude at 32°C. Scale bar, 200 nM, 2 s. H, Average oDA peak amplitudes normalized to 32°C. I, Average oDA decay time constants were
plotted as a function of temperature. *p , 0.05, ***p , 0.001, ###p , 0.001, n.s., not significant.

Indeed, the decay of the DA transients, which is a parameter that
reflects the clearance of extracellular DA by the transporter, was
also affected as the temperature was lowered. The decay time
constant of both DAN-oDA and PFC-oDA increased linearly
with the decrease of the temperature (DAN-oDA: 0.017 6
0.001 s/°C; PFC-oDA: 0.025 6 0.001 s/°C; Fig. 3I), which corre-
sponded with the linear increase in amplitude. However, raising
the temperature from 33°C to 35°C caused a dramatic and selec-
tive drop in the amplitude of the PFC-oDA signals, compared
with DAN-oDA signals (Fig. 3G,H). The slope for PFC-oDA sig-
nals increased threefold from 11% to 31% per °C.
Vidal et al. (1987) have reported on a high dependence
with temperature of the activity of AChE from rat brain,
which reaches maximal rate at ;35°C. Our in vivo and in
vitro pharmacology experiments showed that PFC-oDA sig-
nals require activation of nAChR. Therefore, we speculate
that the observed steep temperature dependence of the sig-
nals is in large part mediated by rate increase in AChE ac-
tivity. Furthermore, at physiological temperatures, the
PFC-evoked DA signals may be very localized near ACh
release sites, presumably overlapping with nAChR expres-
sion in DAN axons.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals J. Neurosci., September 23, 2020 • 40(39):7510–7522 • 7517

A B
Single vs. train
2.5 ****

oDA ratio (5p/1p)

1 pulse 2.0
5 pulses @20 Hz
1.5
1.0
0.5
0.0
eDA DAN-oDA PFC-oDA E-stim DAN PFC

C D Cell-attached E
Amperometry 100
from CIN
80

AP fidelity (%)
1 pulse
60
40

5 pulses @20 Hz 20
0
1p 5p
DAN-oDA PFC-oDA
Figure 4. PFC-oDA shows no summation by train stimulations. A, Representative DA traces of eDA, DAN-oDA, and PFC-oDA transients evoked by single pulse (1p, thin traces) or train of 5
pulses at 20 Hz (5p, thick traces). Scale bar, 200 nM, 2 s. B, Averages with SEM of the DA amplitude ratio (5p/1p) for eDA, DAN-oDA, and PFC-oDA transients were plotted. Open circles repre-
sent individual values. C, Representative amperometric traces for DAN-oDA (left) and PFC-oDA (right) transients evoked by single pulse (gray traces) or train of 5 pulses at 20 Hz (color traces).
PFC-oDA amperometric transients evoked by 50 at 20 Hz were indistinguishable from 1p stimulation, except the large deflection at the stimulation time for 5p pulses. Scale bars: 200 pA,
100 ms. D, Representative cell-attached recordings from CINs with single (top, gray) or train of 5 pulses at 20 Hz (bottom, green). Scale bars: 20 pA, 100 ms. E, Averages with SEM of the action
potential fidelity were plotted for the single pulse and train stimulation. Open circles represent individual values. ****p , 0.0001.

Activity dependent summation of the midbrain and cortical
DA signals
Trains of stimulation pulses have been shown to produce sublin-
ear summation in the amplitude of DA signals evoked by electri-
cal and optogenetic stimulation of DAN axon projections in the
striatum in vitro (Zhang and Sulzer, 2004; Threlfell et al., 2010;
Melchior et al., 2015; Shin et al., 2017). We then tested the degree
of summation of the DA signals in response to short trains of
stimulation pulses (5 pulses at 20 Hz) and compared it with the
transients evoked by a single pulse (Fig. 4A,B). This pattern of
train stimulation was chosen based on burst firing patterns
recorded in vivo for DANs during behavior (Schultz et al., 1993;
Hyland et al., 2002). Indeed, we found that trains of 5 pulses at
20 Hz evoke DA transients 50% larger than those evoked by sin-
gle pulse stimulation for eDA signals and almost double for
DAN-oDA (5p/1p: 1.44 6 0.06, n = 5/n = 3, for eDA; 1.92 6 0.19,
n = 5/n = 2, for DAN-oDA; 1.00 6 0.02, n = 10/n = 7; F(2,2) =
110.6, p = 0.009, repeated-measures one-way ANOVA with
mixed-effects model; p = 0.04 for eDA vs DAN-oDA, p = 0.03 for
eDA vs PFC-oDA, and p = 0.008 for DAN-oDA vs PFC-oDA;
Tukey’s multiple comparisons test; Fig. 4A,B). On the contrary,
PFC-oDA transients evoked by this train stimulation had similar
amplitudes compared with those evoked by a single pulse, indi-
cating no summation of DA signals evoked by PrL/IL inputs. In
order to measure DA signals with better temporal resolution, we
performed amperometric recordings and again compared DA
signals evoked by single pulse and trains (Fig. 4C). In line with
the FSCV results, amperometric recordings showed that DAN-
oDA signals display sublinear summation, whereas PFC-oDA
signals are indistinguishable between single pulse or trains of 5
pulses at 20 Hz. The lack of summation in the PFC-oDA signals
in response to trains resembles the findings obtained with DA
signals evoked by synchronized activation of CINs (Threlfell et
al., 2012; Shin et al., 2017) and further supports the idea that the
PFC-evoked local DA signals are mediated by synchronized
action of CINs in response to stimulation of PrL/IL inputs. The
large transient current deflection in response to each pulse of
stimulation (n = 5/n = 4; Fig. 4C, right) is suggestive of the action
potential firing evoked in CINs and other striatal neurons by the
optogenetic stimulation of PrL/IL inputs. The fact that every
pulse of the train stimulation triggers a current deflection reflects
the ability of PrL/IL inputs to evoke action potentials firing in
downstream neurons and supports the idea that the lack of sum-
mation in DA signals is not likely because of the failure of action
potential firing by CINs. To directly test the contribution of fail-
ure of action potential firing by CINs to the lack of summation
in PFC-oDA signals, we conducted cell-attached recordings
from CINs and measured action potential fidelity in response to
a single pulse and trains of 5 pulses at 20 Hz (single pulse:
96.5 6 2.7%, n = 23/n = 4, t = 1.28, p = 0.21; trains of 5: 81.5 6
7.7%, n = 10/n = 4, t = 2.42, p = 0.04; both with one-sample t test
to 100% value; Fig. 4D,E). Although there was a small but signifi-
cant percentage of failure, the totality of the data indicates that
the lack of summation is not likely caused by action potential
failure, but instead can be attributed to other downstream mech-
anisms, such as desensitization of nAChRs (Threlfell et al., 2012;
Shin et al., 2017).
Input-specific DA signals are evoked by dual optogenetic
stimulation in the same brain slice
Thus far, DA signals evoked by the different inputs were
recorded in brain slices from different mice that expressed ChR2
in either midbrain DANs or in PrL/IL cortex. In order to strictly
test the segregation/distinction of these two pathways, it was im-
portant to examine these two inputs and compare the input-spe-
cific DA signals in the slice preparation from the same mice.
We took advantage of opsins activated by different light wave-
lengths, specifically, ChR2 and the red-shifted opsin ChrimsonR

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 7518 • J. Neurosci., September 23, 2020 • 40(39):7510–7522 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals

A B
1.8 1.2

oDA amplitude (μM)

oDA amplitude (μM)

1.6 1.0

//
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
420 nm 590 nm 420 nm 590 nm 420 nm 590 nm 420 nm 590 nm

ChR2-EYFP in midbrain DAN + ChrimsonR-TdTomato in PrL/IL cortex IR image of brain slice

C D
m
0n
42
LV LV
Carbon
DS DS Fiber

590
nm
ac ac
ac ac
ac
AcC SNC AcC SNC

AcSh VP AcSh VP
SNR SNR
1 mm 200 μm

E DAN-oDA PFC-oDA F oDA peak (nM) oDA tau (sec)

15 15 1500 0.5
Current (nA)

Current (nA)

10 10
5 5
0.4
-5 1.2 -5 1.2 1000
Voltage (V) Voltage (V) 0.3
-0.4 -0.4
15 nA 15 nA
Voltage (V)

0.2
10 10 500
1.2 1.2
5 5 0.1
0 0
-0.4 -0.4 0 0.0
0 2 4 0 2 4 DAN PFC DAN PFC
Time (s) Time (s)

NBQX/CPP DHβE
G H
oDA amplitude (%bsln)

oDA amplitude (%bsln)

ACSF NBQX/CPP ACSF DHβE

100 100

DAN DAN
50 PFC 50 PFC

0 0
0 10 20 30 0 10 20 30
Time (min) Time (min)

Figure 5. Dual-opsin expression to evoke cortical and midbrain DA signals in the same brain slices. A, B, Left, Representative traces of DA signals evoked by either 420 or 590 nm light pulses
from mice expressing (A) only ChR2 in midbrain DANs or (B) only ChrimsonR in PrL/IL cortex. Right, oDA amplitudes were plotted as pairs for each wavelength. Scale bars: 200 nM for A 100
nM for B; 1s C, Example of the fluorescence patterns with filter set for yellow signal (left) or red signal (right) from a sagittal brain slice of a DATIRES-Cre1 mice injected with ChrimsonR-
TdTomato in the PFC and DIO-ChR2-EYFP in the midbrain. D, Configuration to the FSCV DA recording using a carbon fiber and two fiber-optics delivering 420 and 590 nm, respectively.
E, Representative DA transients, CV plots, and color voltammograms of DAN-oDA and PFC-oDA. Scale bar: 100 nM F, Left, Amplitudes were plotted as pairs for DAN-oDA and PFC-oDA recorded
from the same slices. Right, Averages with SEM of decay time constant were plotted for DAN-oDA and PFC-oDA. Dots represent individual values. G, Left, Representative traces of DAN-oDA
(top) and PFC-oDA (bottom) before and after the application of glutamate receptor antagonists NBQX and CPP. Right, Averages with SEM of DAN-oDA and PFC-oDA were plotted as a function
of time as NBQX/CPP was applied. H, Left, Representative traces of DAN-oDA (top) and PFC-oDA (bottom) before and after the application of nAChR antagonists DH b E. Right, Averages with
SEM of DAN-oDA and PFC-oDA were plotted as a function of time as DH b E was applied. Scale bars for G–H: 100 nM; 1 s.

(Klapoetke et al., 2014). We first set up the conditions for selec- delivering pulses of two different light wavelengths without any
tive stimulation of each opsin without cross-activation. In brain detectable cross-activation.
slices expressing only ChR2 in midbrain DANs, pulses of purple Next, in the same DATIRES-Cre1 mice, ChR2-YFP was
light (420 nm) evoked reliable DA signal, whereas similar and expressed in midbrain DANs with a Cre-dependent expression
longer pulses of orange light (590 nm) did not trigger any detect- vector, and ChrimsonR-TdTomato was expressed in PrL/IL cor-
able signals (Fig. 5A). Conversely, in slices expressing only tex. Figure 5C shows examples of the fluorescent expression pat-
ChrimsonR in the PrL/IL cortex, brief orange light pulses evoked tern of ChR2-YFP (green) and ChrimsonR (red) in the same
DA signal, but not purple light (Fig. 5B). Thus, under these ex- sagittal brain slice from this double-injected mouse. Figure 5D
perimental conditions, ChR2 and ChrimsonR can be used in shows the experimental arrangement for the dual-input record-
combination to selectively stimulate two different inputs by ings with the placement of the carbon fiber and the two fiber-

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals J. Neurosci., September 23, 2020 • 40(39):7510–7522 • 7519

optics for delivering the purple and orange Sagital slice with dual opsins
light pulses that were used to activate ChR2
and ChrimsonR, respectively. In these slices
A ChR2 in DAN fibers ChrimsonR in PFC fibers B 1.5
DAN-oDA

Peak Amp (µM)

DAN-oDA PFC-oDA
with dual-opsin expression, brief pulses of ei- 1.0
ther purple or orange light evoked DA signals
0.5
with indistinguishable current–voltage plots,
showing the characteristic peaks for DA oxi- 0.0
0 40 80 120
dation and reduction (Fig. 5E). The DA con-
centration transients had comparable peak PFC-oDA

Peak Amp (µM)

1.5
amplitudes with means of 407 6 32 nM for
1.0
DAN-oDA and 420 6 36 nM for PFC-oDA
(n = 43/n = 13, t = 0.47, df = 42, p = 0.64, paired 0.5
t test; Fig. 5F) and overlapping time courses as 0.0
measured at 10 Hz sample rate of FSCV, in 0 40 80 120
agreement with the results shown in Figure 2. Fluo. intensity (a.u.)
The decay time constant of the DA transients
was similar (0.28 6 0.01 s for DAN-oDA and C PFC-oDA amplitude D PFC input density
0.28 6 0.01 s for PFC-oDA, n = 43/n = 13, t =
0.41, df = 41, p = 0.68 from paired t test; Fig. 5F). 1200 DA transient
800 (nM)
1.0 Fluo. intensity
Then, we tested the pharmacological prop- 400 0.5 (a. u.)
0
0.0
erties of DAN-oDA and PFC-oDA by deliver-
ing alternating purple and orange light pulses
and recording from a single carbon fiber. 152 nM 0.43
Application of the glutamate receptor antago-
200 nM
nists NBQX/CPP (99.2 6 3.1% of the baseline 0.73

for DAN-oDA and 1.6 6 1.1% for PFC-oDA, 8 nM 0.14

n = 5/n = 3; t = 25.3, df = 4, p , 0.0001, two-
tailed paired t test; Fig. 5G) or the nAChR
antagonist DH b E (101.9 6 0.6% of the base-
line for DAN-oDA and 0.8 6 0.8% for PFC-
oDA, n = 5/n = 3; t = 79.13, df = 4, p , 0.0001, E CIN density F PFC-CIN connectivity
two-tailed paired t test; Fig. 5H) completely
abolished PFC-oDA signals while leaving DAN- 6.0 # of CINs per High Probability of
oDA signals in the same location intact. These 5.0 400 x 400 m² Med evoking AP

results confirm the findings from Figure 3, 4.0 Low

which use antagonists and single-input optoge-

netic stimulation to evoke either DAN-oDA or
PFC-oDA in separate slices/mice. These results
also validate the experimental approach by
showing input specificity with no apparent
crossover in the optogenetic stimulation of
ChR2 and ChrimsonR with the described
wavelengths. More importantly, these phar-
macological findings highlight the different
mechanisms and the unique nature of the
input-specific DA signals evoked by PFC and
midbrain inputs.
Cortically evoked signals are spatially
restricted
As shown in Figure 2D, DA signals evoked by
selective stimulation of PrL/IL inputs were
preferentially detected between the NAc core
and dorsal striatum. We then set out to inves-
tigate in more detail the spatial distribution of
the input-specific DA signals using the dual-
opsin expression approach by sampling the
amplitude of DA signals across 18 different
striatal areas of the sagittal brain slice while
evoking responses with purple and orange
light. While DAN-oDA signals were detected throughout the
whole striatum, PFC-oDA signals were spatially restricted (Fig.
6A,B). Both DAN-oDA and PFC-oDA signals showed significant
4.5
5.9
5.3
no AP
Figure 6. Cortically evoked DA signals are spatially restricted. A, The locations of carbon fiber and two fiber-optics
were adjusted to measure DAN-oDA (left) and PFC-oDA (right) from each location where the corresponding oDAs were
superimposed on the fluorescence patterns. Scale bars, 500 mm. B, Peak amplitudes of DAN-oDA (top) and PFC-oDA
(bottom) were plotted as a function of florescence intensity at each location shown in A. Dotted lines indicate linear
regression between the oDA amplitudes and the fluorescence intensities. C, PFC-oDAs were measured from 74 loca-
tions (between ;1.0 and 1.2 mm in mediolateral coordinate) of mice injected with ChR2-EYFP in PrL/IL cortex and
color-coded according to their peak DA concentrations. The average PFC-oDA amplitudes were calculated and color-
coded for the three subregions. D, The PFC input fluorescence intensities were color-coded for the same 74 locations
in C. The averages from the three subregions were calculated and color-coded. E, Left, Fluorescence image of striatal
CINs labeled with td-Tomato. Inset, Examples of identified CINs (red) from the area with the dotted yellow line. Right,
The average numbers of CINs per 400  400 mm2 (area as shown in the inset) were calculated and shown for the
three subregions. F, Using cell-attached patch recording, action potentials evoked by PFC stimulation were observed
from a total of 46 CINs. For each CIN, a minimum light intensity to evoke action potentials was determined to score
the connectivity. White circles with a thicker line represent CINs which did not show any evoked action potentials
even with the maximum light intensity.
correlations with the fluorescent intensity of the projections
(p = 0.03 and r2 = 0.28 for DAN-oDA; p = 0.01, r2 = 0.37 for
PFC-oDA; Fig. 6B). However, the slope of the linear regression
was larger for DAN-oDA signals than PFC-oDA signals (10 vs 6

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 7520 • J. Neurosci., September 23, 2020 • 40(39):7510–7522
nM/AU, respectively; Fig. 6B). Thus, the intensity of the DAN-la-
beled fibers was a good predictor of the magnitude of DAN-oDA
signals. On the other hand, the presence of cortical axon fluores-
cence was not always a predictor of PFC-oDA signals, likely
because of the larger density of passing cortical axons that are
not synaptic terminal run through the striatum.
Averaging the area responses in 5 brain slices from 3 mice, we
found that the mean amplitude of the PFC-oDA signals was larg-
est in the ventral part of the dorsal striatum, which again maps in
gross terms with the location of the brightest intensity of ChR2-
EYFP-labeled projections from PrL/IL cortex (Fig. 6C,D).
However, the correlation between fluorescence intensity and am-
plitude of PFC-oDA signals is not always strong; for example,
the more intensely labeled caudal areas near the globus pallidus
show almost no PFC-oDA signal (Fig. 6C,D). The evidence of
small or no PFC-oDA signals in the more caudal striatum could
possibly reflect greater fiber density rather than innervation of
the caudal portion of the striatum by the axonal projections from
PrL/IL cortex.
The number of CINs is ,1% of the total striatal neurons, and
they are sparsely distributed throughout the striatum (Burke et
al., 2017). We then quantified the density of CINs from the same
area to further determine whether the distribution of CINs con-
tribute to the spatial restriction of PFC-oDA signals. For this
purpose, we took advantage of CIN-tdTomato mouse line, which
fluorescently labels CINs throughout the striatal slices (Fig. 6E).
The quantification analysis of cell numbers in 400  400 mm2
showed that the density of CINs was also highest in the middle
part of the dorsomedial striatum (6 slices from 3 mice), suggest-
ing that an uneven distribution of CINs could also contribute to
the spatial restriction of the local DA signals evoked by PrL/IL
cortical inputs to the ventral portion of the dorsomedial striatum.
To further test whether the spatial profile of PFC-oDA signals is
in part determined by the strength of the innervation of CINs by
PrL/IL cortical neurons, we performed cell-attached recordings
from the same CINs of CIN-TdTomato mice and measured the
ability of PFC optogenetic stimulation to evoke action potentials
in CINs throughout the striatum (46 CINs from 3 mice). We
found that the connectivity followed a similar pattern to the
PFC-oDA profile, and the probability that CINs will fire an
action potential in response to PFC stimulation was higher in the
central part of the striatum than in the NAc and dorsal regions
of the dorsal striatum (Fig. 6F).
In conclusion, this study offers a series of in vitro and in vivo
evidence that stimulation of cortical inputs from the PrL/IL cor-
tex can evoke DA release in the striatum via a local mechanism
that recruits CINs and requires activation of nAChRs. Since
these cortically evoked DA signals are spatially restricted and
have different properties, we speculate that they are engaged in
distinctive striatal functions from those assigned for DA signals
evoked by midbrain DANs.
Discussion
This study provides strong evidence that DA signals in the stria-
tum can occur both in vivo and in vitro in response to stimula-
tion of PrL/IL cortical inputs. These PrL/IL-evoked DA signals
have distinctive pharmacological and physiological properties,
compared with the DA signals evoked by DAN inputs. The in
vitro experiments also introduce a dual optogenetic stimulation
approach with which we can activate two different inputs to the
striatum in the same brain slice, without apparent cross-
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals
activation. This study further reveals the spatial localization of
the DA signals evoked by PFC inputs.
We showed that the PrL/IL-evoked DA signals require activa-
tion of glutamate and nicotinic ACh receptors, in line of a series
of elegant published work. Using in vitro radioactive assays and
in vivo microdialysis, it was first shown that locally administered
ACh and glutamate can trigger DA release in the striatum
(Giorguieff et al., 1976, 1977). Giorguieff et al. (1976) showed
that an nAChR antagonist blocked ACh-evoked DA signals, and
speculated that “. . . the release of DA from dopaminergic termi-
nals can be regulated by cholinergic presynaptic receptors exhib-
iting nicotinic characteristics.” To their credit, their conclusions
agree with our interpretations of the findings from the current
study as well as other recent optogenetic studies where it was
shown that selective stimulation of CINs is sufficient to evoke
DA signals (Cachope et al., 2012; Threlfell et al., 2012; Wang et
al., 2014; Shin et al., 2017). The involvement of CINs in the gluta-
mate-dependent DA signals was also suggested by early work by
Taber and Fibiger (1994) who showed that electrical stimulation
of PFC can increase levels of ACh in the striatum and further
supported by more recent optogenetic studies (Threlfell et al.,
2012; Kosillo et al., 2016; Johnson et al., 2017; Mateo et al., 2017;
Cover et al., 2019).
Here, our experiments confirmed the requirement for iono-
tropic glutamate receptor activation in the PFC-evoked DA sig-
nals as previously shown (Mateo et al., 2017). We also show that
nAChRs are required for the PrL/IL-evoked DA signals in vitro,
similar to other findings when stimulating inputs from other
cortical (Kosillo et al., 2016) and thalamic areas (Threlfell et al.,
2012; Kosillo et al., 2016; Cover et al., 2019) and indirectly in
vivo (Mateo et al., 2017). Our results from the in vivo microdialy-
sis experiments reproduce the original findings from Quiroz et
al. (2016) showing that optogenetic stimulation of PFC inputs
increases striatal levels of DA in vivo (and glutamate, as expected;
Fig. 1). More importantly, we showed that stimulation of PrL/IL
inputs to the striatum also elevates striatal levels of ACh in vivo
and that this cortically evoked DA signals are blocked when a
nAChR antagonist is perfused (Fig. 1C,D). Thus, these findings
support that idea PrL/IL inputs form excitatory synapses on
striatal CINs and can activate then to evoke ACh release (Fig.
1C). In agreement with this conclusion, recent work showed that
optogenetic stimulation of either M1 motor cortex or parafascic-
ular nucleus of the thalamus can induce release of ACh that was
detected using exogenous G-protein-coupled inward rectifying
K-current expressed in medium spiny neurons (Mamaligas et al.,
2019). Together, our findings from both in vitro and in vivo
strongly support the hypothesis of a local mechanism for evoking
DA release in the striatum, in addition to the more conventional
mechanism based on midbrain DAN firing. This local mecha-
nism engages striatal CINs, the targets of inputs from PFC,
which have an essential role in evoking DA release from DAN
fibers.
DA signals evoked by PrL/IL inputs and those evoked by
DAN fibers stimulation share common properties, such as over-
lapping electrochemical profiles of the voltammetric currents
and similar concentration range, time course, and decay time
constant of the signals (Figs. 2, 5E,F). However, there are also
several differences between the input-specific DA signals. First,
DAN-evoked DA signals do not require activation of either iono-
tropic glutamate receptors or nAChRs (Figs. 3C–F, 5G,H), con-
firming the direct nature of this mechanism that is triggered by
ChR2-evoked action potentials in DAN axon fibers. Second,
PrL/IL-evoked DA signals display different temperature
 Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals
sensitivity and a higher threshold compared with DAN-oDA,
requiring a longer duration of light stimulation as determined in
the input-output relationship (Figs. 3, 4). This finding may
reflect the polysynaptic nature of the cortically evoked signals
and the requirement of synchronized activation of CINs to trig-
ger DA release by this local mechanism (Threlfell et al., 2012;
Kosillo et al., 2016; Liu et al., 2018). Third, the amplitude of
DAN-evoked DA signals increases as the number of stimulation
pulses increases, indicating some summation in the DA signals
evoked by a train of stimulation pulses (Fig. 4A,B). In contrast,
the amplitude of DA signals in response to a train of pulses of
cortical input stimulation is similar to the amplitudes obtained
in response to a single pulse of stimulation, indicating no sum-
mation under this condition for the local mechanism (Fig. 4A,B).
ChR2 displays use-dependent inactivation when stimulated at
high frequency (Hass and Glickfeld, 2016). However, the lack of
summation by the train stimulation is unlikely because of this
use-dependent inactivation of ChR2 since 20 Hz train stimulation
was still able to evoke action potentials in CINs recorded in cell-
attached mode (Fig. 4D,E) and amperometry recording showed
large deflections in response to train stimulation (Fig. 4C).
Anecdotally, these current deflections disappeared when the ion-
otropic glutamate receptor antagonists were applied, also sup-
porting the idea that firing in CINs and other downstream
neurons occurs in response to 20 Hz train stimulation. The lack
of summation was also reported in DA signals evoked by direct
optogenetic stimulation of CINs in the dorsal striatum (Threlfell
et al., 2012; Shin et al., 2017), which was suggested to be because
of nAChR desensitization. We speculate then that the lack of
summation we report here by 20 Hz trains in PrL/IL-evoked DA
signals, which also require nAChRs activation, is also because of
desensitization.
To directly study and compare the input-specific DA signals
in the same animal, we established dual optogenetic stimulation
using a red-shifted opsin, ChrimsonR, in combination with
the blue-light activated ChR2. Two wavelengths for excitation
were chosen based on the excitation spectrum of each opsin
(Klapoetke et al., 2014), 590 nm for ChrimsonR and 420 nm for
ChR2, which showed no cross-activation (Fig. 5A,B). This dual
optogenetic approach was further validated using pharmacology
in animals expressing ChR2 in midbrain DANs and ChrimsonR
in PrL/IL neurons. We found that PrL/IL-evoked DA signals
were completely blocked by antagonists of either ionotropic glu-
tamate receptors or nAChRs, without affecting DAN-oDA sig-
nals in the same slice (Fig. 5G,H). Therefore, we determined that
the two inputs can be activated independently using this dual
optogenetic approach in brain slices from the same animal.
Interestingly, using this dual optogenetic approach, an in-
triguing spatial pattern was revealed for the cortically evoked
PrL/IL DA signals. While DAN-oDA signals were recorded
throughout the striatum, PrL/IL oDA signals were spatially re-
stricted to a “hot-spot” region around the boundary between the
NAc core and dorsal medial striatum (Fig. 6). This hot-spot area
for the PrL/IL oDA signals corresponds to the same area where
we found the highest density of CINs in our analysis. Thus, CIN
density could contribute in part to the generation of this hot-
spot area for the PrL/IL-evoked DA signals. Further, we also
found that this area has the highest intensity of fluorescently la-
beled PrL/IL fibers, as well as the highest degree of synaptic con-
nections to CINs, which is also in agreement with reported
patterns of cortical connectivity (Voorn et al., 2004). Our initial
interest was to study the PFC projections to the accumbens.
However, since PrL/IL-evoked DA signals were largest and more
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
J. Neurosci., September 23, 2020 • 40(39):7510–7522 • 7521
reliable in the ventral DS (Fig. 6), recordings were performed in
this area. Together, these findings suggest that PrL/IL inputs and
the synaptic innervation to CINs are preferentially localized to
the dorsomedial subregion, giving rise to the largest cortically
evoked DA signals.
From the mechanism described here for this PrL/IL-evoked
DA signals, we propose that any input to the striatum that
strongly activates CINs (cortical, thalamic, or any other brain
region) can in theory evoke local DA signals in the striatum. We
also speculate, based on ours and other groups’ works, that these
glutamate-driven DA signals can similarly happen in other stria-
tal subregions. The selective localization of PrL/IL-evoked DA
signals suggests the existence of a “topographic map” for the
local DA signals where inputs from different cortical subregions
can trigger local DA signals at the striatal subregion which they
innervate. Spatial segregation of input-specific DA signals in the
striatum is a new concept introduced here that is rarely consid-
ered in the field. This concept may become helpful in under-
standing how DA signals in the striatum can be involved in so
many diverse processes ranging from reinforcement learning
and motivation to motor output and action selection.
References
Adrover MF, Shin JH, Alvarez VA (2014) Glutamate and dopamine trans-
mission from midbrain dopamine neurons share similar release proper-
ties but are differentially affected by cocaine. J Neurosci 34:3183–3192.
Backman CM, Malik N, Zhang Y, Shan L, Grinberg A, Hoffer BJ, Westphal
H, Tomac AC (2006) Characterization of a mouse strain expressing Cre
recombinase from the 39 untranslated region of the dopamine transporter
locus. Genesis 44:383–390.
Burke DA, Rotstein HG, Alvarez VA (2017) Striatal local circuitry: a new
framework for lateral inhibition. Neuron 96:267–284.
Cachope R, Mateo Y, Mathur BN, Irving J, Wang HL, Morales M, Lovinger
DM, Cheer JF (2012) Selective activation of cholinergic interneurons
enhances accumbal phasic dopamine release: setting the tone for reward
processing. Cell Rep 2:33–41.
Chuhma N, Mingote S, Moore H, Rayport S (2014) Dopamine neurons con-
trol striatal cholinergic neurons via regionally heterogeneous dopamine
and glutamate signaling. Neuron 81:901–912.
Cover KK, Gyawali U, Kerkhoff WG, Patton MH, Mu C, White MG,
Marquardt AE, Roberts BM, Cheer JF, Mathur BN (2019) Activation of
the rostral intralaminar thalamus drives reinforcement through striatal
dopamine release. Cell Rep 26:1389–1398.e3.
Floresco SB, Yang CR, Phillips AG, Blaha CD (1998) Basolateral amygdala
stimulation evokes glutamate receptor-dependent dopamine efflux in the
nucleus accumbens of the anaesthetized rat. Eur J Neurosci 10:1241–
1251.
Franklin KB, Paxinos G (2007) The mouse brain in stereotaxic coordinates,
Ed 3. Amsterdam: Elsevier.
Gerfen CR (2000) Dopamine-mediated gene regulation in models of
Parkinson’s disease. Ann Neurol 47:S42–S50; discussion S50-S52.
Gerfen CR, Surmeier DJ (2011) Modulation of striatal projection systems by
dopamine. Annu Rev Neurosci 34:441–466.
Giorguieff MF, Le Floc’h ML, Westfall TC, Glowinski J, Besson MJ (1976)
Nicotinic effect of acetylcholine on the release of newly synthesized (3H)
dopamine in rat striatal slices and cat caudate nucleus. Brain Res
106:117–131.
Giorguieff MF, Kemel ML, Glowinski J (1977) Presynaptic effect of L-glu-
tamic acid on the release of dopamine in rat striatal slices. Neurosci Lett
6:73–77.
Hass CA, Glickfeld LL (2016) High-fidelity optical excitation of cortico-corti-
cal projections at physiological frequencies. J Neurophysiol 116:2056–
2066.
Hill DF, Parent KL, Atcherley CW, Cowen SL, Heien ML (2018) Differential
release of dopamine in the nucleus accumbens evoked by low-versus
high-frequency medial prefrontal cortex stimulation. Brain Stimul
11:426–434.
 7522 • J. Neurosci., September 23, 2020 • 40(39):7510–7522
Hyland BI, Reynolds JN, Hay J, Perk CG, Miller R (2002) Firing modes of
midbrain dopamine cells in the freely moving rat. Neuroscience 114:475–
492.
Johnson KA, Mateo Y, Lovinger DM (2017) Metabotropic glutamate receptor
2 inhibits thalamically-driven glutamate and dopamine release in the dor-
sal striatum. Neuropharmacology 117:114–123.
Klapoetke NC, Murata Y, Kim SS, Pulver SR, Birdsey-Benson A, Cho YK,
Morimoto TK, Chuong AS, Carpenter EJ, Tian Z, Wang J, Xie Y, Yan Z,
Zhang Y, Chow BY, Surek B, Melkonian M, Jayaraman V, Constantine-
Paton M, Wong GK, et al. (2014) Independent optical excitation of dis-
tinct neural populations. Nat Methods 11:338–346.
Kosillo P, Zhang YF, Threlfell S, Cragg SJ (2016) Cortical control of striatal
dopamine transmission via striatal cholinergic interneurons. Cereb
Cortex 26:4160–4169.
Lahiri AK, Bevan MD (2020) Dopaminergic transmission rapidly and persis-
tently enhances excitability of D1 receptor-expressing striatal projection
neurons. Neuron 106:277–290.e6.
Leviel V, Gobert A, Guibert B (1990) The glutamate-mediated release of do-
pamine in the rat striatum: further characterization of the dual excita-
tory-inhibitory function. Neuroscience 39:305–312.
Liu C, Kaeser PS (2019) Mechanisms and regulation of dopamine release.
Curr Opin Neurobiol 57:46–53.
Liu C, Kershberg L, Wang J, Schneeberger S, Kaeser PS (2018) Dopamine
secretion is mediated by sparse active zone-like release sites. Cell
172:706–718.e15.
Luscher C, Robbins TW, Everitt BJ (2020) The transition to compulsion in
addiction. Nat Rev Neurosci 21:247–263.
Madisen L, Zwingman TA, Sunkin SM, Oh SW, Zariwala HA, Gu H, Ng LL,
Palmiter RD, Hawrylycz MJ, Jones AR, Lein ES, Zeng H (2010) A robust
and high-throughput Cre reporting and characterization system for the
whole mouse brain. Nat Neurosci 13:133–140.
Mamaligas AA, Cai Y, Ford CP (2016) Nicotinic and opioid receptor regula-
tion of striatal dopamine D2-receptor mediated transmission. Sci Rep
6:37834.
Mamaligas AA, Barcomb K, Ford CP (2019) Cholinergic transmission at
muscarinic synapses in the striatum is driven equally by cortical and tha-
lamic inputs. Cell Rep 28:1003–1014.e3.
Mateo Y, Johnson KA, Covey DP, Atwood BK, Wang HL, Zhang S, Gildish I,
Cachope R, Bellocchio L, Guzman M, Morales M, Cheer JF, Lovinger
DM (2017) Endocannabinoid actions on cortical terminals orchestrate
local modulation of dopamine release in the nucleus accumbens. Neuron
96:1112–1126.e5.
Matsuda W, Furuta T, Nakamura KC, Hioki H, Fujiyama F, Arai R, Kaneko
T (2009) Single nigrostriatal dopaminergic neurons form widely spread
and highly dense axonal arborizations in the neostriatum. J Neurosci
29:444–453.
Melchior JR, Ferris MJ, Stuber GD, Riddle DR, Jones SR (2015) Optogenetic
versus electrical stimulation of dopamine terminals in the nucleus accum-
bens reveals local modulation of presynaptic release. J Neurochem
134:833–844.
Quinn DM (1987) Acetylcholinesterase: enzyme structure, reaction dynam-
ics, and virtual transition states. Chem Rev 87:955–979.
Quiroz C, Orrú M, Rea W, Ciudad-Roberts A, Yepes G, Britt JP, Ferré S
(2016) Local control of extracellular dopamine levels in the medial nu-
cleus accumbens by a glutamatergic projection from the infralimbic cor-
tex. J Neurosci 36:851–859.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Adrover, Shin et al. · PFC-Driven Striatal Dopamine Signals
Rossi J, Balthasar N, Olson D, Scott M, Berglund E, Lee CE, Choi MJ, Lauzon
D, Lowell BB, Elmquist JK (2011) Melanocortin-4 receptors expressed by
cholinergic neurons regulate energy balance and glucose homeostasis.
Cell Metab 13:195–204.
Schultz W, Apicella P, Ljungberg T (1993) Responses of monkey dopamine
neurons to reward and conditioned stimuli during successive steps of
learning a delayed response task. J Neurosci 13:900–913.
Shimizu N, Duan SM, Hori T, Oomura Y (1990) Glutamate modulates dopa-
mine release in the striatum as measured by brain microdialysis. Brain
Res Bull 25:99–102.
Shin JH, Adrover MF, Alvarez VA (2017) Distinctive modulation of dopa-
mine release in the nucleus accumbens shell mediated by dopamine and
acetylcholine receptors. J Neurosci 37:11166–11180.
Shin JH, Adrover MF, Wess J, Alvarez VA (2015) Muscarinic regulation of
dopamine and glutamate transmission in the nucleus accumbens. Proc
Natl Acad Sci USA 112:8124–8129.
Sulzer D, Cragg SJ, Rice ME (2016) Striatal dopamine neurotransmission:
regulation of release and uptake. Basal Ganglia 6:123–148.
Surmeier DJ, Ding J, Day M, Wang Z, Shen W (2007) D1 and D2 dopamine-
receptor modulation of striatal glutamatergic signaling in striatal medium
spiny neurons. Trends Neurosci 30:228–235.
Taber MT, Fibiger HC (1993) Electrical stimulation of the medial prefrontal cor-
tex increases dopamine release in the striatum. Neuropsychopharmacology
9:271–275.
Taber MT, Fibiger HC (1994) Cortical regulation of acetylcholine release in
rat striatum. Brain Res 639:354–356.
Threlfell S, Clements MA, Khodai T, Pienaar IS, Exley R, Wess J, Cragg SJ
(2010) Striatal muscarinic receptors promote activity dependence of do-
pamine transmission via distinct receptor subtypes on cholinergic inter-
neurons in ventral versus dorsal striatum. J Neurosci 30:3398–3408.
Threlfell S, Lalic T, Platt NJ, Jennings KA, Deisseroth K, Cragg SJ (2012)
Striatal dopamine release is triggered by synchronized activity in cholin-
ergic interneurons. Neuron 75:58–64.
Tritsch NX, Sabatini BL (2012) Dopaminergic modulation of synaptic trans-
mission in cortex and striatum. Neuron 76:33–50.
Tritschler L, Kheirbek MA, Dantec YL, Mendez-David I, Guilloux JP, Faye
C, Doan J, Pham TH, Hen R, David DJ, Gardier AM (2018) Optogenetic
activation of granule cells in the dorsal dentate gyrus enhances dopami-
nergic neurotransmission in the nucleus accumbens. Neurosci Res
134:56–60.
Vidal CJ, Chai MS, Plummer DT (1987) The effect of temperature on the ac-
tivity of acetylcholinesterase preparations from rat brain. Neurochem Int
11:135–141.
Voorn P, Vanderschuren LJ, Groenewegen HJ, Robbins TW, Pennartz CM
(2004) Putting a spin on the dorsal-ventral divide of the striatum. Trends
Neurosci 27:468–474.
Wang L, Shang S, Kang X, Teng S, Zhu F, Liu B, Wu Q, Li M, Liu W, Xu H,
Zhou L, Jiao R, Dou H, Zuo P, Zhang X, Zheng L, Wang S, Wang C,
Zhou Z (2014) Modulation of dopamine release in the striatum by physi-
ologically relevant levels of nicotine. Nat Commun 5:3925.
Zhang H, Sulzer D (2004) Frequency-dependent modulation of dopamine
release by nicotine. Nat Neurosci 7:581–582.
Zhou FM, Liang Y, Dani JA (2001) Endogenous nicotinic cholinergic activity
regulates dopamine release in the striatum. Nat Neurosci 4:1224–1229.
Zhu J, Hexum TD (1992) Characterization of cocaine-sensitive dopamine
uptake in PC12 cells. Neurochem Int 21:521–526.
 Neuron
Article
NMDA Receptors in Dopaminergic Neurons
Are Crucial for Habit Learning
Lei Phillip Wang,1 Fei Li,1,2 Dong Wang,1 Kun Xie,1 Deheng Wang,1 Xiaoming Shen,2 and Joe Z. Tsien1,*
1Brainand Behavior Discovery Institute and Department of Neurology, Georgia Health Sciences University, Augusta, GA 30912, USA
2Department of Developmental and Behavioral Pediatrics of Shanghai Children’s Medical Center & Shanghai Key Laboratory of Children’s
Environmental Health, XinHua Hospital affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai 200092, China
*Correspondence: jtsien@georgiahealth.edu
DOI 10.1016/j.neuron.2011.10.019
SUMMARY
Dopamine is crucial for habit learning. Activities of
midbrain dopaminergic neurons are regulated by
the cortical and subcortical signals among which glu-
tamatergic afferents provide excitatory inputs.
Cognitive implications of glutamatergic afferents in
regulating and engaging dopamine signals during
habit learning, however, remain unclear. Here, we
show that mice with dopaminergic neuron-specific
NMDAR1 deletion are impaired in a variety of
habit-learning tasks, while normal in some other
dopamine-modulated functions such as locomotor
activities, goal-directed learning, and spatial refer-
ence memories. In vivo neural recording revealed
that dopaminergic neurons in these mutant mice
could still develop the cue-reward association res-
ponses; however, their conditioned response robust-
ness was drastically blunted. Our results suggest that
integration of glutamatergic inputs to DA neurons by
NMDA receptors, likely by regulating associative
activity patterns, is a crucial part of the cellular mech-
anism underpinning habit learning.
INTRODUCTION
Many acts, after repetitive practice, would transform from being
goal-directed to automated habits, which can be carried out effi-
ciently and subconsciously. Habits help to free up the cognitive
loads on routine procedures and allow us to focus on new situa-
tions and tasks. Despite breakthroughs unveiling participation of
different anatomical structures in habit formation (Knowlton
et al., 1996; Yin and Knowlton, 2006), the underpinning physio-
logical mechanisms and how different network circuitries inte-
grate relevant information remain unclear.
Dopamine (DA) is an important regulator of synaptic plasticity,
especially in the basal ganglia, a structure essential for habit
learning. In both human patients (Fama et al., 2000; Knowlton
et al., 1996) and rodents (Faure et al., 2005), habit learning is
often found impaired following dopaminergic neuron degenera-
tion. Dopamine has thus been postulated as a main modulator in
the mechanisms subserving habit learning (Ashby et al., 2010).
Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc. 1055
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Despite this importance, the mechanisms modulating dopamine
during habit learning have yet to be fully investigated. Studies
have shown that habit-learning deficits caused by dopamine
deafferentation could not be rescued by simple intrastriatal
injections of DA agonists (Faure et al., 2010). These observations
suggested that dopamine, the modulator itself, might need to be
regulated during normal habit learning. Anatomically, along with
cholinergic inputs, glutamatergic afferents from brain structures
such as pedunculopontine tegmental nucleus (PPTg), subthala-
mic nucleus (STN), and prefrontal cortex (PFC) provide main
forms of excitatory inputs to the midbrain DA neurons (Grace
et al., 2007). NMDA receptors (NMDARs), members of the iono-
tropic glutamate receptor family, are important regulators of DA
neuron activity. First, synaptic plasticity in the glutamatergic
afferents to DA neurons depends on NMDARs (Bonci and Mal-
enka, 1999; Overton et al., 1999; Ungless et al., 2001). This plas-
ticity can be modulated by experiences, environmental factors,
and psychostimulant drugs (Bonci and Malenka, 1999; Kauer
and Malenka, 2007; Saal et al., 2003). Second, iontophoretic
administration of NMDAR antagonists, but not AMPAR-selective
antagonists, attenuated phasic firing of DA neurons, an activity
linked to reward/incentive salience (Schultz, 1998), without
changing the frequency of tonic firing (Overton and Clark,
1992). Third, in drug addiction studies, NMDARs in DA neurons
are essential for developing nicotine-conditioned place prefer-
ence (Wang et al., 2010) and likely also involved in cocaine-
conditioned place preference (Engblom et al., 2008; Zweifel
et al., 2008). Thus, we postulated that modulation of DA neurons
by NMDARs might be important in engaging DA neurons in the
habit learning. Here, we set out to examine the roles of NMDARs
in DA neurons, by generating DA neuron-specific NR1 knockout
mice and testing them in a variety of habit-learning paradigms
(Devan and White, 1999; Dickinson et al., 1983; Packard et al.,
1989; Packard and McGaugh, 1996). In order to understand
the cellular mechanisms, we also recorded the DA neurons in
these mice using multielectrode in vivo neural-recording tech-
niques (Wang and Tsien, 2011).
RESULTS
Production and Basic Characterization of DA
Neuron-Selective NR1 Knockout Mice
These mice, named ‘‘DA-NR1-KO,’’ were produced by crossing
floxed NR1 (fNR1) mice (Tsien et al., 1996) with Slc6a3+/Cre

