REVIEWS

The neural bases for timing

of durations
Albert Tsao1 ✉, S. Aryana Yousefzadeh 2
, Warren H. Meck2, May-Britt Moser 3

and Edvard I. Moser 3 ✉

Abstract | Durations are defined by a beginning and an end, and a major distinction is drawn
between durations that start in the present and end in the future (‘prospective timing’) and
durations that start in the past and end either in the past or the present (‘retrospective timing’).
Different psychological processes are thought to be engaged in each of these cases. The former is
thought to engage a clock-like mechanism that accurately tracks the continuing passage of time,
whereas the latter is thought to engage a reconstructive process that utilizes both temporal and
non-temporal information from the memory of past events. We propose that, from a biological
perspective, these two forms of duration estimation are supported by computational processes
that are both reliant on population state dynamics but are nevertheless distinct. Prospective timing
is effectively carried out in a single step where the ongoing dynamics of population activity directly
serve as the computation of duration, whereas retrospective timing is carried out in two steps: the
initial generation of population state dynamics through the process of event segmentation and
the subsequent computation of duration utilizing the memory of those dynamics.

Prospective timing
The estimation of an ongoing
duration in the present moment.
Events
The basic units of organization
for experience, defined
primarily as perceived time
intervals whose beginnings
and ends are clearly defined.
Retrospective timing
The estimation of duration
based on memory of past
events.
1
Department of Biology,
Stanford University,
Stanford, CA, USA.
2
Department of Psychology
and Neuroscience, Duke
University, Durham, NC, USA.
3
Centre for Neural
Computation, Kavli Institute
for Systems Neuroscience,
Norwegian University of
Science and Technology,
Trondheim, Norway.
✉e-mail: at7@stanford.edu;
edvard.moser@ntnu.no
https://doi.org/10.1038/
s41583-022-00623-3
How the brain naturally perceives and retains time
has been the focus of many studies over the past few
decades1–8. As time is composed of the two concepts
of order and duration9,10, this question may be divided
into separate questions of how the brain perceives and
remembers either order or duration. In this Review, we
focus on duration, and more specifically on whether the
processes for perceiving ongoing duration in the pres-
ent moment (prospective timing) are the same as those
used for estimating duration based on memory of past
events (retrospective timing) (Box 1 and Fig. 1). The most
basic requirement for timing is that each moment in
time can be uniquely represented, and current evidence
suggests that both types of timing achieve this require-
ment through population-level representations of time.
However, how these representations arise and how they
are subsequently used for estimating duration may differ
substantially, and we examine and discuss the evidence
for this. We also discuss a number of potentially impor-
tant ways in which these two types of timing and their
mechanisms might interact, although investigation of
these interactions still remains preliminary.
Prospective timing
The main computational goal for prospective timing
is to accurately track elapsed time within an ongoing
interval. Consequently, a major class of models for pro-
spective timing is based on the concept of an internal
clock, which generates an explicit linear metrical rep-
resentation of time that can be used for estimation of
duration11–13. Early models proposed that the internal
clock would be implemented through a pacemaker–
accumulator mechanism11–13 in which pulses gener-
ated by a central pacemaker would be integrated by an
accumulator to generate estimates of time. However,
outside the circadian system14,15, evidence for biological
implementation of such a mechanism has been miss-
ing. Subsequently, it has been proposed that the internal
clock is implemented through coincidence detection
of oscillatory activity patterns16–18, but, even though
changes in oscillatory activity have been correlated to
timing behaviour19–21, evidence of coincidence detection
has also been missing.
Although pacemaker–accumulator models are intu-
itively familiar owing to their similarity to physical
clocks, a regular pacemaker is not actually needed for
the purpose of timing duration22. Instead, time can be
encoded through changes in neural population activity
over time, or population state dynamics, as long as each
moment in physical time corresponds to a unique popu­
lation activity pattern. This type of timing mechanism
has sometimes been referred to as a ‘population clock’, and
it does not necessarily need to generate a linear metri-
cal representation of time23,24. Additionally, it subsumes
the pacemaker–accumulator mechanism into a single
process — change in neural activity25. In the following