A B
C D
transgenic mice that express Cre recombinase under DA trans-
porter promoter (Zhuang et al., 2005) (Figures 1A and 1B). The
DA neuron-specific deletion of the NR1 gene was confirmed by
both the reporter gene method (Figure 1C) and immunohisto-
chemistry (Figure 1D), which showed that the gene deletion
was restricted to the dopaminergic neurons in regions such as
the VTA and the substantia nigra. No obvious changes were
observed in the expression pattern of tyrosine hydroxylase
(TH), the catecholamine neuronal marker, suggesting that there
was no obvious loss of dopaminergic neurons (see Figure S1
available online).
DA-NR1-KO mice were born in the expected Mendelian ratios
and visually indistinguishable from the controls. Additionally,
they were normal in locomotor activities in a novel open field (Fig-
ure 2A), in learning the rotarod tests (Figure 2B), in an anxiety test
using the elevated plus maze (Figure 2C), and in the novel object
recognition tests (Figure 2D). These results showed that many of
the behavioral functions that were sensitive to dopamine
dysfunctions were preserved in the DA-NR1-KO mice.
DA Neurons in the DA-NR1-KO Showed Normal Tonic
Firing, Reduced Phasic Firing, and Reduced Responses
to the Reward-Predicting Cue
In order to investigate the impact of NR1 deletion on the cellular
properties of DA neurons, we recorded the activities of these
neurons in both the DA-NR1-KO mice and wild-type control litter-
mates. Movable bundles of 8 tetrodes (32 channels) were im-
planted into the ventral midbrain, primarily the VTA. The putative
DA neurons were identified based on their firing patterns and their
sensitivity to dopamine receptor agonist apomorphine (1 mg/kg,
i.p.) at the end of each recording session (Figures 3A–3D).
A total of 14 putative DA neurons from 4 mutant mice and 16
from 6 wild-type controls were recorded and analyzed. Phasic-
firing activities or bursting was defined as a spike train beginning
with an interspike interval (ISI) smaller than 80 ms and termi-
nating with an ISI greater than 160 ms. Compared with the
control neurons, phasic-firing activities were greatly reduced in
1056 Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Neuron
NMDAR Engages DA Neurons in Habit Learning
Figure 1. Generation and Characterization
of DA-NR1-KO Mice
(A) Breeding scheme for making DA-NR1-KO mice
and for DAT-Cre/rosa-stop-lacZ mice.
(B) PCR genotyping of DA-NR1-KO (Cre, fNR1/
fNR1) mice and their littermates.
(C) Immunofluorescent staining of DAT-Cre/Rosa-
Stop-lacZ mouse brain with lacZ antibody, with
anti-TH antibody and overlay of the two stainings.
(D) Immunofluorescent staining of DA-NR1-KO and
wild-type control brain with anti-NR1 antibody,
anti-TH antibodies and overlay of both stainings.
Arrowheads indicate position of some TH-positive
neurons. TH signals colocalize with NR1 in the
control brain, but not in the DA-NR1-KO brains.
Pictures in the larger boxes are the 3-fold enlarge-
ment of the pictures shown in the respective small
boxes.
the DA-NR1-KO neurons. The observed median frequency of
phasic firing decreased from 0.78 ± 0.09 Hz in the control DA
neurons to 0.36 ± 0.09 Hz in KO DA neurons. (Mann-Whitney U
test, p < 0.01) (Figure 3E). A significant reduction was also
observed in the percentages of spikes fired in phasic activities
(34.7% in the controls versus 21.2% in the DA-NR1-KO;
Mann-Whitney U test, p < 0.01) (Figure 3F). The total firing rate
was also reduced in the mutant DA neurons. This appeared to
be correlated with reduced burst set rate (5.18 ± 0.59 Hz, control,
versus 3.85 ± 0.38 Hz, KO; r = 0.7719, Mann-Whitney U test,
p < 0.01) (Figure 3G). No significant difference was observed
in the tonic firing between the mutant and control groups.
(4.42 ± 0.44 Hz in control, versus 3.29 ± 0.36 Hz in KO; Mann-
Whitney U test, p > 0.05) (Figure 3H).
To further evaluate the response of DA neurons in a learning
task, mice were trained 40 trials per day in a Pavlovian-condi-
tioning paradigm in which a 5 KHz tone that lasted 1 s proceeded
immediately before the delivery of a food pellet. DA neurons from
both genotypes were able to associate the tone with phasic
firing, but the conditioned responses were much weaker in the
DA-NR1-KO group (Figure 4A). Although DA-NR1-KO neurons
showed increased firing over the days during the training, their
responses were significantly reduced compared with the
controls on day 1 (19.21 ± 3.24 Hz, control, versus 9.74 ±
0.30 Hz, KO; p < 0.01), day 2 (36.33 ± 4.39 Hz, control, versus
16.43 ± 4.01 Hz, KO; p < 0.01), and day 3 (59.38 ± 3.82 Hz,
control, versus 33.88 ± 4.30 Hz, KO; p < 0.01) (Figure 4B). These
data suggested that while NMDAR1 deletion did not completely
prevent DA neurons from developing conditioned responses
(bursting) toward reward-predicting cues, it did, however,
greatly lower the robustness of the bursting response,
a phenomena that we call DA neuron blunting.
Habit Learning, but Not Goal-Directed Learning,
Was Impaired in the Operant Appetitive Conditioning
To assess habit learning, we first tested the mice in a lever-
pressing operant-conditioning task. In this task an instrumental
 Neuron
NMDAR Engages DA Neurons in Habit Learning

A B

C D

Figure 2. Basic Behavioral Characterization of DA-NR-KO Mice

(A) Locomotor activity tests in a novel environment. Fine movements, ambulatory movements, total movements (total of fine plus ambulatory movements), and
rearing were scored. Post hoc comparisons revealed no differences among the mutant and the three control groups.
(B) Rotarod test. Mice were tested 1 and 24 hr after initial training. All mice performed better (as shown by extended latencies) at 24 hr. Post hoc comparisons
showed no differences among the mutant and control groups at either time points.
(C) Anxiety level tested using elevated plus maze. Percentages of time spent in each arms were calculated. Post hoc comparisons showed no differences among
the genotypes. *p < 0.05, Student’s t test.
(D) Novel object recognition test. At ‘‘24 hour,’’ Time % = time spent exploring novel object/(time spent exploring familiar object + time spent exploring novel
object). At ‘‘Training,’’ Time % = time spent exploring one object/total time spent exploring both objects. All mice spent more time on the novel object at 24th hr
(p < 0.01), with post hoc comparisons revealing no differences among the groups tested. Error bars represent SEM. *p < 0.05, Student’s t test.

action, pressing lever to obtain food, can transform from a goal-
directed to a habitual response after extensive training and
become progressively less sensitive to devaluation of outcome
(Dickinson et al., 1983). The decreased sensitivity can thus be
measured as a behavioral readout of habit learning (Figure 5A).
Both mutant and control mice learned to press the lever on an
extensive training protocol consisting of 4 days of continuous
reinforcement (CRF), 2 days of random interval (RI) 30 s, and
6 days of RI 60 s schedules (Dickinson et al., 1983). Mice in
both groups increased lever-press rates during the training
(CRF days 1–4, RI 30 s days 5 and 6, RI 60 s days 7–12) (Fig-
ure 5B). A two-way ANOVA of repeated measures, with days
and genotype as factors, showed no effect of genotype
(F(1, 231) = 0.07), a main effect of days (F(11, 231) = 51.4; p <
0.01), and no interaction between these factors (F(11, 231) =
0.269). This result suggested that the DA-NR1-KO mice have
normal wanting of the pellet reward and exhibited normal goal-
directed learning.
Lever pressing was then tested after the outcome devaluation.
Mice were prefed with either regular mouse chow to which they
had been exposed in their regular home cages (nondevalued
condition/control), or purified high-energy pellets that are iden-
tical to the rewards earned during lever-press sessions (deval-
ued condition). Feeding with mouse chow was used as a control
for the overall level of satiety, causing little reduction in the
rewarding value of the purified high-energy pellets. Levers
were inserted in the 5 min long probe test that immediately fol-
lowed the 1 hr unlimited food exposure (pellets or chow). No
pellets were given during the tests. Comparing numbers of lever
press during the tests showed that, while no differences were
found between the mutant and the control mice on nondevalued
condition (p = 0.94) or between the devalued and nondevalued
conditions (p = 0.153) in the control group, there was a significant
difference in the mutant mice between devalued and non-
devalued conditions (p < 0.01). Furthermore, there was also
a significant difference between the mutant and control mice
on devalued condition (p < 0.05). A two-way ANOVA of repeated
measures, with treatment and genotype as factors, showed an
interaction between the two factors (F(1, 21) = 4.98; p < 0.05) (Fig-
ure 5C). These suggested that the conditional knockout mice
failed to develop the lever-pressing habit despite extensive
training, and their action stayed goal directed.

Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc. 1057
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

Spatial Navigation Habit, but Not Spatial Memory,
Was Impaired in the Positively Reinforced Plus Maze
Habit learning was then assessed in a navigation-based para-
digm using plus maze place/response-learning tasks (Devan
and White, 1999; Packard, 1999; Packard and McGaugh,
1996). Littermates in genotypes Slc6a3+/Cre; fNR1/+, Slc6a3+/
Cre, and wild-type served as three control groups for the DA-
NR1-KO mice. The maze was built with transparent walls and
placed in a room furnished with spatial cues. The schematic
training and testing schedules are shown in Figure 6A. Naive
animals, always starting from the same location in the maze
(the ‘‘south’’ arm), were trained to find a fixed target site (in the
1058 Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Neuron
NMDAR Engages DA Neurons in Habit Learning
Figure 3. Burst Firing by DA Neurons Is
Impaired in KO Mice
(A and B) Sample waveform of a DA neuron re-
corded from a control (A) and a KO (B) mouse, and
corresponding ISI histogram (10 ms bins).
(C and D) Cumulative spike activity of an example
of a putative DA neuron from the control (C) and
KO (D) mouse in response to the dopamine
receptor agonist apomorphine (1 mg/kg, i.p.).
(E) Burst set rate by DA neurons from control and
KO mice.
(F) Percent spikes fired in bursts by DA neurons
from control and KO mice.
(G) Correlation between burst set rate and firing
rate.
(H) Frequency of nonbursting spikes. See text for
all the statistics.
‘‘east’’ arm) (Training I in Figure 6A). In
order to facilitate developing habit-based
navigation, the north and the west arms
were both closed. It has been shown
that under this paradigm, normal mice
would learn to search the target using
spatial reference memory after moderate
training but would switch to habitual
navigation after extensive training (Pack-
ard and McGaugh, 1996). Probe trials,
during which the start location switched
from the ‘‘south’’ arm to the ‘‘north’’
arm, were given at different time points
to allow dissociation of the spatial and
habitual strategies. Thus, mice using the
‘‘habit strategy’’ were predicted to turn
right (into the ‘‘west’’ arm), whereas the
‘‘spatial’’ mice, guided by distal spatial
cues, were predicted to go to the ‘‘east’’
arm, where the target resided during
training.
All mice were trained in ten trials per
day for 5 consecutive days before the first
probe trial on day 6 (Probe 1 in Figure 6A).
During this probe trial the DA-NR1-KO
group and control mice showed similar
preferences (c2 [3, n = 43] = 0.346; p =
0.951) for the ‘‘spatial’’ strategy, opting
to turn left toward the ‘‘east’’ arm (Figure 6B), suggesting that
they had similarly acquired the spatial memory and that they
shared comparable motivation. All mice were then trained for
10 additional days before the second probe trial (Probe 2 in
Figure 6A) on day 17. During this probe trial no significant differ-
ences were found among the three control groups (c2 [2, n =
29] = 0.499; p = 0.779). As a group, control mice opted to ‘‘turn
right’’ (and into the ‘‘west’’ arm) significantly more on day 17
than on day 6 (c2 [1, n = 29] = 22.587; p = 0.00000201), indicating
a learned ‘‘habit’’-based searching strategy. In contrast, less
than 10% of the DA-NR1-KO mice (compared with 80% of
control mice) (mutants versus controls: c2 = 7.244; p = 0.007)
 Neuron
NMDAR Engages DA Neurons in Habit Learning
A A
Figure 4. Responses of Putative Dopamine Neurons in 3 Day Reward
Test
(A) Peri-event histograms of an example DA neuron recorded from control and
KO mouse in response to the same conditioned tone (5 kHz, 1 s) that predicted
a sugar pellet delivery in a 3 day session.
(B) Excitation peak firing rate of DA neurons from control (red line, n = 16) and
KO (blue line, n = 14) in response to reward during these 3 days. Error bars
represent SD. **p < 0.01, excitation peak rate in KO versus control DA neurons,
in day 1, 2, and 3.
opted to turn ‘‘right’’ on day 17 (Figure 6B), suggesting that they
failed to learn the ‘‘habit’’-based strategies and, instead, kept
using the ‘‘spatial’’ strategy.
To confirm that the deficits in the plus maze tasks were
indeed from habit learning, right after the second probe trial,
mice were further challenged in a ‘‘relearn after 90 rotation’’
procedure (Training II, Figure 6A), three trials a day for 2 days
within the exact same maze and surrounding cues. During
the training, both the west and south arms were blocked. The
start box was placed in the ‘‘east’’ arm, and the food rewards
were in the ‘‘north’’ arm. Mice were tested in a rotation test
on day 19, and accuracies to locate the food were scored.
Mice started from the ‘‘east’’ arm with all arms open during
the test (Figure 6A). For ‘‘habit’’ mice who had learned to
‘‘turn right’’ during previous training sessions (days 1–16), this
new learning was simply a retraining, in which the same habit
response (turning ‘‘right’’) would lead them to the new food
location. However, for the ‘‘spatial’’ mice, switching of target
location from the ‘‘east’’ arm to the ‘‘north’’ arm conflicted
Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc. 1059
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
?!
Figure 5. Habit and Goal-Directed Learning Test with Operant Appe-
titive Conditioning
(A) Paradigm for the training and test.
(B) Lever-pressing speed during training days. No significant difference was
discovered at any day of the training between the mutant and control groups.
(C) Lever pressing in the 5 min extinction test after devaluation of outcome and
nondevaluation of outcome. Mutant mice showed significantly reduced lever
pressing at the devalued condition versus both control mice at devalued
condition and mutants at nondevalued condition. No significant reduction was
detected for the control mice between the devalued and nondevalued
conditions. (See text for all the statistics.) Error bars represent SEM. *p < 0.01,
lever presses by the DA-NR1-KO mice during nondevalued versus devalued
tests; **p < 0.05, lever presses during devalued test by DA-NR1-KO versus by
control.
with the previously learned spatial relationship and, thus, was
predicted to inhibit new learning. As in Figure 6C, the mutants
showed significantly less success (turning ‘‘right’’ or into the
‘‘north’’ arm) (c2 [3, n = 42] = 11.667; p = 0.0006), whereas
no difference was found (c2 [3, n = 42] = 0.73; p = 0.694) among
the three control groups. This supported the notion that mutant
mice failed to learn the habit strategy, even after the extensive
training.

A Figure 6. Habit Learning Analyzed Using Plus
B C
D E
Spatial Navigational Habit Learning, but Not Spatial
Memory, Was Impaired in the Negatively Reinforced
Plus Maze
Because many studies suggested that dopamine is important
for reward pathways, we asked whether habit-learning deficits
seen in the DA-NR1-KO mice hinged on the nature of the rein-
forcement. The aforementioned experiments were replicated in
a water-based plus maze, in which the sole escape from the
water was for mice to locate and climb onto a hidden platform
in the end of one arm. This water-based plus maze behavior
was driven by the desire to escape from the negative environ-
ment and offered an additional opportunity to compare with
habit learning based on positive reinforcement such as the
seeking of a food reward. All parameters such as maze dimen-
sions, cues used, starting and target locations, number of trials
per day, and numbers of days in training remained the same as
those in the previous food-rewarded experiments (Figure 6A).
The first probe trial revealed no significant differences
between any two of the four genotypes (c2 [3, n = 43] = 0.346;
1060 Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Neuron
NMDAR Engages DA Neurons in Habit Learning
Maze
(A) Training and testing schedule for both food reward-
based and water-based plus maze.
(B and C) Plus maze positively reinforced with food
reward.
(D and E) Plus maze negatively reinforced with water.
(B and D) Probe trials on days 6 and 17. Significant
differences were found between days 17 and 6 in the
control mice. Significant differences were also found on
day 17 between the mutant and control groups.
(C and E) Rotation test after 2 days of ‘‘relearn after 90
rotation’’ training. Significant differences were found
between the mutant and control groups. (See text for all
the statistics.) *p < 0.01, mutant versus the controls on day
17 and day 19; **p < 0.01, day 17 versus day 6 in the
control groups.
p = 0.951). The second probe trial showed that
over 80% of the control mice had adopted the
‘‘habit’’ strategy, whereas the mutant mice re-
mained strongly ‘‘spatial’’ (Figure 6D). No differ-
ences were found among the three control
groups (c2 [2, n = 29] = 0.499; p = 0.779). As
a group, the control mice opted for the ‘‘habit’’
strategy significantly more on day 17 than
on day 6 (c2 [1, n = 29] = 22.587; p =
0.00000201). A significantly lower percentage
of DA-NR1-KO mice opted to ‘‘turn right’’
(7.14% versus 80% in the control mice; c2 [1,
n = 43] = 20.904; p = 0.00000483). The deficits
in habit learning were further confirmed in the
rotation test given after 2 days of the ‘‘relearn
after 90 rotation’’ challenge task (Training II,
Figure 6A). A significantly smaller proportion
of the mutant mice (28.6%) in contrast to 80%
of the controls were able to successfully locate
the new platform position (one-tailed proba-
bility = 0.000388, Fisher’s exact test). These
data thus agreed with the findings from the
above food-rewarded tasks suggesting that the learning deficits
were unlikely contingent on the types of reinforcement employed
in the training process.
Due to the significant involvement of spatial learning in the plus
maze task, mice were tested in a spatial version of the plus maze
(Figure 7A). They were trained six trials per day for 6 days to find
a hidden platform in the water-filled plus maze. With all four arms
open, starting points switched between trials in each day
rotating among the distal ends of three arms that did not contain
the platform, following a semi-random order. The platform loca-
tion remained fixed throughout. A probe test was given on day
10, 3 days after the training session ended. During the test,
with the platform removed, mice were released to the center of
the maze and allowed to search for 60 s. Durations spent by
each mouse in each arm were recorded (Figure 7B). Mice from
all four groups spent significantly more time searching in the
target arm (mutants, F(3,32) = 101.292, p < 0.001; Cre, fNR1/+,
F(3,28) = 134.996, p < 0.001; Cre, F(3,36) = 147.806, p < 0.001;
wild-type, F(3, 36) = 294.358, p < 0.001; Newman-Keuls post
 Neuron
NMDAR Engages DA Neurons in Habit Learning
A B
Figure 7. Spatial Memory Test Using Plus Maze
(A) Training and testing paradigms.
(B) Probe trial test. All mice spent more time searching in the target arm. No difference was detected between the mutant and control groups. *p < 0.01, time spent
in the target arm versus in the other three arms. Error bars represent SEM. (See text for all the statistics.)
hoc comparison [the target arm compared to all the other arms],
p < 0.01 for all genotypes). No differences were found between
the mutant and any control groups, suggesting that spatial
learning abilities were unlikely a factor causing the habit-learning
deficits observed in the DA-NR1-KO mice.
Habit Learning in a Nonspatial Zigzag Maze-Based Habit
Task Was Impaired
Instead of compromising habit learning per se, DA-specific NR1
deletion could have skewed the competition between ‘‘spatial’’
and ‘‘habit’’ memory systems in the plus maze task. In order to
investigate this possibility, we designed a nonspatial ‘‘zigzag
maze’’ task as a more direct measurement of habit learning. As
shown in Figure 8A, the water-filled zigzag maze consisted of
eight arms similar in length. Mice were trained to escape onto
a hidden platform. Six different starting points were chosen,
each paired with its own location of the hidden platform. The plat-
form locations were chosen so that they would be reached after
two consecutive right turns from the start point. All mice were
trained 12 trials per day for 10 days. To facilitate developing
the turning habits, some arms were blocked (red lines) so that
mice were only allowed the correct turn at each intersection. A
probe test was given on day 11 in which mice were placed at
a random start location. Some arms in the maze remained
blocked (red lines), but unlike in training, mice were allowed to
choose between turning ‘‘left’’ or ‘‘right’’ at two intersections
(Figure 8A). Mice were scored for whether they finished the two
consecutive right turns (counted as ‘‘successful’’). No differ-
ences were found among the three control genotypes (all
between 90% and 100%, c2 [2, n = 29] = 1.968; p = 0.374) (Fig-
ure 8B), and they were pooled. The conditional knockout mice
showed a significantly lower successful rate in making the two
consecutive right turns (one-tailed probability = 0.000196,
Fisher’s exact test), again suggesting that the DA-NR1-KO
mice are defective in developing the navigation habit.
DISCUSSION
Here, we studied mutant mice with DA neuron-selective NR1
deletion using a set of behavioral tasks as well as in vivo neural-
Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc. 1061
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
recording techniques. Behavioral analysis revealed that the DA-
NR1-KO mice were impaired in several forms of habit learning.
In an operant task where both the mutant and control mice
learned a goal-directed action in the initial training, extensive
training shifted, but only in the control mice, the learned action
from goal directed to habitual. In the mutant mice, this action
remained goal directed and, thus, sensitive to reward devalua-
tion. Similarly, in plus maze tasks, whereas both mutants and
the controls learned to navigate based on spatial cues in initial
training, extensive training shifted navigation from spatial into
habitual also only in the controls, while the mutants’ navigation re-
mained spatially oriented. Such deficits in habit learning were
observed in both positively reinforced and negatively reinforced
tasks. This is consistent with our recent recordings showing
that DA neurons employ a convergent encoding strategy for
processing both positive and negative values (Wang and Tsien,
2011). One notable finding of those in vivo recording experiments
was that some DA neurons exhibit a stimulus-suppression-then-
rebound-excitation type firing pattern in response to negative
experiences (Wang and Tsien, 2011). This offset-rebound excita-
tion may encode information reflecting not only a relief at the
termination of such fearful events but, perhaps, provide some
sort of motivational signals (e.g., motivation to escape).
Therefore, our data strongly suggested that NMDAR functions
in DA neuron be essential for habit learning. A previous study by
Zweifel et al. (2009) reported that the DA neuronal-selective NR1
KO mice were impaired in learning a water maze task and also
impaired in learning a conditioned response in an appetitive T
maze task, seemingly in disagreement with our results of normal
spatial learning and goal-directed learning. The experimental
conditions used in their studies were, however, quite different
from those in ours. The water maze deficit was transient and
detectable only during the very early part (day 2 in a 5 day
session) of their training sessions. The T maze was a goal-
directed paradigm that likely also involved mice learning context
association between landmarks and rewards. Additionally, the
action-reward contingency was also different than that in the
operant paradigm that we used. It is very likely that factors
such as task difficulties, amount of training, cue saliencies,
temporal and spatial contingencies between the CS, and the

A Figure 8. Habit-Learning Test Using Zigzag Maze
B lighted the importance of glutamatergic modula-
rewards can affect the type and amount of involvement by DA
neurons. Using in vivo neural recordings, we observed that
although the response to cue-reward association is much atten-
uated in DA-NR1-KO neurons in term of both response peak
amplitude and duration, these DA neurons, nonetheless, still
could form the cue-reward association. Interaction between
the blunted responsiveness of DA and test conditions may leave
some goal-directed learning impaired by the NR1 deletion,
whereas spare some others. A good example is that in a report
published later by the same group, the authors reported that
the NR1 KO mice were normal in a goal-directed learning para-
digm (Parker et al., 2010). In our study the test conditions and
amount of trainings we used allowed the controls as well as
1062 Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Neuron
NMDAR Engages DA Neurons in Habit Learning
(A) Training and testing paradigms. Red lines indicate
blocked entrances, arrows starting points, and black
circles target locations.
(B) Turning test. Mutant mice showed a significantly
reduced rate of success versus the controls. *p < 0.001,
mutant versus control groups. (See text for all the statis-
tics.)
the mutants to learn goal-directed and spatial
learning normally and indistinguishably. After
extensive training under these conditions, the
mutant mice could not develop the habit
learning, whereas the controls clearly did.
Dopamine is an important modulator for the
habit learning (Wickens et al., 2007; Yin and
Knowlton, 2006). Most of the current under-
standing of its involvement in the habit learning
has so far centered on the downstream path-
ways and structures such as the dorsal striatum
and more recently the PFC (Wickens et al., 2007;
Yin and Knowlton, 2006). Our finding here high-
tions of the DA neuron circuitry itself, in this case
mediated by NMDARs, and suggested that this
upstream pathway should be considered an
integral part of the habit-learning networks.
With perception of the environmental stimuli
likely carried out by glutamatergic signals, it is
conceivable that NMDARs in dopaminergic
neurons participate in the controlling and fine-
tuning of dopaminergic neuron activity patterns
during habit formation. An important part of
this regulation is perhaps to create the cue-rein-
forcement association at an appropriate level in
terms of response robustness and overall DA
neuron network patterns so that DA neurons
would respond accordingly to procedures and
cues with higher incentive salience. NMDARs
are required in mediating synaptic plasticity in
glutamatergic synapses onto DA neurons (Bonci
and Malenka, 1999). Our results showed that
modulation by NMDARs facilitates bursting of
DA neurons toward the learned reward-predict-
ing cues. It is conceivable that the function of NMDARs in regu-
lating phasic firing may be closely linked to its roles in regulating
synaptic plasticity. In fact, studies have shown that enhanced
synaptic strength onto dopamine neurons may act to facilitate
their phasic firing (Stuber et al., 2008).
The blunting of the phasic firing of DA neuron in the mutant
mice can contribute or even result in the habit-learning deficits.
There are several brain regions involved in habit learning that
can be affected by this blunting. The most intuitive one is the
striatum. Dopamine signaling has been postulated as the mech-
anism that trains the striatum, which in turn trains the cortex to
establish the appropriate sensorimotor associations required
for developing habits (Ashby et al., 2010; Wickens et al., 2007).
 Neuron
NMDAR Engages DA Neurons in Habit Learning
Dopamine modulates the plasticity in the corticostriatal
synapses, facilitating induction of LTP in conditions that would
otherwise induce LTD. This facilitation requires dopamine D1
receptor (Calabresi et al., 2000). The low affinity of D1 receptors
toward dopamine coupled with the fast dopamine reuptake
(Cragg et al., 1997) in the striatum likely makes the dopamine
modulation sensitive to the blunting of phasic release. From
a network point of view, it has been reported that dopamine
can cause changes in the coordinated activity of neuronal
ensembles in corticostriatal circuits and by doing so ‘‘gate’’ the
inputs in those downstream regions (Costa et al., 2006). Thus,
when dopamine level is low, such as when bursting activities
are insufficient, it fails to produce and reinforce these networks’
connectivity underlying habit formation. Other than the striatum,
reduced bursting of DA neurons may also affect activities of
structures such as the PFC of which lesion of the medial infralim-
bic area was reported to impair expression of a learned habit
(Coutureau and Killcross, 2003). Studies have shown that tonic
dopamine concentration in the prefrontal area, likely due to the
relatively slower dopamine reuptake (Seamans and Yang,
2004), may be affected by previous phasic dopamine release
(Matsuda et al., 2006). The presence of background dopamine
signal converts LTD to potentiation. This ‘‘priming’’ requires
time to develop and requires D1 and D2 receptors, both of which
have low affinity to dopamine. It is very likely that this phasic
release-induced ‘‘priming’’ could also be affected by the amount
of DA neurons bursting, thus, by blunting of DA response. It will
be of great interest to dissect the various roles of those different
brain regions in habit formation in future studies.
It is also important for future research to further analyze the
contributions of NMDARs within different dopamine subpopula-
tions, and temporally within different phases of habit learning.
The potential subregional circuitry within the DA neuron popula-
tions in the VTA and SNr regions can be highly crucial for inte-
grating distinct cortical and subcortical inputs (Grace et al.,
2007; Lammel et al., 2011; Lisman and Grace, 2005). Thus, it is
conceivable that additional subregional-specific manipulations
and analyses could further elucidate how the glutamatergic
regulation of DA neurons, as revealed by our current study,
modulates habit formation.
In summary our study has provided several important insights
about NMDAR in DA neurons and habit learning. First, NMDARs
in DA neurons are required for learning habits, including appetitive
lever pressing and spatial navigational habits. Second, the
dependence of habit learning on NMDARs in DA neurons was
observed in both positively and negatively reinforced trainings.
Third, DA neurons lacking the NMDARs can still form the cue-
reward association but with greatly reduced phasic activity as
well as conditioned response robustness. Taken together, our
results suggest that the NMDARs in DA neurons are an important
modulator of DA neurons’ response robustness in cue-reward
association and an essential element underpinning habit learning.
EXPERIMENTAL PROCEDURES
Subjects
Mice carrying alleles of NMDAR1 flanked by loxP sites (fNR1) were bred with
Slc63a Cre transgenic mice. Offspring were genotyped by PCR for both the
Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc. 1063
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Cre transgene and for the floxed NMDAR1 (fNR1) locus. Mice used in these
experiments have been bred for at least five generations onto the C57/BL6
background. Animals were maintained on a 12 hr light/dark cycle in the Geor-
gia Health Sciences University animal care facility. Except for when specified
in experiment, such as when food pellets were used as rewards, food and
water were given ad libitum. All procedures relating to animal care and treat-
ment conform to the Institutional and NIH guidelines. For behavioral tests in
the study, we used male mice around 1 year old in age. These animals have
been prescreened to make sure that they have normal vision and hearing
capacity.
Immunohistochemistry
Mice were perfused transcardially with 4% paraformaldehyde (PFA) in
13 PBS followed by a postfixation in 4% PFA overnight. Coronal sections
(50 mm thick) were cut on a vibratome and collected in 0.5% PFA in
13 PBS and stored at 4 C before use. For double-immunofluorescent staining
of b-galactosidase and TH, sections were incubated at 4 C overnight with
gentle shaking in primary antibody (anti-b-galactosidase [pAb] 1/5,000, Invi-
trogen; anti-TH [monoclonal antibody] 1/1,000) in a buffer containing 0.05%
Tween 20, 10% normal goat serum, and 13 PBS following preincubation in
10% normal goat serum and 13 PBS at room temperature for 2 hr. The
sections were then incubated with Alexa-conjugated secondary antibodies
(1/200; Invitrogen) at room temperature for 2 hr. b-Galactosidase IR was
visualized by Alexa 568 and TH IR by Alexa 488. A similar procedure was
employed to double stain NMDAR1 and TH except that anti-NR1 (polyclonal
antibody 1:100; Chemicon, Temecula, CA, USA) was used as the primary anti-
body for NMDAR1. The sections were incubated with Alexa-conjugated
secondary antibodies (1/200; Invitrogen) at room temperature for 2 hr.
NMDAR1 was visualized by Alexa 488 and TH IR by Alexa 594. Fluorescent
images were captured with a confocal laser-scanning microscope and an
epifluorescence microscope.
Elevated Plus Maze
This apparatus consists of a center platform (5 3 5 cm) 37 cm off the ground
with four branching arms (30 cm long and 5 cm wide). Two of the four arms are
open, and the other two arms are enclosed by black walls (20 cm high). Testing
was performed during light phase in a dimly lit room (50 lux). Animals were
placed on the center platform and scored for arm entries and time spent in
each arm. Percentages of time animals spent in the open arms were calculated
as the final readout of anxiety. Unpaired t tests were used to compare the
significance between the different genotypes.
Rotarod
RotaRod analysis was performed using the mouse version of ROTA-ROD
manufactured by San Diego Instruments (San Diego, CA, USA). Mice were
trained by allowing them to run on a rotarod rotating at 30 rpm for a total
time span of 5 min. (Time counting was stopped when mice dropped until
they were put back onto the rotarod again.) During the tests mice were again
placed on top of the rotarod, which rotated at 30 rpm. Durations of each mouse
that stayed on the rotarod (latency to fall) were recorded. Any mice remaining
on the apparatus 300 s after the starts were removed, and the time was scored
as 300 s. Unpaired t tests were used to compare the significance between the
latencies in different genotypes.
Open-Field Activity Test
Locomotor activity was measured by scoring beam breaks in activity cham-
bers (San Diego Instruments). Prior to open-field tests, animals were handled
for 2 consecutive days. Standard rat cages were used as the novel open field
for the mice tested. Locomotor activities were recorded for 1 hr and scored for
both 5 min and 1 hr. Unpaired t tests were used to compare the significance in
fine movements, ambulatory movements, and rearing between the different
genotypes.
Instrumental Training
Mice were placed on a food-deprivation schedule to reduce their weight to
80%–85% of their baseline weight. They were fed for 2 hr with mouse chow