NATuRe RevIeWS | Neuroscience

0123456789();:
Reviews
Box 1 | Terminology
In this Review, we use the terms ‘prospective timing’ and ‘retrospective timing’
as shorthand for duration estimated from ongoing events and duration estimated
from the memory of past events, respectively. Although usage of these terms is well
established160,161,334, there are a few points to clarify regarding their usage. First,
retrospective timing is operationally defined as the estimation of a duration when
subjects are unaware beforehand that they will estimate the duration, whereas
prospective timing is defined as estimation of duration when subjects are aware
beforehand161. The logic of this operational definition is that, particularly during
prospective timing, subjects have the opportunity to explicitly track the passage
of time. Thus, in the case where estimation occurs after the duration has ended but
subjects are already aware beforehand that they will eventually estimate duration,
this would still be considered prospective timing. Second, ‘prospective timing’ and
‘retrospective timing’ refer specifically to the estimation of duration, even though
timing encompasses temporal order in addition to duration. Whether there exists
a similar distinction between temporal order based on ongoing events as opposed
to memory of past events is unclear. Finally, ‘prospective timing’ overlaps with
other commonly used terms such as ‘interval timing’ and ‘reward timing’, whereas
‘retrospective timing’ partially overlaps with ‘episodic timing’, and each has also been
referred to as ‘experienced duration’ and ‘remembered duration’, respectively335.
subsections, we discuss how prospective timing is car-
ried out using population state dynamics, primarily
focusing on a number of recent studies that have actu-
ally observed population state dynamics by recording
from single neurons in animals engaged in duration
estimation tasks.
Explicit prospective timing
Prospective timing can occur either explicitly or implic-
itly26–28. During explicit timing, subjects are aware that they
should specifically attend to the passage of time in order
to either perceive a duration defined by external sen-
sory events (‘sensory timing’) or time their own action
(‘motor timing’). Recent studies recording single-cell
activity from brain areas involved in prospective timing
provide evidence that during both sensory timing29–33
and motor timing34–42, as well as combined sensorimo-
tor timing43,44, neural population activity evolves in a
stereotyped way across a series of population activity
states, defining a stable ‘neural trajectory’ which provides
Population clock
significant and reliable temporal information (Fig. 2).
The encoding of temporal Moreover, inactivation of brain areas exhibiting trajecto-
information through changes ries negatively impacts duration estimation30,35,39,43,45, and
in neural population activity trial-to-trial variability of trajectories has been observed
over time.
to be correlated with animals’ time estimates30,35,37,38,40,
Explicit timing together suggesting a causal link from trajectories to
Prospective timing in which duration estimation. Beyond the timing of a single dura-
subjects are aware that tion, neural trajectories have also been observed during
they should attend to the the timing of multiple durations within more complex
passage of time to either
estimate a duration defined
temporal structures, such as rhythmic sequences46–49.
by external events or generate For example, using a task in which monkeys had to tap
a timed action. at fixed intervals between 450 and 1,000 ms according
to a metronome, Merchant and colleagues observed
Neural trajectory
that over the course of the tapping rhythm, neural
A sequence of population
states over time which trajectories evolved in a circular pattern with interval
describe the evolution of duration effectively being encoded by the radius of the
neural population activity. trajectory47,49, providing an intuitive illustration of how
neural trajectories can function as population clocks for
Implicit timing
Prospective timing in which
prospective timing.
no overt timing behaviour Neural trajectories are capable of encoding tempo-
is required. ral information for durations spanning roughly 1 s to
0123456789();:
at least several minutes. The lower bound of 1 s reflects
the distinction between subsecond and suprasecond
timing mechanisms50–54, although there is not a strict
boundary between the two as they overlap within the
range of roughly 0.5–1 s (refs.52,55). Indeed, potentially
reflecting this overlap, neural trajectories encoding dura-
tions several hundred milliseconds long have also been
observed, with no apparent difference from trajectories
encoding longer durations30,43,56. The upper bound of
several minutes was established by a recent study record-
ing neural activity while rats engaged in a task where
they had to obtain a reward within a 3-s window that
occurred every 5 or 8 min. Stable neural trajectories in
the form of sequential activity were observed in both the
hippocampus and the striatum spanning the entirety of
the minutes-long intervals. Task performance was also
dependent on hippocampal activity57, consistent with a
growing number of studies suggesting that for durations
of tens of seconds and longer, the hippocampus may play
a central role58–61. Whether stable neural trajectories are
able to encode even longer durations beyond several
minutes and whether there exists a maximum limit
remain to be determined.
Neural trajectories reflect the activity of single neu-
rons exhibiting a range of different activity patterns,
some of which contain clear correlations with time.
One of the most common time-related activity patterns
observed in single cells is ramping activity, where a neu-
ron monotonically increases its firing in a linear man-
ner, often across the entirety of an interval22,29,31,34,39,46,62–68
(Fig. 3a,b). Neurons exhibiting unimodal peak activity at
specific moments within the ongoing duration are also
commonly observed30,32,35,36,38,48,69–73 (Fig. 3c). Across a
population of such neurons, their progressive activa-
tions form a sequence of activity that can span an inter-
val. In addition to these two specific types of single-cell
activity pattern, neural trajectories are also made up
of neurons exhibiting more complex activity patterns,
ranging from non-monotonic ramping activity, which
still exhibits clear correlations with time, to patterns
which exhibit no obvious relation to time but are never-
theless stable across repeated trials29,37,43,46,74–76 (Fig. 3a,b).
Whether ramping activity and sequential activity play
special roles within prospective timing remains to be
determined, although recent computational work sug-
gests that sequential activity may be a particularly effec-
tive representation of time for downstream readout36.
For the purpose of defining a stable trajectory capable
of supporting prospective timing however, the only
requirements placed on single-cell activity patterns
are that some subpopulation exhibits stable repeatable
patterns, and that across individual moments in time,
overall population activity states are different. There is
no requirement for single-cell activity patterns to encode
significant temporal information themselves.
Implicit prospective timing
Prospective timing can also occur implicitly, in the
absence of any overt timing behaviour at all26–28,77–79
(implicit timing). Just as in motor timing, where subjects
learn to predict when to initiate an action relative to
an external cue, in a similar manner subjects may also
www.nature.com/nrn

learn to predict when external events will happen within
a stable and repeated temporal structure (‘temporal
expectation’), allowing more efficient and accurate
processing of task-related information26,77–80. As with
sensory and motor timing, there is also evidence that
neural trajectories can encode such learned temporal
expectations. Recording from the premotor cortex of
monkeys during a detection task for a mechanical stim-
ulus that would occur within a fixed temporal window
which was unannounced, Carnevale et al. observed
that neural population activity evolved in a manner
reflecting this window of expectation81. Specifically,
the distance between the neural trajectory and an esti-
mated decision threshold within neural state space was
reduced during the window of expectation. The impact
of temporal expectation can also extend to sensory
representations82–85. In the visual cortex of monkeys
trained to detect a stimulus change whose likelihood
of occurring followed a fixed temporal pattern, neu-
ral responses become modulated by this pattern82.
Similarly, in the primary auditory cortex of rats trained
to discriminate a target sound that occurred with high
probability during a specific period within a trial, sin-
gle neurons exhibit increasingly enhanced frequency
tuning as this period of high expectation approaches,
as well as enhanced tuning during the period itself 83.
a Prospective timing b Retrospective timing
Physical time Physical time
Present Present
moment moment
Elapsed duration
Stable temporal structure Duration can be measured between
defining an interval any pair of past events
Reference point (past event) Reference points (all past events)
End of duration
Perceived time
Perceived time
Reference
event
Physical time Physical time
Fig. 1 | Prospective and retrospective timing. a | During prospective timing, the time
point being measured from (‘reference point’) is established beforehand, allowing the
physical passage of time to be explicitly tracked (top). Prospective timing estimates
are relatively accurate but follow the scalar property, exhibiting increased variability
(grey shading) as physical duration increases (bottom). b | By contrast, retrospective
timing reflects the physical passage of time indirectly through the event structure of
experience, which occurs across time (top). Unlike for prospective timing, which is in
relation to a fixed reference event, the time points defining the interval to be estimated
are freely chosen from past events. Retrospective timing estimates can be more variable
and less accurate (bottom).
NATuRe RevIeWS | Neuroscience
0123456789();:
Reviews
Because the learning of temporal expectation can
occur in the absence of explicit timing behaviour, any task
which contains a stable temporal structure may poten-
tially engage it. Indeed, animal studies recording from
neural populations often use behavioural paradigms that
contain a stable temporal structure but lack overt timing
behaviour, and a growing number of studies have now
observed stable neural trajectories defined by the tempo-
ral structure of these tasks, which appear to encode the
passage of time63,71,72,86,87. Illustrative of this is the obser-
vation of hippocampal time cells during fixed delay peri-
ods within the stable temporal structure of trial-based
non-timing tasks such as the delayed match-to-sample
task6,69,70,88. During this delay period, time cells individu-
ally fire at specific moments of time such that the overall
population of these cells tiles the entire interval, and this
sequence of activity defines a stable neural trajectory
(Fig. 3c). Although these sequences are observed during
tasks that lack any explicit timing behaviour that would
necessitate a representation of time, control analyses and
experiments demonstrate that this sequential activity is
not simply due to obvious time-correlated variables such
as animal movement70,89,90, and suggest that it is in fact a
representation of the passage of time.
The apparent ubiquity of stable trajectories observed
in non-timing tasks across a wide range of brain areas87,91,92
raises the possibility that the presence of stable trajec-
tories and the ability to decode temporal information
accurately do not necessarily imply that this temporal
information is used by the brain86,93. Although this may
be true in many cases, the fact that temporal expectation
acts on internal representations means that its effects on
behaviour may be subtler, and determining whether or
how particular representations of time are actually used
by the brain may require closer examination, similar to
what has been shown recently in the separate context of
motor learning94. As an example, recent studies by Heys
and colleagues demonstrated that sequential activity is
present in the medial entorhinal cortex (MEC) of mice
engaged in a motor timing task95, but inactivation of the
MEC impacts behaviour only if inactivation occurrs dur-
ing the initial learning of the task, resulting in delayed
acquisition of the motor timing behaviour96. Thus, the
apparent use of MEC sequential activity may be limited
to the initial learning phase only. For motor timing tasks
in general, the initial learning of when to generate timed
responses poses a particular challenge, as the range of
times with which a subject may respond is vast. However,
if subjects are able to form hypotheses about when to
act97,98 — temporal expectations related to learning —
this would effectively restrict the time window within
which learning of motor timing occurs and allow more
efficient learning. Sequential activity in the MEC may
serve this purpose, and such a role may generalize to
other temporal associations as well99,100. To be effective,
temporal expectations guiding learning would have to
be generated at a much faster rate than the motor learn-
ing itself, and sequential activity in the MEC has in fact
been observed to arise immediately following exposure to
new environments95. Human functional imaging exper-
iments suggest that the hippocampus may also play a
similar role in guiding the learning of prospective timing
Reviews