in their home cages each day after training. Water was available at all times in
the home cages.
Training and testing took place in eight Med Associates operant chambers
(21.6 cm length 3 17.8 cm width 3 12.7 cm height) housed in boxes with
sound-attenuating walls. Each chamber was equipped with a food magazine,
two retractable levers, one on each side of the magazine, and a 3 W, 24V house
light mounted on the same wall, but above the food magazine. Bio-Serv 20 mg
pellets from a dispenser into the magazine were used as reward. The software
Med-PC-IV from Med Associates was used for equipment control and
behavior recording.
Lever-Press Training
At the beginning of each session, the house light was turned on and the
lever inserted. At the end of each session, the light was turned off and
the lever retracted. Mice were trained in an initial lever-press training con-
sisting of 4 consecutive days of CRF, during which the mice received
a pellet for each lever press. A session would end after 60 min or after
the mouse had collected 30 rewards, whichever came first. After CRF,
mice were trained with RI schedules to generate habitual lever pressing
(Dickinson et al., 1983). The training started with 2 days on RI 30 s, with
a 0.1 probability of reward availability every 3 s contingent on lever press,
and followed by 6 days on the 60 s interval schedule, with a 0.1 probability
of reward availability every 6 s contingent on lever pressing. Repeated-
measures ANOVA was used to compare lever press between the different
genotypes.
Devaluation Tests
A specific satiety procedure was used for outcome devaluation. Mice were
given unlimited access within a fixed duration to either the mouse chow to
which they had been exposed in their home cages (nondevalued condition/
control), or the purified pellets they normally earned during lever-press
sessions (devalued condition). The mouse chow served as a control for overall
level of satiety. This procedure controls the overall level of satiety and motiva-
tional state, while altering the current value of a specific reward. Immediately
after 1 hr of unlimited exposure to the pellets or chow, the mice were subjected
to a 5 min long probe test. During the probe test the lever was inserted, but no
pellet would be delivered in response to lever pressing. This brief extinction
test was designed to test whether the acquired lever pressing of the mice
was controlled by the action-outcome instrumental contingency or habit
(e.g., in response to a antecedent stimuli). On the second day of outcome
devaluation, the same procedure was used, except that those animals that
received mouse chow on day 1 received pellets on day 2, and vice versa.
When grouping, mice were counterbalanced between genotypes and
treatment. Repeated-measures ANOVA and unpaired Student’s t test were
used to compare lever press between the different genotypes as specified in
the text.
Plus Maze
The maze consisted of four arms measuring 35 cm long, 6 cm wide, and 35 cm
deep, with transparent high walls made of clear Plexiglas. For training posi-
tively reinforced with food pellets (20 mg per pellet), animals were maintained
at 80%–75% of their free-feeding weight throughout the experiment. For
training negatively reinforced with water, water was stained opaque and white
with titanium dioxide. A hidden platform was placed 1 inch under the water
surface. The training and testing were as described in the text. For plus
maze assays, littermates in Slc6a3+/Cre, fNR1/+ (control), Slc6a3+/Cre (Cre
control), and wild-type genotypes were chosen as three control groups.
Turning of mice in different tests was compared using chi-square tests, as
specified in the text, to evaluate the performance of mice from different geno-
types. Additionally, repeated-measures ANOVA and unpaired Student’s t test,
as specified in the text, were used to compare time spent in different arms
among mice from the different genotypes.
Zigzag Maze
The shape of the zigzag maze is illustrated in Figure 8A. Each arm measures
about 30 cm long, 6 cm wide, and 35 cm deep. The maze was filled with water
that was stained opaque and white with titanium dioxide. A hidden platform
1064 Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Neuron
NMDAR Engages DA Neurons in Habit Learning
was placed at a designed location 1 inch under the water surface. Training
and tests were done as described in the text. The chi-square test and Fisher’s
exact test were used to compare the performance of mice from different
genotypes.
Surgeries
A 32-channel (a bundle of 8 tetrodes), ultralight (weight <1 g), movable
(screw-driven) electrode array was constructed similar to that described
previously (Lin et al., 2006; Wang and Tsien, 2011). Each tetrode consisted
of four 13 mm diameter Fe-Ni-Cr wires (Stablohm 675, California Fine Wire;
with impedances of typically 2–4 MU for each wire) or 17 mm diameter Plat-
inum wires (90% Platinum 10% Iridium, California Fine Wire; with imped-
ances of typically 1–2 MU for each wire). One week before surgery, mice
(3–6 months old) were removed from the standard cage and housed in
customized home cages (40 3 20 3 25 cm). On the day of surgery, mice
were anesthetized with ketamine/xylazine (80/12 mg/kg, i.p.); the electrode
array was then implanted toward the VTA in the right hemisphere (3.4 mm
posterior to bregma, 0.5 mm lateral and 3.8–4.0 mm ventral to the brain
surface) and secured with dental cement.
Tetrode Recording and Unit Isolation
Two or three days after surgery, electrodes were screened daily for neural
activity. If no dopamine neurons were detected, the electrode array was
advanced 40–100 mm daily, until we could record from a putative dopamine
neuron. Multichannel extracellular recording was similar to that described
previously (Lin et al., 2006; Wang and Tsien, 2011). In brief, spikes (filtered
at 250–8000 Hz; digitized at 40 kHz) were recorded during the whole experi-
mental process using the Plexon Multichannel Acquisition Processor System.
Mice behaviors were simultaneously recorded using the Plexon CinePlex
tracking system. Recorded spikes were isolated using the Plexon Offline
Sorter software: multiple spike sorting parameters (e.g., principle component
analysis, energy analysis) were used for the best isolation of the tetrode-re-
corded spike waveforms. Combining the stability of multi-tetrode recording
and multiple unit-isolation techniques available in Offline Sorter (e.g., principle
component analysis, energy analysis), individual VTA neurons can be studied
in great detail, in many cases for days.
Reward Conditioning of DA Neurons
Mice were slightly food restricted before reward association training. In
reward conditioning, mice were placed in the reward chamber (45 cm in
diameter, 40 cm in height). Mice were trained to a tone (5 kHz, 1 s) with
subsequent sugar pellet delivery for at least 3 days (40 trials per day;
with an interval of 1–2 min between trials). The tone was generated by the
A12-33 audio signal generator (5 ms shaped rise and fall; about 80 dB at
the center of the chamber) (Coulbourn Instruments). A sugar pellet (14 mg)
was delivered by a food dispenser (ENV-203-14P; Med. Associates) and
dropped into one of two receptacles (12 3 7 3 3 cm) at the termination of
the tone (the other receptacle was used as control, where a sugar pellet
was never received).
Analysis of In Vivo Recording Data
Sorted neural spikes were processed and analyzed in NeuroExplorer (Nex
Technologies) and MATLAB. Dopamine neurons were classified based on
the following three criteria.
(1) Low baseline firing rate (0.5–10 Hz).
(2) Relatively long ISI (all the classified putative dopamine neurons are with
ISIs >4 ms within a R99.8% confidence level). The shortest ISI we re-
corded was 4.1 ms under any conditions in our experiment (only well-
isolated units with amplitude R0.4mV were used for calculation of the
shortest ISI). The averaged shortest ISI was 6.8 ± 2.2 ms (mean ± SD;
n = 36). In contrast the ISI for nondopamine neurons can be as short as
1.1 ms.
(3) Regular firing pattern when mice were freely behaving (fluctuation
<3 Hz). Here, fluctuation represents the SD of the firing rate histogram
bar values (bin = 1 s; recorded for at least 600 s).
 Neuron
NMDAR Engages DA Neurons in Habit Learning
SUPPLEMENTAL INFORMATION
Supplemental Information includes one figure and Supplemental Experimental
Procedures and can be found with this article online at doi:10.1016/j.neuron.
2011.10.019.
ACKNOWLEDGMENTS
We thank Fengying Huang and Brianna Klein for technical assistance. This
work was supported by funds from NIMH, NIA, and Georgia Research Alliance
(all to J.Z.T.). L.P.W., F.L., X.S., and J.Z.T. conceived and designed the exper-
iments. L.P.W., F.L., D.W., K.X., and D.H.W. performed the experiments.
L.P.W., F.L., D.W., K.X., and J.Z.T. analyzed the data. L.P.W., F.L., X.S., and
J.Z.T. contributed reagents/materials/analysis tools. L.P.W., F.L., K.X., and
J.Z.T. wrote the paper. All experiments involving animals were performed
according to procedures approved by the Georgia Health Sciences University
IACUC review board.
Accepted: October 10, 2011
Published: December 21, 2011
REFERENCES
Ashby, F.G., Turner, B.O., and Horvitz, J.C. (2010). Cortical and basal ganglia
contributions to habit learning and automaticity. Trends Cogn. Sci. (Regul. Ed.)
14, 208–215.
Bonci, A., and Malenka, R.C. (1999). Properties and plasticity of excitatory
synapses on dopaminergic and GABAergic cells in the ventral tegmental
area. J. Neurosci. 19, 3723–3730.
Calabresi, P., Gubellini, P., Centonze, D., Picconi, B., Bernardi, G., Chergui, K.,
Svenningsson, P., Fienberg, A.A., and Greengard, P. (2000). Dopamine and
cAMP-regulated phosphoprotein 32 kDa controls both striatal long-term
depression and long-term potentiation, opposing forms of synaptic plasticity.
J. Neurosci. 20, 8443–8451.
Costa, R.M., Lin, S.C., Sotnikova, T.D., Cyr, M., Gainetdinov, R.R., Caron,
M.G., and Nicolelis, M.A. (2006). Rapid alterations in corticostriatal ensemble
coordination during acute dopamine-dependent motor dysfunction. Neuron
52, 359–369.
Coutureau, E., and Killcross, S. (2003). Inactivation of the infralimbic prefrontal
cortex reinstates goal-directed responding in overtrained rats. Behav. Brain
Res. 146, 167–174.
Cragg, S., Rice, M.E., and Greenfield, S.A. (1997). Heterogeneity of electrically
evoked dopamine release and reuptake in substantia nigra, ventral tegmental
area, and striatum. J. Neurophysiol. 77, 863–873.
Devan, B.D., and White, N.M. (1999). Parallel information processing in the
dorsal striatum: relation to hippocampal function. J. Neurosci. 19, 2789–2798.
Dickinson, A., Nicholas, D.J., and Adams, C.D. (1983). The effect of the instru-
mental training contingency on susceptibility to reinforcer devaluation. Q. J.
Exp. Psychol. B 35B, 35–51.
Engblom, D., Bilbao, A., Sanchis-Segura, C., Dahan, L., Perreau-Lenz, S.,
Balland, B., Parkitna, J.R., Luján, R., Halbout, B., Mameli, M., et al. (2008).
Glutamate receptors on dopamine neurons control the persistence of cocaine
seeking. Neuron 59, 497–508.
Fama, R., Sullivan, E.V., Shear, P.K., Stein, M., Yesavage, J.A., Tinklenberg,
J.R., and Pfefferbaum, A. (2000). Extent, pattern, and correlates of remote
memory impairment in Alzheimer’s disease and Parkinson’s disease.
Neuropsychology 14, 265–276.
Faure, A., Haberland, U., Condé, F., and El Massioui, N. (2005). Lesion to the
nigrostriatal dopamine system disrupts stimulus-response habit formation.
J. Neurosci. 25, 2771–2780.
Faure, A., Leblanc-Veyrac, P., and El Massioui, N. (2010). Dopamine agonists
increase perseverative instrumental responses but do not restore habit forma-
tion in a rat model of Parkinsonism. Neuroscience 168, 477–486.
Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc. 1065
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Grace, A.A., Floresco, S.B., Goto, Y., and Lodge, D.J. (2007). Regulation of
firing of dopaminergic neurons and control of goal-directed behaviors.
Trends Neurosci. 30, 220–227.
Kauer, J.A., and Malenka, R.C. (2007). Synaptic plasticity and addiction. Nat.
Rev. Neurosci. 8, 844–858.
Knowlton, B.J., Mangels, J.A., and Squire, L.R. (1996). A neostriatal habit
learning system in humans. Science 273, 1399–1402.
Lammel, S., Ion, D.I., Roeper, J., and Malenka, R.C. (2011). Projection-specific
modulation of dopamine neuron synapses by aversive and rewarding stimuli.
Neuron 70, 855–862.
Lin, L., Chen, G., Xie, K., Zaia, K.A., Zhang, S., and Tsien, J.Z. (2006). Large-
scale neural ensemble recording in the brains of freely behaving mice.
J. Neurosci. Methods 155, 28–38.
Lisman, J.E., and Grace, A.A. (2005). The hippocampal-VTA loop:
controlling the entry of information into long-term memory. Neuron 46,
703–713.
Matsuda, Y., Marzo, A., and Otani, S. (2006). The presence of background
dopamine signal converts long-term synaptic depression to potentiation in
rat prefrontal cortex. J. Neurosci. 26, 4803–4810.
Overton, P., and Clark, D. (1992). Iontophoretically administered drugs acting
at the N-methyl-D-aspartate receptor modulate burst firing in A9 dopamine
neurons in the rat. Synapse 10, 131–140.
Overton, P.G., Richards, C.D., Berry, M.S., and Clark, D. (1999). Long-term
potentiation at excitatory amino acid synapses on midbrain dopamine
neurons. Neuroreport 10, 221–226.
Packard, M.G. (1999). Glutamate infused posttraining into the hippocampus or
caudate-putamen differentially strengthens place and response learning.
Proc. Natl. Acad. Sci. USA 96, 12881–12886.
Packard, M.G., and McGaugh, J.L. (1996). Inactivation of hippocampus or
caudate nucleus with lidocaine differentially affects expression of place and
response learning. Neurobiol. Learn. Mem. 65, 65–72.
Packard, M.G., Hirsh, R., and White, N.M. (1989). Differential effects of fornix
and caudate nucleus lesions on two radial maze tasks: evidence for multiple
memory systems. J. Neurosci. 9, 1465–1472.
Parker, J.G., Zweifel, L.S., Clark, J.J., Evans, S.B., Phillips, P.E., and Palmiter,
R.D. (2010). Absence of NMDA receptors in dopamine neurons attenuates
dopamine release but not conditioned approach during Pavlovian condi-
tioning. Proc. Natl. Acad. Sci. USA 107, 13491–13496.
Saal, D., Dong, Y., Bonci, A., and Malenka, R.C. (2003). Drugs of abuse and
stress trigger a common synaptic adaptation in dopamine neurons. Neuron
37, 577–582.
Schultz, W. (1998). Predictive reward signal of dopamine neurons.
J. Neurophysiol. 80, 1–27.
Seamans, J.K., and Yang, C.R. (2004). The principal features and mecha-
nisms of dopamine modulation in the prefrontal cortex. Prog. Neurobiol. 74,
1–58.
Stuber, G.D., Klanker, M., de Ridder, B., Bowers, M.S., Joosten, R.N.,
Feenstra, M.G., and Bonci, A. (2008). Reward-predictive cues enhance excit-
atory synaptic strength onto midbrain dopamine neurons. Science 321, 1690–
1692.
Tsien, J.Z., Huerta, P.T., and Tonegawa, S. (1996). The essential role of hippo-
campal CA1 NMDA receptor-dependent synaptic plasticity in spatial memory.
Cell 87, 1327–1338.
Ungless, M.A., Whistler, J.L., Malenka, R.C., and Bonci, A. (2001). Single
cocaine exposure in vivo induces long-term potentiation in dopamine neurons.
Nature 411, 583–587.
Wang, D.V., and Tsien, J.Z. (2011). Convergent processing of both positive
and negative motivational signals by the VTA dopamine neuronal populations.
PLoS One 6, e17047.
Wang, L.P., Li, F., Shen, X., and Tsien, J.Z. (2010). Conditional knockout of
NMDA receptors in dopamine neurons prevents nicotine-conditioned place
preference. PLoS One 5, e8616.

Wickens, J.R., Horvitz, J.C., Costa, R.M., and Killcross, S. (2007).
Dopaminergic mechanisms in actions and habits. J. Neurosci. 27, 8181–
8183.
Yin, H.H., and Knowlton, B.J. (2006). The role of the basal ganglia in habit
formation. Nat. Rev. Neurosci. 7, 464–476.
Zhuang, X., Masson, J., Gingrich, J.A., Rayport, S., and Hen, R. (2005).
Targeted gene expression in dopamine and serotonin neurons of the mouse
brain. J. Neurosci. Methods 143, 27–32.
1066 Neuron 72, 1055–1066, December 22, 2011 ª2011 Elsevier Inc.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Neuron
NMDAR Engages DA Neurons in Habit Learning
Zweifel, L.S., Argilli, E., Bonci, A., and Palmiter, R.D. (2008). Role of NMDA
receptors in dopamine neurons for plasticity and addictive behaviors.
Neuron 59, 486–496.
Zweifel, L.S., Parker, J.G., Lobb, C.J., Rainwater, A., Wall, V.Z., Fadok, J.P.,
Darvas, M., Kim, M.J., Mizumori, S.J., Paladini, C.A., et al. (2009). Disruption
of NMDAR-dependent burst firing by dopamine neurons provides selective
assessment of phasic dopamine-dependent behavior. Proc. Natl. Acad. Sci.
USA 106, 7281–7288.
 Personality and Social Psychology
Bulletin http://psp.sagepub.com/

The Pull of the Past: When Do Habits Persist Despite Conflict With Motives?
David T. Neal, Wendy Wood, Mengju Wu and David Kurlander
Pers Soc Psychol Bull 2011 37: 1428 originally published online 22 August 2011
DOI: 10.1177/0146167211419863

The online version of this article can be found at:

http://psp.sagepub.com/content/37/11/1428

Published by:

http://www.sagepublications.com

On behalf of:

Society for Personality and Social Psychology

Additional services and information for Personality and Social Psychology Bulletin can be found at:

Email Alerts: http://psp.sagepub.com/cgi/alerts

Subscriptions: http://psp.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://psp.sagepub.com/content/37/11/1428.refs.html

>> Version of Record - Sep 19, 2011

OnlineFirst Version of Record - Aug 22, 2011

What is This?

Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 This version has been modified from the original version, which was included in the print issue of
PSPB 37:11 (November 2011), to correct Figure 2.

In the print version, Figure 2 included an incorrect legend within the second graph. On p. 1434 of this
corrected version, the incorrect legend has been removed from Figure 2.

Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

419863
et al.Personality and Social Psychology Bulletin
PSPXXX10.1177/0146167211419863Neal

Article
Personality and Social Psychology Bulletin

The Pull of the Past: When Do Habits 37(11) 1428–1437

© 2011 by the Society for
Personality and Social Psychology, Inc
Persist Despite Conflict With Motives? Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0146167211419863
http://pspb.sagepub.com

David T. Neal1, Wendy Wood1, Mengju Wu2,

and David Kurlander3

Abstract
To identify the factors that disrupt and maintain habit performance, two field experiments tested the conditions under which
people eat out of habit, leading them to resist motivational influences. Habitual popcorn eaters at a cinema were minimally
influenced by their hunger or how much they liked the food, and they ate equal amounts of stale and fresh popcorn. Yet,
mechanisms of automaticity influenced habit performance: Participants ate out of habit, regardless of freshness, only when
currently in the context associated with past performance (i.e., a cinema; Study 1) and only when eating in a way that allowed
them to automatically execute the response cued by that context (i.e., eating with their dominant hand; Study 2). Across all
conditions, participants with weaker cinema-popcorn-eating habits ate because of motivations such as liking for the popcorn.
The findings reveal how habits resist conflicting motives and provide insight into promising mechanisms of habit change.

Keywords
attitudes, automatic processes, health, self-regulation

Received August 27, 2010; revision accepted April 10, 2011

Habits develop when people give a response repeatedly in a
particular context and thereby form associations in memory
between the response and recurring context cues. Theories of
automaticity illuminate the cognitive mechanisms behind hab-
its. When people have a strong habit, perception of recurring
context cues activates the response in memory and may addi-
tionally deactivate alternative responses (Danner, Aarts, &
de Vries, 2008; Neal, Labrecque, Wood, & Lally, 2011). For
example, when habitual runners were subliminally primed
with words relating to the contexts in which they ran (e.g.,
gym, forest), the act of running became strongly accessible in
thought (Neal et al., 2011). Suggesting that this context cuing
process may be minimally influenced by motivations, sub-
liminally priming participants’ personal goals for running
did not make running habits more accessible.
Habits initially are likely to form through goal pursuit—
as people repeatedly perform actions that yield desired
outcomes. However, once habits have developed, they are
performed with only limited influence from supporting moti-
vations (Triandis, 1977, 1979). Nonetheless, experimental
tests of how these mechanisms play out with real-world
behaviors have not yet been provided. The present research
provides these tests by independently manipulating motiva-
tional and contextual factors across two field experiments
with a socially significant behavior, habitual eating.
The basic question in our experiments is: What disrupts
habit performance? The design pitted negative attitudes
toward stale food against established habits to eat in a given
context. If people continue to perform habits despite this
substantial conflicting motive, then habits are resistant to
changes in attitudes and goals. Although habits may not be
easily influenced by motives, their dependence on context
cues should make them vulnerable to disruptions in automa-
ticity. That is, habits should not be activated automatically
outside of their typical performance context and should not
be executed automatically when responses are performed in
novel ways (e.g., using the nondominant hand). By manipu-
lating the factors that do and do not control habits, we can
test whether strong habits become responsive to current
motivations when automaticity is disrupted.
Evidence of the Automaticity
Guiding Habitual Behavior
Social psychological experiments have largely focused on
types of automaticity that flexibly accommodate to current
1
University of Southern California, Los Angeles, CA, USA
2
University of California, San Diego, CA, USA
3
Duke University, Durham, NC, USA
Corresponding Authors:
David T. Neal or Wendy Wood, Department of Psychology, University of
Southern California, Los Angeles, CA 90086
Email: david@empiricaresearch.com.au or wendy.wood@usc.edu

Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Neal et al.
goals and attitudes (Kruglanski et al., 2002). Yet, research in
neuroscience, animal learning, and correlational studies of
everyday behavior has focused primarily on cued actions
that are not influenced by current motives (reviewed in
Wood & Neal, 2007, 2009).
In neuroscience research on the brain systems activated
during behavioral tasks, performance of stimulus–response
habits is localized in the basal ganglia, especially the puta-
men, whereas control of goal-directed actions is localized in
other brain regions, often involving the prefrontal cortex
(Balleine & O’Doherty, 2010; Yin & Knowlton, 2006). Thus,
habit formation involves neural shifts from “largely evalua-
tion-driven circuits to those engaged in performance”
(Graybiel, 2008, p. 362). Furthermore, the different brain sys-
tems controlling habits and goal-directed behavior often
compete in guiding action such that activity associated with
goal-directed control may be actively suppressed during per-
formance of habitual behavior (Balleine & O’Doherty, 2010;
Poldrack et al., 2001). Animal learning research also differen-
tiates habits from goal-directed actions, primarily in terms of
behavioral reactions to rewards. Specifically, extended train-
ing at a task such as lever pressing for a food reward produces
habitual, continued responding to the lever cue even after the
reward is devalued (e.g., the animal is sated, the food has
been associated with a toxin; Dickinson & Balleine, 1995).
The minimal influence of motivations on habit perfor-
mance also is a central finding of correlational studies of
everyday behavior, including behavior prediction (Triandis,
1977) and habit discontinuity (Verplanken, Walker, Davis,
& Jurasek, 2008). Across a range of everyday behaviors—
from bus travel, to fast food consumption, to exercise—
motivational factors (e.g., attitudes, intentions, self-concept,
attitude accessibility) predicted future performance for
nonhabitual behaviors but had limited predictive power for
strong habits (Ouellette & Wood, 1998). Furthermore,
consistent with the idea that habits are cognitively repre-
sented as context–response associations, habit strength had
these effects only when habits were assessed from frequent
performance in stable contexts—the specific conditions that
enable the formation of habit associations in memory
(Aldrich, Montgomery, & Wood, in press; Danner et al.,
2008). Discontinuity studies that assess naturally occurring
context changes find that participants continue to perform
habits with minimal influence from goals, but only so long as
they continue to live in the same context (Verplanken et al.,
2008; Wood, Tam, & Witt, 2005). When participants move
house to a new location without cues to habit performance,
they apparently are freed to act on their goals.
The findings from neuroscience, animal learning, and
correlational research on everyday behavior suggest that,
once developed, habits are performed with limited influence
from motivations. It may be that strong habits are directly
brought to mind by the context cues (e.g., settings, preceding
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
1429
actions) associated with prior performance. Then, through
ideomotor processes, people may act on the accessible
behavior in mind (Bargh, 1994). According to this view, for
example, a person’s habit to snack on crackers at night may
come to be directly cued by sitting in front of the TV, and
changes in motivational states of hunger and liking for food
may have little influence on how much the person eats.
Attitudes and Goals Flexibly Guide Behavior
The insensitivity of habit performance to current motives
outlined in the preceding research might seem surprising to
social psychologists given the field’s focus on the flexible
operation of attitudes and goals. Although strong goals and
attitudes are known to have enduring influence, they usually
guide responding in a given context depending on the
expected outcomes of an action and the value placed on
those outcomes (Eagly & Chaiken, 1993). Thus, people act
on strong food preferences in a given context depending on
what food is available and how tasty it is. Even automatic
goal pursuit—in which goals are activated and guide respond-
ing outside of awareness—yields “not a static behavioral
response, but an automated strategy for dealing with the
environment” (Bargh & Barndollar, 1996, p. 461, italics in
original). Similarly, automatically activated attitudes often
are malleable and context dependent (Dijksterhuis, Smith,
van Baaren, & Wigboldus, 2005).
The flexible influence of implicit goals and attitudes is
evident in research that opposes implicit against explicit
motives. For example, Macrae and Johnston’s (1998) par-
ticipants responded to implicit priming of a helping goal by
offering assistance to someone in need, except when the
implicit helping goal was inconsistent with their explicit
goal of being on time. In other research, participants carried
out implementation intentions to study—a form of auto-
matic goal pursuit in which people plan to perform a par-
ticular response upon encountering a particular cue—only
when they held the explicit goal of studying (Sheeran,
Webb, & Gollwitzer, 2005). In an experiment related to our
present focus on eating habits, implicit goals relevant to
thirst increased the amount that participants drank during a
beverage taste test, but only when they were thirsty (Strahan,
Spencer, & Zanna, 2002). Nonetheless, conflicts between
explicit and implicit goals are not always resolved in favor
of explicit ones (Shah & Kruglanski, 2002). Instead, these
dispositions combine in dynamic ways to shape responding
in goal pursuit.
Habits contrast with the flexible pattern of responding char-
acteristic of motives. Unlike automatic goals and attitudes,
performance of strong habits may not be dynamically
responsive to current motives. Thus, a strong habit to eat a
particular food in a given context may be insensitive to cur-
rent hunger levels or the palatability of the food.
 1430
The Present Research
To test the factors that can alter habit performance, we con-
ducted two field experiments varying the conditions under
which people consume popcorn at a movie cinema. Some
participants in our research had strong habits to eat popcorn
at the cinema (i.e., a history of frequent popcorn consump-
tion in that setting), whereas others had weaker habits. We
chose to study eating behavior in part because it is a signifi-
cant social problem and in part because eating provides a
strong test of the sensitivity of habits to changes in motiva-
tions. People believe that their eating behavior is largely a
motivated activity in response to the way food tastes (e.g.,
Vartanian, Herman, & Wansink, 2008). Indeed, 17 of 22
participants in a pilot study provided “tastes good” or a close
synonym when asked why they ate popcorn at the cinema.
Thus, in the present experiments, we manipulated attitudes
toward the food through its palatability. For some partici-
pants, the popcorn was fresh, whereas for others it was 7 days
old and stale (adapted from Wansink & Kim, 2005).
To test the effects of changes in motivations, we first
assessed whether participants with strong and weak habits
evaluated the popcorn in the same way. It is possible that
strong-habit participants would report more favorable atti-
tudes even toward stale popcorn. By assessing attitudes
toward the specific popcorn served and not toward popcorn
in general, we evaluated the immediate, proximal attitude
relevant to participants’ eating. We also assessed partici-
pants’ reported hunger and, in Study 2, their perceptions of
social norms. Our second test of the susceptibility of habits
to shifts in motivations involved assessing how much pop-
corn participants ate. Specifically, we evaluated whether
participants with strong habits ate more of the liked, fresh
popcorn than the disliked, stale popcorn and ate more when
they were hungry.
Even if habits are not responsive to changes in motives,
they should be disrupted by challenges to habit cuing. In
Study 1, participants were tested either in the context rele-
vant to their habit (watching movies in a cinema) or in a
novel context (watching music videos in a conference room).
We anticipated that habits would be automatically brought to
mind only in the cinema context. In Study 2, all participants
ate in the habit-related context of a cinema, and we manipu-
lated whether they ate in their typical way (with their domi-
nant hand) or in a novel way (using their nondominant hand).
Using one’s nondominant hand has been shown to impede
the automatic, smooth performance of habitual responses
(Sakai, Kitaguchi, & Hikosaka, 2003), presumably because
it brings the action under intentional control. Thus, eating
habits should be automatically executed only when eating
with the dominant hand.
If habits resist changes in motives but can be disrupted by
changes in automatic cuing, then we anticipate the following
three-way interaction: Participants at the cinema with strong
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Personality and Social Psychology Bulletin 37(11)
popcorn habits should be guided by the action in mind and
eat similar amounts of fresh and stale popcorn. In contrast,
participants with weak or moderate habits should eat more of
the fresh popcorn that they like. Alternatively, when auto-
matic cuing is disrupted by eating in a novel context or eat-
ing in a cinema but in a novel manner, then even strong-habit
participants are likely to fall back on their attitudes and con-
sume more fresh than stale popcorn.
Study 1
Method
Design and participants. Study 1 used a Context (cinema vs.
meeting room) × Food Attitudes (fresh vs. stale) between-
participants design. Habit strength for eating popcorn at the
movies was an additional, continuous predictor. For the
cinema context, 98 participants (57 males) were recruited
immediately before the showing of a regularly scheduled
feature film at a campus cinema. For the meeting room con-
text, 60 participants (35 males) were recruited through flyers
and participated in a campus meeting room near the cinema.
Participants were paid $6 for completing the study.1
Procedure
Participants in the cinema setting were told that the study
examined personality differences in movie interests and that
they would rate a series of movie trailers. Sessions were
conducted in groups of around 15. Before entering the the-
ater, participants rated their current feelings (hunger, time
since last meal, several filler items) and were given a 591-ml
bottle of water and a box of popcorn. Participants were ran-
domly assigned to receive popcorn that was either fresh
(popped 1 hr before the session) or stale (popped 7 days
before the session). Each box was discretely numbered so
that it could be matched to participants’ survey responses.
Boxes were weighed before and after distribution on a digi-
tal scale, with initial amount averaging 61.73 g (SD = 6.74).
Participants entered the theater with their popcorn and
water and, to reduce potential social influence, sat as far as
possible from other participants. The lights immediately
dimmed, and a series of six movie trailers for unreleased
films was shown, totaling 15 min of viewing time. All pop-
corn boxes and water containers were collected immediately
after the final trailer. Participants then moved to the cinema
foyer and completed the survey, which, to maintain the cover
story, first assessed their interest in seeing each film and then
included a personality inventory and the liking and habit
strength measures (see below).
Participants in the meeting room condition followed a
similar procedure with one critical modification. The study
was conducted in a campus meeting room and involved
watching and rating music videos, a novel setting not associated
 Neal et al.
Table 1. Mean Liking of Popcorn Received in Studies 1 and 2 as a
Function of Experimental Condition
Fresh
Habit cuing manipulation M SD M SD
Study 1
Cinema 3.51 (0.93) 2.60
Meeting room 3.60 (0.96) 3.03
Study 2
Eating with dominant hand 4.18 (1.42) 3.10
Eating with nondominant hand 4.50 (1.31) 2.72
Higher numbers reflect more positive ratings of the popcorn on a 7-point
scale ranging from very bad (1) to very good (7).
with prior popcorn eating. A pilot test (N = 22) confirmed
that the music videos were equivalent to the movie trailers
in terms of their interest value and attentional involvement
(ts < 1). Instead of rating their interest in watching each
movie, participants rated their interest in owning each music
video. All other aspects of the design were identical to
the cinema setting (e.g., lighting levels, popcorn boxes, time
since last major meal, number and visibility of other
participants).
Hunger ratings. Before watching the videos, participants
rated their current hunger on a 5-point scale (anchors not at
all to extremely).
Ratings of popcorn attitude. On a 7-point scale, participants
rated their liking for the popcorn they received (anchors very
bad to very good).
Habit strength. On a 7-point scale, participants indicated
how frequently in the past they ate popcorn in movie theaters
(anchors always to never, reverse scored).
Results and Discussion
Our primary hypotheses were tested via regression models
with the following predictors: context (cinema vs. meeting
room), popcorn freshness (stale vs. fresh), habit strength to
eat popcorn in theaters, and all two- and three-way interac-
tions. Gender was a control variable in all analyses.
Reported attitudes and hunger. To establish that the experi-
ment was constructed appropriately to test our hypotheses,
we first used the regression model to predict participants’
liking of the popcorn they received. As expected, the model
yielded only a main effect of the freshness manipulation,
B = 0.54, t(141) = 1.98, SE = 0.27, p = .05, establishing that
fresh popcorn was liked significantly more than stale. Impor-
tantly, this effect was not qualified by any interactions with
context or habit strength.2 Mean liking of fresh popcorn was
close to the scale midpoint (M = 3.56, SD = 1.23), and mean
liking of stale popcorn was below the midpoint (M = 2.72,
SD = 0.91), demonstrating that participants decidedly dis-
liked the stale popcorn (see Table 1). No other differences in
liking or in self-reported hunger emerged across conditions.
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
1431
Thus, participants in the meeting room had similar levels of
hunger to those in the cinema, t(144) = 1.57, SE = 0.16,
p = .117. Furthermore, the comparable ratings of popcorn in
Stale
both contexts suggests that participants in the meeting room
did not pay any closer attention to the popcorn than those at
the cinema, t(144) = 1.30, SE = 0.18, p = .194.
(1.16) Factors influencing performance of eating habits. The per-
(0.89) centage of popcorn participants consumed was then analyzed
with the regression model. Significant main effects emerged
(1.42) for gender (men ate more), B = 14.37, t(142) = 3.32, SE = 4.32,
(1.36) p = .001; context, B = 27.20, t(142) = 4.53, SE = 6.00,
p = .001; and freshness, B = 15.96, t(142) = 2.27, SE = 7.05,
p = .025. The latter two effects were qualified by the pre-
dicted three-way interaction between habit strength, context,
and freshness, B = 11.36, t(142) = 2.04, SE = 5.56, p = .043,
and no other effects were significant. To test whether strong-
habit participants’ eating was activated automatically in the
cinema context with little regard for freshness, we calculated
simple regression slopes between percent of popcorn con-
sumed and freshness of the popcorn separately for partici-
pants with weak, moderate, and strong consumption habits in
the cinema (Cohen, Cohen, West, & Aiken, 2003; see Figure 1).
In the cinema context, participants with weak habits—who
infrequently ate popcorn at cinemas—ate significantly less
popcorn in the stale than in the fresh conditions, B = –30.03,
t(90) = 4.28, SE = 7.01, p = .001. Those with moderate pop-
corn eating habits also ate less stale than fresh popcorn,
B = –19.87, t(90) = 4.94, SE = 4.02, p = .001. Importantly,
as predicted, those with strong popcorn eating habits ate the
same percentage of stale as fresh popcorn when located in
the cinema environment, B = –9.72, t(90) = 1.41, SE = 6.89,
p = .162.
In the meeting room context, an effect of freshness was
evident at every level of habit strength. Specifically, people
ate (at least marginally) less stale than fresh popcorn, whether
they had weak habits, B = –11.24, t(90) = 1.86, SE = 6.04,
p = .074; moderate habits, B = –17.21, t(90) = 4.09, SE = 4.21,
p = 0.001; or strong habits, B = –23.17, t(90) = 3.03, SE = 7.65,
p = .001. Critically, this pattern demonstrates that when
tested in a novel environment, participants with strong
popcorn consumption habits ate less stale than fresh
popcorn.
Habit performance did not depend on motives. We con-
ducted separate analyses to evaluate directly whether habit
performance depended on participants’ hunger and liking for
the popcorn. To evaluate hunger as a moderator, we included
this factor as a predictor in the model. Although hungrier
participants ate marginally more in general, B = 4.01,
t(142) = 1.7, SE = 2.36, p = .092, the critical interaction
between hunger and habit strength was not significant (t < 1),
suggesting that habit effects did not depend on this motive.
Even more importantly, after including hunger and the
Hunger × Habit Strength interaction, the predicted three-way
interaction maintained between habit strength, context, and
freshness, B = 12.27, t(142) = 2.17, SE = 5.66, p = .032.
 1432
70
Percent of popcorn box consumed
60
50
40
30
20
10
Stale
Cinema context
Figure 1. Percentage of popcorn eaten during 15 min of movie trailers in the cinema and 15 min of music videos in the meeting room
(Study 1)
Simple slopes depict percent consumed of stale and fresh popcorn for those with strong (mean + 1 SD), moderate (mean), and weak (mean – 1 SD) habits
for eating popcorn at cinemas.
We then conducted analyses evaluating liking as a mod-
erator. Again, this motivation had a main effect, with partici-
pants eating more popcorn when they liked it more, B = 8.32,
t(142) = 3.95, SE = 2.11, p < .001, but the interaction between
liking and habit strength was not significant (t < 1), sug-
gesting that habit effects did not depend on this attitude.
Furthermore, the three-way interaction between habit strength,
context, and freshness remained close to significant after
including these effects in the model, B = 9.69, t(142) = 1.80,
SE = 5.35, p = .07.
In general, the overall pattern suggests that habit perfor-
mance resists shifts in attitudes but can be disrupted by shifts
in habit cuing. It is important to note that strong-habit par-
ticipants experienced the shift in motivation as a function of
the food’s freshness. They rated the fresh popcorn as more
likable than the stale, and thus they did not value the popcorn
more than those with weak habits. Because we assessed lik-
ing for the specific popcorn in the experiment and not pop-
corn in general, we evaluated the attitude toward the object
most closely linked to consumption of this particular food.
Although strong-habit participants experienced this motiva-
tional shift, their eating was not guided by it. That is, when
in the cinema so that eating was activated automatically by
the context, these participants consumed about 63% of the
popcorn regardless of its palatability. Furthermore, tests of
whether the habit effects depended on liking for the popcorn
or current hunger revealed that these factors did not moder-
ate the effects of habit strength on eating.
The eating behavior of strong-habit participants was,
however, disrupted by features associated with automatic
cuing. When in the meeting room watching movie videos,
cinema-popcorn habits were not automatically activated, and
strong-habit participants’ behavior was brought under inten-
tional control. Then, like the weak-habit participants in all
conditions, they ate about half as much stale popcorn
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Personality and Social Psychology Bulletin 37(11)
70
Percent of popcorn box consumed
60
50
40
30
Low habit
Moderate habit 20
High habit
10
Fresh Stale Fresh
Meeting room context
compared with fresh popcorn. Thus, participants’ habitual
eating was resistant to changes in motivational states but
influenced by variations in habit automaticity.
Study 2
Study 2 replicated the test of insensitivity of habit perfor-
mance to motivational states and tested an additional disrup-
tion of habit automaticity—whether the response could be
repeated as in the past or had to be performed in a novel
way. We anticipated that participants performing the
response in a novel way would be unable to act automati-
cally on any habitual response triggered by the cinema con-
text. To test this idea, half of the participants were instructed
to use their nondominant hand to eat. This disruption in the
fluid execution of habitual eaters’ consumption should not
affect their judgments of the popcorn. Instead, the manipula-
tion should hinder the automatic execution of the habitual
response and bring behavior under intentional control—so
that participants can be guided by their motivations and eat
more fresh than stale popcorn. Thus, we tested whether par-
ticipants with strong cinema-popcorn habits, when eating
with their nondominant hand at the cinema, would be hin-
dered in automatically repeating their past behavior and
instead would become responsive to food freshness, con-
suming only popcorn they liked.
The study also assessed several potential alternative
explanations for the predicted effects. To see whether the
hand manipulation affected consumption by prompting peo-
ple to attend more to the popcorn’s freshness (and perhaps
especially so for those with strong habits), we tested people’s
recall for specific details of the movie trailers. We reasoned
that participants attending closely to the popcorn would, by
definition, be attending less to the trailers and thus would
recall fewer specific details. Also, based on prior work in
 Neal et al.
action identification theory (Vallacher & Wegner, 1989) and
construal level theory (Trope & Liberman, 2010), we tested
whether the hand manipulation encouraged people to con-
strue popcorn consumption at a lower, more concrete level,
thereby highlighting lower level, immediate outcomes such
as taste. Finally, we assessed participants’ perceived norms
to eat popcorn and tested whether variation in normative
beliefs could explain the effects of habit. Thus, Study 2
tested for changes in attention, construal level, and perceived
norms as possible explanations for the predicted effects.
Method
Design and participants. The design of Study 2 was identi-
cal to Study 1 except that the context manipulation was
replaced by a manipulation to eat with a particular hand.
Thus, the design was a Hand Used to Eat (dominant vs.
nondominant) × Food Freshness (fresh vs. stale) between-
participants design. Eighty-nine participants (47 males) were
paid $6 for completing the study.3
Procedure
Participants were recruited and participated just before the
showing of a feature film at a campus cinema. Procedural
details mirrored the cinema condition of Study 1, with the
exception of the manipulation of the hand participants used
to eat. For this manipulation, we designed a modified
popcorn box with a vertically aligned handle on one side.
Participants were instructed to slide one hand in-between the
handle and the box and hold the box in that manner through-
out the session. This instruction was embedded in several
other methodological details and framed as necessary for
experimental control. Testing sessions were randomly
assigned such that, in half, participants were instructed to
hold the box in their left (vs. right) hand.
In the follow-up survey, participants completed the
Edinburgh Handedness Inventory (Oldfield, 1971). This
enabled us to code participants according to whether they
had eaten with their typical, dominant hand or their atypical,
nondominant hand. Ambidextrous participants (N = 4) were
coded as dominant hand eaters, but excluding this group did
not alter the results reported below. At the study conclusion,
participants were asked for an anonymous, honest estimate
regarding the percentage of eating time in which they had
actually used the instructed hand to eat. Estimates were made
on an 11-point scale (0% of the time to 100% of the time).
Compliance with the hand manipulation was good. Ninety
percent of the sample reported complying 100% of the time.
Those who deviated did so an average of only 10% of the
time. Excluding these participants or including compliance
as a control variable did not alter the reported results.
Normative beliefs. To evaluate the extent to which habit
performance depends on social norms, in the final survey
participants rated on a 7-point scale the extent to which
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
1433
“other people important to them think that it’s normal to
eat popcorn at the cinema” (anchors strongly disagree to
strongly agree). Because this item was not a close-to-significant
predictor of consumption, it is not discussed further.
Attention to the trailers. To assess how much attention par-
ticipants paid to the movie trailers, they answered nine dif-
ficult questions about the trailers (e.g., “What color did the
main character in ‘Whip It’ dye her hair?”). The only close-
to-significant effect in the analyses on factual recall was a
trend for slightly worse recall among strong-habit partici-
pants, B = –0.41, t(80) = 1.78, SE = 0.23, p = .07. This pattern
counters the possibility that strong-habit participants were
not attending to the popcorn and were able to attend more
closely to the movie trailers.
Level of construal of popcorn eating. Three questions
assessed whether people construed the act of eating popcorn
in the study at a higher, more abstract level versus a lower,
more concrete level. Modeled after Vallacher and Wegner’s
(1989) research, each question involved choosing between a
more concrete description of eating popcorn (i.e., chewing,
holding a box, placing food in your mouth) and a more
abstract one (i.e., part of the movie experience, sharing
something with other people, a healthful snack). The sum of
abstract choices (range = 0-3) served as the measure of con-
strual level. Consistent with action identification theory and
construal level theory, more habitual eaters preferred more
abstract descriptions of popcorn consumption, B = 0.14,
t(81) = 2.51, SE = 0.07, p = .036. However, the interaction
between habit strength and hand used to eat did not signifi-
cantly predict construal (t > 1.2). Thus, the predicted effects
of the manipulation cannot be explained by changes in con-
strual level.
Results and Discussion
The hypotheses were tested using a regression model with
the following predictors: hand used to eat, popcorn fresh-
ness, habit strength to eat popcorn in theaters, and all two-
and three-way interactions. Gender was a control variable in
all analyses.
Reported attitudes and hunger. The model predicting par-
ticipants’ liking of the popcorn they received yielded only a
main effect, reflecting that participants disliked the stale
popcorn (M = 3.43, SD = 1.59) more than the fresh (M = 4.35,
SD = 1.59), B = 1.02, t(81) = 3.44, SE = 0.30, p = .001.
Importantly, this effect was not qualified by any interactions
with context or habit strength.4 As can be seen from the lik-
ing means in Table 1, no other differences emerged across
conditions. Thus, mean liking did not vary with the hand
used to eat, t(85) = 0.78, SE = 0.33, p = .44. Prestudy hunger
levels also were identical among those eating with their
dominant versus nondominant hand, t(85) = 000, SE = 0.26,
p = 1.00.
Factors influencing performance of eating habits. The model
predicting percentage of popcorn consumed yielded a main
 1434
50
Percent of popcorn box consumed
40
30
20
10
Stale Fresh
Ate with dominant hand
Figure 2. Percentage of popcorn eaten during 15 min of movie trailers in the cinema when participants were eating with their typical,
dominant hand or atypical, nondominant hand (Study 2)
Simple slopes depict percent consumed of stale and fresh popcorn for those with strong (mean + 1 SD), moderate (mean), and weak (mean – 1 SD) habits
for eating popcorn at cinemas.
effect of habit strength, B = 6.74, t(81) = 2.55, SE = 2.64,
p = .013, and a marginal effect of hand used to eat, B = 8.27,
t(81) = 1.67, SE = 4.95, p = .094. These main effects were
qualified by the predicted three-way interaction between
habit strength, hand used to eat, and freshness, B = 8.52,
t(81) = 2.22, SE = 3.84, p = .029. To test whether strong hab-
its were executed automatically— regardless of freshness—
only when participants used their dominant hand, we
calculated simple regression slopes between percent of pop-
corn consumed and freshness of the popcorn separately for
participants with weak, moderate, and strong consumption hab-
its as a function of the hand they used to eat (Cohen et al., 2003).
As predicted (see Figure 2), using the dominant versus
nondominant hand to eat popcorn apparently enabled or dis-
rupted habit cuing. That is, habitual eaters using their domi-
nant hand automatically executed the response activated in
the cinema context. The same amount of fresh and stale
popcorn was eaten by participants with moderate habits,
B = –1.41, t(81) = 0.46, SE = 3.07, p = .647, and strong hab-
its, B = 8.81, t(81) = 1.55, SE = 5.68, p = .13. Only partici-
pants with weak habits ate significantly less stale than fresh
popcorn when eating with the dominant hand, B = –11.64,
t(81) = 1.99, SE = 5.84, p = .053.
Participants eating with their nondominant hand, in con-
trast, did not respond habitually to the cinema context and
were influenced by popcorn freshness. Specifically, those
with moderate habits ate significantly less stale than fresh
popcorn, B = –6.44, t(81) = 2.04, SE = 3.16, p = .048, and
those with strong habits ate marginally less, B = –12.89,
t(81) = 1.91, SE = 6.75, p = .063. Unexpectedly, those with
weak habits did not eat less stale than fresh popcorn when
eating with their nondominant hand, B = 0.02, t(81) = 0.01,
SE = 2.00, p = .998. However, inspection of the consumption
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Personality and Social Psychology Bulletin 37(11)
50
Low habit
Percent of popcorn box consumed
Moderate habit
40
High habit
30
20
10
Stale Fresh
Ate with non-dominant hand
rates for this group suggests a floor effect, such that eating
with the nondominant hand was at such a low rate among
these individuals that the stale popcorn could not induce fur-
ther declines.
Habit performance did not depend on motives. As in Study1,
we tested directly whether habit performance depended on
participants’ hunger and liking for popcorn. Although hun-
grier participants ate marginally more in general, B = 2.72,
t(79) = 1.76, SE = 1.55, p = .08, the critical interaction
between hunger and habit strength was not significant (t < 1),
suggesting that habit effects did not depend on this motive.
Even more importantly, after including hunger and the
Hunger × Habit Strength interaction in the model, the predicted
three-way interaction maintained between habit strength, con-
text, and freshness, B = 7.75, t(79) = 2.01, SE = 3.86, p = .048.
In the analyses on liking as a moderator, this motivation
yielded a main effect, with participants eating more popcorn
if they liked it more, B = 3.90, t(79) = 2.52, SE = 1.34,
p = .014. However, the interaction between liking and habit
strength was not significant (t < 1), suggesting that the habit
effects did not depend on this attitude. After adding these
additional terms to the model, the critical three-way interac-
tion between habit strength, context, and freshness dropped
below significance, B = 5.54, t(79) = 1.50, SE = 3.69, p = .13.
However, the low power to detect effects in this model may
have contributed to this nonsignificant finding.
In general, just like Study 1, habitual cinema-popcorn eat-
ers did not like the popcorn any more or less than nonhabit-
ual eaters. Everyone disliked the stale popcorn and expressed
neutral attitudes toward the fresh. It is also interesting that
strong-habit participants expressed the same normative
beliefs about eating popcorn in the cinema as weak-habit
participants. Also, the slightly lower recall of the movie
 Neal et al.
trailers among strong-habit participants suggests that their
attention was not unduly captured by these presentations,
and thus they could attend to the quality of the popcorn as
much as weak-habit participants. Finally, although strong-
habit participants did favor more abstract descriptions of
popcorn consumption, this tendency was not altered by the
hand use manipulation, making changes in construal level an
unlikely explanation for the results.
Despite the comparable subjective judgments of partici-
pants with strong and weak habits, striking differences
emerged in their eating patterns. Strong-habit participants,
when they could eat automatically with their dominant hand,
repeated their past responses regardless of the palatability of
the food. However, when strong-habit participants were
forced to eat in an atypical way, their behavior was brought
under intentional control, and—like the weak-habit partici-
pants in all conditions—they ate little of the stale popcorn.
General Discussion
Two field experiments tested the factors controlling a
socially significant habit, eating behavior. In both experi-
ments, participants’ eating habits were resistant to changes
in attitudes and goals. Strong habits persisted regardless of
whether participants were hungry and whether the popcorn
was fresh and palatable or stale and distasteful. Thus, when
in the context associated with frequent past consumption,
strong-habit participants rigidly carried out their past
responses. Unlike the motivated types of automaticity stud-
ied most often in social psychology, habit responses were
not sensitive to participants’ current motivational states.
Instead, strong-habit participants acted on habitual responses
in memory to such an extent that they ate even food they
disliked.
Habitual eating, although not influenced by motivations,
was disrupted by factors that we anticipated would block the
processes of habit automaticity. That is, participants in an
environment not associated with past consumption (a meet-
ing room) should not have the habitual response automati-
cally brought to mind. Also, participants at the cinema eating
in a novel way (with their nondominant hand) should not
have been able to automatically carry out the response in
mind. Our experiments thus directly manipulated the critical
mechanisms that enable habit performance. When habit
automaticity was apparently disrupted, participants’ eating
came under intentional control. That is, participants ate more
when they were hungry and when they liked the popcorn—
yielding greater consumption of fresh than stale popcorn.
The findings thus echo prior research showing that people
relying on their habits were less responsive to the outcomes
of their behavior (Ji & Wood, 2007; Verplanken, Aarts, &
van Knippenberg, 1997). By disrupting habitual eaters’ auto-
matic activation and execution of their past behavior, the
present research reveals further that responsiveness to out-
comes can be restored.
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
1435
Importantly, the different eating patterns of strong- and
weak-habit participants could not be explained by differ-
ences in the subjective evaluation of the popcorn. Habitual
eaters recognized that the stale popcorn was not palatable,
and they reported disliking it just as extremely as did nonha-
bitual eaters. It is striking, then, that these negative attitudes
only curtailed habitual eaters’ consumption when habit auto-
maticity was disrupted. At the process level, the results do
not appear to be due to changes in attention, social norms, or
construal level across experimental conditions. Our check on
recall of the details of the movie trailers in Study 2 revealed
that the hand use manipulation did not change how much
attention people paid to the popcorn, either as a main effect
or through an interaction with habit strength. Also, all par-
ticipants held similar norms to eat popcorn in the cinema.
Finally, the hand use manipulation did not alter whether
people thought about their popcorn consumption at a more
abstract versus a more concrete level. Thus, it appears
unlikely that the manipulation to disrupt habit automaticity
worked through these alternative mechanisms.
The present study contributes to growing evidence that
performance settings can serve as direct cues to habitual
behavior. Habitual runners have been found to have strong
cognitive associations between the location in which they
typically trained and mental representations of running (Neal
et al., 2011). Similarly, in behavior prediction studies, strong-
habit participants tend to repeat responses independently of
intentions when they are in familiar performance contexts (Ji
& Wood, 2007). Based on this earlier research, we anticipated
that the actual location of the theater might serve as a trigger
to habit performance. Yet, the actual triggers might include a
variety of aspects of the cinema setting and the evening show
time. Because we had participants sit some distance from each
other in the cinema, we doubt that movie companions were
triggers for consumption in the present studies. Furthermore,
because we used the same popcorn and the same bags in the
cinema and meeting rooms, these features cannot have trig-
gered habit performance. However, the exact nature of the
triggering cues is a question for future research.
In general, our findings expand on work conducted by
Wansink and colleagues in which the quantity of food people
consume is influenced by simple manipulations of environ-
mental cues such as plate and serving size (e.g., Sobal &
Wansink, 2007; Wansink & Kim, 2005). Our results suggest
that such environmental cues are likely to be most influential
when people have developed habits to respond in particular
ways to the cue. The current results also suggest a possible
role for eating habits in Schachter’s (1968) externality
hypothesis, which stipulated that obese individuals are more
driven by external cues and less by internal cues than are
nonobese individuals (see also Wansink, Payne, & Chandon,
2007). Obesity has been linked to eating patterns such as
nocturnal snacking that are repeated consistently at particu-
lar times of day (Berg et al., 2009). Thus, obese individuals
may rely less than nonobese people on internal cues partly
 1436
because their eating behavior is more habitual, leading them
to respond as cued by environments of past consumption.
Finally, our findings suggest ways to gain traction on the
difficult problem of habit change (see Rothman, Sheeran,
& Wood, 2009; Verplanken & Wood, 2006). Consistent
with the idea that habits are a form of non-goal-dependent
automaticity (see Bargh, 1994; Moors & De Houwer, 2006),
habits may not be especially sensitive to alterations in moti-
vational states but instead may best be disrupted by changes
in triggering contexts. Habit change may thus require
impeding habit activation or interrupting fluid habit execu-
tion. Although our findings suggest that both avenues are
effective, it is not always possible for dieters to avoid or
alter the environments in which they typically overeat (see
Quinn, Pascoe, Wood, & Neal, 2010). More feasible, per-
haps, is for dieters to actively disrupt the execution of the
activated eating sequence by simple manipulations such as
eating with the nondominant hand and, in so doing, bring
their eating under their personal control.
Authors’ Note
The authors are grateful to Meredith McAdams, Connie Wang, and
Fade Eadeh for their help with data collection.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to
the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research,
authorship, and/or publication of this article.
Notes
1. In Study 1, 16 additional participants (7 in the cinema context,
9 in the meeting room context) were excluded from the analysis
because they either declined the popcorn or did not eat any and
thus were not exposed to the freshness manipulation.
2. To further test for possible effects of habit strength on liking,
we created high- and low-habit groups via a median split of the
habit variable. As with the continuous measure, strong-habit par-
ticipants liked the fresh popcorn (M = 3.66, SD = 0.77) signifi-
cantly more than the stale popcorn (M = 2.70, SD = 1.11), and
weak-habit participants also liked the fresh popcorn (M = 3.33,
SD = 1.14) more than the stale popcorn (M = 2.83, SD = 1.05). An
ANOVA on this dichotomous measure of habit strength yielded
only a main effect of freshness, F(1, 146) = 18.58, p < .001, and
the Habit × Freshness interaction term was not significant (F < 2).
3. In Study 2, 15 additional participants (8 in the right-hand condi-
tion, 7 in the left-hand condition) were excluded because they
either declined the popcorn box or did not eat any popcorn and
thus were not exposed to the freshness manipulation.
4. As in Study 1, we used a median split of habit strength to explore
potential differences in liking. Again, strong-habit participants
liked the fresh popcorn (M = 4.38, SD = 1.76) significantly more
than the stale popcorn (M =3.00, SD = 1.60), and weak-habit par-
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Personality and Social Psychology Bulletin 37(11)
ticipants also liked the fresh popcorn (M = 4.25, SD = 1.12) more
than the stale popcorn (M = 2.84, SD = 1.18). An ANOVA on
this dichotomous measure of habit strength yielded only a main
effect of freshness, F(1, 83) = 19.16, p < .001, and the habit and
Habit × Freshness interaction terms were not significant (Fs > 1).
References
Aldrich, J. H., Montgomery, J., & Wood, W. (in press). Turnout as
a habit. Political Behavior.
Balleine B. W., & O’Doherty, J .P. (2010). Human and rodent
homologies in motor control: Cortico-striatal determinants of
goal-directed and habitual action. Neuropsychopharmacology,
35, 48-69. doi:10.1038/npp.2009.131
Bargh, J. A. (1994). The Four Horsemen of automaticity: Aware-
ness, efficiency, intention, and control in social cognition. In
R. S. Wyer Jr. & T. K. Srull (Eds.), Handbook of social cognition
(2nd ed., pp. 1-40). Hillsdale, NJ: Erlbaum.
Bargh, J. A., & Barndollar, K. (1996). Automaticity in action:
The unconscious as repository of chronic goals and motives.
In P. M. Gollwitzer & J. A. Bargh (Eds.), The psychology of
action (pp. 457-471). New York, NY: Guilford.
Berg, C., Lappas, G., Wolk, A., Strandhagen, E., Tóren, K.,
Rosengren, A., . . . & Lissner, L. (2009). Eating patterns and
portion size associated with obesity in a Swedish population.
Appetite, 52, 21-26. doi:10.1016/j.appet.2008.07.008
Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied
multiple regression/correlation analysis for the behavior sci-
ences. Mahwah, NJ: Erlbaum.
Danner, U. N., Aarts, H., & de Vries, N. K. (2008). Habit vs.
intention in the prediction of future behavior: The role of
frequency, context stability and mental accessibility of past
behavior. British Journal of Social Psychology, 47, 245-265.
doi:10.1348/014466607X230876
Dickinson, A., & Balleine, B. (1995). Motivational control of
instrumental action. Current Directions in Psychological Sci-
ence, 4, 162-167. doi:10.1111/1467-8721.ep11512272
Dijksterhuis, A., Smith, P. K., van Baaren, R. B., & Wigboldus, D. H. J.
(2005). The unconscious consumer: Effects of environment
of consumer choice. Journal of Consumer Psychology, 15,
193-202. doi:10.1207/s15327663jcp1503_3
Eagly, A. H., & Chaiken, S. (1993). The psychology of attitudes.
Fort Worth, TX: Harcourt, Brace & Jovanovich.
Graybiel, A. M. (2008). Habits, rituals, and the evaluative brain.
Annual Review of Neuroscience, 31, 359-387. doi:10.1146/
annurev.neuro.29.051605.112851
Ji, M., & Wood, W. (2007). Purchase and consumption habits: Not
necessarily what you intend. Journal of Consumer Psychology,
17, 261-276. doi:10.1016/S1057-7408(07)70037-2
Kruglanski, A. W., Shah, J. Y., Fishbach, A., Friedman, R.,
Chun, W. Y., & Sleeth-Keppler, D. (2002). A theory of goal
systems. Advances in Experimental Social Psychology, 34,
331-378.
Macrae, C. N., & Johnston, L. (1998). Help, I need somebody:
Automatic action and inaction. Social Cognition, 16, 400-417.
 Neal et al.
Moors, A., & De Houwer, J. (2006). Automaticity: A theoretical
and conceptual analysis. Psychological Bulletin, 132, 297-326.
doi:10.1037/0033-2909.132.2.297
Neal, D. T., Labrecque, J., Wood, W., & Lally, P. (2011). How do
habits guide behavior? Perceived and actual triggers of habits
in daily life. Manuscript under review, University of Southern
California, Los Angeles.
Oldfield, R. C. (1971). The assessment and analysis of handed-
ness: The Edinburgh Inventory. Neuropsychologia, 9, 97-113.
doi:10.1016/0028-3932(71)90067-4
Ouellette, J. A., & Wood, W. (1998). Habit and intention in every-
day life: The multiple processes by which past behavior pre-
dicts future behavior. Psychological Bulletin, 124, 54−74.
doi: 10.1037/0033-2909.124.1.54
Poldrack, R. A., Clark, J., Pa’re-Blagoev, J., Shohamy, D., Creso
Moyano, J., Myers, C., & Gluck, M. A. (2001). Interactive
memory systems in the human brain. Nature, 414, 546−550.
doi:10.1038/35107080.
Quinn, J. M., Pascoe, A. M., Wood, W., & Neal, D. T. (2010).
Can’t control yourself? Monitor those bad habits. Per-
sonality and Social Psychology Bulletin, 36, 499-511.
doi:10.1177/0146167209360665
Rothman, A. J., Sheeran, P., & Wood, W. (2009). Reflective and
automatic processes in the initiation and maintenance of food
choices. Annals of Behavioral Medicine, 28(Suppl.), 4-17.
doi:10.1007/s12160-009-9118-3
Sakai, K. S., Kitaguchi, K., & Hikosaka, O. (2003). Chunking
during human visuomotor sequence learning. Experimental
Brain Research, 152, 229-242. doi:10.1007/s00221-003-
1548-8
Schachter, S. (1968). Obesity and eating. Science, 161, 751-756.
doi:10.1126/science.161.3843.751
Shah, J. Y., & Kruglanski, A. W. (2002). Priming against your will:
How goal pursuit is affected by accessible alternatives. Journal
of Experimental Social Psychology, 38, 368-383. doi:10.1016/
S0022-1031(02)00005-7
Sheeran, P., Webb, T. L., & Gollwitzer, P. M. (2005). The interplay
between goal intentions and implementation intentions. Person-
ality and Social Psychology Bulletin, 31, 87-98. doi:10.1177/
0146167204271308
Sobal, J., & Wansink, B. (2007). Kitchenscapes, tablescapes,
platescapes, and foodscapes: Influences of microscale built
environments on food intake. Environment and Behavior, 39,
124-142. doi:10.1177/0013916506295574
Strahan, E., Spencer, S. J., & Zanna, M. P. (2002). Subliminal
priming and persuasion: Striking while the iron is hot. Journal
Downloaded from psp.sagepub.com at UNIV OF ILLINOIS URBANA on October 19, 2014
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
1437
of Experimental Social Psychology, 38, 556-568. doi:10.1016/
S0022-1031(02)00502-4
Triandis, H. C. (1977). Interpersonal behavior. Monterey, CA:
Brooks/Cole.
Triandis, H. C. (1979). Values, attitudes, and interpersonal behavior.
Nebraska Symposium on Motivation, 27, 195-259.
Trope. Y., & Liberman, N. (2010). Construal-level theory of psy-
chological distance. Psychological Review, 117, 440-463.
doi:10.1037/a0018963.
Vallacher, R. R., & Wegner, D. M. (1989). Levels of personal
agency: Individual variation in action identification. Journal of
Personality and Social Psychology, 57, 660-671.
Vartanian, L. R., Herman, C. P., & Wansink, B. (2008). Are we aware
of the external factors that influence our food intake? Health
Psychology, 27, 533-538. doi:10.1037/0278-6133.27.5.533
Verplanken, B., Aarts, H., & van Knippenberg, A. (1997). Habit,
information acquisition, and the process of making travel mode
choices. European Journal of Social Psychology, 727, 539-560.
doi:10.1002/(SICI)1099-0992(199709/10)27:5<539::AID-
EJSP831>3.0.CO;2-A
Verplanken, B., Walker, I., Davis, A., & Jurasek, M. (2008). Context
change and travel mode choice: Combining the habit disconti-
nuity and self-activation. Journal of Environmental Psychol-
ogy, 28, 121-127. doi:10.1016/j.jenvp.2007.10.005
Verplanken, B., & Wood, W. (2006). Interventions to break and
create consumer habits. Journal of Public Policy and Market-
ing, 25, 90-103. doi:10.1509/jppm.25.1.90
Wansink, B., & Kim, J. (2005). Bad popcorn in big buckets: Portion
size can influence intake as much as taste. Journal of Nutrition
Education and Behavior, 37, 242-245. doi: 10.1016/S1499-
4046(06)60278-9
Wansink, B., Payne, C. R., & Chandon, P. (2007). Internal and
external cues of meal cessation: The French paradox redux?
Obesity, 15, 2920-2924. doi:10.1038/oby.2007.348
Wood, W., & Neal, D. T. (2007). A new look at habits and the
habit-goal interface. Psychological Review, 114, 843-863.
doi:10.1037/0033-295X.114.4.843
Wood, W., & Neal, D. T. (2009). The habitual consumer. Jour-
nal of Consumer Psychology, 19, 579-592. doi:10.1016/j.jcps
.2009.08.003
Wood, W., Tam, L., & Witt, M. G. (2005). Changing circum-
stances, disrupting habits. Journal of Personality and Social
Psychology, 88, 918-933. doi:10.1037/0022-3514.88.6.918
Yin, H., & Knowlton, B. (2006). The role of the basal ganglia in
habit formation. Nature Reviews Neuroscience, 7, 464-476.
doi:10.1038/nrn1919
 Debate & Analysis
Making health habitual:
the psychology of ‘habit-formation’ and general practice
MAKING HEALTH HABITUAL
The Secretary of State recently proposed
that the NHS:
‘... take every opportunity to prevent poor
health and promote healthy living by making
the most of healthcare professionals’
contact with individual patients.’ 1
Patients trust health professionals as
a source of advice on ‘lifestyle’ (that is,
behaviour) change, and brief opportunistic
advice can be effective.2 However, many
health professionals shy away from giving
advice on modifying behaviour because they
find traditional behaviour change strategies
time-consuming to explain and difficult for
the patient to implement.2 Furthermore,
even when patients successfully initiate the
recommended changes, the gains are often
transient3 because few of the traditional
behaviour change strategies have built-in
mechanisms for maintenance.
Brief advice is usually based on
advising patients on what to change and
why (for example, reducing saturated fat
intake to reduce the risk of heart attack).
Psychologically, such advice is designed to
engage conscious deliberative motivational
processes, which Kahneman terms ‘slow’
or ‘System 2’ processes.4 However, the
effects are typically short-lived because
motivation and attention wane. Brief advice
on how to change, engaging automatic
(‘System 1’) processes, may offer a valuable
alternative with potential for long-term
impact.
Opportunistic health behaviour advice
must be easy for health professionals to give
and easy for patients to implement to fit into
routine health care. We propose that simple
advice on how to make healthy actions into
habits — externally-triggered automatic
responses to frequently encountered
contexts — offers a useful option in the
behaviour change toolkit. Advice for creating
habits is easy for clinicians to deliver and
easy for patients to implement: repeat a
chosen behaviour in the same context, until
it becomes automatic and effortless.
HABIT FORMATION AND HEALTH
While often used as a synonym for frequent
or customary behaviour in everyday
parlance, within psychology, ‘habits’ are
defined as actions that are triggered
automatically in response to contextual
664 British Journal of General Practice, December 2012
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
cues that have been associated with their
performance:5,6 for example, automatically
washing hands (action) after using the toilet
(contextual cue), or putting on a seatbelt
(action) after getting into the car (contextual
cue). Decades of psychological research
consistently show that mere repetition of
a simple action in a consistent context
leads, through associative learning, to the
action being activated upon subsequent
exposure to those contextual cues (that is,
habitually).7–9 Once initiation of the action is
‘transferred’ to external cues, dependence
on conscious attention or motivational
processes is reduced.10 Therefore habits
are likely to persist even after conscious
motivation or interest dissipates.11 Habits
are also cognitively efficient, because
the automation of common actions frees
mental resources for other tasks.
A growing literature demonstrates the
relevance of habit-formation principles
to health.12,13 Participants in one study
repeated a self-chosen health-promoting
behaviour (for example, eat fruit, go for a
walk) in response to a single, once-daily
cue in their own environment (such as, after
breakfast). Daily ratings of the subjective
automaticity of the behaviour (that is, habit
strength) showed an asymptotic increase,
with an initial acceleration that slowed to a
plateau after an average of 66 days.9 Missing
the occasional opportunity to perform the
behaviour did not seriously impair the habit
formation process: automaticity gains soon
resumed after one missed performance.9
Automaticity strength peaked more quickly
for simple actions (for example, drinking
water) than for more elaborate routines (for
example, doing 50 sit-ups).
Habit-formation advice, paired with
a ‘small changes’ approach, has been
tested as a behaviour change strategy.14,15
In one study, volunteers wanting to lose
weight were randomised to a habit-based
intervention, based on a brief leaflet listing
“Advice for creating habits is easy for clinicians to
deliver and easy for patients to implement: repeat
a chosen behaviour in the same context, until it
becomes automatic and effortless.”
10 simple diet and activity behaviours and
encouraging context-dependent repetition,
or a no-treatment waiting list control. After
8 weeks, the intervention group had lost
2 kg compared with 0.4 kg in the control
group. At 32 weeks, completers in the
intervention group had lost an average of
3.8 kg.14 Qualitative interview data indicated
that automaticity had developed: behaviours
became ‘second nature’, ‘worming their
way into your brain’ so that participants
‘felt quite strange’ if they did not do
them.10 Actions that were initially difficult
to stick to became easier to maintain. A
randomised controlled trial is underway to
test the efficacy of this intervention where
delivered in a primary care setting to a
larger sample, over a 24-month follow-up
period.16 Nonetheless, these early results
indicate that habit-forming processes
transfer to the everyday environment, and
suggest that habit-formation advice offers
an innovative technique for promoting long-
term behaviour change.13
MAKING HEALTHY HABITS
We suggest that professionals could
consider providing habit-formation advice
as a way to promote long-term behaviour
change among patients. Habit-formation
advice is ultimately simple — repeat an
action consistently in the same context.12
The habit formation attempt begins at the
‘initiation phase’, during which the new
behaviour and the context in which it will be
done are selected. Automaticity develops
in the subsequent ‘learning phase’, during
which the behaviour is repeated in the
chosen context to strengthen the context-
behaviour association (here a simple
ticksheet for self-monitoring performance
may help; Box 1). Habit-formation
culminates in the ‘stability phase’, at which
the habit has formed and its strength has
plateaued, so that it persists over time with
minimal effort or deliberation.
 and settings to maintain interest (trying
different fruits with or between different
“... typical [behaviour change] advice ... emphasises meals). Variation may stave off boredom,
variation in behaviours and settings ... . Variation may but is effortful and depends on maintaining
stave off boredom, but is effortful and ... incompatible motivation, and is incompatible with
development of automaticity.6
with development of automaticity.” Patients should choose the target
behaviour themselves. Progress towards a
self-determined behavioural goal supports
patients’ sense of autonomy and sustains
Initiation requires the patient to be located within an existing daily routine (for interest,17 and there is evidence that a
sufficiently motivated to begin a habit- example, ‘when I go on my lunch break’) behaviour change selected on the basis of
formation attempt, but many patients would provides a convenient and stable starting its personal value, rather than to satisfy
like to eat healthier diets or take more point.10 external demands such as physicians’
exercise, for example, if doing so were Keeping going during the learning recommendations, is an easier habit
easy. Patients must choose an appropriate phase is crucial. The idea of repeating a target.18 Patients need to select a new
context in which to perform the action. The single specific action (for example, eating behaviour (for example, eat an apple) rather
‘context’ can be any cue, for example, an a banana) in a consistent context (with than give up an existing behaviour (do not
event (‘when I get to work’) or a time of cereal at breakfast) is very different from eat fried snacks) because it is not possible
day (‘after breakfast’), that is sufficiently typical advice given to people trying to to form a habit for not doing something. The
salient in daily life that it is encountered and take up new healthy behaviours, which automaticity of habit means that breaking
detected frequently and consistently. A cue often emphasises variation in behaviours existing habits requires different and
altogether more effortful strategies than
making new habits.12
Patients should be encouraged to aim for
small and manageable behaviour changes,
Box 1. A tool for patients because failure can be discouraging. A
Make a new healthy habit sedentary person, for example, would be
1. Decide on a goal that you would like to achieve for your health. more appropriately advised to walk one or
2. Choose a simple action that will get you towards your goal which you can do on a daily basis. two stops more before getting on the bus
3. Plan when and where you will do your chosen action. Be consistent: choose a time and place that you
than to walk the entire route — at least
encounter every day of the week.
4. Every time you encounter that time and place, do the action. for their first habit goal. Small changes
5. It will get easier with time, and within 10 weeks you should find you are doing it automatically without can benefit health: slight adjustments to
even having to think about it. dietary intake can aid long-term weight
6. Congratulations, you’ve made a healthy habit!
management,19 and small amounts of light
My goal (e.g. ‘to eat more fruit and vegetables’) _________________________________________________ physical activity are more beneficial than
none.20 Moreover, simpler actions become
My plan (e.g. ‘after I have lunch at home I will have a piece of fruit’) habitual more quickly.9 Additionally,
(When and where) ___________________________ I will ___________________________ behaviour change achievements, however
Some people find it helpful to keep a record while they are forming a new habit. This daily tick-sheet can be small, can increase self-efficacy, which
used until your new habit becomes automatic. You can rate how automatic it feels at the end of each week, can in turn stimulate pursuit of further
to watch it getting easier. changes.21 Forming one ‘small’ healthy
Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 9 Week 10 habit may thereby increase self-confidence
for working towards other health-promoting
Monday
habits.
Tuesday
Unrealistic expectations of the duration
Wednesday of the habit formation process can lead
Thursday the patient to give up during the learning
Friday phase. Some patients may have heard that
Saturday
habits take 21 days to form. This myth
appears to have originated from anecdotal
Sunday
evidence of patients who had received
Done on plastic surgery treatment and typically
>5 days, adjusted psychologically to their new
yes or no
appearance within 21 days.22 More relevant
How research found that automaticity plateaued
automatic on average around 66 days after the first
does it feel?
daily performance,9 although there was
Rate from
considerable variation across participants
1 (not at all)
to10
and behaviours. Therefore, it may be helpful
(completely) to tell patients to expect habit formation
(based on daily repetition) to take around

British Journal of General Practice, December 2012 665

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
 ADDRESS FOR CORRESPONDENCE
Benjamin Gardner
Health Behaviour Research Centre, Department of
Epidemiology and Public Health, University College
London, Gower Street, London, WC1E 6BT, UK.
E-mail: b.gardnersood@ucl.ac.uk
10 weeks. Our experience is that people
are reassured to learn that doing the
behaviour gets progressively easier; so they
only have to maintain their motivation until
the habit forms. Working effortfully on a
new behaviour for 2–3 months may be an
attractive offer if it has a chance of making
the behaviour become ‘second nature’.
CONCLUSION
Psychological theory and evidence
around habit-formation generates
recommendations for simple and
sustainable behaviour change advice. We
acknowledge that health professionals do
not always find it appropriate to offer lifestyle
counselling to patients: some patients can
become annoyed when advised to change
their behaviour, and this reaction can
threaten patients’ trust in and satisfaction
with the doctor–patient relationship.2
However, in settings where professionals
feel able to offer behaviour advice, we
suggest that they consider providing
guidance on habit-formation. Habit-
formation advice can be delivered briefly, it
is simple for the patient to implement, and it
has realistic potential for long-term impact.
It offers health professionals a useful tool
for incorporating evidence-based health
promotion into encounters with patients. A
sample tool for health professionals to use
with patients to encourage habit formation
is provided in Box 1.
Benjamin Gardner,
Lecturer in Health Psychology, Health Behaviour
Research Centre, Department of Epidemiology and
Public Health, University College London, London.
Phillippa Lally,
ESRC Postdoctoral Research Fellow, Health
Behaviour Research Centre, Department of
Epidemiology and Public Health, University College
London, London.
Jane Wardle,
Professor of Clinical Psychology, Health Behaviour
Research Centre, Department of Epidemiology and
Public Health, University College London, London.
Provenance
Freely submitted; externally peer reviewed.
DOI: 10.3399/bjgp12X659466
666 British Journal of General Practice, December 2012
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
REFERENCES
1. Lansley A. Response to NHS Future Forum’s
second report. London: Department of Health,
2012. http://www.dh.gov.uk/prod_consum_dh/
groups/dh_digitalassets/documents/digitalasset/
dh_132088.pdf (accessed 16 Oct 2012).
2. Lawlor DA, Keen S, Neal RD. Can general
practitioners influence the nation’s health
through a population approach to provision of
lifestyle advice? Br J Gen Pract 2000; 50(455):
455–459.
3. Jeffery RW, Drewnowski A, Epstein LH,
Stunkard AJ, et al. Long-term maintenance of
weight loss: current status. Health Psychol 2000;
19(Suppl1): 5–S16.
4. Kahneman D. A perspective on judgment and
choice. Am Psychol 2003; 58(9): 697–720.
5. Neal DT, Wood W, Labrecque JS, Lally P. How
do habits guide behavior? Perceived and actual
triggers of habits in daily life. J Exp Soc Psychol
2012; 48: 492–498.
6. Wood W, Neal DT. A new look at habits and the
habit-goal interface. Psychol Rev 2007; 114(4):
843–863.
7. Bayley PJ, Frascino JC, Squire LR. Robust
habit learning in the absence of awareness
and independent of the medial temporal lobe.
Nature 2005; 436(7050): 550–553.
8. Hull CL. Principles of behavior: an introduction
to behavior theory. New York, NY: Appleton-
Century-Crofts, 1943.
9. Lally P, van Jaarsveld CHM, Potts HWW, Wardle
J. How are habits formed: modelling habit
formation in the real world. Euro J Soc Psychol
2010; 40: 998–1009.
10. Lally P, Wardle J, Gardner B. Experiences of
habit formation: a qualitative study. Psychol
Health Med 2011; 16(4): 484–489.
11. Gardner B, de Bruijn GJ, Lally P. A systematic
review and meta-analysis of applications of the
Self-Report Habit Index to nutrition and physical
activity behaviours. Ann Behav Med 2011; 42(2):
174–187.
12. Lally P, Gardner B. Promoting habit
formation. Health Psychol Rev . In press: DOI:
10.1080/17437199.2011.603640.
13. Rothman AJ, Sheeran P, Wood W. Reflective
and automatic processes in the initiation and
maintenance of dietary change. Ann Behav Med
2009; 38(Suppl1): S4–17.
14. Lally P, Chipperfield A, Wardle J. Healthy habits:
Efficacy of simple advice on weight control based
on a habit-formation model. Int J Obes 2008;
32(4): 700–707.
15. McGowan L, Cooke LJ, Croker H, et al. Habit-
formation as a novel theoretical framework for
dietary change in pre-schoolers. Psychol Health
2012; 27(Suppl1): 89.
16. Beeken RJ, Croker H, Morris S, et al. Study
protocol for the 10 Top Tips (10TT) Trial:
Randomised controlled trial of habit-based
advice for weight control in general practice.
BMC Public Health 2012; 12(1): 667.
17. Deci EL, Ryan RM. The support of autonomy and
the control of behavior. J Pers Soc Psychol 1987;
53(6): 1024–1037.
18. Gardner B, Lally P. Does intrinsic motivation
strengthen physical activity habit? Modeling
relationships between self-determination, past
behaviour, and habit strength. J Behav Med
[Epub ahead of print].
19. Hill JO. Can a small-changes approach help
address the obesity epidemic? A report of the
Joint Task Force of the American Society for
Nutrition, Institute of Food Technologists, and
International Food Information Council. Am J
Clin Nutr 2009; 89(2): 477–484.
20. Warburton DER, Nicol CW, Bredin SSD. Health
benefits of physical activity: the evidence. Can
Med J 2006; 174(6): 801–809.
21. Bandura A. Social cognitive theory: an agentic
perspective. Annu Rev Psychol 2001; 52: 1–26.
22. Maltz M. Psycho-cybernetics. New York, NY:
Prentice Hall, 1960.
 AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 145:382–389 (2011)
Novelty-Seeking DRD4 Polymorphisms are Associated
with Human Migration Distance Out-of-Africa After
Controlling for Neutral Population Gene Structure
Luke J. Matthews1* and Paul M. Butler2
1
Department of Human Evolutionary Biology, Harvard University, 11 Divinity Ave., Cambridge, MA 02138
2
Department of Neurology, Boston University School of Medicine, 72 East Concord St., 5th Floor, B528, Boston, MA 02118
KEY WORDS functional polymorphisms; novelty-seeking trait (NS), natural selection; cultural
neuroscience; human variation
ABSTRACT Numerous lines of evidence suggest that
Homo sapiens evolved as a distinct species in Africa by
150,000 years before the present (BP) and began major
migrations out-of-Africa
50,000 BP. By 20,000 BP, our
species had effectively colonized the entire Old World, and
by 12,000 BP H. sapiens had a global distribution. We pro-
pose that this rapid migration into new habitats selected
for individuals with low reactivity to novel stressors. Cer-
tain dopamine receptor D4 (DRD4) polymorphisms are
associated with low neuronal reactivity and increased ex-
ploratory behavior, novelty seeking, and risk taking, collec-
tively considered novelty-seeking trait (NS). One previous
report (Chen et al.: Evol Hum Behav 20 (1999) 309–324)
demonstrated a correlation between migratory distance
Dopamine neurotransmission is fundamental to human
appetitive and reward-seeking behaviors, movement, cog-
nition, and sociability. DRD4 is a metabotropic inhibitory
receptor, which acts to decrease intracellular adenylyl
cyclase production following dopamine binding (Van Tol
et al., 1991; Oldenhof et al., 1998). Neuroanatomically,
DRD4 is differentially expressed, with the prefrontal cor-
tex containing the highest density of receptors that act to
inhibit downstream neural firing (Dulawa et al., 1999).
Humans exhibit several DRD4 polymorphisms, which
include a well-described 48 base-pair (bp) variable num-
ber tandem repeat (VNTR) in exon 3 (Ding et al., 2002;
Grady et al., 2003). The most common VNTR alleles are
2-, 4-, and 7-repeats (2R, 4R, 7R), which together
account for over 90% of all variation at this locus (Chang
et al., 1996; Ding et al., 2002). Across all human popula-
tions 4R allele frequencies are highest, 2R is common in
Asians but rare in Native Americans, and 7R is common
in the Americans but rare in Asians and Africans
(Chang et al., 1996; Wang et al., 2004). 7R and 2R alleles
are associated with a partial loss of DRD4-mediated pre-
frontal inhibition because of blunted second messenger
response. Compared with the 4R, the 2R and 7R alleles
result in 40 and 80% reduction in intracellular second
messenger response, respectively (Asghari et al., 1995;
Wang et al., 2004).
Promoter polymorphisms, including a 2521 C/T single
nucleotide polymorphism (SNP) and a 120-bp tandem
duplication located 1.2 kilo-bp upstream of the initiation
codon, are associated with similar neurophysiologic
downstream effects as 2R and 7R exon 3 VNTR alleles
(Kamakura et al., 1997; Seaman et al., 1999; Lowe et al.,
C 2011
V WILEY-LISS, INC.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
and the seven-repeat (7R) VNTR DRD4 allele at exon 3 for
human populations. This study, however, failed to account
for neutral genetic processes (drift and admixture) that
might create such a correlation in the absence of natural
selection. Furthermore, additional loci surrounding DRD4
are now recognized to influence NS. Herein we account for
neutral genetic structure by modeling the nonindepend-
ence of neutral allele frequencies between human popula-
tions. We retest the DRD4 exon 3 alleles, and also test
two other loci near DRD4 that are associated with NS. We
conclude there is an association between migratory dis-
tance and DRD4 exon 3 2R and 7R alleles that cannot be
accounted for by neutral genetic processes alone. Am J
Phys Anthropol 145:382–389, 2011. V 2011 Wiley-Liss, Inc.
C
2004). There is a well-supported correlation between the
DRD4 7R, 2R, 120-bp promoter duplication, and 2521
C/T SNP and the personality trait novelty seeking (NS)
(Tomitaka et al., 1999; Tsai et al., 2004; Reist et al., 2007;
Munafò et al., 2008). High NS individuals are considered
exploratory, impulsive, excitable, quick-tempered, and
extravagant, whereas low NS individuals tend to be rigid,
prudent, stoic, reflective, staid, and slow-tempered (Clo-
ninger et al., 1991; Ebstein et al., 1996).
Chen et al. (1999) suggested that global population
patterning of 4R and 7R frequency variation can be
explained by increased migratory distances of NS indi-
viduals (7R variant) compared with that of low NS indi-
viduals (4R variant). They proposed NS individuals rep-
resented a behavioral phenotype that functioned better
under the novel ecological and social circumstances
imposed during migration such that migration effectively
Grant sponsors: Department of Neurology and Program in Behav-
ioral Neuroscience, Boston University School of Medicine, Depart-
ment of Human Evolutionary Biology, Harvard University.
*Correspondence to: Luke Matthews, Department of Human Evo-
lutionary Biology, Harvard University, 11 Divinity Ave., Cambridge,
MA 02138. E-mail: ljmatth@fas.harvard.edu
Received 27 August 2010; accepted 11 January 2011
DOI 10.1002/ajpa.21507
Published online 5 April 2011 in Wiley Online Library
(wileyonlinelibrary.com).
 DRD4 ALLELES AND HUMAN MIGRATION PATTERNS 383

Fig. 1. (Redrawn from Coop et al., Genetics 2010, 185:1411–1423) The effect of statistical nonindependence of population genetic
data. The frequencies shown are for a single SNP allele, AGT M235T, in various populations plotted against latitude. Note how popula-
tions from the same region frequently cluster on the same side of the regression line, which indicates their statistical nonindependence.
Independent data points would scatter randomly across the regression line. The situation is highly analogous to the more well-character-
ized nonindependence that arises at the species level in a phylogenetic context (Rendall and Di Fiore, 1995; Rohlf, 2006).

selected for NS alleles. They found no association
between 7R frequencies and recent migration to the
United States by Europeans and East Asians, suggesting
that 7R carriers are not prone to migrate, but that they
may experience higher fitness in response to many gen-
erations of migration. This selection hypothesis proposes
that migration presents novel ecological and social cir-
cumstances which NS individuals deal with more effec-
tively, with less stress response, and thereby experience
greater fitness relative to low NS individuals.
Alternatively, if decisions by individuals in recent
migrations are fundamentally different from those in an-
cient migrations, then it could be that NS individuals
during ancient migrations were actually more prone to
migrate. This nonselection hypothesis still specifies a cor-
relation between migration distance and NS alleles, but
the correlation comes about through genetically based
tendencies in the decisions of NS individuals.
In this study, we do not attempt to distinguish
between these alternative processes (selection or biased
decisions based on genotype) that would result in a cor-
relation of NS alleles and migratory distance. Instead,
we seek to determine if the correlation itself can be
accounted for by neutral genetic processes alone; that is,
is it necessary to invoke one of the above non-neutral
processes patterning variation in NS alleles? We retested
Chen et al.’s (1999) findings for the 7R allele in conjunc-
tion with three other alleles (exon 3 VNTR 2R allele,
2521 C/T SNP, 120-bp duplication), while controlling for
the effects of population history and admixture.
We controlled for neutral genetic processes by using
the results of a major recent global genetic survey
(Tishkoff et al., 2009) to construct a correlation structure
that expressed the correlations expected to arise by drift
and admixture in gene frequencies among human popu-
lations. Because the out-of-Africa migration proceeded
though a process of serial colonization, significant popu-
lation size bottlenecks have occurred throughout the his-
tory of human migration across the globe (Ramachan-
dran et al., 2005; Li et al., 2008; DeGiorgio et al., 2009;
Deshpande et al., 2009; Hunley et al., 2009). Thus,
observed associations of migratory distance and particu-
lar allelic frequencies may result simply from cumulative
effects of genetic drift that skew allele frequencies more
dramatically with increasing distance from Africa (Klopf-
stein et al., 2006). This problem exists for any associa-
tion of a particular allele frequency and an ecological
trait, not just migratory distance (Coop et al., 2010; Han-
cock et al., 2010). In particular, Coop et al., (2010) have
proposed that because drift does not systematically affect
the frequency of any particular allele, human popula-
tions out of Africa only show shared patterns of allelic
differentiation because they share the same stochastic
drift events. This renders human populations as statisti-
cally nonindependent (see Fig. 1). We return to this
point in more detail in the Discussion. As pursued by
other authors (Coop et al., 2010; Hancock et al., 2010)
we attempted to control for neutral genetic processes by
attacking the fundamental problem of data point nonin-
dependence with a correlation structure in a generalized

American Journal of Physical Anthropology

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 384 L.J. MATTHEWS AND P.M. BUTLER
linear model. The correlation structure expressed the
expected correlations among population data points
under a model of neutral genetic change, and trans-
formed the DRD4 data to remove neutral genetic effects
through a likelihood-based linear regression model. We
assessed the performance of our correlation structure
with a set of previously published neutral loci.
We also used the same neutral loci in an alternative
test of the specific association of migratory distance and
NS alleles by assessing whether the residual variation of
migratory distance regressed on DRD4 alleles was less
than exhibited by similar regressions with neutral
genetic loci. We asked if frequencies of NS alleles at
DRD4 showed a tighter increase with migratory distance
than we would expect by chance, chance being reflected
in this case by neutrally evolving loci.
MATERIALS AND METHODS
Functional gene data
We searched the published literature for loci surround-
ing DRD4 with well-characterized in vitro effects on
expression or binding, well-characterized effects on person-
ality traits, and with population level sampling from
groups whose neutral genetic structure was known. We
found 3 loci that were available across 15 or more human
populations (Chang et al., 1996; Chen et al., 1999; Osier
et al., 2002; Rajeevan et al., 2003; ALFRED, http://alfred.
med.yale.edu). These loci were DRD4 exon 3 VNTRs, 2521
C/T SNP, and the promoter region 120-bp duplication.
Migratory distance
We derived population locations from the midpoint of
the geographic range reported in ALFRED (http://alfred.
med.yale.edu; midpoint calculations–http://www.movable-
type.co.uk/scripts/latlong.html). We calculated the migra-
tory distance of each population in our study as the
great circle distances connecting them across the globe
and along known routes of human migration out-of-
Africa (Prugnolle et al., 2005; Liu et al., 2006, Mellars,
2006; Long et al., 2009; calculations performed in Google
Earth). We set the San as migratory point zero because
they were the most anciently diverged population of
those included in our study and because they are geneti-
cally close to the origin within Africa of human genetic
diversity as inferred through coalescent models (Tishkoff
et al., 2009). Thus, within Africa we constructed the
following routes (San ? Mbuti, San ? Bantu ? Hausa/
Biaka). All routes out of Africa went through the Mbuti
in central Africa, and then the Druze in the Levant.
European and Near Eastern routes led from the Druze
directly to Sardinian and Adygei. We constructed a south-
ern route through Asia as Druze ? Cambodian ? Pap-
uan ? Melanesian. A more Northern route through Asia
led to the Han followed by the Japanese. The route to the
Americans went through the Han and then direct to the
Yakut, followed by the New World populations in a North
to South sequence. These migration values thus attempt
to capture the minimum migratory distance of popula-
tions from the origins of our species to their present loca-
tions, relative to the other populations in the study. They
may underestimate the actual distances that populations
have migrated since their inception as distinct demes.
American Journal of Physical Anthropology
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Neutral genetic structure
We controlled for neutral genetic processes of drift,
admixture, and the history of population separations by
incorporating published D2 genetic distances into a gen-
eralized least squares linear model (see Comparative
Analyses below). The D2 expresses the expected allele
frequency differentiation between two populations under
an evolutionary model with no selection or mutation
(Reynolds et al., 1983). We obtained distances from a
recent global study of human genetic differentiation that
used data from 1327 polymorphic and neutrally evolving
loci from 3945 individuals (Tishkoff et al., 2009).
Comparative analyses
We constructed a generalized least squares (gls) statis-
tical model with migratory distance as the dependent
variable and with three independent variables (imple-
mented in R, package ‘‘nlme’’, Pinheiro et al., 2009). One
independent variable was the summed frequency for
DRD4 2R and 7R alleles, because higher frequencies of
either allele should produce similar phenotypic effects in
the form of increased NS behavior. The other two inde-
pendent variables were the frequencies of the T SNP at
position 2521, and the frequencies of the promoter region
120-bp duplication. All independent variables were coded
as fixed effects, with the prediction that higher frequen-
cies of each would be associated with greater migratory
distance. Because the variance distribution of frequency
data rarely conform to assumptions of linear modeling,
we arcsine transformed the raw frequency data for subse-
quent analysis (Sokal and Rohlf, 1995, pp 419–422).
We included a correlation structure in our model to
control for the nonindependence of data points in this
study (Dow et al., 1984; Pinheiro and Bates, 2000; Rohlf,
2006; Dow, 2007; Coop et al., 2010; Hancock et al. 2010).
The allele frequencies of human populations are ren-
dered statistically nonindependent because they share
complex histories of drift and admixture that stochasti-
cally alter the frequencies of all alleles, including those
that are selectively neutral. Chen et al., (1999)
recognized this nonindependence, but dealt with it insuf-
ficiently by simply scoring whether pairs of populations
participated in the same migration routes. Here, we
account for the nonindependence with robust measures
of genetic distance between populations from neutrally
evolving loci. We converted the D2 distances into
expected correlations within the study sample by apply-
ing the linear transformation:
ðmaxðD2 Þ  D2 ij Þ= maxðD2 Þ
where max(D2) is the maximum distance in the matrix
and D2ij is the observed distance between a given pair i,
j. This transformation produced a matrix of expected cor-
relations among data points ranging from 0 for the two
most divergent populations to 1 for each population to
itself. This correlation matrix is thus directly analogous
in its range of expected correlations to the matrices used
in phylogenetic generalized least squares analyses
between species (Pagel, 1999; Garland and Ives, 2000;
Rohlf, 2006; Coop et al., 2010), but it does not presume
bifurcating relationships among the study taxa.
Using our correlation structure to control for neutral
population genetic processes, we constructed least squares
regression models from each possible combination of the
three independent variables, and used Akaike Information
 DRD4 ALLELES AND HUMAN MIGRATION PATTERNS 385
TABLE 1. Allele frequencies as percentages and migratory distance out of Africa
Population exon 3 R2 1 R7a 120 duplicationb T SNPb Migratory Distance (km) N-exon 3 N - 120 dup. & T SNP
Adygei 17 74 57 7524 104 108
Bantu 24 NA NA 2431 80 NA
Biaka 14 41 66 2692 124 140
Cambodian 32 87 74 13540 50 46
Columbian 62 NA NA 29997 26 NA
Druze 12 69 60 6235 50 140
Han 18 68 66 13123 182 94
Hausa NA 75 67 3792 NA 72
Japanese 18 71 53 15774 102 100
Karitiana 60 79 95 31751 108 102
Maya 40 52 86 27593 100 102
Mbuti 16 46 53 2627 74 68
Melanesian 30 43 57 19602 46 44
Papuan 25 NA NA 17682 40 NA
Pima 23 28 94 24824 70 100
San 0 NA NA 0 44 NA
Sardinian 15 NA NA 8746 26 NA
Surui 83 70 89 32124 90 92
Yakut 10 59 70 17728 78 102
Yoruba NA 60 61 3950 NA 152

NA 5 missing data point.

a
From Chang et al., 1996.
b
From ALFRED. Sample size, N, given as total chromosomes sampled (2X individuals).

Criterion (AIC) and likelihood ratio tests (LRT) to select
the best model. We then assessed the significance of the
remaining independent variables through t-tests of the
slope residuals.
To assess how well our D2-derived correlation matrix
controlled for neutral genetic processes in our final
model, we also ran our model with a subset of the neu-
tral alleles used by Tishkoff et al., (2009) to calculate
genetic distances. We used the 405 microsatellite and
duplication loci provided online by Tishkoff et al., (2009)
that were variable for the populations in our study. The
full data set used by Tishkoff et al., (2009) for the D2 cal-
culation includes all alleles at 1327 loci. For each of the
405 loci, we selected one allele at random from individu-
als in the populations of our final model. We calculated
the frequency of these random alleles within each popu-
lation, applied the arcsine transform to the frequencies,
and then ran the same gls analyses for each allele both
with and without our correlation structure. When using
our correlation structure to control for neutral genetic
processes, ideally only 5% of these neutral loci would
show a significant association with migratory distance at
the 0.05 nominal level for significance. Because the
human populations are nonindependent data points, we
expected much higher rates of type 1 error (false posi-
tives) when the correlation structure was not used.
We also conducted an alternative test of NS alleles that
were deemed significant in our linear model with a corre-
lation structure. In this test, we ran a standard least
squares regression of migratory distance on the NS allele
frequency. If the observed association in fact arose by neu-
tral evolutionary processes, then the goodness of fit
between increases in migratory distance and NS allele fre-
quencies should be no stronger than that observed
between migratory distance and each of the 405 neutral
loci. One standard measure of goodness of fit in a regres-
sion model is the residual variation from the best fit
regression line. We calculated the residual variation for
the best fit model of each of the 405 neutral loci and mi-
gratory distance. We subtracted the absolute residual
deviations of each of the 405 neutral loci from the absolute
residuals from the NS allele frequencies. If the association
of NS allele frequencies and migratory distance were no
different than that expected under neutral evolution, then
on average about half of the neutral loci would have
larger residuals and half would have smaller residuals
than from the NS allele frequencies. If the association of
NS allele frequencies and migratory distance is more than
expected to arise from neutral processes, then the differ-
ence in residuals should be significantly negative, as the
NS residuals should be much smaller than typical of a
neutral allele—indicative of a tighter than expected asso-
ciation with migratory distance. We tested whether the
calculated differences in residuals were significantly nega-
tive with a one-sample Wilcoxon signed rank test.
RESULTS
We assessed which independent variables should be
included in the model for migratory distance through a
likelihood analysis of all the data points without any
missing values (Table 1). The best-supported model to
explain out-of-Africa migratory distance included the
composite 2R/7R DRD4 allele frequencies (Table 2). The
model with this single predictor variable had a better
AIC value than all other models. The next most complex
models that additionally included either the 120-bp
duplication, or the 2521 T SNP (see Table 2), did not
sufficiently improve the likelihood of the data under the
likelihood ratio test (P 5 0.193). Within the most com-
plex model that included all three predictor variables,
variance inflation factors for colinearity among the inde-
pendent variables were low (2R/7R 5 1.4, 120-bp dupli-
cation 5 1.3, 2521 T SNP 5 1.3; R package ‘‘Design’’,
Harrel, 2009). This indicates that allele frequencies at
the loci are largely independent of one another and the
alleles at alternative loci can be regarded as unlinked.
The t-test of slope residuals in our best model showed a
significant positive relationship between the DRD4 2 1 7
repeat elements and migratory distance (b 5 18687, P 5
0.018, two-tailed, 16 df; Fig. 2). Thus, the summed fre-
quency of 2 and 7 repeat element alleles was more associ-
ated with migratory distance than expected by chance.

American Journal of Physical Anthropology

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 386 L.J. MATTHEWS AND P.M. BUTLER
TABLE 2. Models ordered by AIC value
Model AIC log Likelihood
r 265.0 2129.5
r,d 265.3 2128.7
r,s 265.3 2128.7
r,s,d 266.3 2128.1
s 266.6 2130.3
d,s 268.6 2130.3
d 269.7 2131.9
r 5 DRD4 2 1 7 repeat frequency; d 5 120-bp duplication fre-
quency; s 5 2521 T SNP frequency.
Our tests of 405 neutral markers from Tishkoff et al.,
(2009) revealed that Type I error rates (false positives)
were somewhat elevated under the assumption that the
these markers were neutral with respect to selection pres-
sures correlated with migratory distance (observed error 5
0.12, expected error 5 0.05, z-test of proportions, P \
0.001). However, the slope estimates for the neutral loci
were not biased up or down, rather they were centered on
zero (mean b 5 99.3, t-test for difference from 0, P 5 0.833,
two-tailed). The correlation structure, however, greatly
improved the error rate when compared with gls models of
the neutral markers that lacked any correction for the non-
independence of the data points (observed error 5 0.457).
Although the mean of slope estimates from these reduced
gls models was more extreme (mean b 5 2241) than in the
models that included a correlation structure, the estimated
slopes again were not significantly different from zero
(P 5 0.84, two-tailed t-test). This can occur because the var-
iance of the slopes across the models without a correlation
structure (r2 5 6.05 3 108) was an order of magnitude
greater than the variance from slope estimates that used a
correlation structure (r2 5 8.94 3 107, Fig. 3).
Given this evidence for slightly inflated Type I error
(0.118 instead of 0.05), we corrected the P-value for the
significant 2R/7R result based on the observed inflation
of error in the neutral loci (a factor of 2.36). The 2R/7R
result remained significant after this correction (cor-
rected P 5 0.042, two-tailed).
The residual variation migratory distance and the
DRD4 2 1 7 allele frequency was significantly less than
that observed at the 405 neutral loci (median difference
in NS and neutral residuals 5 23040 km, Wilcoxon
signed rank test, P \ 0.001)
DISCUSSION
This study supports that higher frequencies of novelty
seeking (NS) 2 and 7 R alleles at DRD4 exon 3 are asso-
ciated with population histories of migration, which con-
firms previous findings specific to the 7R alleles (Chen et
al., 1999). Because the out-of-Africa migration proceeded
though a process of serial colonization, significant and
demographically shared population size bottlenecks have
occurred throughout the history of human migration
across the globe (Ramachandran et al., 2005; Li et al.,
2008; DeGiorgio et al., 2009; Deshpande et al., 2009;
Hunley et al., 2009). Thus, observed associations of mi-
gratory distance and allelic frequencies may result sim-
ply from cumulative effects of genetic drift that skew
allele frequencies more dramatically with increasing
distance from Africa (Klopfstein et al., 2006). Our study
is the first to demonstrate that neutral genetic processes
of drift and/or admixture, which produce statistical non-
independence among population data points, cannot
American Journal of Physical Anthropology
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Fig. 2. Migratory distance out-of-Africa versus summed 2R
and 7R DRD4 exon 3 alleles.
explain the association between higher frequencies of NS
alleles in DRD4 exon 3 and migration distance.
The gls statistical model with a correlation structure
to remove the effects of shared genetic ancestry, admix-
ture, and drift was largely successful. The low variance
inflation factors indicate that, with respect to the associa-
tion with migratory distance, the loci we tested can be
regarded as statistically unlinked. Although the correla-
tion structure we used did not reduce the rate of Type I
errors (false positive results) to their expected value of
0.05, it did substantially improve the Type I error rate.
This marked improvement of the error rate strongly sug-
gests, as proposed recently by other researchers, that sta-
tistical associations of allelic frequencies and ecological
traits caused by neutral evolution arise primarily from
the statistical nonindependence of human populations as
data points (Coop et al., 2010; Hancock et al., 2010).
Although derived alleles may show a tendency for higher
frequency outside Africa as a result of range expansion
(Klopfstein et al., 2006), the association in a linear model
of the frequencies of neutral alleles and distance from
Africa is not biased, as is reflected by the slopes of the
neutral loci being centered on zero. As shown in Figure 3,
even when we did not control for neutral population
structure, the estimated slopes were centered on zero.
The fact that using the correlation structure reduced the
slope variance while keeping it centered on zero shows
that the statistical nonindependence of intraspecific popu-
lations behaves much the same as the nonindependence
of interspecific datapoints in a phylogenetic context (Mar-
tins and Garland, 1991; Rendall and Di Fiore, 1995; Gar-
land et al., 2005; Rohlf, 2006; see also Coop et al., 2010).
Phylogenetic nonindependence is well-known to increase
the variance of slope estimates and thus inflate Type I
error, but it does not systematically bias these estimates;
they remain centered on zero (Rohlf, 2006).
This being said, our correlation structure did not
reduce the Type I error to its expected level. This can
occur if the correlation structure is somewhat miss-speci-
fied, in that it does not fully reflect the true process that
generated the nonindependence of the data (Martins and
 DRD4 ALLELES AND HUMAN MIGRATION PATTERNS 387

Fig. 3. Histograms of the slope estimates from migratory distance regressed on each of the 405 neutral alleles. The right panel
shows the estimates from a least squares model without any adjustment for population data point nonindependence. The left panel
shows slopes from the same loci from a model that used our correlation structure to control for the nonindependence that arises
from drift and admixture. Note that all estimates are centered on zero (i.e., they are neutral), and that the use of the correlation
structure greatly reduces the variance of the estimated slopes.

Garland, 1991; Rohlf, 2006). For example, one factor that
may have contributed to miss-specification could be that
the D2 distance does not allow for frequency differences
to arise from mutation or from types of selection that
may be difficult to detect within the largely neutral
markers (e.g., purifying and balancing selection). We
expect that future research will enable such miss-specifi-
cations to be corrected by using optimizing transforma-
tions of the correlation structure based on the dependant
and independent variables used in the linear model. Such
procedures have recently become routine in the context of
phylogenetic generalized least squares (Garland et al.,
2005). Some common transformations of the phylogenetic
correlation structure to reflect the inferred evolutionary
process include the multiplier lambda and the exponen-
tial transform rho, both of which control for reduced
phylogenetic signal, as well as the exponential transform
kappa that reflects speciational evolution (Grafen, 1989;
Pagel, 1999). The inflation of Type I error in our study is
only slightly greater than observed in phylogenetic simu-
lations before the implementation of such transformations
of correlation structures (Martins and Garland, 1991).
Developing appropriate transformations for intraspecific
correlation structures is beyond the scope of this study.
Thus, we elected to correct the P-value from the DRD4 2R/
7R result by the inflation of Type I error observed among
the neutral loci. The 2R/7R result remained significant af-
ter this correction. The residual variation from a least
squares regression of migratory distance on DRD4 2 1 7
repeat allele frequency was also significantly less than the
residual variation observed among neutral loci.
These results also suggest a strong warning to recent
publications by psychologists and others that have her-
alded the advent of a new ‘‘cultural neuroscience’’
research program (Chiao and Blizinsky, 2009; Chiao et
al., 2010; Goldman et al., 2010; Kitayama and Park,
2010; Kitayama and Tompson, 2010; Way and Lieber-
man, 2010). This program is based on finding associa-
tions of gene frequencies with behavioral and ecological
traits across human groups but does not incorporate the
effects of neutral genetic processes. One example, pub-
lished recently in AJPA, purported to find evidence of
NS alleles of DRD4 being linked to hunter-gatherer life-
styles among Native South Americans (Tovo-Rodrigues
et al., 2010). We agree that such a research program,
properly conducted, will yield important scientific
insights. None of these studies, however, have accounted
adequately for the nonindependence of human popula-
tions. Tovo-Rodrigues et al. (2010) made no accommoda-
tion for the role of neutral evolution of allele frequencies.
Both Chiao and Blizinsky (2009) and Way and Lieber-
man (2010) attempted to deal with nonindependence by
lumping data points into similar cultural regions.
Anthropologists have demonstrated, however, that such
approaches fail to remove the nonindependence that still
exists among the lumped cultural regions (Dow, 1984;
Mace and Pagel, 1994; Eff, 2004; Dow, 2007). Cultural
regions, furthermore, express cultural and linguistic
linkages among people and are not designed to reflect
the population genetic structure that determines the
nonindependence of allele frequencies.
In short, treating human populations as statistically
independent treats them as something that profoundly
they are not. Populations exhibit complex histories of
separation, drift, and admixture, and ignoring this is a
serious methodological error that will inflate Type I
error, reduce statistical power, and thereby produce
many erroneous conclusions.
We found no support for an association between migra-
tory distance and two promoter region loci at DRD4 that
alter the expression levels of this gene and mediate NS
as measured through psychological surveys. Although
the inclusion of more data points may change this result,
these negative results are not likely the product of low
power, as the power to detect the observed simple Pear-
son correlation of the 2R/7R frequencies and migratory
distance was 0.96 (R package ‘‘pwr’’, Champely, 2009).
These promoter loci may not, in fact, correlate with dis-
persal distance if their phenotypic effects are more
greatly influenced by gene X gene and gene X environ-

American Journal of Physical Anthropology

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 388 L.J. MATTHEWS AND P.M. BUTLER
ment interactions than are the effects of the 2R/7R allele
located within exon 3. Given that convergent phenotypic
change can often reflect nonconvergent underlying
genetic changes (Tishkoff et al., 2006), studies like ours
may be able to detect only those loci with highly robust
phenotypic effects. Though few, such loci are important
to medical intervention, because doctors in their clinical
practice are able at present to test for genetic variation
at target alleles, whereas diagnoses based on complete
individual genotyping may still lie far in the future.
We hope our positive results for DRD4 exon 3 will mo-
tivate more research on this topic that will include sam-
pling of both functional and neutral markers from human
populations of known provenience. Unfortunately, no
such sample exists at present for DRD4 alleles nor for
other loci that would be good candidates for selection by
migration effects, specifically the serotonin transporter 5-
HTT and monoamine oxidase A (Nakamura et al., 2000;
Lea and Chambers, 2007). The paucity of data has health
consequences because all these loci are linked to a variety
of psychopathologies (Caspi et al., 2003; Canli and Lesch,
2007; Guo et al., 2008). Population-level sampling and
the resultant better understanding of the functionality of
these loci would likely result in better treatment out-
comes for patients than do the current common practices
of genetic profiling based on broad ethnic labels. Such
labels confound genetic and cultural forms of variation
(e.g., Munafò et al., 2003; Chiao and Blizinsky, 2009).
Genetic and cultural identities frequently overlap, but
just as frequently they are disjoint or intersect in com-
plex ways (Risch et al., 2002; Hunley et al., 2009; Long et
al., 2009). We can only understand these interactions if
we first sample both genetic and cultural systems in
ways that are most appropriate to their particular modes
of inheritance and distributions of variation.
In summary, we found significant support for dispersal
distance during the out-of-Africa migrations being asso-
ciated with higher frequencies of NS alleles in DRD4
exon 3. Given the findings of Chen et al., (1999) that
these alleles did not induce migration within a single
generation, the likely best explanation of our observed
association is that NS individuals experience higher fit-
ness during migrations. Migration distance, thus, effec-
tively selects for NS phenotypes and associated alleles
even though NS individuals may be no more prone to
migrate than are other individuals. The atypical pattern
of variation in the 7R VNTR of DRD4 exon 3 suggests it
emerged as a recent mutational event
40–50 Ka BP
(Ding et al., 2002) and is under positive Darwinian selec-
tion (Wang et al., 2004). This coalescence time is roughly
coincident with the first observed wave of migration of
modern humans from Africa and into Europe and Asia.
The coalescence time, genetic signature of selection, and
observed association with migratory distance, are all
highly consistent with the hypothesis that increased fre-
quencies of 2R and 7R exon 3 VNTRs have been selected
by repeated generations of migration.
ACKNOWLEDGMENTS
The authors report no conflicts of interest. The authors
thank Patrick McNamara, Charles Nunn, the lab group of
Richard Wrangham, and especially Amanda Lobell for help-
ful discussions on this research. Amanda Lobell and Jason
Hodgson provided helpful comments on the manuscript and
analysis. The authors also thank Sarah Tishkoff for provid-
ing the D2 distances. The authors thank the Editor-in-Chief,
American Journal of Physical Anthropology
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
the Associate Editor, and the two anonymous reviewers for
their helpful comments on the manuscript.
LITERATURE CITED
Asghari V, Sanyal S, Buchwaldt S, Paterson A, Jovanovic V,
Van Tol HH. 1995. Modulation of intercellular cyclic AMP
levels by different human dopamine D4 receptor variants.
J Neurochem 65:1157–1165.
Canli T, Lesch K-P. 2007. Long story short: the serotonin trans-
porter in emotion regulation and social cognition. Nat Neuro-
sci 10:1103–1109.
Caspi A, McClay J, Moffitt TE, Mill J, Martin J, Braithwaite A,
Poulton R. 2003. Influence of life stress on depression: modera-
tion by a polymorphism in the 5-HTT gene. Science 80:504–517.
Champely, S. 2009. pwr: Basic functions for power analysis. R
package version 1.1.1. Available at: http://CRAN.R-project.org/
package5pwr.
Chang FM, Kidd JR, Livak KJ, Pakstis AJ, Kidd KK. 1996. The
worldwide distribution of allele frequencies at the human
dopamine D4 receptor locus. Hum Genet 98:91–101.
Chen C, Burton M, Greenberger E, Dmitrieva J. 1999. Popula-
tion migration and the variation of dopamine D4 receptor
(DRD4) allele frequencies around the globe. Evol Hum Behav
20:309–324.
Chiao JY, Blizinsky KD. 2009. Culture-gene coevolution of indi-
vidualism-collectivism and the serotonin transporter gene.
Proc R Soc B 277:529–537.
Chiao JY, Hariri AR, Harada T, Mano Y, Sadato N, Parrish TB,
Iidaka T. 2010. Theory and methods in cultural neuroscience.
Soc Cogn Affect Neurosci 5:356–361.
Cloninger CR, Przybeck TR, Svrakic DM. 1991 The tridimen-
sional personality questionnaire: U.S. normative data. Psychol
Rep 69(3Part 1):1047–1057.
Coop G, Witonsky D, Di Rienzo A, Pritchard JK. 2010. Using
environmental correlations to identify loci underlying local
adaptation. Genetics 185:1411–1423.
DeGiorgio M, Jakobsson M, Rosenberg NA. 2009. Out of Africa:
modern human origins special feature: explaining worldwide
patterns of human genetic variation using a coalescent-based
serial founder model of migration outward from Africa. Proc
Natl Acad Sci USA 106:16057–16062.
Deshpande O, Batzoglou S, Feldman MW, Cavalli-Sforza LL.
2009. A serial founder effect model for human settlement out
of Africa. Proc Biol Sci 276:291–300.
Ding YC, Chi HC, Grady DL, Morishima A, Kidd JR, Kidd KK,
Flodman P, Spence MA, Schuck S, Swanson JM, Zhang YP,
Moyzis RK. 2002. Evidence of positive selection acting at the
human dopamine receptor D4 gene locus. Proc Natl Acad Sci
USA 99:309–314.
Dow MM. 2007. Galton’s problem as multiple network autocor-
relation effects–cultural trait transmission and ecological
constraint. Cross-Cultural Res 41:336–363.
Dow MM, Burton ML, White DR, Reitz KP. 1984. Galton’s prob-
lem as network autocorrelation. Am Ethnol 11:754–770.
Dulawa SC, Grandy DK, Low ML, Paulus MO, Geyer MA. 1999.
Dopamine D4 receptor-knock-out mice exhibit reduced explo-
ration of novel stimuli. J Neurosci 19:9550–9556.
Ebstein R, Novick O, Umansky R, Priel B, Osher Y, Blaine D,
et al. 1996. Dopamine D4 receptor (DRD4) exon III polymor-
phism associated with the human personality trait of novelty
seeking. Nat Genet 12:78–80.
Eff EA. 2004. Does Mr. Galton still have a problem? Autocorrelation
in the standard cross-cultural sample. World Cult 15:153–170.
Garland T Jr., Bennet AF, Rezende EL. 2005. Phylogenetic
approaches in comparative physiology. J Exp Biol 208:3015–3035.
Garland T Jr., Ives AR. 2000. Using the past to predict the pres-
ent: confidence intervals for regression equations in phyloge-
netic comparative methods. Am Nat 155:346–364.
Goldman N, Glei DA, Lin Y-H, Weinstein M. 2010. The sero-
tonin transporter polymorphism (5-HTTLPR): allelic variation
and links with depressive symptoms. Depress Anxiety
27:260–269.
 DRD4 ALLELES AND HUMAN MIGRATION PATTERNS
Google Inc. Google Earth (Version 5.0.11733.9347).( 2009) Moun-
tain View, CA: Google Inc. Available at: http://earth.google.com/.
Grady DL, Chi HC, Ding YC, Smith M, Wang E, Schuck S, Flod-
man P, Spence MA, Swanson JM, Moyzis RK. 2003. High preva-
lence of rare dopamine receptor D4 alleles in children diagnosed
with attention-deficit hyperactivity disorder. Mol Psych 8:536–545.
Grafen A. 1989. The phylogenetic regression. Phil Trans R Soc
Lond B 326:119–157.
Guo G, Ou XM, Roettger M, Shih JC. 2008. The VNTR 2 repeat
in MAOA and delinquent behavior in adolescence and young
adulthood: associations and MAOA promoter activity. Eur J
Hum Genet 16:626–634.
Hancock AM, Witonsky DB, Ehler E, and 8 additional authors.
2010. Human adaptations to diet, subsistence, and ecoregion
are due to subtle shifts in allele frequency. Proc Natl Acad Sci
USA 107:8924–8930.
Harrell FE Jr. 2009. Design: Design Package. R package version
2.3–0. Available at http://CRAN.R-project.org/package5Design.
Hunley KL, Healy ME, Long JC. 2009. The global pattern of
gene identity variation reveals a history of long-range migra-
tions, bottlenecks, and local mate exchange: implications for
biological race. Am J Phys Anthropol 139:35–46.
Kamakura S, Iwaki A, Matsumoto M, Fukumaki Y. 1997. Clon-
ing and characterization of the 50 flanking region of the
human dopamine D4 receptor gene. Biochem Biophys Res
Commun 235:321–326.
Kitayama S, Park J. 2010. Cultural neuroscience of the self:
understanding the social grounding of the brain. Soc Cogn
Affec Neurosci 5:111–129.
Kitayama S, Tompson S. 2010. Envisioning the future of cul-
tural neuroscience. Asian J Social Psychol 13:92–101.
Klopfstein S, Currat M, Excoffier L. 2006. The fate of mutations surf-
ing on the wave of a range expansion. Mol Biol Evol 23:482–490.
Lea R, Chambers G. 2007. Monoamine oxidase, addiction, and
the ‘‘warrior’’ gene hypothesis. N Z Med J 120:1250.
Li JZ, Absher DM, Tang H, Southwick AM, Casto AM, Ramachan-
dran S, Cann HM, Barsh GS, Feldman M, Cavalli-Sforza LL,
Myers RM. 2008. Worldwide human relationships inferred from
genome-wide patterns of variation. Science 319:1100–1104.
Liu H, Prugnolle F, Manica A, Balloux F. 2006. A geographically
explicit genetic model of worldwide human-settlement history.
Am J Hum Genet 79:230–237.
Long JC, Li J, Healy ME. 2009. Human DNA sequences: more
variation and less race. Am J Phys Anthropol 139:23–34.
Lowe N, Kirley A, Mullins C, Fitzgerald M, Gill M, Hawi Z. 2004.
Multiple marker analysis at the promoter region of the DRD4
gene and ADHD: evidence of linkage and association with the
SNP -616. Am J Med Genet B Neuropsych Genetics 131:33–37.
Mace R, Pagel M. 1994. The comparative method in anthropol-
ogy. Curr Anthropol 35:349–357.
Martins EP, Garland T Jr. 1991. Phylogenetic analyses of the
correlated evolution of continuous characters: a simulation
study. Evolution 45:534–557.
Mellars P. 2006. Going East: new genetic and archaeological
perspectives on the modern human colonization of Eurasia.
Science 313:796–800.
Munafò MR, Clark TG, Moore LR, Payne E, Walton R, Flint J.
2003. Genetic polymorphisms and personality in healthy adults:
as systematic review and meta-analysis. Mol Psych 8:471–484.
Munafò M, Yalcin B, Willis-Owen S, Flint J. 2008. Association
of the dopamine D4 receptor (DRD4) gene and approach-
related personality traits: meta-analysis and new data. Biol
Psych 63:197–206.
Nakamura M, Ueno S, Sano A, Tanabe H. 2000. The human se-
rotonin transporter gene linked polymorphism (5-HTTLPR)
shows ten novel allelic variants. Mol Psych 5:32–38.
Oldenhof J, Vickery R, Anafi M, Oak J, Ray A, Schoots O, Paw-
son T, von Zastrow M, Van Tol HH. 1998. SH3-binding domains
in the dopamine D4 receptor. Biochemistry 37:15726–15736.
Osier MV, Cheung KH, Kidd JR, Pakstis AJ, Miller PL, Kidd
KK. 2002. ALFRED: an allele frequency database for anthro-
pology. Am J Phys Anthropol 119:77–83.
Pagel M. 1999. Inferring the historical patterns of biological
evolution. Nature 401:877–884.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
389
Pinheiro J, Bates D. 2000. Mixed-effects models in S and
S-PLUS. New York: Springer Verlag.
Pinheiro J, Bates D, DebRoy S, Sarkar D, andthe R Core team.
2009. nlme: linear and nonlinear mixed effects models. R
package version 3.1–96.
Prugnolle F, Manica A, Bolloux F. 2005. Geography predicts neutral
genetic diversity of human populations. Curr Biol 15:R159–R160.
Rajeevan H, Osier MV, Cheung KH, Deng H, Druskin L, Hein-
zen R, Kidd JR, Stein S, Pakstis AJ, Tosches NP, Yeh CC,
Miller PL, Kidd KK. 2003. ALFRED–the ALlele FREquency
Database–update. Nucleic Acids Res 31:270–271.
Ramachandran S, Deshpande O, Roseman CC, Rosenberg NA,
Feldman MW, Cavalli-Sforza LL. 2005. Support from the rela-
tionship of genetic and geographic distance in human popula-
tions for a serial founder effect originating in Africa. Proc
Natl Acad Sci USA 102:15942–15947.
Rendall D, Di Fiore A. 1995. The road less traveled: phyloge-
netic perspectives in primatology. Evol Anthropol 4:43–52.
Reist C, Ozdemir V, Wang E, Hashemzadeh M, Mee S, Moyzis
R. 2007. Novelty seeking and the dopamine D4 receptor gene
(DRD4) revisited in Asians: haplotype characterization and
relevance of the 2-repeat allele. Am J Med Genet Part B Neu-
ropsychiatric Genet 144B:453–457.
Reynolds J, Weir BS, Cockerham CC. 1983. Estimation of the
coancestry coefficient: basis for a short-term genetic distance.
Genetics 105:767–779.
Risch N, Burchard E, Ziv E, Tang H. 2002. Categorization of
humans in biomedical research: genes, race and disease. Ge-
nome Biol 3: comment 2007.
Rohlf FJ. 2006. A comment on phylogenetic correction. Evolu-
tion 60:1509–1515.
Seaman M, Fisher J, Chang F, Kidd KK. 1999. Tandem duplication
polymorphism upstream of the dopamine D4 receptor gene
(DRD4). Am J Med Genet. Part B Neuropsych Genet 88:705–709.
Sokal RR, Rohlf FJ. 1995. Biometry: the principles and practice
of statistics in biological research. New York: W.H. Freeman
and Company.
Tishkoff SA, Reed FA, Friedlaender FR, Ehret C, Ranciaro A,
Froment A, Hirbe JB, Awomoyi AA, Bodo JM, Coumbo O,
Ibrahim M, Juma AT, Kotze MJ, Lema G, Moore JH, Morten-
sen H, Nyambo TB, Omar SA, Powell K, Pretorius GS, Smith
MW, Thera MA, Wambebe C, Weber JL, Williams SM. 2009.
The genetic structure and history of Africans and African
Americans. Science 324:1035–1044.
Tishkoff SA, Reed FA, Ranciaro A, Voight BF, Babbitt CC, Sil-
verman JS, Powell K, Mortensen HM, Hirbo JB, Osman M,
Ibrahim M, Omar SA, Lema G, Nyambo TB, Ghori J, Bump-
stead S, Pritchard JK, Wray GA, Deloukas P. 2006. Conver-
gent adaptation of human lactase persistence in Africa and
Europe. Nature Genet 39:31–40.
Tomitaka M, Tomitaka S, Otuka Y, Kim K, Matuki H, Sakamoto
K, Tanaka A. 1999. Association between novelty seeking and
dopamine receptor D4 (DRD4) exon III polymorphism in Japa-
nese subjects. Am J Med Genet 88:469–471.
Tovo-Rodrigues L, Callegari-Jacques SM, Petzl-Erler ML,
Tsuneto L, Salzano FM, Hutz MH. 2010. Dopamine receptor
D4 allele distribution in Amerindians: a reflection of past
behavior differences? Am J Phys Anthropol 143:458–464.
Tsai S, Hong C, Younger W, Chen T. 2004. Association study of
catechol-o-methyltransferase gene and dopamine D4 receptor
gene polymorphisms and personality traits in healthy young
Chinese females. Neuropsychobiology 50:153–156.
Van Tol H, Bunzow JR, Guan HC, Sunahara RK, Seeman P,
Niznik HB, Civelli O. 1991. Cloning of the gene for a human
dopamine D4 receptor with high affinity for the antipsychotic
clozapine. Nature 350:610–614.
Wang E, Ding Y-C, Flodman P, Kidd JR, Kidd KK, Grady DL,
Ryder OA, Spence MA, Swanson JM, Moyzis RK. 2004. The
genetic architecture of selection at the human dopamine re-
ceptor D4 (DRD4) gene locus. Am J Hum Genet 74:931–944.
Way BM, Lieberman MD. 2010. Is there a genetic contribution
to cultural differences? Collectivism, individualism and
genetic markers of social sensitivity. Soc Cogn Affect Neurosci
5:203–211.
American Journal of Physical Anthropology
 FROM THE ACADEMY: SACKLER LECTURE
Addiction: Beyond dopamine reward circuitry
Nora D. Volkowa,b,1, Gene-Jack Wangc, Joanna S. Fowlerc, Dardo Tomasib, and Frank Telangb
a
National Institute on Drug Abuse, National Institutes of Health, Bethesda, MD 20892; bNational Institute on Alcohol Abuse and Alcoholism,
National Institutes of Health, Bethesda, MD 20892; and cMedical Department, Brookhaven National Laboratory, Upton, NY 11973

Edited by Donald W. Pfaff, The Rockefeller University, New York, NY, and approved November 9, 2010 (received for review August 31, 2010)

Dopamine (DA) is considered crucial for the rewarding effects of drugs of abuse, but its role in addiction is much less clear. This review
focuses on studies that used PET to characterize the brain DA system in addicted subjects. These studies have corroborated in humans the
relevance of drug-induced fast DA increases in striatum [including nucleus accumbens (NAc)] in their rewarding effects but have
unexpectedly shown that in addicted subjects, drug-induced DA increases (as well as their subjective reinforcing effects) are markedly
blunted compared with controls. In contrast, addicted subjects show significant DA increases in striatum in response to drug-conditioned
cues that are associated with self-reports of drug craving and appear to be of a greater magnitude than the DA responses to the drug. We
postulate that the discrepancy between the expectation for the drug effects (conditioned responses) and the blunted pharmacological
effects maintains drug taking in an attempt to achieve the expected reward. Also, whether tested during early or protracted withdrawal,
addicted subjects show lower levels of D2 receptors in striatum (including NAc), which are associated with decreases in baseline activity in
frontal brain regions implicated in salience attribution (orbitofrontal cortex) and inhibitory control (anterior cingulate gyrus), whose
disruption results in compulsivity and impulsivity. These results point to an imbalance between dopaminergic circuits that underlie reward
and conditioning and those that underlie executive function (emotional control and decision making), which we postulate contributes to the
compulsive drug use and loss of control in addiction.

prefrontal cortex | dorsal striatum | substance use disorders | stimulant drugs | brain imaging