a b
Cell 1 T3 Cell 1
T1 T2 T3
T1

Firing rate

Firing rate

Firing rate

Dim 2
Cell 2 Cell 2 Dimensionality
Cell 3 reduction
1 2 3 1 2 3 1 2 3
...
Cell no. Cell no. Cell no.
T2 Cell 3 Cell N Dim 1

c
Sensory timing Motor timing Temporal expectation
Stimulus start Stimulus end Interval start Interval end Interval start Cue

Duration 1 Duration 1
Duration 2 Duration 2
Duration 3 Duration 3
Elapsed time
Elapsed time Elapsed time

Expectation-
related
internal state

Stimulus
Dim 2

Dim 2

Dim 2
end Interval end

Stimulus start Interval start Interval start

Dim 1 Dim 1 Dim 1

d Reward
Production epoch
tp 1,000
Estimation epoch
ts
800
–0.15 0 0.15

tp (ms)
Go
(tp –ts) / ts
640
Set

Ready Ready Set 480

Time
480 640 800 480 640 800
ts (ms) ts (ms)

e Set + 200 ms
Ready 1
800
0
PC3
–1
–1
Set
Set
0 –1
PC3 PC2
0 480 0
PC1
1
1 Go
–1
0 1
1 –1 0
PC1
PC2
Estimation epoch (sensory timing) Production epoch (motor timing)

behaviours101, and hippocampal time cells also show fast Neural trajectories as a mechanism
adaptation to new temporal contexts70,102. More gener- Although similar time-varying population activity is
ally, the fact that sequential activity is observed across observed across sensory timing, motor timing and tem-
a wide range of brain areas under similar experimental poral expectation, this does not necessarily mean that
conditions may reflect the role of sequential activity as the way in which these activity patterns are subsequently
a general scaffold for supporting subsequent learning of used for time estimation is also shared across all three
time-related responses36,103. types of prospective timing. Indeed, sensory timing

www.nature.com/nrn

0123456789();:

◀ Fig. 2 | Neural trajectories during prospective timing. a | The activity of a population of
neurons, or population state, at a given moment can be depicted as a point within a state
space where each axis corresponds to the firing rate of a specific neuron. In this example,
there are three neurons, so the state space is three-dimensional. As the firing rates of the
three neurons change across time, so does the population state, tracing out a trajectory,
in this case spanning three time points (T1, T2 and T3). b | For a population consisting
of N cells, the state space is N-dimensional. Dimensionality reduction can be used
to reduce noise and to project data onto specific subspaces, and often it is helpful to
apply dimensionality reduction techniques to more easily identify important features of
neural trajectories. In this cartoon example, dimensionality reduction reveals a simpler
trajectory within a two-dimensional space. c | Different trajectory features are used for
sensory timing, motor timing and temporal expectation. For sensory timing of different
durations (indicated by different shades of grey), a common trajectory progresses at a
constant speed (starting at a common population state and reaching a common
intermediate state over the same amount of elapsed physical time, indicated in blue)
but stops at different terminal states for specific durations (indicated by different shades
of red). For motor timing of different durations (indicated by different shades of grey),
a common trajectory progresses at different speeds (starting at a common population
state and reaching different intermediate population states over the same amount of
elapsed physical time, indicated in blue) to reach a common terminal state (different
durations indicated by different shades of red). For temporal expectation, a trajectory
begins to evolve following the start of a fixed temporal structure, and reaches a specific
state in accordance with this temporal structure. In the illustrated example, the temporal
structure is defined by a cue which occurs at a fixed interval following a start signal, and
the trajectory reaches a particular set of states (indicated in purple) at the time of the
cue. This expectation-related internal state may then be reflected in an altered response
criterion or modulation of sensory representations, for example. d | Left, structure of
‘ready, set, go’ sensorimotor timing task. Monkeys are presented with a sample interval ts,
which they reproduce as tp through the timing of a motor action. Top centre, reward
(green shading) is a function of the relative error between ts and tp. Bottom centre, the set
of sample intervals used. Right, behaviour from an example monkey illustrating that
animals are able to carry out both sensory timing and motor timing accurately in order
to reproduce presented intervals. Dots indicate individual trials, and circles indicate
average tp per ts. e | Neural trajectories during estimation (left) and production (right)
epochs, based on activity recorded from the dorsomedial frontal cortex. These two sets
of trajectories illustrate how prospective timing is accomplished through population
state dynamics: specific durations are associated either with specific population states
along the trajectory (estimation epoch) or specific trajectory speeds (production epoch).
Triangles indicate ‘ready’, circles indicate ‘set’, squares indicate ‘go’ and small circles
indicate neural states at 40-ms increments. Dim, dimension; PC, principal component.
Parts d and e adapted with permission from ref.56, Elsevier.
differs in a basic way from both motor timing and tem-
poral expectation in that the end of a duration during
sensory timing is defined by an external event rather
than being internally determined23. Thus, the terminal
state of a neural trajectory during sensory timing is dic-
tated by external input, whereas it would be internally
generated during motor timing or temporal expectation.
Experimental observations are consistent with this
difference between sensory timing versus motor timing
and temporal expectation. In sensory and motor tim-
ing tasks where multiple durations are discriminated
or generated, the way in which the different durations
are encoded differs. During discrimination of different
durations for sensory timing, neural population activity
evolves along a common trajectory for all of the dura-
tions but reaches different terminal states corresponding
to specific physical durations29–31,56 (Fig. 2c,e). In contrast,
during generation of an action at different durations for
motor timing, a similar trajectory evolves for all dura-
tions, reaching a common terminal state that triggers
action, and the durations are instead represented by
the speed at which the trajectory evolves35,39,40,43,44,56,104
(Fig. 2c,e). As a result, temporal scaling of activity patterns
NATuRe RevIeWS | Neuroscience
0123456789();:
Reviews
reflective of different durations can be observed at both
the population level and the single-cell level (Figs. 2c,e
and 3b). Finally, for temporal expectation, the learned
period of expectation is reflected through a shift in
popu­lation state that reflects the predictable external
temporal structure82–85. Thus, as with motor timing,
trajectory speed plays a central role in the timing of
expectation as it determines when this shift is reached.
It is important to note that for all three types of pro-
spective timing, estimation using neural trajectories
implies that the terminal state of a trajectory directly
serves as the duration estimate — the computation of the
duration is carried out by the population state dynamics
that lead to a specific terminal state105. In other words, the
evolution of activity along a neural trajectory is itself
the estimation process. If this is the case, then speed-
ing up or slowing down of population state dynamics
should have a predictable impact on timing. The fact
that trial-to-trial fluctuations in trajectory speed are
correlated with trial-to-trial behavioural variability
suggests this30, but two studies have taken a somewhat
more direct approach by controlling the temperature
of brain areas which are involved in prospective timing
and exhibit stable neural trajectories. Xu et al. found
that cooling the medial prefrontal cortex (which pre-
sumably slowed down population state dynamics) of
rats engaged in a motor timing task resulted in longer
response times37. Similarly, Monteiro et al. bidirection-
ally manipulated the temperature of the striatum in rats
engaged in an interval categorization task, and found
that duration judgments were shorter when the temper-
ature was increased (presumably speeding up population
state dynamics) and vice versa106. Although these studies
provide support for the idea that prospective timing is
carried out through population state dynamics, more
precise manipulations are still needed.
During sensory timing, because durations are
defined by external input, it is possible for the estimation
of duration to be separated from the duration itself. This
raises the possibility of a separate, cognitively different
type of sensory timing. If metrical representations of
specific moments in physical time corresponding to the
start and end of the duration are generated and main-
tained in memory, it would be possible to alternatively
determine duration through temporal arithmetic of
these representations, rather than through population
state dynamics. This version of sensory timing would
also differ from sensory timing in which estimation is
carried out through population state dynamics but then
stored in working memory in order to guide behav-
iour at a later time point. Neurons encoding specific
moments in physical time have been observed, and
often do appear to reflect a metrical representation of
time72,107, while neurons encoding the difference between
two durations have also been observed108, suggesting that
the brain is capable of carrying out arithmetic operations
on abstract temporal quantities. Nevertheless, direct evi-
dence for estimation of duration through temporal arith-
metic remains to be observed. Moreover, in an interval
categorization task where monkeys had to categorize
durations as belonging to either long or short categories,
neural recordings reveal that the task is solved through
Reviews