D
rugs of abuse (including alco-
hol) are inherently rewarding,
which is why they are consumed
by humans or self-administered
by laboratory animals (1). Only a small
percentage of individuals exposed to drugs
will become addicted, that is, shift from
controlled drug use to compulsive drug
use with loss of control over intake despite
adverse consequences, however (2). Fac-
tors that determine who becomes addicted
include genetic (50% of risk), develop-
mental (risk is higher in adolescence), and
environmental (e.g., drug access, stress)
factors (2).
The mesolimbic dopamine (DA) path-
way [DA cells in ventral tegmental area
(VTA) projecting into nucleus accumbens
(NAc)] seems to be crucial for drug reward
(1). Other DA pathways [mesostriatal
(DA cells in substantia nigra {SN} pro-
jecting into dorsal striatum) and meso-
cortical (DA cells in VTA projecting into
frontal cortex)] are now also recognized to
contribute to drug reward and addiction
(1). The mode of DA cell firing also dif-
ferently modulates the rewarding and
conditioning effects, of drugs (predomi-
nantly phasic DA cell firing) vs. the
changes in executive function that occur in
addiction (predominantly tonic DA cell
firing) (3, 4).
This review summarizes studies that
used PET to evaluate DA’s role in drug
reward and addiction. These findings show
that addiction affects not only the DA
reward circuit but circuits involved with
conditioning/habits, motivation, and exec-
utive functions (inhibitory control, sa-
lience attribution, and decision making).
Other neurotransmitters (and neuropep-
Downloaded by guest on February 17, 2022
roadaptations from repeated drug use
(i.e., glutamate, opioids, GABA, cortico-
tropic-releasing factor). These are not
discussed here (except for glutamate), but
several reviews address them (5, 6).
DA and Acute Drug Reward
All drugs that can lead to addiction in-
crease DA in NAc, which is achieved
through their interaction with different
molecular targets by the various drug
classes (6) (Table 1). In humans, PET
studies have shown that several drugs
[stimulants (7, 8), nicotine (9), alcohol
(10), and marijuana (11)] increase DA in
dorsal and ventral striatum (where NAc is
located). These studies used a radiotracer
that binds to DA D2 receptors (D2Rs)
but only when these are not occupied by
DA (i.e., [11C]raclopride). By comparing
binding after placebo and after the drug,
these studies estimate the decreases in D2R
availability induced by the drug, which are
proportional to DA increases (12). Most
studies have reported that participants who
display the greatest DA increases with the
drug also report the most intense “high”
or “euphoria” (reviewed ref. 13).
PET studies have also shown that the
speed with which a drug enters and leaves
the brain (pharmacokinetic profile) is
crucial for its reinforcing effects. Specifi-
cally, PET studies of brain pharmacoki-
netics of drugs labeled with positron
emitters show that peak levels in human
brain are reached within 10 min after i.v.
administration and that this fast drug
uptake is associated with the high (13)
(Fig. 1). Indeed, for an equivalent level of
cocaine reaching the brain (assessed as
equivalent level of DA transporter block-
elicited a more intense high than when it
entered the brain more slowly (snorted)
(14). This is consistent with preclinical
studies showing that the faster the drug’s
entry into the brain, the stronger are its
reinforcing effects (15). This probably re-
flects the fact that abrupt and large DA
increases triggered by drugs mimic the fast
and large DA increases associated with
phasic DA firing that are associated in the
brain with conveying information about
reward and saliency (16).
Drug-induced DA increases in NAc
occur in nonaddicted as well as addicted
subjects, which raises the question of how
they relate to addiction.
To start with, there is increasing evidence
that DA’s role in reinforcement is more
complex than just coding for reward per se
(hedonic pleasure) and that stimuli that
induce fast and large DA increases also
trigger conditioned responses and elicit in-
centive motivation to procure them (17).
Through conditioning, a neutral stimulus
that is linked with the reinforcer (i.e., nat-
ural reinforcer, drug) acquires the ability by
This article arises from the Sackler Lecture, “Addiction:
Conflict Between Brain Circuits,” presented by Nora Volkow
on June 11 at the AAAS Auditorium in Washington, DC.
The lecture opened the Arthur M. Sackler Colloquium of
the National Academy of Sciences, on “Quantification of
Behavior.” The complete program and audio files of most
presentations are available on the NAS Web site at www.
nasonline.org/quantification. See all papers from this col-
loquium in supplement 3 of volume 108.
Author contributions: N.D.V., G.-J.W., and J.S.F. designed
research; N.D.V., G.-J.W., J.S.F., D.T., and F.T. performed
research; N.D.V., G.-J.W., J.S.F., D.T., and F.T. analyzed data;
and N.D.V. wrote the paper.
The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

tides) are involved with drug reward (i.e., ade), when cocaine entered the brain 1
To whom correspondence should be addressed. E-mail:
cannabinoids, opioids) and with the neu- rapidly (smoked and i.v. administration), it nvolkow@nida.nih.gov.

www.pnas.org/cgi/doi/10.1073/pnas.1010654108 PNAS | September 13, 2011 | vol. 108 | no. 37 | 15037–15042

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Table 1. Main classes of drugs of abuse, their main molecular targets, and some of the mechanism(s) by which they increase DA in NAc (44)
Drug Target Mechanism for DA increases

Stimulant drugs (cocaine,
amphetamine,
methamphetamine)
Opioids (heroin, opioid
analgesics)
Nicotine (cigarettes and other
tobacco products)
Alcohol and inhalants
Cannabinoids (marihuana)
DAT
MOR
Nicotinic receptors
(predominantly α4β2 subtype)
Multiple targets, including
GABA and glutamate
receptors
Cannabinoid CB1 receptors
Blocks DAT on the terminals of DA projecting neurons from VTA to NAc
(cocaine) or releases DA from the vesicles of DA terminals
(methamphetamine, amphetamine)
Disinhibits VTA DA neurons by inhibiting GABA interneurons that contain MOR
in the VTA or directly activates NAc neurons that contain MOR
Directly activates VTA DA neurons by stimulating their nicotine receptors and
indirectly activates them by stimulating the nicotine receptors in
glutamatergic terminals to VTA DA neurons
Facilitates GABAergic neurotransmission, which may disinhibit VTA DA neurons
from GABA interneurons or may inhibit glutamate terminals that regulate
DA release in Nac
Regulates dopaminergic signaling through CB1R in NAc neurons and in GABA
and glutamate terminals to NAc

DAT, DA transporter; MOR, μ-opioid receptor.

itself to increase DA in striatum (including inforcer (food) when exposed to a condi- to which a similar process occurs in response
NAc) in anticipation of the reward, and tioned stimulus (CS), the DA neurons stop to drugs of abuse is unclear, however, be-
this is associated with drug seeking (17). responding to the primary reinforcer and, cause drugs, through their pharmacological
In animals trained to expect a natural re- instead, respond to the CS (16). The extent actions, can directly activate DA neurons

Fig. 1. Pharmacokinetics of stimulant drugs in the human

brain and relationship to the “high.” (A) Axial brain im-
ages of the distribution of [11C]cocaine, [11C]MP, and [11C]
methamphetamine at different times (minutes) after its
administration. (B) Time activity curve for the concentra-
tion of [11C]cocaine, [11C]MP, and [11C]methamphetamine
in striatum alongside the temporal course for the high
experienced after i.v. administration of these drugs. Note
that the fast brain uptake for these drugs corresponds
to the temporal course of the high, which suggests that
the high is associated with the “rate of DA increases.” In
contrast, their clearance shows a correspondence with the
high for cocaine and for methamphetamine but not for
MP. The difference between MP and cocaine may reflect
the differences in their rate of clearance and that between
MP and methamphetamine may reflect their different
mechanisms of action. Specifically, because MP and co-
caine increase DA by blocking DA transporters, the DA
increases are terminated by autoreceptor activation,
which inhibits DA release. For cocaine, its fast rate of
clearance (20-min half-life in brain) results in short-lasting
autoreceptor activation, whereas for MP, its slower clear-
ance (60-min half-life) results in long-lasting inhibition of
DA release by autoreceptors, which terminates the high
even though the drug is still in the brain. In contrast,
methamphetamine, which is a DA releaser, is not sensitive
to autoreceptor activation; thus, DA increases are not
terminated by this mechanism, accounting for the longer
lasting duration of the high. Modified from ref. 18.
Downloaded by guest on February 17, 2022

Copyright (1995) American Medical Association. All rights

reserved. Reprinted from ref. 43. Copyright (2008), with
permission from Elsevier.

15038 | www.pnas.org/cgi/doi/10.1073/pnas.1010654108 Volkow et al.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 (i.e., nicotine) or increase DA release (i.e.,
amphetamine) (Table 1).
To answer this, we compared DA in-
creases induced by the stimulant drug
methylphenidate (MP) between cocaine-
addicted subjects and controls. Like co-
caine, MP increases DA by blocking DA
transporters; both drugs have a similar
distribution in human brain and have
similar behavioral effects when given
i.v. (18). In detoxified cocaine-addicted
subjects (n = 20, detoxified 3–6 wk),
we showed marked attenuation of MP-
induced DA increases in striatum (50%
lower) and of the increases in self-reports
of high, compared with non–drug-abusing
controls (n = 23). Similar findings were Fig. 2. DA changes induced by i.v. MP in controls and in active cocaine-addicted subjects. (A) Average
reported after administration of i.v. am- nondisplaceable biding potential (BPND) images of [11C]raclopride in active cocaine-addicted subjects (n =
phetamine (another stimulant drug) in 19) and in controls (n = 24) tested after placebo and after i.v. MP. (B) D2R availability (BPND) in caudate,
putamen, and ventral striatum after placebo (blue) and after MP (red) in controls and in cocaine-
recently detoxified cocaine abusers (de-
addicted subjects. MP reduced D2R in controls but not in cocaine-addicted subjects. Note that cocaine
toxified 2 wk), who also showed decreased abusers show both decreases in baseline striatal D2R availability (placebo measure) and decreases in DA
DA release in striatum and attenuated release when given i.v. MP (measured as decreases in D2R availability from baseline). Although one could
self-reports of euphoria (19). Because question the extent to which the low striatal D2R availability in cocaine-addicted subject limits the ability
a confound in these studies was the pos- to detect further decreases from MP, the fact that cocaine-addicted subjects show reductions in D2R
sibility that drug withdrawal accounted for availability when exposed to cocaine cues (Fig. 3) indicates that the attenuated effects of MP on [11C]
the attenuated DA responses, we repeated raclopride binding reflect decreased DA release.
this study in active cocaine-addicted sub-
jects (n = 19, nondetoxified) (20). In ac- largest cue-induced DA increases in dorsal drug, this suggests that conditioned re-
tive cocaine abusers, MP-induced DA striatum also had the highest scores on sponses may drive the DA signaling that
changes did not differ from placebo and measures of addiction severity. Because triggers and maintains the motivation to
the DA changes were 80% lower than in the dorsal striatum is implicated in habit take the drug. To the extent that the drug
controls (n = 24); the self-reports of high learning, this association is likely to reflect (even when its DA-enhancing effects are
were also attenuated (Fig. 2). Marked the strengthening of habits as chronicity of attenuated) predicts reward, the act of its
blunting of striatal DA increases second- addiction progresses. This suggests that administration (e.g., injection, smoking)
ary to MP or to amphetamine has also a basic disruption in addiction might be may become a conditioned cue and, as
been documented in detoxified alcoholics DA-triggered conditioned responses that such, may increase DA. Thus, although
(reviewed in ref. 13). If, as is currently result in habits leading to compulsive drug drugs may initially lead to DA release in
believed, drug-induced DA increases in consumption. Inasmuch as in cocaine- striatum (signaling reward), with repeated
NAc underlie drug reward, why would addicted subjects, the DA increases trig- administration and as habits develop, there
cocaine-addicted subjects, who show gered by conditioned cues appear to be appears to be a shift in the DA increases
a marked attenuation of drug-induced DA larger than those produced by a stimulant from the drug to the CS, as reported for
increases, compulsively take the drug?

DA and Conditioning to Drug Cues

The explanation may arise from the process
of conditioning, which is one of the initial
neuroadaptations that follow exposure to
drugs and involves DA phasic signaling
(predominantly D1Rs) and synaptic changes
in NMDA and AMPA receptors (modulated
by glutamate) (21, 22). These conditioned
responses are believed to underlie the in-
tense desire for the drug (craving) and the
compulsive use that occurs when addicted
subjects are exposed to drug cues.
To assess if drug conditioned cues would
increase DA in cocaine-addicted subjects,
we tested active cocaine-addicted sub-
jects (n = 18) when subjects watched a
neutral video (nature scenes) vs. when
they watched a cocaine-cue video (scenes
of subjects procuring and smoking co-
Fig. 3. DA changes induced by conditioned cues in active cocaine-addicted subjects. (A) Average non-
caine) (23). Cocaine cues significantly in-
displaceable biding potential (BPND) images of [11C]raclopride in cocaine-addicted subjects (n = 17) tested
creased DA in dorsal striatum, and the while viewing a neutral video (nature scenes) and while viewing a cocaine-cues video (subjects admin-
magnitude of this increase was correlated istering cocaine). (B) D2R availability (BPND) in caudate, putamen, and ventral striatum for the neutral
with the subjective experience of craving
Downloaded by guest on February 17, 2022

video (blue) and the cocaine-cues video (red). The cocaine cues decreased D2R in caudate and putamen.
(Fig. 3); similar findings were reported by (C) Correlations between changes in D2R (reflecting DA increases) and self-reports of cocaine craving
another laboratory (24). Subjects with the induced by the cocaine-cues video. Modified from ref. 23.

Volkow et al. PNAS | September 13, 2011 | vol. 108 | no. 37 | 15039

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 PET studies have shown that addicted
subjects have significant reductions in D2R
availability in striatum that persist months
after protracted detoxification (reviewed
in ref. 13). To investigate the functional
significance of the striatal D2R reductions,
we have assessed their relationship to
baseline measures of brain glucose me-
tabolism (marker of brain function). We
have shown that reductions in striatal D2R
are associated with decreased metabolism
in orbitofrontal cortex (OFC), anterior
cingulate gyrus (ACC), and dorsolateral
prefrontal cortex (DLPFC) (26–28) (Fig.
4). Because OFC, CG, and DLPFC are
involved with salience attribution, inhi-
bitory control/emotion regulation, and
decision making, we had postulated that
Fig. 4. Correlations between striatal D2R availability and metabolism in prefrontal brain regions. (A) Axial their improper regulation by DA in ad-
brain images for a control and for a cocaine-addicted subject for baseline images of D2R availability in dicted subjects could underlie the en-
striatum (obtained with [11C]raclopride) and of brain glucose metabolism in OFC (obtained with [18F]FDG). hanced motivational value of drugs in their
(B) Correlations between striatal D2R and metabolism in OFC in cocaine-addicted and methamphetamine- behavior and loss of control over drug in-
addicted subjects. Reprinted from ref. 13, Copyright (2009), with permission from Elsevier.
take (29). In addition, because impair-
ments in OFC and ACC are associated
with compulsive behaviors and impulsivity,
natural reinforcers (16). Preclinical studies DA and Inhibitory Control in Addiction we postulated that DA’s impaired modu-
have revealed that glutamatergic projec- The capacity to inhibit prepotent responses lation of these regions could underlie the
tions from prefrontal cortex into VTA/SN is likely to contribute to an individual’s ability compulsive and impulsive drug intake seen
and NAc mediate these conditioned re- to restrain from taking drugs, and thus his or in addiction (30, 31). Indeed, in metham-
sponses (5). her vulnerability to addiction (25). phetamine abusers, low striatal D2R was

Fig. 5. Model proposing a network of interacting circuits underlying addiction: reward (nucleus accumbens, VTA, and ventral pallidum), conditioning/memory
(amygdala, medial OFC for attribution of saliency, hippocampus, and dorsal striatum for habits), executive control (DLPFC, ACC, inferior frontal cortex, and
lateral OFC), and motivation/drive (medial OFC for attribution of saliency, ventral ACC, VTA, SN, dorsal striatum, and motor cortex). Nac, nucleus accumbens.
(A) When these circuits are balanced, this results in proper inhibitory control and decision making. (B) During addiction, when the enhanced expectation value
of the drug in the reward, motivation, and memory circuits overcomes the control circuit, this favors a positive-feedback loop initiated by the consumption of
the drug and perpetuated by the enhanced activation of the motivation/drive and memory circuits. These circuits also interact with circuits involved in mood
regulation, including stress reactivity (which involves the amygdala and hypothalamus) and interoception (which involves the insula and ACC and contributes
Downloaded by guest on February 17, 2022

to awareness of craving). Several neurotransmitters are implicated in these neuroadaptations, including glutamate, GABA, norepinephrine, corticotropic-
releasing factor, and opioid receptors. CRF, corticotropic-releasing factor; NE, norepinephrine. Modified with permission from ref. 35; permission conveyed
through Copyright Clearance Center, Inc.

15040 | www.pnas.org/cgi/doi/10.1073/pnas.1010654108 Volkow et al.

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 associated with impulsivity (32), and
low striatal D2R was associated with im-
pulsivity and predicted compulsive cocaine
administration in rodents (33).
It is also possible that the initial vulner-
ability for drug use occurs in prefrontal
regions and that repeated drug use triggers
the decreases in striatal D2R. Indeed, in
a study done in subjects who, despite having
a high risk for alcoholism (positive family
history of alcoholism), were not alcoholics,
we showed higher than normal striatal D2R
availability that was associated with normal
metabolism in OFC, ACC, and DLPFC
(25). We interpreted this to suggest that
normal prefrontal function may have pro-
tected these subjects from alcohol abuse.
DA and Motivation in Addiction
DA is also involved in motivation (i.e.,
vigor, persistence, effort toward the pursuit
of reinforcing stimuli) through its regula-
tion of several target regions, including
NAc, ACC, OFC, DLPFC, amygdala,
dorsal striatum, and ventral pallidum (34).
The enhanced motivation to procure
drugs is a hallmark of addiction. Drug-
addicted individuals will go to extreme
behaviors to obtain drugs, even at the ex-
pense of seriously adverse consequences
(2). Drug seeking and drug taking become
their main motivational drives, which dis-
place other activities (35). Thus, the ad-
dicted subject is aroused and motivated
when seeking to procure the drug but
tends to be withdrawn and apathetic when
exposed to non–drug-related activities.
This shift has been studied by comparing
the brain activation patterns occurring
with exposure to conditioned cues with
those occurring without such cues.
In contrast to the decreases in prefrontal
activity reported in detoxified cocaine
abusers when not stimulated with drug
or drug cues (reviewed in ref. 13), these
prefrontal regions become activated when
cocaine abusers are exposed to craving-
1. Wise RA (2009) Roles for nigrostriatal—not just
mesocorticolimbic—dopamine in reward and addiction.
Trends Neurosci 32:517–524.
2. Volkow N, Li TK (2005) The neuroscience of addiction.
Nat Neurosci 8:1429–1430.
3. Wanat MJ, Willuhn I, Clark JJ, Phillips PE (2009) Phasic
dopamine release in appetitive behaviors and drug
addiction. Curr Drug Abuse Rev 2:195–213.
4. Grace AA (2000) The tonic/phasic model of dopamine
system regulation and its implications for understanding
alcohol and psychostimulant craving. Addiction 95(Suppl
2):S119–S128.
5. Kalivas PW (2009) The glutamate homeostasis hypothesis
of addiction. Nat Rev Neurosci 10:561–572.
6. Koob GF (1992) Neural mechanisms of drug rein-
forcement. Ann N Y Acad Sci 654:171–191.
7. Volkow ND, et al. (1999) Reinforcing effects of
psychostimulants in humans are associated with
increases in brain dopamine and occupancy of D(2)
receptors. J Pharmacol Exp Ther 291:409–415.
Downloaded by guest on February 17, 2022
8. Drevets WC, et al. (2001) Amphetamine-induced do-
pamine release in human ventral striatum correlates
with euphoria. Biol Psychiatry 49:81–96.
Volkow et al.
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
inducing stimuli (either drugs or cues)
(36–39). Similarly, cocaine abusers studied
shortly after an episode of cocaine binging
showed increased metabolic activity in
OFC and ACC (also dorsal striatum) that
was associated with craving (40).
Moreover, when we compared the
response to i.v. MP between cocaine-
addicted and nonaddicted subjects, we
showed that MP increased metabolism in
ventral ACC and medial OFC (an effect
associated with craving) only in addicted
subjects, whereas it decreased metabolism
in these regions in nonaddicted subjects
(41). This suggests that the activation of
these prefrontal regions with drug expo-
sure may be specific to addiction and as-
sociated with the enhanced desire for the
drug. Moreover, in a subsequent study
in which we prompted cocaine-addicted
subjects to inhibit craving purposefully
when exposed to drug cues, we showed
that subjects who were successful in in-
hibiting craving decreased metabolism
in medial OFC (processes motivational
value of reinforcer) and NAc (predicts
reward) (42).
These findings corroborate the involve-
ment of OFC, ACC, and striatum in the
enhanced motivation to procure the drug
in addiction.
Systems Model of Addiction
As summarized above, several brain cir-
cuits are relevant in the neurobiology of
addiction. Here, we highlighted four of
these circuits: reward/saliency, motivation/
drive, conditioning/habits, and inhibitory
control/executive function (Fig. 5). The
mood regulation circuit (contributes to
regulation of stress reactivity) and the in-
teroception circuit (contributes to aware-
ness of drug craving and mood) also
participate in addiction, but their involve-
ment in the human brain has been much
less investigated. Consequences of the
disruption of these circuits are an en-
9. Brody AL, et al. (2009) Ventral striatal dopamine release
in response to smoking a regular vs a denicotinized
cigarette. Neuropsychopharmacology 34:282–289.
10. Boileau I, et al. (2003) Alcohol promotes dopamine
release in the human nucleus accumbens. Synapse 49:
226–231.
11. Bossong MG, et al. (2009) Delta 9-tetrahydrocannabinol
induces dopamine release in the human striatum. Neuro-
psychopharmacology 34:759–766.
12. Breier A, et al. (1997) Schizophrenia is associated with
elevated amphetamine-induced synaptic dopamine
concentrations: Evidence from a novel positron emis-
sion tomography method. Proc Natl Acad Sci USA 94:
2569–2574.
13. Volkow ND, Fowler JS, Wang GJ, Baler R, Telang F
(2009) Imaging dopamine’s role in drug abuse and
addiction. Neuropharmacology 56(Suppl 1):3–8.
14. Volkow ND, et al. (2000) Effects of route of administration
on cocaine induced dopamine transporter blockade in the
human brain. Life Sci 67:1507–1515.
15. Balster RL, Schuster CR (1973) Fixed-interval schedule
of cocaine reinforcement: Effect of dose and infusion
duration. J Exp Anal Behav 20:119–129.
PNAS | September 13, 2011 | vol. 108 | no. 37 | 15041
hanced motivational value of the drug
(secondary to learned associations through
conditioning and habits) at the expense of
other reinforcers (secondary to decreased
sensitivity of the reward circuit) and an
impaired ability to inhibit the intentional
actions associated with the strong desire
to take the drug (secondary to impaired
executive function) that result in compul-
sive drug taking in addiction (35).
Although it is likely that DA changes
underlie some aberrant behaviors in
addiction, it is also possible that some
DA changes may reflect attempts to
compensate for deficits in other neuro-
transmitters, particularly because DA is
modulated by glutamate (GABA has been
less investigated). Corticostriatal gluta-
matergic terminals are responsible for
learning well-established behaviors and
for changing these behaviors when they
are no longer adaptive, and neuro-
adaptations in these projections (and in
amygdalostriatal glutamate pathways)
with repeated drug use (including im-
paired regulation of glutamate synaptic
release) are implicated in the enhanced
motivation for drug seeking that occurs in
addiction (5). Impairments in glutamate-
induced neuroplasticity with chronic
drug exposure are also likely to be in-
volved in the prefrontal function deficits
reported in addicted individuals that re-
sult in impairments in inhibitory control
and in inability to change maladaptive
behaviors and to learn from the adverse
consequences of drug use.
This model suggests a multipronged
therapeutical approach to addiction to
decrease the reinforcing properties of
drugs, enhance the rewarding properties of
natural reinforcers, inhibit conditioned-
learned associations, enhance motivation
for non–drug-related activities, and stren-
gthen inhibitory control.
ACKNOWLEDGMENTS. We thank Linda Thomas
for editorial assistance.
16. Schultz W (2010) Dopamine signals for reward value
and risk: Basic and recent data. Behav Brain Funct 6:
24.
17. Owesson-White CA, et al. (2009) Neural encoding of
cocaine-seeking behavior is coincident with phasic
dopamine release in the accumbens core and shell.
Eur J Neurosci 30:1117–1127.
18. Volkow ND, et al. (1995) Is methylphenidate like cocaine?
Studies on their pharmacokinetics and distribution in the
human brain. Arch Gen Psychiatry 52:456–463.
19. Martinez D, et al. (2007) Amphetamine-induced
dopamine release: Markedly blunted in cocaine depen-
dence and predictive of the choice to self-administer
cocaine. Am J Psychiatry 164:622–629.
20. Wang G-J, et al. (2010) Decreased brain dopaminergic
responses in active cocaine dependent subjects. J Nucl
Med 51(Suppl 2):74.
21. Zweifel LS, et al. (2009) Disruption of NMDAR-
dependent burst firing by dopamine neurons provides
selective assessment of phasic dopamine-dependent
behavior. Proc Natl Acad Sci USA 106:7281–7288.
22. Kauer JA, Malenka RC (2007) Synaptic plasticity and
addiction. Nat Rev Neurosci 8:844–858.
 23. Volkow ND, et al. (2006) Cocaine cues and dopamine in
dorsal striatum: Mechanism of craving in cocaine
addiction. J Neurosci 26:6583–6588.
24. Wong DF, et al. (2006) Increased occupancy of
dopamine receptors in human striatum during cue-
elicited cocaine craving. Neuropsychopharmacology
31:2716–2727.
25. Volkow ND, et al. (2006) High levels of dopamine D2
receptors in unaffected members of alcoholic families:
Possible protective factors. Arch Gen Psychiatry 63:
999–1008.
26. Volkow ND, et al. (1993) Decreased dopamine D2
receptor availability is associated with reduced frontal
metabolism in cocaine abusers. Synapse 14:169–177.
27. Volkow ND, et al. (2001) Low level of brain dopamine D2
receptors in methamphetamine abusers: Association
with metabolism in the orbitofrontal cortex. Am J
Psychiatry 158:2015–2021.
28. Volkow ND, et al. (2007) Profound decreases in dopamine
release in striatum in detoxified alcoholics: Possible
orbitofrontal involvement. J Neurosci 27:12700–12706.
29. Volkow ND, Fowler JS (2000) Addiction, a disease of
compulsion and drive: Involvement of the orbitofrontal
cortex. Cereb Cortex 10:318–325.
Downloaded by guest on February 17, 2022
15042 | www.pnas.org/cgi/doi/10.1073/pnas.1010654108
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
30. Volkow ND, Ding YS, Fowler JS, Wang GJ (1996)
Cocaine addiction: Hypothesis derived from imaging
studies with PET. J Addict Dis 15:55–71.
31. Goldstein RZ, Volkow ND (2002) Drug addiction and its
underlying neurobiological basis: Neuroimaging
evidence for the involvement of the frontal cortex.
Am J Psychiatry 159:1642–1652.
32. Lee B, et al. (2009) Striatal dopamine d2/d3 receptor
availability is reduced in methamphetamine dependence
and is linked to impulsivity. J Neurosci 29:14734–14740.
33. Everitt BJ, et al. (2008) Review. Neural mechanisms
underlying the vulnerability to develop compulsive
drug-seeking habits and addiction. Philos Trans R Soc
Lond B Biol Sci 363:3125–3135.
34. Salamone JD, Correa M, Farrar A, Mingote SM (2007)
Effort-related functions of nucleus accumbens dopamine
and associated forebrain circuits. Psychopharmacology
(Berl) 191:461–482.
35. Volkow ND, Fowler JS, Wang GJ (2003) The addicted
human brain: Insights from imaging studies. J Clin
Invest 111:1444–1451.
36. Volkow ND, et al. (1999) Association of methylphenidate-
induced craving with changes in right striato-orbitofrontal
metabolism in cocaine abusers: Implications in addiction.
Am J Psychiatry 156:19–26.
37. Wang G-J, et al. (1999) Regional brain metabolic
activation during craving elicited by recall of previous
drug experiences. Life Sci 64:775–784.
38. Grant S, et al. (1996) Activation of memory circuits
during cue-elicited cocaine craving. Proc Natl Acad Sci
USA 93:12040–12045.
39. Daglish MR, Nutt DJ (2003) Brain imaging studies in
human addicts. Eur Neuropsychopharmacol 13:453–458.
40. Volkow ND, et al. (1991) Changes in brain glucose
metabolism in cocaine dependence and withdrawal.
Am J Psychiatry 148:621–626.
41. Volkow ND, et al. (2005) Activation of orbital and me-
dial prefrontal cortex by methylphenidate in cocaine-
addicted subjects but not in controls: Relevance to
addiction. J Neurosci 25:3932–3939.
42. Volkow ND, et al. (2010) Cognitive control of drug
craving inhibits brain reward regions in cocaine
abusers. NeuroImage 49:2536–2543.
43. Fowler JS, et al. (2008) Fast uptake and long-lasting
binding of methamphetamine in the human brain:
Comparison with cocaine. NeuroImage 43:756–763.
44. Nestler EJ, et al. (2005) Is there a common molecular
pathway for addiction? Nat Neurosci 8:1445–1449.
Volkow et al.
 Molecular Psychiatry (2000) 5, 467–475
 2000 Macmillan Publishers Ltd All rights reserved 1359-4184/00 $15.00
www.nature.com/mp

MILLENNIUM ARTICLE

Is there an evolutionary mismatch between the normal

physiology of the human dopaminergic system and
current environmental conditions in industrialized
countries?
L Pani
CNR Center for Neuropharmacology, ‘BB Brodie’ Department of Neuroscience, University of Cagliari and Neuroscienze
Scarl, Cagliari, Italy

A large body of evidence has recently defined a field theory known as ‘evolutionary mismatch’,
which derives its attributes largely from the fact that current environmental conditions are
completely different from those in which the human central nervous system evolved. Current
views on the evolutionary mismatch theory lack, however, any attempts to define which brain
areas or neuronal circuits should be mostly involved in coding such misevolved traits and to
what extent our neurobiological knowledge can be applied to the topographical localization
of a specific psychopathology. In this respect the mesocorticolimbic dopaminergic circuits
have long been misconceptualized as simple reward or reinforcement systems. Instead, they
motivate and coordinate the functions of the higher brain areas that mediate planning and
foresight and direct finalized movement in both animals and humans. These systems make
animals intensely interested in exploring the world around them, but by the same means they
also make them susceptible to the environmental stimuli that have been sought and con-
sumed. It is has been speculated that the cortical dopamine targets that developed most
recently in phylogeny are of particular functional value, and that the mesocorticolimbic dopa-
minergic system is involved in more complex integrative functions than previously assumed.
In the present paper I will argue that some mental disorders may have their deep roots in the
evolutionary mismatch between the normal physiology of the mesocorticolimbic dopami-
nergic system and the current environmental conditions in affluent societies. Molecular Psy-
chiatry (2000) 5, 467–475.
Keywords: evolution; limbic system; dopamine; stress; depression; emotions; Darwinian medicine

Introduction Table 1 Steps in brain evolution with increasing environ-

mental interaction
The essential steps in the evolution of the brain from
early mammals1 to humans2 show an increasing motor3
1. Bipedalism
and cognitive-emotional interaction4 with the external 2. Development of a cooling system
world (Table 1). This ‘environmental pressure’ 3. Trichromatic vision
required the development of a neuronal system which 4. Hemispheric dominance/use of tools
was able to decipher and cope with the signals and 5. Nursing and maternal care
stimuli coming from the brain’s surroundings. 6 Voluntary control of speech
In this sense, the utilization of ‘the principle of 7. Separation call
reward’5 by linking outcomes of behaviors to either 8. Self/other feelings awareness
reinforcing or aversing sensations (and memorizing 9. Development of synthatic ability
them), which is part of the normal physiology of the 10. Pressure to speak
dopaminergic system, was absolutely instrumental in
the course of evolution.
Several authors, however, consider such a view an
also motivate and coordinate the functions of higher
over-simplification of the role of the mesolimbic and
brain areas that mediate planning, foresight and pro-
mesocortical dopamine circuits, which not only direct
mote states of eagerness.6,7 By these functions animals
finalized movement in both animals and humans, but
become intensely interested in exploring the world
around them, but by the same means they are exposed
Correspondence: Dr L Pani, CNR Center for Neuropharmacology, and susceptible to the environmental stimuli that have
Via Porcell, 4, 09124-I, Cagliari, Italy. E-mail: panil얀unica.it been sought and consumed. In mammals, and
Received 3 January 2000; revised and accepted 10 April 2000 especially in primates, the mesocortical dopaminergic

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Evolutionary mismatch of the dopaminergic system
L Pani
468
system generates and sustains curiosity, and it is quite
efficient at facilitating learning. It is particularly effec-
tive in memorizing information about where material
resources are located and the best way to obtain the
things needed for survival, including food, water,
warmth and sex.8
While telencephalic dopaminergic nuclei, such as
the nucleus accumbens and the caudate-putamen, have
been conserved in their gross and microscopic struc-
ture throughout evolution from rodents9 to humans,10
the structural organization of the cortical dopaminergic
circuits, most strictly involved in human pathologies
such as depression and psychosis, shows significant
differences between rodents and primates.11,12
Recent data show that the major development of the
cerebral cortex in primates was accompanied (or
produced?) by a similar expansion of the dopaminergic
input to all cortical layers, with a specific and laminar
redistribution of the dopaminergic terminals.11 It has
been speculated that the cortical dopamine targets that
developed most recently in phylogeny are to be con-
sidered of particular functional value, and that the
mesocorticolimbic dopaminergic system is involved in
more complex integrative functions than previously
assumed.
In the present paper I will argue that the evolved
physiology of the mesocorticolimbic dopaminergic sys-
tem may be mismatched with respect to current
environmental conditions and may contribute to the
precipitation of human psychiatric disorders.
Evolution of the dopaminergic system
Enzymes
The acquisition of the behaviors modulated by dopami-
nergic innervation has been so indispensable in sur-
vival that the enzymatic machinery needed to synthe-
size dopamine appeared very early in phylogenesis.
One of the most rudimentary nervous systems, that of
the sea pansy Renilla koellikeri (Cnidaria), is able to
synthesize dopamine; such synthesis is inhibited by
alpha-methyl-p-tyrosine.13 In this invertebrate, the role
of dopamine is that of giving bioluminescence and pro-
ducing muscular contractions, which are most likely to
be used to capture food or avoid danger; insufficient
to result in such superior animal characteristics as the
ability to feel pleasure and pain. The neurochemical
basis of the effects of dopamino-mimetic drugs in other
distant organisms has not been fully elucidated, but the
hyperkinesia induced by cocaine, a drug which blocks
the dopamine transporter, in Planaria specimens sug-
gests that the selective site for pre-synaptic dopamine
re-uptake is already present in this flatworm.14 Gold-
fish show place preference for the water bowls where
they have received amphetamine, a drug of abuse that
increases dopaminergic transmission.15 In the cephalo-
chordate amphioxus, a sister group to vertebrates, the
presence of dopamine (but not of noradrenaline) has
been assayed. In contrast, two distinct genes homolo-
gous to the jawed vertebrate dopamine D1 and beta
adrenergic receptor genes were shown to be extant in
Molecular Psychiatry
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
representatives of the earliest craniates, lamprey and
hagfish, paralleling a high dopamine and noradrena-
line content throughout their brain.16
Receptor family
The evolving vertebrate nervous system was always
accompanied by major gene duplication events which
generated novel organs along with a definite sympath-
etic system which was needed to monitor and control
them. Vertebrate neural pathways synthesizing
monoamine neurotransmitters (mostly dopamine and
noradrenaline) were subsequently recruited in order to
process increased information demands by mediating
psychomotor functions, such as selective
attention/predictive reward and emotional drive, via
the activation of multiple G-protein linked catechola-
mine receptor subtypes.
At least 1000 million years ago (MYA), all the classi-
cal transmitter ligand molecules evolved the ability to
activate a wide range of ion channels, resulting in exci-
tation, inhibition and biphasic or multiphasic
responses. In this respect, the invertebrate receptors
which have so far been cloned show striking hom-
ologies with mammalian receptors, indicating that
many of the basic receptor subtypes evolved at an early
period, probably around 800 MYA.17
The evolution of these receptor-mediated events was
similarly driven by forces of gene duplication, at the
cephalochordate/vertebrate transition. In amphioxus, a
single catecholamine receptor gene was found which,
based on molecular phylogeny and functional analysis,
forms a monophyletic group with both vertebrate dopa-
mine D1 and beta adrenergic receptor classes.18 These
data suggest that a dopamine D1/beta receptor gene
duplication was required for the elaboration of novel
catecholamine psychomotor adaptive responses, and
that a noradrenergic system specifically emerged at the
origin of vertebrate evolution. Recent investigations
have confirmed that the dopamine D1A, D1B, and D1C
receptors share molecular, pharmacological, and func-
tional attributes that unambiguously allow for their
evolutionary classification as distinct D1 receptor sub-
types of the G-protein linked super-family, in the ver-
tebrate phylum.18
All these receptors conserve a fairly stable secondary
structure, a moderate and reasonably steady rate of
sequence change, and usually lack introns within their
coding sequence. Several results indicate that the first
event to occur within this gene family was the diver-
gence of the catecholamine receptors from the muscar-
inic acetylcholine receptors, which occurred prior to
the divergence of the arthropod and vertebrate lin-
eages. Subsequently, the ability to activate specific
second-messenger pathways diverged independently
in both the muscarinic and the catecholamine recep-
tors. This appears to have occurred after the divergence
of the arthropod and vertebrate lineages, but before the
divergence of the avian and mammalian lineages. How-
ever, the second-messenger pathways activated by
adrenergic and dopamine receptors did not diverge
independently. Rather, the ability of the catecholamine

receptors to bind to specific ligands, such as epine-
phrine, norepinephrine, dopamine, or octopamine, was
repeatedly modified in evolutionary history, and in
some cases was modified after the divergence of the
second-messenger pathways.19
When the distribution of the dopamine D1 and D2
receptors in the brain of a mouse, rat, or cat is com-
pared with that of a monkey, similarities are found
only in the basal ganglia; the highest density being
present in the caudate-putamen, in the nucleus accum-
bens, in the olfactory tubercle, and in the substantia
nigra. Apart from these nuclei, and particularly in more
recent structures such as the cerebellum and the frontal
cortex, the absolute and relative values of the dopa-
mine receptor subtypes is completely different in pri-
mates, suggesting that the transcriptional control of the
gene encoding for the dopaminergic receptors has also
undergone profound changes throughout evolution.20
The dopaminergic system and the mismatch
theory
Large sets of data (for a review see Nesse and
Williams)21 have recently allowed the definition of a
field theory called ‘evolutionary mismatch’, which
derives its attributes largely from the fact that current
environmental conditions in industrialized countries
are completely different from those in which the
human central nervous system evolved. The mismatch
between evolved traits and contemporary lifestyle, the
presence of evolutionary trade-offs (eg selection of
some traits over others), and historical constraints (eg
how a trait has evolved), has resulted in the poor spe-
cialization of selected traits, as well as a particular sus-
ceptibility to mental disorders.
To explain the somatic correlates of such an evo-
lutionary view of diseases, it is assumed that pathogens
that attack humans evolve faster than we do. If the
ever-mutating environmental conditions, where
change is mostly due to fast technological advance-
ment (Table 2), could now be assimilated to organic
pathogens, their rapid rate of evolution and change
would impose a pressure that is not paralleled by an
evolution of the brain structures which are able to ana-
lyze, process, and elaborate on such rapid variations.
For example, a poor fit between strongly evolved traits
(eg dependency) and the current urban social environ-
ment may account for the increase in frequency of anx-
iety and depressive disorders in modern societies.22
Current views on the evolutionary mismatch theory
lack, however, a definition of which brain areas or neu-
ronal circuits should be involved in coding such mis-
evolved traits, and to what extent our neurobiological
knowledge can be applied to the topographical localiz-
ation of a specific psychopathology.
The appetitive and consummatory phases of
behaviors triggered by environmental cues which are
associated with survival properties (whether rewarding
or aversive for the individual) are known to be
mediated by sensory-specific dopaminergic neuronal
circuits. These dopaminergic systems are involved in
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Evolutionary mismatch of the dopaminergic system
L Pani
469
Table 2 Years of significant inventions and technological
advancement
1550 Screwdriver
1690 Steam engine
1770 Electrical battery
1853 Hypodermic needle
1876 Telephone
1885 Gas-engine auto
1886 Coca-Cola
1903 Airplane
1920 Radio broadcast
1926 Television
1948 Transistors
1962 Minicomputer
1969 Moon landing
1972 Videogames
1978 Test tube baby
1988 Patented animal life
1992 Global internet
2000 Human genome sequenced
those mechanisms which determine satiety to food,23
sex,24 and drugs of abuse.25 In all these conditions, the
most important dopamine systems involved are that of
the terminal fields of the prefrontal cortex,26 and that
of the nucleus accumbens,27 both originating from neu-
rons located in the ventral tegmental area of the mid-
brain. Whether the stimulation of these neurons
mediates the incentive salience (reviewed in Berridge
and Robinson),28 or the associative learning29
capacities of an event, is still a matter of intense debate.
According to the first hypothesis, rewarding (and thus
also averse) stimuli that have acquired motivational
significance (salience) are part of a phenomenon which
is neither unitary nor subjective; dopamine-related
neural circuits mediating only a specific component of
reward, ie the attribution of incentive salience to an
otherwise neutral event. In contrast, others have argued
that mesolimbic dopaminergic pathways do not code
for such an ability, but serve only to respond to specific
stimuli which possess high motivational impact, as a
result of their novelty, unpredictability, specific sen-
sory modalities, or occurrence under a deprivation
state;29 in these conditions, dopamine neurons define
the substrate for precise associative learning abilities.
From an evolutionary perspective, the two mech-
anisms must play an equally important role in the
attribution of an incentive value to ethologically
relevant stimuli, and must facilitate, by means of both
incentive-salience and associative-learning, the selec-
tion and the memory of appropriate behavioral
responses. The conservation of such principles
throughout evolution explains why pharmacological
manipulations of the dopaminergic transmission are
similarly able to modify basic behaviors in distant
species such as lizards and rats (Table 3). Stimulation
of the dopaminergic transmission will increase these
behaviors, while the blocking of dopaminergic recep-
tors or lesions of the dopaminergic pathways will
decrease them. This confirms that dopamine is an
Molecular Psychiatry
 Evolutionary mismatch of the dopaminergic system
L Pani
470
Table 3 Special forms of basic behaviorsa
1. Establishing a ‘territory’
2. Showing place preference
3. Hunting
4. Greeting
5. Social grouping
6. Feeding and drinking
7. Grooming
8. Courtship
9. Mating
10. Breeding
11. Maternal behavior
12. Play (mammals only)
a
Adapted from Reference 6.
important neurotransmitter coding for sensory-specific
responses, in the sense that a perceived environmental
clue is acted on cognitively, either with an increase in
exploration (by means of curiosity and reward) or a
decrease (through fear and aversion),30 but also show
that the physiological responses of the dopaminergic
system are strictly ‘environment sensitive’. Electrophy-
siological recordings have indicated that key cortical
and subcortical dopaminergic regions are involved in
these motivated behaviors. Among these, a particular
role in primates is given to the connections between
the nucleus accumbens and the frontal cortex,31 which
I will consider as primary sites for the evolutionary
mismatch between the dopaminergic system and the
ability to process significant environmental-driven
change-related information that guides coherent
behavioral responses.
Expansion and regression of dopaminergic
innervation
Animals with big brains are scarce, because bigger neu-
ronal structures are more costly in terms of energy
expenditure, development time, and anatomical com-
plexity. It was always supposed that a much better diet
(in terms of caloric value) must have been the starting
point for the development of a much larger cerebral
mass. Recently it was described how, about 40 MYA,
a duplication of the gene coding for a retinal cone pig-
ment in an ancestor of prosimians resulted in the
development of trichromatic vision.32 This ability to
see in color was instrumental in detecting ripe fruit,
and gave an immediate evolutionary advantage over
green-leaf eaters. At about the same time, the system
for emotional communication via facial expression
expanded and the olfactory bulb completed its
regression in primates.
Recently, this hypothesis was confirmed, showing
that a highly predictable relationship exists between
the relative size of the neocortex and that of the total
brain.33 When the sizes of 10 measured brain divisions
from 131 species are plotted as a function of total brain
size, the two most expanded structures in all species
are, in fact, the neocortex and the striatum (caudate-
Molecular Psychiatry
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
putamen; nucleus accumbens and internal capsule), at
variance with that of the olfactory bulb. Thus, two
heavily innervated dopaminergic structures—one
coupled to taste related reward and satiety, and the
other to olfactory satiety—seemed to be mutually
exclusive in their development. These complementary
characteristics have been conserved in modern
humans.34 The neural mechanisms of sensory-specific
satiety in primates are implemented in the orbitofron-
tal cortex, which has an important integrative function.
Neurons in this region of the cortex were indeed found
to respond to stimulation of the taste, olfactory, or vis-
ual systems. In addition, some neurons were found to
possess a bimodal response; responding, for example,
to both taste and olfactory, or taste and visual stimuli.
Since these multimodal neurons were found in very
close proximity to unimodal neurons, and since the
unimodal sensory neurons were intermingled, it is
possible that the orbitofrontal cortex represents the first
cortical area of convergence for these three modalities
in primates.35
Conditions that have changed through evolution
and which interfere with the function of the
dopaminergic system
Empirical data suggest a causal relationship between
evolutionary-new environmental factors, specific dys-
regulation of the physiological ability of the dopami-
nergic system to process environmental stimuli and
specific psychiatric disorders. Accordingly, the use of
psychotropic medication in outpatient medical prac-
tice rose dramatically during the last 10 years,
especially in industrialized countries.
Around 10 years ago, Gershon et al showed that the
rates of bipolar, schizoaffective, and unipolar disorders
were higher in the cohorts born after 1940 than in the
cohorts born earlier.36 It was suggested that an ominous
trend could be present, leading to an increase in preva-
lence of a broad spectrum of familial affective disorders
in the coming decades. This prediction proved to be
correct and confirmed by more recent reports where
changes in the prescribing patterns of psychotropic
medications by office-based United-States physicians
were examined.37 In the 9 years period between 1985
and 1994, the number of visits during which a psycho-
tropic medication was prescribed increased from 32.73
million to 45.64 million. Antianxiety or hypnotic drug
visits, previously the largest category, decreased and
were surpassed by antidepressant visits, which
doubled from 10 to 20 million in the last 5 years per-
iod; and stimulant drug visits increased more than five-
fold in the same period. These dramatic trends are in
agreement with our findings (Table 4), where we con-
sider the Italian sales of psychotropic medications
classified into three major categories: anxiolytics, anti-
depressants, and antipsychotics and confirmed a sig-
nificant tendency toward increase for antidepressants
(+34%) and antipsychotics (+15%) but not for anxio-
lytics (+4%) in the last 5 years.
Thus, it appears that in current affluent societies the

Table 4 The Italian market for psychotropic compounds
Millions of prescriptions per year
1995 1996 1997 1998 1999
Antipsychotics 11 675 12 432 12468 12 940 13 428
Antidepressants 16 750 18 275 19413 20 524 22 493
Anxiolytics 33 143 34 589 34093 34 917 34 487
frequency of psychiatric disorders requiring medi-
cations which modulate the function of (and not only
of) the dopaminergic system is increasing. In addition,
the outcomes of acute affective disorders are consider-
ably more favorable in developing countries than in
comparable studies in developed settings38 where at
least four areas of evolutionary novelty can contribute
and will be briefly discussed below.
Drugs of abuse
The ability to extract, purify and lately synthesize psy-
choactive compounds has increased their availability
to the general population in quantities and qualities
never experienced before in the history of mankind. In
addition, the development of new routes of adminis-
tration (eg hypodermic needles, crack smoking, organic
solvents) has increased the penetration of drugs of
abuse into the brain, and has contributed to a change
in the traditional relationship between the use of the
natural source of the drug (eg the chewing of coca-
leaves) to that of its active component (eg the endoven-
ous injection of cocaine hydrochloride).
Drugs of abuse offer a remarkable example of evo-
lutionary mismatch at the level of the dopaminergic
system, and help to clarify its role in the general physi-
ology of the central nervous system. If drugs of abuse
that act on the dopaminergic system (that is all of them,
directly or indirectly) were used solely for their
hedonic and ‘reward producing’ potential, then this
capacity, ie the pure ability to experience pleasure,
would be the only one to be mainly affected by their
chronic use. Instead, after a variable period of use, all
drugs which determine a state of dependence interfere
with the global adaptation of an individual to its
environment, producing not only an impairment of
his/her hedonic capacities, but also a more disruptive
effect on the cognitive and emotional abilities that are
necessary for an effective interaction with the external
world (maladaptation). The limbic portion of the dopa-
minergic system is essential for the generation of the
basic emotions that mediate various pro-social
behaviors, such as maternal care, playfulness, and
friendship, and which have been learned and have
evolved to influence motivational states and, ulti-
mately, to increase fitness. All these behaviors, along
with the emotions which are related to them and their
cognitive implications, are profoundly altered in a drug
addict, much more than just the capacity to experience
pleasure or avoid pain (whether physical or
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Evolutionary mismatch of the dopaminergic system
L Pani
471
psychological). Drugs of abuse that convey a signal that
falsely indicates the arrival of a huge fitness benefit
(much higher than food, water, sex, etc), actively
change behavioral hierarchies. As a consequence, drug-
seeking increases in frequency to the point where it
becomes the only dominant behavior, displacing all the
other adaptive traits.39
Stress
The definition of the term stress has gone from ‘a non-
specific response of the body to any demands made
upon it’,40 to ‘both a survival mechanism and an indi-
cator of internal and external cues’.41 Both acute and
chronic stress have a detrimental impact on the normal
function of the dopaminergic system (for a recent
review see Pani et al).42 The response of the dopami-
nergic system to stress is neither homogenous nor gen-
eralized, and numerous results suggest that mesopre-
frontal cortex dopaminergic projections are selectively
activated by sudden changes in environmental con-
ditions, whether expected or not.43
The fine tuning of dopamine extracellular concen-
tration in the medial prefrontal cortex seems to be
essential for the decision-taking protocols of the medial
prefrontal cortex in the rat and the monkey.44 This sup-
ports the idea of a critical range of dopamine turnover
for optimal prefrontal cortical cognitive functioning,
with reduced or excessive dopamine turnover leading
to cognitive impairment. Several data point to ventral
tegmental area projections,45 dopamine receptors,46
and a loss of inhibitory tone on ventral tegmental area
dopamine cells47 and lateral-basolateral amygdala (see
below) as important regulatory sites for maintaining
superior cognitive functions. The increased ability to
remove dopamine in chronically stressed animals has
also indicated that altered dopamine clearance may
serve as an adaptive mechanism in the medial pre-
frontal cortex.48 It has been further suggested that
increased dopamine D1 receptor stimulation during
stress may serve to take the medial prefrontal cortex
‘off-line’, in order to allow posterior cortical and sub-
cortical structures to regulate more ‘primitive’ forms of
behavior.49 Since dopaminergic innervation of the
medial prefrontal cortex is able to regulate the activity
of subcortical dopamine innervations, disruption of the
medial prefrontal cortex dopamine fibers may result in
the altered biochemical responsiveness of the dopa-
mine subcortical innervations.50
In humans, the balance between cortical and subcort-
ical dopaminergic activity may serve as a protection
against psychotic decompensation from chronic
endogenous or exogenous insult,51 and the failure of
this coping mechanism may well contribute to the vul-
nerability of the medial prefrontal cortex in many
neuropsychiatric disorders related to the stress
response.52
Chronic emotional reactions
Emotional responses can be difficult to differentiate
from generic reactions to stress, in laboratory animals
as well as in humans. In addition, emotions such as
Molecular Psychiatry
 Evolutionary mismatch of the dopaminergic system
L Pani
472
empathy seem to be uniquely human, and therefore
impossible to mimic in animal models. In spite of this,
several animal behaviors such as freezing, self-
grooming, exploration and defecation have been con-
sidered as valid somatic correlates to human emotional
responses in which the dopaminergic system is
involved. In all the models, the chronic nature of the
stimulus is essential to produce a stable change in the
normal physiology of the dopaminergic system. In
modern times this can be obtained by generic stress
due to sleep disruption, time shifts, poor physical
activity, increased physical constraints, and unnatural
social rules. In diseases in which dopamine function
is compromised, humans exhibit a constellation of
emotional symptoms, suggesting that the dopamine
system plays an important role in the integration of
superior cortical functions. A dorsal tier of dopamine
neurons receives input from the nucleus accumbens
and from the amygdala, and projects widely through-
out the cortex. Through these projections, the dopami-
nergic limbic system has an enormous influence on
cortical output and can therefore affect the emotional
and motivational ‘coloring’ of a wide range of
behaviors.53
In this respect, the lateral-basolateral amygdala
receives sensory input from environmental stimuli
and, along with the central amygdala, mediates con-
ditional fear. Through its projections to forebrain, mid-
brain and hindbrain areas, the central amygdala gov-
erns the behavioral, autonomic, and endocrine
responses that characterize a central fear state.54
Accumulating revelations about the amygdala-based
fear system has led to considerable progress in deline-
ating the neural connections and cellular mechanisms
that underlie aversive emotionality, and very recently
electrophysiological evidence has shown that neural
discharge in the central amygdala is a consequence of
repeated low-current, high-frequency electrical stimu-
lation of dopaminergic nuclei of the ventral tegmen-
tal area.55
The introduction of Roman high-avoidance (RHA)
and Roman low-avoidance (RLA) rats, two selectively
bred lines that differ in their level of emotionality, into
an unfamiliar environment, the application of a high-
intensity loud noise, or immobilization, are all associa-
ted with an increase in extracellular cortical dopamine
metabolite levels in the RHA but not in the RLA line,
suggesting that the differently evoked emotional reac-
tions may produce a common activation of cognitive
processes.56 This difference in reaction to a stressful
stimulus, according to the genetic makeup of an indi-
vidual, is an essential function of the mesencephalic
dopaminergic neurons, and of their role in the organi-
zation of behavioral responses which are the reflection
of a heightened attention of the animal attempting to
cope with the stressor.57 The mesocortical dopami-
nergic innervations seem to facilitate the functions of
integrative structures, and to have a coordinated role
in regulating the information flow between cortical
structures.58
The study of fear and the role of the dopaminergic
Molecular Psychiatry
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
system in rodents have, however, produced inconsist-
ent data. While some results suggest an important
relationship between social and motor reactivity, as
well as an important, albeit again strain-dependent,
role for dopamine D1 receptors in mediating specific
emotional behaviors,59 others demonstrate that long-
term administration of low, pre-synaptic doses of the
dopamine D2 antagonist l-sulpiride is highly effective
in an animal model of anticipatory anxiety/panic
behavior.60 In contrast, recent findings suggest that
quinpirole (a dopamine D2 agonist) decreases fear by
blocking the retrieval of a learned association between
a conditioned and an unconditioned stimulus.61
Chronic sleep deprivation
Humans have conserved mechanisms, like those that
exist in other animals, which detect changes in day
length and make corresponding adjustments in the dur-
ation of nocturnal periods. Several studies from Wehr
and collaborators62 have suggested that modern men’s
use of artificial light after dark and artificial darkness
during the daytime suppresses responses63 to seasonal
changes in the duration of the natural scoto- and photo-
period that might otherwise occur at a given latitude,
producing a chronic disturbance in the sleep pattern
and decreasing, overall, the total amount of sleep. It is
difficult to conceive that chronic sleep deprivation in
humans could be obtained and maintained without any
activation in the stress axis.
Sleep deprivation in rats has been extensively stud-
ied as a possible animal model of mania.64 Interest-
ingly, at the end of the period of sleep deprivation
(approximately 72 h), the rat does not fall asleep as
soon as it is returned to its home cage, but shows a
period of wakefulness of about 30 min, during which
the animal presents a cohort of symptoms that appear
to mimic those present in idiopathic mania. In parti-
cular, during this period the animal displays a high
degree of hyperactivity, irritability, aggressiveness,
hyper-sexuality, and stereotypy. The model allowed
the discovery of an active role of limbic dopamine in
the generation of arousal and insomnia related to sleep
deprivation-induced stress.65 Accordingly, direct evi-
dence demonstrates that acute and chronic sleep depri-
vation may trigger a manic episode in otherwise heal-
thy individuals,66 and that recurrent brief hypomania
which lasts 1–3 days and belongs to the bipolar spec-
trum, is increasing in young adults.67 These findings
are indirectly confirmed by the increased use of neuro-
leptics (Table 4) which, although their use has been
widely discouraged for mood disorders, are still the
standard treatment for acute bipolar mania.68
Conclusions
The brain of Homo sapiens sapiens developed on the
African plains, in populations of a few hundred thou-
sand individuals; today there are six billion of us
worldwide.
It is time to ask ourselves whether the structural
organization and the neuronal plasticity of the human

brain are still able to control the evolution of the
environment he has created. Some influential neuro-
biologists have expressed strong doubts.69
The dopaminergic innervation to the prefrontal cor-
tex is essential for processing and elaborating on the
‘expectancy’ of events, and in the working memory
related to them, in both rats and monkeys.44 Environ-
mentally induced changes in the activity of the dopa-
minergic system, either by conventional (food, water,
sex etc) or non-conventional (drugs of abuse, chronic
stress, emotional reactions etc) stimuli could mismatch
the natural codes for reward-aversion, sensory-motor,
incentive-salience mechanisms which have been con-
served in the last millions of years and which are
needed to associate a learned outcome of a given stim-
uli with a definite emotional state of the organism.
In this sense, the dorsolateral portion of the pre-
frontal cortex is important in monitoring the context of
the situation in relation to how, and what, reward is
given.70 Patients with prefrontal lesion are poor in
assessing how actions relate to goals. The results sug-
gest that pre-frontal cortical lesions provoke a selective
impairment in managerial knowledge that may contrib-
ute to difficulties in the formulation and execution of
plans.71
Neurons which are critical for the computation of
sensory-specific satiety in the monkey are located in
the orbitofrontal cortex and, in humans, decision-
making processes depend on signals arising in the
prefrontal cortex in association with peripheral
somato-visceral signals that can introduce biases in the
reasoning processes.72 When normal subjects
experience situations that they associate with the
possibility of negative consequences, they sense a
peripheral reaction (mainly an adrenergic response)
and learn to make decisions accordingly. Damage to
the orbitofrontal cortex in adulthood as well as in early
life73 disrupts the capacity to make these physiological
responses, and impairs this ability to make crucial
social decisions.
Considering the profound influence of prefrontal cor-
tex damage on several adaptive behaviors in rats and
primates, another question arises: is it possible that in
the absence of organic pathologies, a ‘functional
uncoupling’ of the cognitive capacities (eg working
memory, decision-making, associative learning) of the
prefrontal cortex could be the site of the dopaminergic
evolutionary mismatch?
In this respect, the temperament (as the genetic
model of personality)74 of an individual may strongly
relate to this evolutionary model. If this theory holds
true it should be possible to study by imaging tech-
niques whether or not patients with different tempera-
ments, obviously selected by natural selection for con-
ferring evolutionary trade-off, such as the bipolar-
anxious temperament types,75,76 would show any dif-
ference in the functions of their prefrontal cortex
activity.
A better understanding of the mechanisms, origins
and functions of the dopaminergic system as well as
other neurotransmitters and modulators, will also
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Evolutionary mismatch of the dopaminergic system
L Pani
473
enhance our ability to cope with stress-related mental
disorders in the modern environment, and sharpen our
wisdom in making decisions about the therapeutic use
of psychoactive drugs.
Acknowledgements
I am grateful to Anna Porcella for her critical comments
and to David Webb for his precise editorial work on the
manuscript. This work was supported by a CNR grant.
References
1 Divac I. Monotremunculi and brain evolution. TINS 1995; 18: 2–4.
2 Wilson DR. Evolutionary epidemiology. Darwinian theory in the
service of medicine and psychiatry. Acta Biotheoretica 1993; 41:
205–218.
3 Commentary on Greenfield PM, Language tools and brain: the
ontogeny and phylogeny of hierarchically organized sequential
behavior. Behav Brain Sci 1994; 17: 357–365.
4 Povinelli DJ, Preuss TM. Theory of mind: evolutionary history of
a cognitive specialization. TINS 1995; 18: 418–424.
5 Wise RA. The brain and reward. In: Liebman J, Cooper SJ (eds). The
Neuro-pharmacological Basis of Reward. Oxford University Press:
Oxford, 1989, pp 248–262.
6 MacLean PD. Brain evolution relating to family, play and the separ-
ation call. Arch Gen Psychiatry 1985; 42: 405–417.
7 Le Moal M, Simon H. Mesocorticolimbic dopaminergic network:
functional and regulatory roles. Physiol Rev 1991; 71: 155–234.
8 Panksepp J. Affective Neuroscience: the Foundations of Human
and Animal Emotions. Oxford Univ Press: 1998, pp 53–55.
9 Bertler A, Rosengren E. Occurrence and distribution of catecholam-
ine in brain. Acta Physiol Scand 1959; 47: 350–361.
10 Sano I, Gamo T, Kakimoto Y, Taniguchi K, Takesada M, Nishinuma
K. Distribution of catechol compounds in human brain. Biochem
Biophys Acta 1959; 32: 586–588.
11 Berger B. Dopaminergic innervation of the frontal cerebral cortex.
Evolutionary trends and functional implications. In: Chauvel P,
Delgado-Escueta (eds). Advances in Neurology, Vol 57, Raven
Press: New York, 1992, pp 525–544.
12 Berger B, Gaspar P, Verney C. Dopaminergic innervation of the cer-
ebral cortex: unexpected differences between rodents and primates
[erratum appears in TINS 1991; 14: 119]. TINS 1991; 14: 21–27.
13 Pani AK, Anctil M. Evidence for biosynthesis and catabolism of
monoamines in the sea pansy Renilla koellikeri (Cnidaria). Neuro-
chem Int 1994; 25: 465–474.
14 Palladini G, Ruggeri S, Stocchi F, De Pandis MF, Venturini G, Mar-
gotta V. A pharmacological study of cocaine activity in planaria.
Comp Biochem Physiol C Pharmacol Toxicol Endocrinol 1996; 115:
41–45.
15 Lett BT, Grant VL. The hedonic effects of amphetamine and pento-
barbital in goldfish. Pharmacol Biochem Behav 1989; 32: 355–356.
16 Vincent JD, Cardinaud B, Vernier P. Evolution of monoamine
receptors and the origin of motivational and emotional systems in
vertebrates. Bull Acad Natl Med 1998; 182: 1505–1516.
17 Walker RJ, Brooks HL, Holden-Dye L. Evolution and overview of
classical transmitter molecules and their receptors. Parasitology
1996; 113: S3–S33.
18 Cardinaud B, Sugamori KS, Coudouel S, Vincent JD, Niznik HB,
Vernier P. Early emergence of three dopamine D1 receptor subtypes
in vertebrates. Molecular phylogenetic, pharmacological, and func-
tional criteria defining D1A, D1B, and D1C receptors in European
eel Anguilla anguilla. J Biol Chem 1997; 272: 2778–2787.
19 Fryxell KJ. The evolutionary divergence of neurotransmitter recep-
tors and second-messenger pathways. J Mol Evol 1995; 41: 85–97.
20 Camps M, Kelly PH, Palacios JM. Autoradiographic localization of
dopamine D1 and D2 receptors in the brain of several mammalian
species. J Neural Transm Gent Sect 1990; 80: 105–127.
21 Nesse RM, Williams GC. Why We Get Sick. Random House: New
York, 1994.
22 McGuire MT, Troisi A. Darwinian Psychiatry. Oxford University
Press: New York, 1998.
Molecular Psychiatry
 Evolutionary mismatch of the dopaminergic system
L Pani
474 23 Ahn S, Phillips AG. Dopaminergic correlates of sensory-specific
satiety in the medial prefrontal cortex and nucleus accumbens of
the rat. J Neurosci 1999; 19: RC29.
24 Fiorino DF, Coury A, Phillips AG. Dynamic changes in nucleus
accumbens dopamine efflux during the Coolidge effect in male rats.
J Neurosci 1997; 17: 4849–4855.
25 Fibiger HC. Role of catecholamine transmitters in reward systems:
implications for the neurobiology of affect. In: Oreland E (ed). Brain
Reward Systems and Abuse. New York Press: New York, 1987, pp
61–74.
26 Cenci MA, Kalén P, Mandel RJ, Björklund A. Regional differences
in the regulation of dopamine and noradrenaline release in medial
frontal cortex, nucleus accumbens and caudate-putamen: a
microdialysis study in the rat. Brain Res 1992; 581: 217–228.
27 Phillips AG, Atkinson LJ, Blackburn JR, Blaha CD. Increased extra-
cellular dopamine in the nucleus accumbens of the rat elicited by
a conditioned stimulus for food: an electrochemical study. Can J
Physiol Pharmacol 1993; 71: 387–393.
28 Berridge KC, Robinson TE. What is the role of dopamine in reward:
hedonic impact, reward learning, or incentive salience? Brain Res
Rev 1998; 28: 309–369.
29 Bassareo V, Di Chiara G. Differential influence of associative and
nonassociative learning mechanisms on the responsiveness of pre-
frontal and accumbal dopamine transmission to food stimuli in rats
fed ad libitum. J Neurosci 1997; 17: 851–861.
30 Blackburn JR, Pfaus JG, Phillips AG. Dopamine functions in appeti-
tive and defensive behaviors. Prog Neurobiol 1992; 39: 247–279.
31 Rolls ET. The Brain and Emotions. Oxford University Press: New
York, 1999.
32 Jacobs GH, Neitz M, Deegan JF, Neitz J. Trichromatic colour vision
in New World monkeys. Nature 1996; 382: 156–158.
33 Finlay BL, Darlington RB. Linked regularities in the development
and evolution of mammalian brains. Science 1995; 268: 1578–1584.
34 Rolls ET, Rolls JH. Olfactory sensory-specific satiety in humans.
Physiol & Behav 1997; 3: 461–473.
35 Rolls ET, Baylis LL. Gustatory, olfactory, and visual convergence
within the primate orbitofrontal cortex. J Neurosci 1994; 14:
5437–5452.
36 Gershon ES, Hamovit JH, Guroff JJ, Nurnberger JI. Birth-cohort
changes in manic and depressive disorders in relatives of bipolar
and schizoaffective patients. Arch Gen Psychiatry 1987; 44: 314–
319.
37 Pincus HA, Tanielian TL, Marcus SC, Olfson M, Zarin DA, Thomp-
son J et al. Prescribing trends in psychotropic medications: primary
care, psychiatry, and other medical specialties. JAMA 1998, 279:
526–531.
38 Brown AS, Varma VK, Malhotra S, Jiloha RC, Conover SA, Susser
ES. Course of acute affective disorders in a developing country set-
ting. J Nerv Ment Dis. 1998; 186: 207–213.
39 Nesse RM, Berridge KC. Psychoactive drugs in evolutionary per-
spective. Science 1997; 278: 63–66.
40 Selye H. The evolution of the stress concept. Am Sci 1973; 61:
692–696.
41 Akil HA, Morano IM. Stress. In: Kupfer D, Bloom F (eds). Psycho-
pharmacology the Fourth Generation of Progress. Raven Press: New
York, 1995, pp 773–785.
42 Pani L, Porcella A, Gessa GL. The role of stress in the pathophysiol-
ogy of the dopaminergic system. Mol Psychiatry 2000; 5: 14–21.
43 Kaneyuki H, Yokoo H, Tsuda A, Yoshida M, Mizuki Y, Yamada M
et al. Psychological stress increases dopamine turnover selectively
in mesoprefrontal dopamine neurons of rats: reversal by diazepam.
Brain Res 1991; 557: 154–161.
44 Murphy BL, Arnsten AF, Goldman-Rakic PS, Roth RH. Increased
dopamine turnover in the prefrontal cortex impairs spatial working
memory performance in rats and monkeys. Proc Natl Acad Sci USA
1996; 93: 1325–1329.
45 Murphy BL, Arnsten AF, Jentsch JD, Roth RH. Dopamine and spa-
tial working memory in rats and monkeys: pharmacological rever-
sal of stress, induced impairment. J Neurosci 1996; 16: 7768–7775.
46 Steketee JD, Kalivas PW. Sensitization to psychostimulants and
stress after injection of pertussis toxin into the A10 dopamine
region. J Pharmacol Exp Ther 1991; 259: 916–924.
47 Sorg BA, Steketee JD. Mechanisms of cocaine-induced sensitiz-
Molecular Psychiatry
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
ation. Prog Neuropsychopharmacol Biol Psychiatry 1992; 16:
1003–1012.
48 Meiergerd SM, Schenk JO, Sorg BA. Repeated cocaine and stress
increase dopamine clearance in the rat medial prefrontal cortex.
Brain Res 1997; 773: 203–207.
49 Zahrt J, Taylor JR, Mathew RG, Arnsten AF. Supranormal stimu-
lation of D1 dopamine receptors in the rodent prefrontal cortex
impairs spatial working memory performance. J Neurosci 1997; 17:
8528–8535.
50 Deutch AY, Clark WA, Roth RH. Prefrontal cortical dopamine
depletion enhances the responsiveness of mesolimbic dopamine
neurons to stress. Brain Res 1990; 521: 311–315.
51 Friedhoff AJ, Carr KD, Uysal S, Schweitzer J. Repeated inescapable
stress produces a neuroleptic-like effect on the conditioned avoid-
ance response. Neuropsychopharmacology 1995; 13: 129–138.
52 Finlay JM, Zigmond MJ. The effect of stress on central dopami-
nergic neurons: possible clinical implications. Neurochem Res
1997; 22: 1387–1394.
53 Haber SN, Fudge JL. The interface between dopamine neurons and
the amygdala: implications for schizophrenia. Schizophr Bull 1997;
23: 471–482.
54 Davis M. The role of the amygdala in conditioned fear. In: Aggleton
JP (ed). The Amygdala: Neurobiological Aspects of Emotion, Mem-
ory, and Mental Dysfunction. Wiley: New York, 1992, pp 255–305.
55 Gelowitz DL, Kokkinidis L. Enhanced amygdala kindling after elec-
trical stimulation of the ventral tegmental area: implications for
fear and anxiety. J Neurosci 1999; 19: RC41.
56 Bertolucci-D’Angio M, Serrano A, Scatton B. Mesocorticolimbic
dopaminergic systems and emotional states. J Neurosci Meth 1990;
34: 135–142.
57 D’Angio M, Serrano A, Driscoll P, Scatton B. Stressful environmen-
tal stimuli increase extracellular DOPAC levels in the prefrontal
cortex of hypoemotional (Roman high-avoidance) but not hyper-
emotional (Roman low-avoidance) rats. An in vivo voltammetric
study. Brain Res 1988; 451: 237–247.
58 Louilot A, Taghzouti K, Simon H, Le Moal M. Limbic system, basal
ganglia, and dopaminergic neurons. Executive and regulatory neu-
rons and their role in the organization of behavior. Brain Behav
Evol 1989; 33: 157–161.
59 Gendreau PL, Gariépy JL, Petitto JM, Lewis MH. D1 dopamine
receptor mediation of social and nonsocial emotional reactivity in
mice: effects of housing and strain difference in motor activity.
Behav Neurosci 1997; 111: 424–434.
60 Cavazzuti E, Bertolini A, Vergoni AV, Arletti R, Poggioli R, For-
gione A et al. l-Sulpiride, at a low, non-neuroleptic dose, prevents
conditioned fear stress-induced freezing behavior in rats. Psycho-
pharmacology (Berl) 1999; 143: 20–23.
61 Nader K, LeDoux J. The dopaminergic modulation of fear: quinpir-
ole impairs the recall of emotional memories in rats. Behav Neuro-
sci 1999; 113: 152–165.
62 Wehr TA, Giesen HA, Moul DE, Turner EH, Schwartz PJ. Sup-
pression of men’s responses to seasonal changes in day length by
modern artificial lighting. Am J Physiol 1995; 269: R173–R178.
63 Wehr TA. Effect of seasonal changes in daylength on human neuro-
endocrine function. Horm Res 1998; 49: 118–124.
64 Gessa GL, Pani L, Fadda P, Fratta W. Sleep deprivation in the rat:
an animal model of mania. Eur Neuropsychopharmacol 1995; 5:
S89–S93.
65 Fadda P, Martellotta MC, Gessa GL, Fratta W. Dopamine and
opioids interactions in sleep deprivation. Prog Neuropsychopharm-
acol Biol Psychiatry 1993; 17: 269–278.
66 Wright JB. Mania following sleep deprivation. Br J Psychiatry 1993;
163: 679–680.
67 Angst J. The emerging epidemiology of hypomania and bipolar II
disorder. J Affect Disord 1998; 50: 143–151.
68 Chou JC, Zito JM, Vitrai J, Craig TJ, Allingham BH, Czobor P. Neur-
oleptics in acute mania: a pharmacoepidemiologic study. Ann
Pharmacother 1996; 30: 1396–1398.
69 Changeux JP. L’Homme Neuronal. Fayard: Paris, 1983.
70 Watanabe M. Reward expectancy in primate prefrontal neurons.
Nature 1996; 382: 629–632.
71 Sirigu A, Zalla T, Pillon B, Grafman J, Agid Y, Dubois B. Selective
impairments in managerial knowledge following pre-frontal cortex
damage. Cortex 1995; 31: 301–316.
 Evolutionary mismatch of the dopaminergic system
L Pani

72 Damasio AR. The somatic marker hypothesis and the possible func-
tions of the prefrontal cortex. Philos Trans R Soc Lond B Biol Sci
1996; 351: 1413–1420.
73 Anderson SW, Bechara A, Damasio H, Tranel D, Damasio AR.
Impairment of social and moral behavior related to early damage
in human prefrontal cortex. Nature Neurosci 1999; 11: 1032–1037.
74 von Zerssen D, Akiskal HS. Personality factors in affective dis-
orders: historical developments and current issues with special ref-
erence to the concepts of temperament and character. J Affect Dis-
475
ord 1998; 51: 1–5.
75 Akiskal HS. Toward a definition of generalized anxiety disorder as
an anxious temperament type. Acta Psychiatr Scand Suppl 1998;
393: 66–73.
76 Perugi G, Toni C, Akiskal HS. Anxious-bipolar comorbidity. Diag-
nostic and treatment challenges. Psychiatr Clin North Am 1999; 22:
565–583.

Molecular Psychiatry

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 Reviews and Overviews

PET Imaging of Dopamine D2 Receptors

in Monkey Models of Cocaine Abuse:
Genetic Predisposition Versus Environmental Modulation

Michael A. Nader, Ph.D. Objective: Animals self-administer many
of the drugs that humans abuse, including
cocaine. This article describes studies using
Paul W. Czoty, Ph.D.
preclinical animal models to differentiate
the influences of neurobiological predis-
position from environmental modulation
of cocaine addiction, including studies
from the authors’ laboratory using nonhu-
man primates.
Method: Addiction is described in terms
of vulnerability, maintenance, and absti-
nence. This review focuses on dopamine
receptor function, in particular that of the
D2-like receptors, as measured by the non-
invasive imaging procedure positron emis-
sion tomography. Findings from human
studies of addiction and animal models
are reviewed.
Results: There appears to be an inverse
relationship between D2 receptor availabil-
ity and vulnerability to the reinforcing ef-
fects of cocaine. Environmental variables
can increase or decrease D2 receptor bind-
ing in an orderly fashion, and the resulting
changes in D2 function influence the vul-
nerability to abuse cocaine. In mainte-
nance, chronic cocaine exposure produces
decreases in D2 receptor binding, which
may be a mechanism that contributes to
continued drug use. Finally, during absti-
nence there are individual differences in
rates of recovery of D2 receptor availability.
Conclusions: The goal of the preclinical
research described in this review is to
achieve a better understanding of individ-
ual differences in susceptibility and vulner-
ability to the reinforcing effects of cocaine.
It is clear that the development of novel
animal models will extend our under-
standing of the neurobiological basis of
drug addiction to include a greater appre-
ciation of the role of environmental factors
in affecting predisposition, mediating con-
tinued drug use, and triggering relapse.