a Sensory timing c
Interval start Interval end Ramping Sequential activity

Cell 1 Cell 3 Cell 1

Firing rate
Cell 2 Cell 4
Cell 2

Firing rate
b Motor timing Ramping
Cell 3
Cell 1 Cell 3
Firing rate

Cell 2 Cell 4 Cell 4

Time (s) Time (s) Time (s)

Fig. 3 | single-cell activity underlying stable neural trajectories during prospective timing. a | During sensory timing,
single-cell activity evolves along the same pattern for different durations, reaching different end points dependent
on the specific duration (two example durations shown here). Grey traces show evolution of activity over the shorter
duration, and black traces show the continued evolution of activity for the longer duration. Activity patterns can be
complex and non-monotonic, as illustrated by the two example cells in the left column, or ramping, as illustrated by the
two example cells in the right column. b | During motor timing, single-cell activity exhibits temporal scaling, or stretching
of responses for longer durations (black trace compared with grey trace). This scaling occurs both for cells with complex
non-monotonic activity patterns, as illustrated by the two example cells in the left column, and for cells with ramping
activity, as illustrated in by the two example cells in the right column. c | In addition to ramping activity and more complex
non-monotonic activity patterns (as in parts a and b), single cells (four examples are shown here) can also exhibit peak
activity at specific points in time within an interval such that as a population, their peak activity defines a sequence which
spans the interval (indicated by the diagonal arrow).

Recurrent neural networks
Neural networks in which each
unit can receive input from
other units in the network
in addition to external input.
population dynamics akin to motor timing rather than
comparison of abstract representations of duration that
would be generated by temporal arithmetic31. Thus, the
use of abstract arithmetic operations for sensory timing
may be limited to specific conditions109.
How stable neural trajectories arise
Computational insights. What are the key factors that
shape population state dynamics into stable neural
trajectories capable of supporting prospective timing?
Computational studies using recurrent neural networks to
reproduce prospective timing behaviours suggest that
the generation of stable trajectories can arise simply
through interaction between recurrent dynamics within
a local circuit and external sensory input related to the
duration110–112. Specifically, recurrent networks lacking
any specialized structure113 can recapitulate both exper-
imentally observed timing behaviour and neural activity
patterns at single-cell and population levels when trained
to perform prospective timing tasks. For sensory timing
tasks, recurrent networks generate population activity
which evolves along a common trajectory to reach spe-
cific terminal states for specific durations31,114–116. For
motor timing tasks, recurrent networks produce specific
durations by modulating the speed at which a common
trajectory evolves43,115,117,118. For temporal expectation,
recurrent networks trained on non-timing tasks with
fixed temporal structures generate trajectories which
not only encode significant temporal information but
are also capable of supporting temporal expectation115.
In further support of the apparent capability of local
recurrent circuits to generate stable trajectories support-
ing prospective timing, in vitro studies using stimula-
tion following fixed temporal patterns have found that
local circuits are capable of learning these structures and
generating stable trajectories119,120.
Beyond illustrating that recurrent networks lacking
specialized structure are generally capable of performing
prospective timing tasks and that they do so using activ-
ity patterns similar to those which have been observed
experimentally, these recurrent network models may
also provide additional insights into the mechanisms
for sensory and motor timing. A key consideration for
sensory timing is how external input is able to influence
the ongoing progression of the trajectory121. Within a
recurrent network, if recurrent synaptic weights are suf-
ficiently weak, the network’s ability to propagate activity
through recurrent dynamics will be limited, and pro-
gression of the trajectory will predominantly be driven
by external input. Models based on state-dependent
networks 122 demonstrate that networks with weak
recurrent weights are in fact capable of performing sen-
sory timing on the subsecond timescale52,122,123. In the
case where recurrent weights are not sufficiently weak,
however, and the network is able to generate a trajec-
tory on its own, progression of the trajectory can still be

www.nature.com/nrn

0123456789();:

influenced by external input as long as the input is suffi-
ciently strong114. For motor timing, a central question is
how trajectory speed modulation occurs. Speed modu-
lation begins with the influence of duration input on the
network that determines which duration should be pro-
duced. Assuming that duration input is transient, recur-
rent network models suggest that the influence of this
input may occur in a population activity subspace which
is orthogonal to the subspace within which the sub­
sequent recurrent dynamics that drive evolution of the
motor timing trajectory play out43. When duration input
is assumed to be tonic, a similar separation of recurrent
dynamics and input into orthogonal subspaces is also
observed117. Recurrent network models also suggest that
within the recurrent dynamics that drive evolution of
the trajectory, the speed at which the trajectory evolves
may be due to the interaction of duration input with the
nonlinear activation functions of single neurons43.
Experimental insights. Although local circuits may be
capable of generating stable trajectories, experimental
observations demonstrate that interareal interactions
also play an important role in the generation of stable
trajectories. Human functional MRI studies consis­
tently observe that networks of multiple brain areas are
activated during prospective timing tasks2,124–128, and
specific areas within these networks are often engaged
during particular phases of a given task129–133. Thus, the
generation of stable trajectories may arise in part through
interaction between multiple brain areas playing differ-
ent complementary roles2,125. Single-unit recordings from
animals are beginning to provide more detailed insight
into what this interaction may look like. Corticostriatal
interactions, in particular, may play an important role in
prospective timing17,76,125, and several studies recording
from both the striatum and the cortex during the same
timing task32,36,38,39,134 have observed dissociations which
suggest that the striatum may act to refine cortical tra-
jectories representing the passage of time135,136. In motor
timing tasks and sensory timing tasks, respectively, the
striatum was found to have greater temporal informa-
tion than the orbitofrontal cortex38 or the medial prefron-
tal cortex32. For the dorsolateral striatum in particular,
activity was observed to be more sequential than that
in either the secondary motor cortex36 or the medial
prefrontal cortex32, suggesting that refinement within
the dorsolateral striatum may in part be reflected in the
generation of trajectories which are primarily defined by
sequential activity of cells which fire at specific moments
in time within a duration136. Consistent with this, in a
model trained to generate timed responses using activ-
ity patterns from either the dorsolateral striatum or sec-
ondary motor cortex as input, striatal input in the form
of sequential activity allowed more consistent timing
behaviour36. However, given that both the striatum and
the cortex are composed of distinct areas involved in
different functions, whether such refinement of cortical
representations of time is a general function that the stri-
atum carries out on the different cortical inputs it receives
still remains to be determined. Interactions across areas
may reflect the distribution of different computational
roles across separate populations of neurons, and similar
NATuRe RevIeWS | Neuroscience
0123456789();:
Reviews
distribution may also occur across different functionally
defined subpopulations of neurons (or subspaces) within
a single area such that within-area interactions between
specific subpopulations of neurons may also play a role
in the generation of stable trajectories43.
Interactions with neuromodulatory systems also play
a critical role in prospective timing. Manipulation of
dopaminergic signalling, in particular, has been shown to
generate shifts in duration estimates, which suggest that
dopaminergic neurons may control the speed at which
neural trajectories evolve125,137–141. Consistent with this,
direct recordings of dopaminergic neuron activity in the
substantia nigra pars compacta show that their popu­
lation activity is correlated with animals’ estimates of
duration142,143. Additionally, optogenetic manipulation
of the activity of these dopaminergic neurons shifts
animals’ duration estimates in a manner consistent with
control over the speed of trajectories: increased dopa-
minergic activity drives overestimation of time and vice
versa142,143. Beyond control of the speed at which trajec-
tories evolve, dopaminergic signalling may also have a
substantial influence on the initial formation of trajec-
tories through its role in reinforcement learning142,144–147.
In addition to the dopaminergic pathways, cholinergic
pathways have also been shown to play an important role
in prospective timing. Disruption of cholinergic signalling
has a substantial impact on timing behaviour125,137,140,148,
but this effect typically lags by several sessions following
the start of the manipulation, suggesting that its effect
is related to memory of the learned intervals and their
trajectories, rather than the ongoing evolution of trajec-
tories in the present moment. Additionally, cholinergic
input has been shown to be necessary and sufficient
for learning of reward timing intervals in the primary
visual cortex149,150. Overall, a key question that remains
regarding the generation of stable trajectories is whether
there exist any general principles spanning sensory tim-
ing, motor timing and temporal expectation, or whether
this process is instead better understood in terms of the
specific individual contexts in which it occurs.
As we continue to learn more about how stable tra-
jectories arise, one question which may be informative is
whether stable trajectories arise specifically for the pur-
pose of timing, or instead for some other function which
occurs across time4,151. It has long been debated whether
timing is supported by a dedicated central clock or dis-
tributed local representations of time2,5,124,125,152,153, and,
if timing is supported by distributed local representa-
tions of time reflecting the inherent temporal nature of
many different brain functions152,154, then understand-
ing these functions themselves may also inform us about
how representations of time arise. For an area such as
the motor cortex where the temporal nature of motor
action is inherent, understanding how population state
dynamics underlie motor action may be in large part the
same as understanding motor timing. A similar overlap
may also exist for areas such as the striatum and the fron-
tal cortex which are involved in reward prediction146,147,
given that stable temporal structures are what allow
reward prediction to occur. But time is also present in
less obvious functions, as illustrated by the example of
path integration, carried out in part by the hippocampus
Reviews
and the MEC. Path integration is a process defined by
movement over time, and the computations generating
a representation of elapsed distance may also support
the estimation of elapsed time in several ways. First, an
explicit representation of both elapsed distance and aver-
age velocity over a period of time can be used to directly
estimate elapsed time. Elapsed distance by itself may also
directly serve as a rough estimate of elapsed time without
any additional computation. Second, the central com-
putation for path integration and for estimating elapsed
time is similar — counting steps for path integration,
and counting units of time for elapsed time — such that
the mechanisms which allow path integration to occur
may also support estimation of elapsed time155. Finally,
a central behaviour supported by path integration —
homing — shares substantial similarities with the pro-
cess of prospective timing. In each case, either elapsed
time or distance is tracked with respect to a fixed tem-
poral or physical landmark, which is set at the start of the
behaviour. Although it remains unclear for both homing
and tracking of elapsed time how reference points are
set156,157, the similarity of the computational goal in each
case raises the possibility that they may share a similar
mechanism. Beyond the computational overlap between
prospective timing and path integration, path integration
also occurs on longer timescales than the time-related
functions of other areas such as the motor cortex or
the striatum. Thus, the hippocampus and other areas
involved in path integration offer an existing solution
to the challenge of estimating duration across different
timescales, specifically for longer durations.
Retrospective timing
The major difference between prospective timing and
retrospective timing is that in prospective timing, the start
of the interval is set beforehand, allowing active tracking of
the ongoing passage of time (Fig. 1). In contrast, retro­
spective timing can be computed only from memory of
the interval and events occurring within it158,159, and the
main computational goal for retrospective timing is
the conversion of memory representations into a duration
value. In this way, it is similar to the special instance of
sensory timing in which the estimation process is sepa-
rated in time from the interval being timed. However, for
this particular form of sensory timing, it is still possible to
generate an explicit representation of the ongoing passage
of time and use memory of this explicit temporal infor-
mation to estimate duration, whereas such a possibility
does not exist during retrospective timing. More impor-
tantly, a large number of human behavioural experiments
explicitly comparing prospective timing and retrospective
timing have established that they generate significantly
different duration estimates for the same physical time
intervals, strongly suggesting that different mechanisms
underlie each type of timing. First, in a foundational
study by Hicks et al., participants estimated either pro-
spectively (asked before the interval) or retrospectively
(asked following the interval) a 42-s duration during
which they dealt a deck of cards160. The average pro-
spective duration estimate was 42.25 s, while the average
retro­spective duration estimate was 32.28 s. The temporal
delay between the end of the interval and retrospective
0123456789();:
estimation was approximately 15 s, and consisted only of
the time it took to ask participants to make the estimate,
limiting the potential for poor memory recall to impact
retrospective estimates. Subsequent variations of this par-
adigm across a wide range of conditions have consis­tently
observed that prospective estimates are longer and tend
to be less variable161,162. Second, acting on the hypo­thesis
that prospective and retrospective timing engage different
cognitive processes, a number of studies have examined
the effects of manipulating different cognitive varia-
bles and observed dissociations between prospective
timing and retrospective timing161,163. For example, Block
et al. modulated participants’ cognitive load by presenting
them with a list of words which they either simply had to
pay attention to (low-load condition) or had to name an
associated action for (high-load condition). Prospective
estimates of how long the list presentation lasted under
the high-load condition were significantly shorter than
prospective estimates under the low-load condition,
while retrospective estimates were unaffected by task
difficulty and cognitive load164. This dissociation is likely
to reflect the importance attention to the passage of time
plays for prospective timing specifically165.
Immediate and remote retrospective timing
Just as prospective timing is separated into different types
(implicit timing, explicit timing and sensory, motor or
temporal expectation), retrospective timing is also likely
to encompass more than one type, particularly since the
timescales that it encompasses span a few seconds to dec-
ades. However, systematic examination of retrospective
timing across different timescales to identify these poten-
tially different types remains to be done. Nevertheless,
one separate distinction that has been drawn within
retrospective timing is between immediate estimation
(when estimation directly follows the interval) and
remote estimation (when estimation is separated by some
delay from the interval). Estimates made immediately fol-
lowing the interval are shorter than those made following
a delay166–171, and this effect is true across a wide range of
interval lengths (from tens of seconds to 15 min) as well
as delay lengths (from seconds to 1 month). Although it
seems plausible that this difference may in part be due to
the respective involvement of either short-term memory
(for immediate estimation) or long-term memory (for
remote estimation), it still remains to be determined
whether this difference may be attributed to specific
memory processes (for example, memory encoding or
memory recall). Several studies additionally show that
the length of the delay has an impact on retrospective
timing166,167,170, but a more detailed understanding of
this interaction is also lacking. Because the categori-
zation of retrospective timing is currently very limited
and does not clearly suggest any mechanistic differences
between immediate estimation and remote estimation,
for the purposes of our discussion, we do not distinguish
between different types of retrospective timing.
Retrospective timing and events
A simple mechanism for retrospective timing would
be to directly encode explicit temporal information
(for example, timestamps in physical time) during the
www.nature.com/nrn