(Am J Psychiatry 2005; 162:1473–1482)

R ecent estimates indicate that the number of new co-

caine users per year more than doubled from 1992 (0.5 mil-
lion) to 2002 (1.1 million) (1). From 1994 to 2001, visits to
hospital emergency rooms that involved cocaine increased
by 22% (2). Despite more than three decades of research
into the molecular, cellular, and behavioral effects of co-
caine, no pharmacotherapy for cocaine abuse has demon-
strated sufficient safety and effectiveness for widespread
clinical use (3, 4). As astutely pointed out by Leshner (5),
most people see drug abuse and addiction as a social prob-
lem, and as such, the problem is frequently handled only
with social solutions, in particular through the criminal jus-
tice system. Recent reports indicate that the 40%–60% of
people seeking treatment are ordered to do so by the courts
or corrections system (see reference 6), suggesting that
treatment does not occur until drug use has led to substan-
tial adverse consequences to the individual. However, if
drug addiction is viewed as a brain disease, then addiction
should be amenable to treatments, just as depression,
schizophrenia, and anxiety are treatable. As Leshner stated,
Understanding addiction as a brain disease ex-
plains in part why historic policy strategies focusing
solely on the social or criminal justice aspects of drug
use and addiction have been unsuccessful. They are
missing at least half of the issue. If the brain is the core
of the problem, attending to the brain needs to be a
core part of the solution. (5, p. 47)
In response to growing evidence, the World Health Or-
ganization recently issued a report focusing on the role of
the brain in mediating drug dependence (7). The goal of
the present review is to highlight some of the brain imag-
ing studies of cocaine abuse with a special emphasis on
the use of animal models to help us understand the hu-
man condition.
Use of Brain Imaging Techniques
in Cocaine Abuse Research
Cocaine is an indirectly acting monoamine agonist, bind-
ing to the dopamine, serotonin, and norepinephrine trans-
porters (8, 9). The majority of research on cocaine has in-
volved the dopamine system, which will be the focus of this
review. Cocaine binds to the dopamine transporter and
blocks uptake of dopamine from the synapse. The elevated
levels of dopamine activate two superfamilies of dopamine

Am J Psychiatry 162:8, August 2005 http://ajp.psychiatryonline.org 1473

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 MONKEY MODELS OF COCAINE ABUSE
receptors, D1- and D2-like. This review will focus on D2-like
receptors, which have been intimately linked to drug abuse
(for instance, see references 10–12). The positron emission
tomography (PET) studies that we will describe involve ra-
diotracers that do not differentiate among subtypes of the
D2 superfamily (i.e., D2, D3, and D4 receptor subtypes).
To study the brain, powerful imaging techniques have
been developed that allow for the noninvasive exploration
of brain function in humans and in animals. In this review,
we will highlight studies utilizing PET. A clear strength of
PET is the ability to examine the brain repeatedly in longi-
tudinal studies. PET has been described as a functional
measure of brain activity because it records the uptake
and washout of a radioactive marker that competes with
endogenous neurotransmitters (13). For example, the ra-
diotracers [11C]raclopride and [18F]fluoroclebopride bind
to D2-like receptors with similar affinities. The primary
dependent variable in PET imaging studies is the ratio of
the distribution of radioligand in the region of interest
compared to its distribution in a region devoid of recep-
tors. This ratio, termed the distribution volume ratio, pro-
vides a unitless number representing the ratio of receptor
density (Bmax) to receptor affinity. In theory, changes in the
distribution volume ratio noted in longitudinal studies re-
flect changes in Bmax. However, changes in the distribu-
tion volume ratio may also reflect changes in extracellular
dopamine levels. Increases in extracellular dopamine will
decrease radioligand binding, whereas decreases in extra-
cellular dopamine will increase radioligand binding (for
example, see references 13–15). Thus, in addition to being
affected by the actual number of receptors in the tissue,
the binding of these radioligands is influenced by the
amount of dopamine present in the synapse. As a result,
when we describe PET data, we use interchangeably
“binding” and “availability,” since both are being assessed.
In contrast to PET imaging studies, in vitro receptor au-
toradiography provides a measure of receptor density in a
particular brain structure that is not influenced by the lev-
els of a neurotransmitter. The limitation, however, is that
the procedures are terminal and, consequently, only one
time point can be studied. These are important consider-
ations for longitudinal PET studies because the possibility
that changes in PET measures are due to changes in neu-
rotransmitter levels rather than the density of receptors
cannot be ruled out unless additional studies are con-
ducted. In this review, we attempt to address this issue by
also describing in vitro studies using receptor autoradiog-
raphy procedures in which circulating dopamine levels
are not influencing receptor density assessments.
Animal Models of Drug Abuse:
One of the Most Powerful Models
of Human Disease
One strategy implemented by the National Institute on
Drug Abuse is the development of novel animal models to
1474 http://ajp.psychiatryonline.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
better understand the neurochemical mechanisms medi-
ating the high abuse potential of cocaine. A greater un-
derstanding of the neuropharmacology of cocaine will
ultimately lead to improved treatment for cocaine depen-
dence. Animal models of drug self-administration have
proven to be valid predictors of human drug abuse (3, 16–
20). Animals will self-administer most of the drugs that
humans abuse, including cocaine, alcohol, nicotine, and
heroin, by the routes (intravenous, inhaled, and oral) used
by humans (17). More recently, these models have been
extended to examine behavioral and physiological re-
sponsiveness to environmental variables as indicators of
vulnerability to drug use (21). In one of the earliest studies
on individual differences and vulnerability to drug abuse,
Piazza et al. (22) initially characterized rats’ behavior in an
open-field apparatus on the basis of how much locomotor
activity was observed prior to any drug exposure. They
found that rats characterized as “high responders” in an
open field had higher basal corticosterone levels and were
more likely to self-administer d-amphetamine than rats
characterized as “low responders.” Consistent with this
characterization, noncontingent electric foot shock, which
increased corticosterone levels in rats, facilitated ac-
quisition of cocaine self-administration (23). In other ex-
periments, bilateral adrenalectomy or administration of
metyrapone, which blocks corticosterone synthesis, com-
pletely abolished the acquisition of cocaine self-adminis-
tration (24). Thus, it appears that individual differences in
vulnerability to drug abuse can be observed in animal
models and that these differences may be due, in part, to
stress-hormone responsiveness.
In addition, animal models of drug abuse have begun to
incorporate more sophisticated designs, including the in-
clusion of alternative reinforcers and the study of cocaine
choice, that may provide greater validity than models that
simply examine the reinforcing effects of drugs (see refer-
ences 25–27). For example, when monkeys are given a
choice between food and cocaine, there is a dose-depen-
dent relationship between drug dose and preference (see,
for instance, references 26–28). That is, when low cocaine
doses are the alternative to food, monkeys choose food; at
higher cocaine doses, a monkey’s preference shifts to al-
most exclusively cocaine choices. Manipulations of envi-
ronmental variables can increase or decrease cocaine
choice. For example, increases in the magnitude of the
nondrug alternative (i.e., food reinforcement) decrease
cocaine choice. Similarly, increases in the response re-
quirement for cocaine (analogous to increases in the cost
of the drug) decrease cocaine choice (26, 27, 29). The con-
verse is also true: increases in the cost of the preferred
nondrug alternative increase the frequency with which
cocaine is chosen (26, 27, 29).
These results emphasize that drug reinforcement is not
simply mediated by actions in the brain but that the envi-
ronment can alter drug reinforcement and, as we will
show, brain function. A growing body of preclinical re-
Am J Psychiatry 162:8, August 2005

search supports the hypothesis that environmental stres-
sors can enhance and environmental enrichment can at-
tenuate the reinforcing effects of drugs. This hypothesis
has clear implications for clinical outcomes. As we will dis-
cuss, these environmental stimuli affect brain function,
including processes mediated by dopamine receptors.
Compared to enrichment, more work has focused on the
role of stress in drug abuse (see reference 30), with stres-
sors including foot shock or more ethologically relevant
variables such as social stress and social defeat. We will de-
scribe studies that utilize a novel animal model of drug
abuse in which male monkeys live in social groups and
have access to cocaine daily.
To relate the preclinical findings to a more global goal of
translational research, it is important to point out several
advantages of the animal models highlighted in this arti-
cle. First, drugs can have different effects on brain func-
tion depending on whether they are administered by the
investigator or self-administered by the animal (31, 32).
Because we are interested in studying addictive behavior,
drug self-administration has face validity as an animal
model of a human condition. Second, in our studies of
nonhuman primates we also examine complex social be-
haviors and stable individual behavioral characteristics
(e.g., aggressive and affiliative behaviors in particular) that
closely model human social interactions (see reference
33). In macaque monkeys, the formation of social hierar-
chies is determined by the outcome of physical confronta-
tions (i.e., fights). The monkey that wins all of the fights in
the social group is the most dominant, the monkey that
wins all encounters except with the most dominant ani-
mal is the monkey ranked number 2, and so forth. This
hierarchy is transitory and linear, such that if number 2
wins a battle with number 3 and number 3 is ranked above
number 4, then number 2 is above number 4. We view the
linear hierarchy that forms, with the dominant monkeys at
the top and the subordinate animals at the bottom, as a
continuum from environmental enrichment at one end to
high levels of stress at the other. Socially derived stressors
have high ethological validity with regard to the study of
human drug abuse. Finally, nonhuman primates have
many advantages over other animal species. There is
abundant evidence that rodent and primate brains differ
in the anatomy, physiology, and neurochemistry of brain
dopamine systems (34–37). Furthermore, the nonhuman
primate brain differs substantially from the rodent brain
in terms of cocaine-induced changes in brain metabolism
(38, 39). Nonhuman primates can be studied in long-term
experiments with cocaine self-administration (over years)
that use within-subject designs, and they are capable of
learning complex behaviors in order to obtain drugs. The
use of PET imaging to examine neuroadaptations allows
for longitudinal examination of brain changes due to envi-
ronmental and/or pharmacological manipulations (e.g.,
references 40–42). The studies described in the following
Am J Psychiatry 162:8, August 2005
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
MICHAEL A. NADER AND PAUL W. CZOTY
sections will focus on three phases of drug addiction: vul-
nerability, maintenance, and abstinence/relapse.
Relation of D2 Receptor Binding
to Vulnerability to Cocaine Abuse
Perhaps one of the most challenging issues in drug
abuse research involves understanding the etiology of ad-
diction. Epidemiological studies suggest that approxi-
mately 17% of the people who use cocaine become de-
pendent on the drug (43, 44). Studies in humans can
generate hypotheses about potential brain markers that
may indicate a predisposition to become addicted. For
example, Volkow et al. (45) studied 23 non-drug-abusing
male subjects and used the dopamine transport inhibitor
methylphenidate as a tool to study stimulant abuse. First,
each subject was scanned with the D2 receptor ligand
[11C]raclopride. On another day, they were administered
0.5 mg/kg of methylphenidate and asked to complete an
analogue self-rating scale for pleasant drug effects. Ap-
proximately one-half (N=12) of the 23 subjects reported
liking the dose of methylphenidate, while nine of them
described it as “unpleasant” (two were indifferent to the
drug effect). Subjects who found methylphenidate “pleas-
ant” had significantly lower levels of D2 receptor binding
than subjects who reported the drug as “unpleasant.”
While these findings suggest an inverse relationship be-
tween D2 receptor availability and “vulnerability” to stim-
ulant reinforcement, because of ethical concerns humans
cannot be studied prior to drug exposure and then al-
lowed to self-administer the drug for an extended period
of time. Animal models, however, can be used to test hy-
potheses generated from studies of human drug abusers.
We explored further the relationship between D2 recep-
tors and sensitivity to psychostimulants in a novel model
of drug abuse in monkeys housed together (40). Previous
work from our group had shown a relationship between
D2 receptor availability, as measured with PET, and the so-
cial rank of female monkeys (46). Subordinate monkeys
had significantly lower levels of D2 receptor binding than
dominant monkeys. This relationship between social rank
and D2 receptors generated an interesting series of ques-
tions related not only to drug abuse but also to trait theory.
Our first question was whether D2 receptor availability in-
fluenced social rank. That is, were monkeys genetically
predisposed to a particular position in the social hierarchy
according to basal D2 receptor availability? To answer this
question, we studied 20 individually housed male mon-
keys by using the D2 receptor ligand [18F]fluoroclebopride
(47) before the monkeys were housed together. The level
of D2 receptor binding determined during individual
housing did not predict eventual social rank. That is, D2
availability was not a trait variable influencing dominance
hierarchies (40).
The next question was whether formation of the social
hierarchies influenced D2 receptor availability. When
http://ajp.psychiatryonline.org 1475
 MONKEY MODELS OF COCAINE ABUSE
TABLE 1. Relation of Dopamine D2 Function to Social Rank
in Monkeys While Individually Housed and Again After
Social Group Formationa
Distribution Volume Ratio (Bmax/Kd)
From PET Study With [18F]Fluoroclebopride
Individual Social % Overall, the findings in rodents are in the same direction
Housing Housing Change
Social Rank Mean SE Mean SE Mean SE
Dominant (N=5) 2.49 0.08 3.04 0.23 22.0 8.8
Subordinate (N=5) 2.40 0.06 2.49 0.10 3.9 5.3
a
The data were originally presented by Morgan et al. (40) in Nature
Neuroscience (http://www.nature.com/neuro) and are reprinted
with permission of Nature Publishing Group.
these animals were rescanned after 3 months of social
housing, there were significant differences between
groups, with the subordinate monkeys having signifi-
cantly lower D2 receptor binding than the dominant mon-
keys (40). This finding replicated our earlier work with fe-
male monkeys (46) and extended it to males. However,
when we compared each monkey’s [18F]fluoroclebopride
distribution volume ratio when they were individually
housed to the new ratio after social group formation, a
profound effect was noted: the average distribution vol-
ume ratios for dominant monkeys increased by over 20%,
while those for the subordinate animals were nearly un-
changed from their original baselines (Table 1). These re-
sults suggest that becoming the dominant monkey in the
social group produced large changes in dopamine recep-
tor function.
One hypothesis that could account for the observed
changes is that being the dominant monkey is analogous
to living in an enriched environment. Dominant animals
have access to treats in the pen, they are groomed more
often than subordinate monkeys, and they move about
freely (33). Enrichment could affect the PET signal (i.e.,
produce increases in the distribution volume ratio) by in-
creasing D2 receptor densities and/or decreasing levels of
extracellular dopamine in dominant monkeys. Studies us-
ing rodents have clearly shown that environmental en-
richment can affect dopamine neurotransmission in an
orderly and reliable fashion consistent with both potential
mechanisms. For example, Bowling et al. (48)) examined
the effects of different rearing environments on dopamine
synthesis and metabolism in rats. The group in the “en-
riched” condition lived together, with 12 or 13 rats per
cage, and was exposed to novel toys. Another group lived
under “impoverished” conditions consisting of individual
housing without toys in the cage. In vitro studies indicated
that the rats in the enriched condition had lower striatal
concentrations of dopamine than the rats in the impover-
ished condition. In another study, Hall et al. (49) used in
vivo microdialysis to compare rats reared in isolation with
socially reared rats, and they found that dopamine levels
in the nucleus accumbens were higher in the isolation-
reared rats than in the socially reared rats. Furthermore,
D2 receptor densities were lower in the isolation-reared
1476 http://ajp.psychiatryonline.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
rats than in the socially reared rats (49). These findings
clearly show that environmental variables as seemingly
subtle as housing conditions can produce profound
changes in the functional status of dopamine systems.
as the effects we observed after formation of social groups
by monkeys.
An important question is whether these environmentally
induced brain changes have behavioral consequences. In
particular, how do these variables influence vulnerability
to drug abuse? If, as previously noted in humans, there is a
relationship between D2 receptor availability and stimu-
lant reinforcement, then there should be differences in
drug reinforcement between dominant and subordinate
monkeys and subordinate animals should be more sensi-
tive to the effects of cocaine. This is, in fact, what we found
(40). When subsequently allowed access to cocaine, sub-
ordinate animals self-administered cocaine at higher rates
and had larger intakes than those for dominant monkeys
(Figure 1). Overall, these findings confirmed an inverse re-
lationship between D2 receptor availability and cocaine
reinforcement and suggested that environmental vari-
ables could affect brain function and vulnerability to co-
caine abuse.
Decrease of D2 Receptor Binding
With Continued Cocaine Use
Human PET imaging studies have shown lower D2 re-
ceptor availability in cocaine abusers than in age-matched
comparison subjects (11). However, as already described,
it is not clear whether the lower D2 receptor binding was a
predisposing trait or a consequence of cocaine exposure
(see reference 11). Understanding how long-term cocaine
exposure affects dopamine receptor function could ulti-
mately lead to better treatment strategies. One hypothesis
is that D2 receptor down-regulation occurs as an adapta-
tion to the chronic elevation in extracellular dopamine
due to chronic blockade of dopamine uptake. A strength of
using animal subjects is the ability to study drug-naive
subjects and to observe changes in receptor availability
with exposure to a drug by means of a within-subjects,
longitudinal design (see reference 50, for example).
We recently extended our studies of socially housed
monkeys to examine the effects of chronic cocaine expo-
sure in dominant and subordinate monkeys (51). Whereas
dominant monkeys were initially protected from the re-
inforcing effects of cocaine by elevated D2 availability,
chronic exposure to self-administered cocaine resulted in
D2 measures that were no longer different from those of
subordinate monkeys (Figure 2). These findings suggested
that exposure to cocaine attenuated or reversed the pow-
erful effects of environmental context on dopamine re-
ceptor availability. Studies utilizing in vitro receptor auto-
radiography of D2 receptors have consistently shown
lower receptor densities in monkeys with long-term his-
Am J Psychiatry 162:8, August 2005

FIGURE 1. Relationship of Cocaine Self-Administration to Social Rank in Monkeysa
Injections (per session)
50
Mean Number of
40
30
20
10
0
Saline 0.003
2.0
(mg/kg per session)
Mean Intake
1.5
1.0
0.5
0.0
0.003
a The values are averages from the last 3 days that each dose was available for self-administration. The data are adapted from an article by
Morgan et al. (40) in Nature Neuroscience (http://www.nature.com/neuro) and are reprinted with permission of Nature Publishing Group.
b Significantly different from value for dominant group (p<0.05).
tories of cocaine self-administration; many of these ef-
fects were directly related to dose and duration of expo-
sure (52–54).
A central issue for this phase of addiction is one that was
raised for the first phase, namely individual differences.
For example, one issue to consider in addition to the dose
of a drug taken in the lifetime of the individual is the pat-
tern and duration of drug use. Volkow et al. (11) noted that
there was not a relationship between D2 receptor binding
and the dose of cocaine used, but there was a significant
correlation between D2 binding and duration of cocaine
use. The effects of duration and cocaine intake can be ad-
dressed with animal models in which cocaine availability
is the primary independent variable. The question could
be phrased as, Does administering a large amount of co-
caine over a short period of time produce greater long-
term effects on dopamine receptor function than does ad-
ministering moderate doses over longer periods? In a
study not utilizing cocaine self-administration, in vitro re-
ceptor autoradiography was used to assess the conse-
quences of chronic cocaine use and abstinence on D2 re-
ceptor levels (55). Monkeys were treated four times per
day with cocaine for 2 consecutive weeks, followed by a 2-
week withdrawal period. The investigators found no dif-
ferences between cocaine-treated and control monkeys in
D2 receptor densities in the caudate nucleus, prefrontal
cortex, substantia nigra, and nucleus accumbens. How-
ever, this amount of cocaine administered has since been
shown to produce robust decreases in D2 receptor densi-
ties when self-administered over longer periods of time
(see, for instance, references 52 and 53). Such findings
suggest that the duration of exposure is as important as
the actual dose of drug administered and reinforce the
Am J Psychiatry 162:8, August 2005
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
MICHAEL A. NADER AND PAUL W. CZOTY
Dominant monkeys (N=5) Subordinate monkeys (N=4)
b
b
0.01 0.03 0.10
b
b
0.01 0.03 0.10
Dose of Cocaine (mg/kg per injection)
importance of using contingent drug administration in
animal models of addiction. Future studies using in vivo
imaging techniques will better address this extremely im-
portant question.
As we discussed for the vulnerability phase, an important
question is whether there are functional consequences to
cocaine-induced changes in D2 receptor binding. Of par-
ticular interest are issues related to cognitive function, de-
cision making, and choice behavior. For example, Grant et
al. (56) compared a group of drug abusers (opiate and/or
stimulant) and non-drug-abusing comparison subjects on
a series of neuropsychological tests to examine the long-
term consequences of drug use on decision making. In
one set of studies, subjects were exposed to the “gambling
task” (57), which has strong face validity for evaluating
cognitive deficits (56). The task involved four different
decks of cards that differed along three dimensions: im-
mediate gain, long-term expected gain, and schedule of
penalties. Two of the four decks had smaller gains and
punishers but higher overall “yields”; the comparison sub-
jects chose from those decks most often. In contrast, the
majority of the drug abusers chose from the low-yield
decks, indicating poor decision making. To control for
possible differences in IQ, Grant et al. (56) also used the
Wisconsin Card Sorting Test and found no differences be-
tween the groups. The authors concluded that impair-
ments in decision making could certainly account for con-
tinued use of drugs in the presence of adverse social and
personal consequences.
Studies with cocaine abusers cannot clarify whether
such cognitive impairments are a consequence of exces-
sive cocaine use or reflect preexisting decrements that pre-
dispose certain individuals to drug abuse. Animal studies
http://ajp.psychiatryonline.org 1477
 MONKEY MODELS OF COCAINE ABUSE
FIGURE 2. PET Images of [18F]Fluoroclebopride in Dominant and Subordinate Monkeys With Extensive Histories of Cocaine
Self-Administrationa
C-6528
Dominant
a The images were coregistered with MRIs from the same monkeys and are shown at the level of the striatum (caudate and putamen). From
experiments described by Czoty et al. (51).
can address this issue. Using our model with group-housed
monkeys, we studied choice behavior involving cocaine
and a nondrug reinforcer—banana-flavored pellets (27).
Subordinate monkeys were more sensitive to the reinforc-
ing effects of cocaine than were dominant monkeys. That
is, subordinate monkeys chose cocaine over food at a lower
cocaine dose than that for dominant monkeys. Such find-
ings suggest that environmental context can substantially
influence the preference to choose drug over nondrug al-
ternatives even when D2 binding does not differ between
groups.
Variable Recovery of D2 Receptor
Binding During Cocaine Abstinence
The final phase of drug addiction to be discussed is ab-
stinence and the variables that influence relapse. As with
vulnerability and maintenance, we will focus on dopa-
mine D2 receptor availability. From a clinician’s viewpoint,
abstinence is the critical phase because the population
seeking professional guidance and treatment are in this
phase; it is worth pointing out that this is the phase of drug
addiction about which we know and understand the least.
The preceding issues related to vulnerability (predisposi-
tion) and maintenance (i.e., changes in brain function due
to chronic drug exposure) are critical in understanding
brain changes associated with abstinence. However, un-
derstanding these variables may provide only a glimpse of
the complexity involved in brain changes during absti-
nence. Perhaps Childress et al. stated it best: “This search
[for a cocaine medication] has been complicated by the
heterogenous target: drug desire that emerges during co-
caine cessation…may well have a different brain substrate
than desire induced by cocaine itself…and the cues that
signal it” (58, p. 11).
1478 http://ajp.psychiatryonline.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
C-6627
Subordinate
Over a decade ago, Volkow and colleagues (59) reported
significantly lower D2 receptor binding in cocaine abusers
abstinent from cocaine for up to 4 months. These reduc-
tions in D2 receptor binding were associated with de-
creased glucose utilization in the orbitofrontal cortex and
with self-ratings of dysphoria (11). Thus, there was an as-
sociation between dopamine receptor availability and
negative subjective effects during abstinence. The study
ended 4 months after the start of abstinence because all
subjects had relapsed. Animal models, on the other hand,
can provide valuable insight into the neuropharmacology
of abstinence and relapse because abstinence can be
studied for years.
Our first efforts to study D2 receptor function during ab-
stinence involved using PET imaging to compare D2 re-
ceptor binding in monkeys with extensive histories of co-
caine self-administration and cocaine-naive comparison
monkeys (i.e., a group design). For example, the first mon-
key we studied had self-administered cocaine for over 3
years and had a lifetime cocaine intake of approximately
20 g. PET scans were conducted at various time points
during abstinence. Differences in [18F]fluoroclebopride
binding between this cocaine-experienced monkey and a
cocaine-naive monkey were apparent at all time points
and, as shown in Figure 3, did not dissipate even after 7
months of abstinence. The image from a comparison
monkey shows excellent uptake of [18F]fluoroclebopride
into D2-rich regions of the basal ganglia. In the abstinent
cocaine-experienced monkey, D2 receptor binding was
more than 20% lower. While it is possible that the lower D2
signal in the cocaine-experienced monkey was due to ele-
vated levels of dopamine, it is unlikely that such an effect
would persist for 7 months. The other possibility is that
chronic cocaine exposure resulted in persistent reduc-
tions in D2 receptor densities. This hypothesis has been
Am J Psychiatry 162:8, August 2005
 MICHAEL A. NADER AND PAUL W. CZOTY

FIGURE 3. PET Images of [18F]Fluoroclebopride in Cocaine-Naive and Cocaine-Experienced Monkeysa

Cocaine-experienced monkey
Cocaine-naive monkey
after 227 days of abstinence

a Images are shown at the level of the basal ganglia. From an unpublished study by M.A. Nader and R.H. Mach.

confirmed by means of in vitro receptor autoradiography
(52, 53; see reference 54 for further discussion).
It is necessary to comment on animal models and “crav-
ing.” As defined by Hommer, craving “is a term derived
from popular psychology that is used to describe one of the
mental states—namely, the intense desire for a certain ob-
ject or experience” (60, p. 187). Assessing a mental state in
an animal is not a realistic undertaking. However, we can
measure behavior and hypothesize about mental states.
Describing the multitude of animal models of craving is
beyond the scope of this review. In brief, the strategies can
involve persistence of drug seeking in the presence of stim-
uli that signal no drug is available (61), escalation of drug
seeking with changes in drug availability (62), responding
in the presence of conditioned stimuli (63), or a return of
responding that had been previously extinguished (i.e., re-
instatement) (64).
The reinstatement procedure has been widely used as a
model of the ability of priming injections of drugs, drug-
related cues, and stress to precipitate relapse into drug
taking (reviewed in reference 65). Although the predictive
and construct validity of the procedure remain to be firmly
established (66), the ability of noncontingent administra-
tion of cocaine to reinstate extinguished responding previ-
ously maintained by cocaine has been well established
(65). Furthermore, it has been demonstrated that drugs
that directly or indirectly activate the dopamine system
can reinstate cocaine seeking under a food-drug choice
procedure (67). Thus, the reinstatement procedure al-
lowed us another means to assess potential differences in
dopamine receptor function between dominant and sub-
ordinate monkeys. Socially housed monkeys were allowed
to self-administer cocaine under conditions in which they
also could choose food reinforcement. Under these condi-
tions, as the cocaine dose increased, the percentage of to-
tal responses involving the cocaine-associated lever in-
creased (27). When saline was substituted for cocaine, the
FIGURE 4. Effect of a Noncontingent Cocaine Injection on
Individual Dominant (D1–D4) and Subordinate (S1–S4)
Monkeys’ Responses on a Lever Previously Associated With
Cocaine but Currently Producing Saline Injectionsa
Dominant Subordinate
Percentage of Cocaine-Lever Responses
100
75
50
25
0
After D1 D2 D3 D4 After S1 S2 S3 S4
Noncon- After Noncon- After
tingent Noncontingent tingent Noncontingent
Saline 0.56 mg/kg Saline 0.56 mg/kg
(N=4) Cocaine (N=4) Cocaine
a The first bar for each group represents the average injection-lever re-
sponding following noncontingent saline injection. A significant dif-
ference was observed after noncontingent injection of 0.56 mg/kg co-
caine between the dominant and subordinate monkeys. The data are
from an unpublished study by P.W. Czoty, C. McCabe, and M.A. Nader.
monkeys shifted their responses to the food-associated le-
ver (Figure 4). We next examined the ability of noncontin-
gent cocaine to reinstate responses on the cocaine-associ-
ated lever, although these responses continued to produce
saline injections. Noncontingent cocaine produced dose-
related increases in responses on the cocaine-associated
lever at doses up to 0.30 mg/kg in all monkeys (data not
shown). However, dominant and subordinate monkeys
differed in their responses to a higher cocaine dose (0.56
mg/kg). All four subordinate monkeys responded prima-

Am J Psychiatry 162:8, August 2005 http://ajp.psychiatryonline.org 1479

Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159

 MONKEY MODELS OF COCAINE ABUSE
rily on the injection lever following this dose of cocaine
(Figure 4). Greater variability was observed in the domi-
nant monkeys’ responses to this dose. These data are con-
sistent with our earlier results demonstrating that subor-
dinate monkeys are, on average, more sensitive to the
reinforcing strength of cocaine and/or less sensitive to the
aversive effects of cocaine than are dominant monkeys.
Conclusions
In this review we have attempted to briefly highlight the
use of animal models and brain imaging procedures (PET
and in vitro receptor autoradiography) to convey a better
understanding of the neuropharmacology of drug addic-
tion. The focus on dopamine D2 receptors can be viewed
as a strength as well as a limitation. The strength of this fo-
cus is the understanding of how the level of this receptor
superfamily serves as a trait influencing vulnerability to
drug abuse and how it serves as a potential state variable
showing malleability as a result of environmental or phar-
macological manipulations. The weakness of this focus on
D2 receptors is the fact that drug addiction is not medi-
ated by one receptor subtype. Future studies must exam-
ine how other neurotransmitter and neurohormone sys-
tems change in concert with brain dopamine systems, as
well as the development of novel radioligands for specific
receptor subtypes (e.g., D3 receptors). In addition, in-
creased anatomical selectivity with higher-resolution PET
cameras will better address the contribution of particular
brain circuitry to behavioral effects.
The use of animal models to study human disease is an
extremely important tool. We have described models that
could be characterized as predictive, isomorphic, and ho-
mologous of the human condition. In particular, the use of
socially housed nonhuman primates to better understand
the neuropharmacology and the behavioral and social
consequences of drug abuse and to evaluate potential
treatment strategies has unequivocal potential. It is our
hope that clinicians reading this review will gain a better
understanding of the research questions being asked and
the potential importance of the results for the develop-
ment of therapies (behavioral and pharmacological) for
drug addiction.
Finally, this review has provided evidence that the envi-
ronment can profoundly affect drug abuse vulnerability,
maintenance, and relapse. Environmental stress and en-
richment can influence brain function and vulnerability
to drug abuse. Furthermore, even following long-term
drug use, environmental variables can affect the behav-
ioral effects of cocaine. The research described in this
review represents the growing number of studies doc-
umenting the benefits of environmental enrichment, ir-
respective of genetic predispositions to abuse drugs. Such
preclinical findings suggest that outcome measures will
be enhanced not only by taking individuals out of a stress-
ful environment but also by providing alternative rein-
1480 http://ajp.psychiatryonline.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
forcers—whether these are better living conditions, jobs,
or other activities (68).
Presented at the 157th annual meeting of the American Psychiatric
Association, New York, May 1–6, 2004. Received Sept. 15, 2004; revi-
sion received Feb. 14, 2005; accepted Feb. 22, 2005. From the Center
for the Neurobiological Investigation of Drug Abuse, Department of
Physiology and Pharmacology and Department of Radiology, Wake
Forest University School of Medicine. Address correspondence and
reprint requests to Dr. Nader, Department of Physiology and Phar-
macology, Wake Forest University School of Medicine, Winston-Sa-
lem, NC 27157; mnader@wfubmc.edu (e-mail).
Supported by National Institute on Drug Abuse grants DA-10584,
DA-14637, DA-09085, and P50 DA-06634.
References
1. 2003 National Survey on Drug Use and Health (NSDUH). Rock-
ville, Md, US Department of Health and Human Services, Sub-
stance Abuse and Mental Health Services Administration, Na-
tional Institute on Drug Abuse, 2004
2. Emergency Department Trends From the Drug Abuse Warning
Network, Final Estimates 1994–2001. Rockville, Md, US Depart-
ment of Health and Human Services, Substance Abuse and
Mental Health Services Administration, National Institute on
Drug Abuse, 2002
3. Mello NK, Negus SS: Preclinical evaluation of pharmacothera-
pies for treatment of cocaine and opioid abuse using drug self-
administration procedures. Neuropsychopharmacology 1996;
14:375–424
4. Platt DM, Rowlett JK, Spealman RD: Behavioral effects of co-
caine and dopaminergic strategies for preclinical medication
development. Psychopharmacology (Berl) 2002; 163:265–282
5. Leshner AI: Addiction is a brain disease, and it matters. Science
1997; 278:45–47
6. McLellan AT: Crime and punishment and treatment: latest
findings in the treatment of drug-related offenders (editorial). J
Subst Abuse Treat 2003; 25:187–188
7. World Health Organization: Neuroscience of Psychoactive Sub-
stance Use and Dependence. Geneva, WHO, 2004
8. Madras BK, Fahey MA, Bergman J, Canfield DR, Spealman RD:
Effects of cocaine and related drugs in nonhuman primates, I:
[3H]cocaine binding sites in caudate-putamen. J Pharmacol
Exp Ther 1989; 251:131–141
9. Ritz MC, Lamb RJ, Goldberg SR, Kuhar MJ: Cocaine receptors on
dopamine transporters are related to self-administration of co-
caine. Science 1987; 237:1219–1223
10. Blum K, Cull JC, Braverman ER, Comings DE: Reward deficiency
syndrome. Am Sci 1996; 84:132–145
11. Volkow ND, Fowler JS, Wang GJ, Hitzemann R, Logan J, Schlyer
DJ, Dewey SL, Wolf AP: Decreased dopamine D2 receptor avail-
ability is associated with reduced frontal metabolism in co-
caine abusers. Synapse 1993; 14:169–177
12. Young RM, Lawford BR, Nutting A, Noble EP: Advances in mo-
lecular genetics and the prevention and treatment of sub-
stance misuse: implications of association studies of the A1 al-
lele of the D2 dopamine receptor gene. Addict Behav 2004;
29:1275–1294
13. Laruelle M: Imaging synaptic neurotransmission with in vivo
binding competition techniques: a critical review. J Cereb
Blood Flow Metab 2000; 20:423–451
14. Dewey SL, Smith GS, Logan J, Brodie JD, Yu DW, Ferrieri RA, King
PT, MacGregor RR, Martin TP, Wolf AP, Volkow ND, Fowler JS,
Meller E: GABAergic inhibition of endogenous dopamine re-
lease measured in vivo with 11C-raclopride and positron emis-
sion tomography. J Neurosci 1992; 12:3773–3780
Am J Psychiatry 162:8, August 2005

15. Mach RH, Nader MA, Ehrenkaufer RL, Line SW, Smith CR, Gage
HD, Morton TE: Use of positron emission tomography to study
the dynamics of psychostimulant-induced dopamine release.
Pharmacol Biochem Behav 1997; 57:477–486
16. Spealman RD, Goldberg SR: Drug self-administration by labora-
tory animals: control by schedules of reinforcement. Annu Rev
Pharmacol Toxicol 1978; 18:313–339
17. Griffiths RR, Bigelow GE, Henningfield JE: Similarities in animal
and human drug-taking behavior, in Advances in Substance
Abuse, vol 1. Edited by Mello NK. Greenwich, Conn, JAI Press,
1980, pp 1–90
18. Johanson CE, Fischman MW: The pharmacology of cocaine re-
lated to its abuse. Pharmacol Rev 1989; 41:3–52
19. Woolverton WL, Nader MA: Experimental evaluation of the re-
inforcing effects of drugs, in Testing and Evaluation of Drugs of
Abuse. Edited by Adler MW, Cowan A. New York, Wiley-Liss,
1990, pp 165–192
20. Koob GF, Sanna PP, Bloom FE: Neuroscience of addiction. Neu-
ron 1998; 21:467–476
21. Koob GF, Le Moal M: Drug addiction, dysregulation of reward,
and allostasis. Neuropsychopharmacology 2001; 24:97–129
22. Piazza PV, Deminiere JM, Le Moal M, Simon H: Factors that pre-
dict individual vulnerability to amphetamine self-administra-
tion. Science 1989; 245:1511–1513
23. Goeders NE, Guerin GF: Non-contingent electric footshock facil-
itates the acquisition of intravenous cocaine self-administra-
tion in rats. Psychopharmacology 1994; 114:63–70
24. Goeders NE, Guerin GF: Effects of surgical and pharmacological
adrenalectomy on the initiation and maintenance of intrave-
nous cocaine self-administration in rats. Brain Res 1996; 722:
145–152
25. Paronis CA, Gasior M, Bergman J: Effects of cocaine under con-
current fixed ratio schedules of food and IV drug availability: a
novel choice procedure in monkeys. Psychopharmacology
(Berl) 2002; 163:283–291
26. Negus SS: Rapid assessment of choice between cocaine and
food in rhesus monkeys: effects of environmental manipula-
tions and treatment with d-amphetamine and flupenthixol.
Neuropsychopharmacology 2003; 28:919–931
27. Czoty PW, McCabe C, Nader MA: Assessment of the relative re-
inforcing strength of cocaine in socially housed monkeys using
a choice procedure. J Pharmacol Exp Ther 2005; 312:96–102
28. Nader MA, Wolverton WL: Effects of increasing the magnitude
of an alternative reinforcer on drug choice in a discrete-trials
choice procedure. Psychopharmacology (Berl) 1991; 105:169–
174
29. Nader MA, Woolverton WL: Effects of increasing response re-
quirement on choice between cocaine and food in rhesus
monkeys. Psychopharmacology (Berl) 1992; 108:295–300
30. Goeders NE: Stress, the hypothalamic-pituitary-adrenal axis,
and vulnerability to drug abuse. NIDA Res Monogr 1998; 169:
83–104
31. Dworkin SI, Mirkis S, Smith JE: Response-dependent versus re-
sponse-independent presentation of cocaine: differences in
the lethal effects of the drug. Psychopharmacology (Berl) 1995;
117:262–266
32. Bradberry CW: Acute and chronic dopamine dynamics in a
nonhuman primate model of recreational cocaine abuse. J
Neurosci 2000; 20:7109–7115
33. Kaplan JR, Manuck SB, Clarkson TB, Lusso FM, Taub DM: Social
status, environment, and atherosclerosis in cynomolgus mon-
keys. Arteriosclerosis 1982; 2:359–368
34. Berger B, Gaspar P, Verney C: Dopaminergic innervation of the
cerebral cortex: unexpected differences between rodents and
primates. Trends Neurosci 1991; 14:21–27
35. Haber SN, McFarland NR: The concept of the ventral striatum
in nonhuman primates. Ann NY Acad Sci 1999; 877:33–48
Am J Psychiatry 162:8, August 2005
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
MICHAEL A. NADER AND PAUL W. CZOTY
36. Joel D, Weiner I: The connections of the dopaminergic system
with the striatum in rats and primates: an analysis with respect
to the functional and compartmental organization of the stria-
tum. Neuroscience 2000; 96:451–474
37. Cragg SJ, Hille CJ, Greenfield SA: Dopamine release and uptake
dynamics within nonhuman primate striatum in vitro. J Neuro-
sci 2000; 20:8209–8217
38. Lyons D, Friedman DP, Nader MA, Porrino LJ: Cocaine alters ce-
rebral metabolism within the ventral striatum and limbic cor-
tex of monkeys. J Neurosci 1996; 16:1230–1238
39. Porrino LJ, Lyons D, Miller MD, Smith HR, Friedman DP, Daunais
JB, Nader MA: Metabolic mapping of the effects of cocaine dur-
ing the initial phases of self-administration in the nonhuman
primate. J Neurosci 2002; 22:7687–7694
40. Morgan D, Grant KA, Gage HD, Mach RH, Kaplan JR, Prioleau O,
Nader SH, Buchheimer N, Ehrenkaufer RL, Nader MA: Social
dominance in monkeys: dopamine D2 receptors and cocaine
self-administration. Nat Neurosci 2002; 5:169–174
41. Howell LL, Hoffman JM, Votak JR, Landrum AM, Wilcox KM,
Lindsey KP: Cocaine-induced brain activation determined by
positron emission tomography neuroimaging in conscious
rhesus monkeys. Psychopharmacology (Berl) 2002; 159:154–
160
42. Lindsey KP, Wilcox KM, Votak JR, Goodman MM, Plisson C, Car-
roll FI, Rice KC, Howell LL: Effects of dopamine transporter in-
hibitors on cocaine self-administration in rhesus monkeys: re-
lationship to transporter occupancy determined by positron
emission tomography neuroimaging. J Pharmacol Exp Ther
2004; 309:959–969
43. Anthony JC, Warner LA, Kessler RC: Comparative epidemiology
of dependence on tobacco, alcohol, controlled substances,
and inhalants: basic findings from the National Comorbidity
Survey. Exp Clin Psychopharmacol 1994; 2:244–268
44. Wagner FA, Anthony JC: From first drug use to drug depen-
dence: developmental periods of risk for dependence upon
marijuana, cocaine, and alcohol. Neuropsychopharmacology
2002; 26:479–488
45. Volkow ND, Wang G-J, Fowler JS, Logan J, Gatley SJ, Gifford A,
Hitzemann R, Ding Y-S, Pappas N: Prediction of reinforcing re-
sponses to psychostimulants in humans by brain dopamine D2
receptor levels. Am J Psychiatry 1999; 156:1440–1443
46. Grant KA, Shively CA, Nader MA, Ehrenkaufer RL, Line SW, Mor-
ton TE, Gage HD, Mach RH: The effect of social status on striatal
dopamine D2 receptor binding characteristics in cynomolgus
monkeys assessed with positron emission tomography. Syn-
apse 1998; 29:80–83
47. Mach RH, Luedtke RR, Unsworth CD, Boundy VA, Nowak PA,
Scripko JG, Elder ST, Jackson JR, Hoffman PL, Evora PH, Rao AV,
Moilinoff PB, Childers SR, Ehrenkaufer RLE: 18F-Labeled radioli-
gands for studying the dopamine D2 receptor with positron
emission tomography. J Med Chem 1993; 36:3707–3720
48. Bowling SL, Rowlett JK, Bardo MT: The effect of environmental
enrichment on amphetamine-stimulated locomotor activity,
dopamine synthesis and dopamine release. Neuropharmacol-
ogy 1993; 32:885–893
49. Hall FS, Wilkinson LS, Humby T, Inglis W, Kendall DA, Marsden
CA, Robbins TW: Isolation rearing in rats: pre- and postsynaptic
changes in striatal dopaminergic systems. Pharmacol Biochem
Behav 1998; 59:859–872
50. Thanos PK, Volkow ND, Freimuth P, Umegaki H, Ikari H, Roth G,
Ingram DK, Hitzemann R: Overexpression of dopamine D2 re-
ceptors reduces alcohol self-administration. J Neurochem
2001; 78:1094–1103
51. Czoty PW, Morgan D, Shannon EA, Gage HD, Nader MA: Charac-
terization of dopamine D1 and D2 receptor function in socially
housed cynomolgus monkeys self-administering cocaine. Psy-
chopharmacology (Berl) 2004; 174:381–388
http://ajp.psychiatryonline.org 1481
 MONKEY MODELS OF COCAINE ABUSE
52. Moore RJ, Vinsant SL, Nader MA, Porrino LJ, Friedman DP: Ef-
fect of cocaine self-administration on dopamine D2 receptors
in rhesus monkeys. Synapse 1998; 30:88–96
53. Nader MA, Daunais JB, Moore T, Nader SH, Moore RJ, Smith HR,
Friedman DP, Porrino LJ: Effects of cocaine self-administration
on striatal dopamine systems in rhesus monkeys: initial and
chronic exposure. Neuropsychopharmacology 2002; 27:35–46
54. Porrino LJ, Lyons D, Letchworth SR, Freedland CS, Nader MA:
Structural and functional neuroimaging of the effects of co-
caine in human and nonhuman primates, in Handbook of
Neurotoxicology, vol 2. Edited by Massaro EJ. Totowa, NJ, Hu-
mana Press, 2002, pp 413–435
55. Farfel GM, Kleven MS, Woolverton WL, Seiden LS, Perry BD: Ef-
fects of repeated injections of cocaine on catecholamine re-
ceptor binding sites, dopamine transporter binding sites and
behavior in rhesus monkeys. Brain Res 1992; 578:235–243
56. Grant S, Contoreggi C, London ED: Drug abusers show im-
paired performance in a laboratory test of decision making.
Neuropsychologia 2000; 38:1180–1187
57. Bechara A, Damasio AR, Damasio H, Anderson SW: Insensitivity
to future consequences following damage to human prefron-
tal cortex. Cognition 1994; 50:7–15
58. Childress AR, Mozley PD, McElgin W, Fitzgerald J, Reivich M,
O’Brien CP: Limbic activation during cue-induced cocaine crav-
ing. Am J Psychiatry 1999; 156:11–18
59. Volkow ND, Fowler JS, Wolf AP, Schlyer D, Shiue C-Y, Alpert R,
Dewey SL, Logan J, Bendriem B, Christman D, Hitzemann R,
1482 http://ajp.psychiatryonline.org
Tatiane Bezerra - tatyoshizawa@hotmail.com - IP: 179.214.167.159
Henn F: Effects of chronic cocaine abuse on postsynaptic
dopamine receptors. Am J Psychiatry 1990; 147:719–724
60. Hommer DM: Functional imaging of craving. Alcohol Res
Health 1999; 23:187–196
61. Deroche-Gamonet V, Belin D, Piazza PV: Evidence for addic-
tion-like behavior in the rat. Science 2004; 305:1014–1017
62. Ahmed SH, Koob GF: Transition from moderate to excessive
drug intake: change in hedonic set point. Science 1998; 282:
298–300
63. Vanderschuren LJMJ, Everitt BJ: Drug seeking becomes compul-
sive after prolonged cocaine self-administration. Science 2004;
305:1017–1019
64. de Wit H, Stewart J: Reinstatement of cocaine-reinforced re-
sponding in the rat. Psychopharmacology (Berl) 1981; 75:134–
143
65. Shaham Y, Shalev U, Lu L, De Wit H, Stewart J: The reinstate-
ment model of drug relapse: history, methodology and major
findings. Psychopharmacology (Berl) 2003; 168:3–20
66. Katz JL, Higgins ST: The validity of the reinstatement model of
craving and relapse to drug use. Psychopharmacology (Berl)
2003; 168:21–30
67. Gasior M, Paronis CA, Bergman J: Modification by dopaminer-
gic drugs of choice behavior under concurrent schedules of in-
travenous saline and food delivery in monkeys. J Pharmacol
Exp Ther 2004; 308:249–259
68. Higgins ST: The influence of alternative reinforcers on cocaine
use and abuse: a brief review. Pharmacol Biochem Behav
1997; 57:419–427
Am J Psychiatry 162:8, August 2005