Event segmentation
The parcellation of continuous
ongoing experience into
discrete events.
Event trajectory
A neural trajectory defined
through the process of event
segmentation.
Contextual drift
A dynamic in which
representations of context
gradually change over time, as
a natural result of time-varying
inputs.
NATuRe RevIeWS | Neuroscience
initial experience of an interval, such that the time of
an event could be directly recalled from memory rather
than having to be estimated from the contents of mem-
ory later. Although this does happen in some circum-
stances, the normal mechanism for retrospective timing
is instead a reconstructive process that brings together
and utilizes temporal information from a number of dif-
ferent sources158,171–178. This point is clearly demonstrated
by several studies which observed that retrospective
duration estimates for both public and autobiographi-
cal events have similar accuracy regardless of whether
participants actually remember the events or do not
remember them158,173,174. If temporal information was
directly retrieved from memory, then duration estimates
would necessarily be less accurate for events which
were not remembered. Instead, these results suggest
that duration can be actively computed using temporal
information inferred from the events which constitute
an experience179, even in the extreme case when the
experience is not one’s own. In addition to this infer-
ential construction of temporal information, retrospec-
tive timing can also draw on certain explicit forms of
temporal information, such as temporal landmarks and
temporal schema. Temporal landmarks are events for
which temporal information is explicitly attached, such
as birthdays or flashbulb memories177, and can aid recon-
struction in a manner similar to how physical landmarks
reduce error during path integration. Temporal infor-
mation for these particular memories may be retained
over time because of their centrality to the experience180.
Temporal schema are general temporal structures within
experience158. These can be defined by temporal features,
as in the days of the week or the seasons, but they can
also be defined by non-temporal features, such as the
progression through grade levels of primary school181,182.
The events making up an experience also provide a
natural source of temporal information, as they both
occur across time and define intervals of time. Here we
adopt the definition of ‘event’ provided by Zacks and
Tversky183 in the context of event segmentation theory:
“a segment of time at a given location that is conceived
by an observer to have a beginning and an end.” Many
studies have shown that retrospective time estimates
are correlated with the number of remembered events
from the timed interval159,184–189. Moreover, manipulat-
ing the event structure of an experiment (for example,
by showing a sequence of stimuli in either random or
fixed order) impacts the subsequent estimated dura-
tion of the experiment159,184–188. In an illustrative study
by Poynter, participants were presented with a list con-
sisting of 27 unrelated words and three high-priority
events (last names of US presidents) that they were
instructed to pay particular attention to186. High-priority
events were either placed as the first three items of the
list (unsegmented condition) or otherwise distributed
approximately evenly across the list (segmented con-
dition). Participants’ retrospective estimates for seg-
mented intervals were longer than their retrospective
estimates for unsegmented intervals. Notably, there
were no differences in recall or recognition across the
two conditions, indicating that the difference in duration
estimates was not caused by differences in memory but
0123456789();:
Reviews
instead reflected the substantial influence that the event
structure of the experience (defined by high-priority
events) had on retrospective timing. Similarly, additional
studies have shown that manipulating event structure
more indirectly through changes in context impact the
estimated duration. For example, if a basic sequence of
events such as the presentation of a list of words occurs
within a changing context, the estimated duration will
be longer than for the same sequence of events within
a constant context. In these studies, context spanned a
wide range, from environmental190–192 to cognitive164,193
to emotional190,194,195, and the increase in estimated dura-
tion may be interpreted as reflecting the more complex
event structure of the experiment. Thus, the event struc-
ture of ongoing experience has a defining influence on
retrospective timing.
Neural event trajectories
The close relation between event structure and retro-
spective timing suggests that the neural mechanisms
underlying the perception of events may also play a
critical role in retrospective timing. Human behavioural
experiments196,197 and functional imaging experiments198
examining the process by which events are perceived
have established that continuous ongoing experience is
segmented into discrete events183,199. Subsequent stored
representations of these discrete events within long-term
memory may thus serve as the basis for retrospective
timing. Consistent with this, both human behavioural
experiments196,200–206 and human functional imaging
experiments207–210 demonstrate that the temporal units
defined during initial online segmentation of an experi-
ence have a substantial influence on the structure of sub-
sequent long-term memory for that experience, while
the event boundaries determined during initial seg-
mentation subsequently serve as anchors for long-term
memory211,212. Moreover, ongoing experience is in fact
simultaneously segmented into events on multiple
timescales183,213–216 such that the stored representations
of all segmented events would be able to flexibly sup-
port time estimation across the full range of timescales
present within episodic memory.
Returning to the idea from prospective timing that
neural population activity can be used to encode the
passage of physical time as long as each moment in time
corresponds to a unique neural population activity state,
assuming events serve as the basis for retrospective tim-
ing, it may likewise be the case that neural population
activity can be used for retrospective timing if each event
within continuous ongoing experience is represented by
a unique neural population state. Accurate retrospective
timing across a series of events would then depend on
population activity evolving along a trajectory composed
of sufficiently distinct population states for each event.
We refer to such a trajectory defined through the pro-
cess of event segmentation as a neural ‘event trajectory’
(Fig. 4a) . Many models of episodic memory propose
the existence of a similar time-varying temporal con-
text signal, often referred to as ‘contextual drift’, which
would allow different episodes of experience to be
linked together and organized by temporal order217–220.
Neural event trajectories do not necessarily have to
Reviews

a Encoding of ongoing Evidence for event trajectories

experience into memory
Human functional imaging. A number of human func-
Event
tional imaging experiments provide evidence that event
1 trajectories occur and influence retrospective timing.

Dimension 1
2 Jenkins and Ranganath presented participants with
3
4 a sequence of objects and subsequently asked them in a
5
surprise test afterwards to indicate when specific objects
Physical time
had appeared during the overall sequence223. They found
that in the prefrontal cortex, neural population states
Dimension 2 became increasingly dissimilar as the experiment pro-
gressed. Moreover, the dissimilarity between a given
b B1 W1 W2 B2 B3 W3 B4 W4 B5 B6 W5 W6 object and temporally adjacent objects was greater for
250 s those objects whose time of occurrence was accurately
trials estimated. Thus, if the initial presentation of each object
is taken as a discrete event, accurate retrospective timing
Black walls White walls
Population state for 10 s bin was dependent on the representation of each event being
Mean state for 250 s trial
B1 B6 W1 W6 sufficiently distinct. Using a similar paradigm, Ezzyat
and Davachi presented participants with sequences
LEC CA3 MEC
100 of visual objects placed over a background that either
100 remained constant or changed within a sequence,
50 thus defining a putative event boundary209. After each
sequence, participants were asked to judge whether two
LD2

0 0 0
–50
–100
–100
–100 0 100 –50 0 50
LD1 LD1
Fig. 4 | Neural event trajectories encode event structure of experience. a | An event
structure occurring across continuous physical time is encoded by discretized
representations in neural population state space reflecting event segmentation. Left,
a sequence of events occurs across physical time. Each block corresponds to a constant
interval of physical time, different-coloured blocks correspond to different events and
shading of blocks corresponds to the progression of a given event across physical time.
Right, the neural population states evoked during the sequence of events. Each dot
corresponds to the population state defined over a constant interval of physical time,
different-coloured dots correspond to different events and shading of dots corresponds
to the progression of a given event across physical time. Arrows illustrate the progression
of the event trajectory, with shading indicating the progression of physical time. Neural
population activity is shown in a two-dimensional population activity space. Discretized
representations arise in part through the greater rate of change of population states
around event boundaries, which may be driven by a combination of changes in the
external environment as well as internal states. The event trajectory defined across
the overall sequence of events can be used for retrospective timing. b | Visualization
of dimensionality-reduced population activity in the lateral entorhinal cortex (LEC),
hippocampus (CA3) and the medial entorhinal cortex (MEC) allows observation of event
trajectories. Dimensionality reduction was performed by means of linear discriminant
analysis. Axes correspond to the first two linear discriminants (LD1 and LD2). During
a session rats foraged freely in an environment whose wall colour switched between
black and white over 12 trials, each approximately 250 s long, interleaved with intertrial
periods outside the environment. The generation of distinct population states for
each individual trial period in the LEC but not CA3 may reflect event segmentation
on different timescales in each respective area. Panel b reprinted from ref.241,
Springer Nature Limited.
evolve gradually221 and are separate222 from ‘representa-
tional drift’, which is not driven by the event structure of
ongoing experience. Finally, just as how a linear metri-
cal representation is not required for neural trajectories
supporting prospective timing, event trajectories also do
not imply a literal timestamping of events such that the
specific population state associated with a given event
directly indicates when it happened in physical time.
of the objects from the sequence were close together or
far apart in time. Whether the two objects spanned an
event boundary was variable, but the actual temporal
distance was kept constant. Similar to the findings of
–50 0 50 Jenkins and Ranganath, population states in both the
LD1 hippocampus and object-selective visual cortex were less
similar across pairs that were judged as far apart (Fig. 5a).
Taking a more naturalistic approach, Nielson et al. used
participants’ own individual memories of autobiograph-
ical events to examine the relationship between neural
activity during remote recall of these events and the
actual physical durations between them when they were
originally experienced224. For a month, participants wore
a smartphone that would take photographs and record
the time and location at regular intervals. At the end
of the month, participants were imaged while recalling
experiences cued by logged photographs. Once again,
the amount of time separating two events when they
were originally experienced was correlated with how
different their respective population states were during
subsequent recall (Fig. 5b). This correlation was present
in the hippocampus but only for events which the par-
ticipants could recall and was altogether absent from the
primary visual cortex, probably reflecting the fact that
these population states are generated through episodic
memory recall.
The aforementioned studies each involved behav-
iours that do not necessarily engage retrospective timing
specifically but are still at the very least closely related.
Lositsky et al. combined functional imaging with an
explicit retrospective timing task. Participants first lis-
tened to a 25-min long story and then were asked to make
estimates of how much time spanned different pairs of
clips from the story225. Consistent with findings of the
other studies using simpler stimuli, larger changes in
population state during the initial encoding of the story
subsequently resulted in more accurate duration esti-
mates. However, because the event structure of the story
was more complex and in part subjective, Lositsky et al.
could also examine whether the perceived event

www.nature.com/nrn

0123456789();:
 Reviews

structures of individual participants were correlated with link across event perception, event trajectories and
their retrospective duration estimates. Lositsky et al. retrospective timing altogether.
asked participants to mark event boundaries, and found There have been very few computational models for
that the number of marked boundaries was in fact retrospective timing226, but recent work from Fountas et al.
correlated with duration estimates, providing a direct demonstrates that an event-based model for retrospective

a Encoding Left hippocampus b

Changing scene
(event boundary)
4s 6
5

Neural distance
4
Constant scene
Physical 3
time 2
0.10 1
P < 0.025

Pattern correlation (z)

0
1 day 1 week 1 month
0.05
Temporal distance test log10(time)
Rate temporal proximity:
0.00
‘Close’ ‘Far’

Very close/close/far/very far –0.05

c
Encoding of ongoing Reinstatement of population Convert neural distances into
experience into memory states from memory subjective duration estimates

d1 d1
Dimension 2

Dimension 2

d2
d2

Dimension 1 Dimension 1

Fig. 5 | relation between duration estimates and event trajectories. a | Left, structure of associative encoding
task. During an initial encoding phase, human participants were presented with multiple sequences of four images
consisting of a face or object simulus placed over a scene, which either remained constant or changed after two images,
thus defining an event boundary. Subsequently, they were presented with a pair of items (face or object) from the
encoding phase and asked to judge how close together in time they appeared during encoding, using a scale of ‘very
close’, ‘close’, ‘far’ and ‘very far’. Temporal distance estimates reflected subjective time estimates, as object pairs were
always separated by a constant physical temporal distance. Right, with use of functional MRI, hippocampal activity
was measured during the encoding phase. Similarity of population states was lower for pairs judged to be temporally
far apart than for pairs judged to be close in time, demonstrating that subjective estimates of time are related to
population states arising during initial encoding of experience. b | Measurement of the similarity of neural population
states during recall of past events occurring over a time span of 1 month (using human functional imaging) reveals that
population states for events become increasingly more separated with increased temporal distance. Thus, retrospective
timing based on population states from reinstated event trajectories is, in principle, capable of being accurate with
respect to the actual physical duration. Each individual blue dot indicates the neural distance for a pair of events, black
lines indicate regressions for individual participants of neural distance against the logarithm of physical temporal
distance and the red line indicates the average of these regressions across participants. c | During retrospective timing,
population states from an event trajectory arising during the initial encoding of experience (left) are reinstated (centre)
and subjective duration is estimated by a process in which some feature defined across the reinstated population states
(for example, distance in population state space) is converted into units of physical time (right). Left, each coloured
circle corresponds to the population state defined across a constant amount of physical time, such that events which
span longer intervals of physical time correspond to a greater number of circles and the shading of circles indicates the
progression of physical time across an individual event. Arrows illustrate the progression of the event trajectory, with
shading indicating the progression of physical time. Centre, each circle corresponds to the recalled population state
for a given event. The dashed lines indicate potential durations that could be estimated between pairs of events, and
the solid lines indicate the two specific durations (d1 and d2) which are being estimated and compared. Part a adapted
with permission from ref.209, Elsevier. Part b adapted with permission from ref.224, PNAS.

NATuRe RevIeWS | Neuroscience

0123456789();:
Reviews
timing is able to reproduce key features of human retro-
spective timing227. Behavioural data for retrospective tim-
ing were generated by presenting participants with a video
of a natural scene and then asking them to estimate its
duration. The same video was then presented to a neural
network trained to perform image classification, and the
frame-by-frame unit activations were recorded across dif-
ferent layers of the network to simulate event trajectories
across different brain areas, as might be measured by func-
tional imaging. The number of salient events (defined as
events poorly predicted by a generative model) was then
used to generate a duration estimate. Using this compu-
tational approach (previously proposed by Roseboom
et al.228) to explicitly test the relation between events and
the estimated duration of an experience, Fountas et al.
found that the model not only generated accurate duration
estimates but also recapitulated several well-established
human estimation biases, including the judgement of
scenes with more events as longer than scenes with fewer
events, despite both being of the same physical duration.
Single-cell recordings. Single-cell recordings from
animals may allow us to gain a more detailed mecha-
nistic understanding of event trajectory dynamics and
the underlying process for event segmentation within the
context of retrospective timing. However, whereas many
well-established experimental paradigms exist for stud-
ying prospective timing in animals, experimental para-
digms for retrospective timing are still at a more nascent
stage229–231. Retrospective timing is ideally studied using a
one-shot paradigm in which subjects are completely una-
ware of the timing task until after the stimulus duration
has been presented, thus ensuring that no attention can
be explicitly focused on the passage of time and the influ-
ence of processes for estimating present duration can be
minimized. This is challenging in animal experiments
due to the inability to directly ask animals to make time
estimations232 — in the vast majority of experimental par-
adigms, animals must learn how to respond over repeated
trials, and in doing so may also learn to pay attention to
the passage of time. Nevertheless, a number of studies
have established behavioural paradigms which capture
a key element of retrospective timing in that animals
cannot predict when to begin timing or what specific
temporal features to pay attention to a priori229–231,233–235.
One paradigm, first developed using scrub jays229, is to
associate different temporal characteristics with specific
food items (for example, different types of food that spoil
at different durations). Animals are allowed to forage for
this food and are observed to see whether they make
use of this temporal information across different time
points. Because the foraging time points occur at inter-
vals determined by the experimenter, the animals cannot
pre-emptively time their behaviour such that they explic-
itly attend to the passage of time. Additionally, both the
longer time intervals used in these experiments, span-
ning up to multiple days, and the engagement of animals’
natural foraging strategies, allowing them to learn the
task across relatively few trials, may limit the likelihood
that animals learn to track time explicitly for the task.
Given the close link between retrospective timing
and episodic memory, many animal studies236–241 have
0123456789();:
taken an alternative approach to understanding retro-
spective timing, studying it in the context of episodic
memory and focusing on the medial temporal lobe. The
medial temporal lobe has long been a point of focus for
studies of episodic memory using single-cell recordings
in animals242–245, and the results from human functional
imaging studies209,223,225,246,247 provide evidence that the
mechanisms for retrospective timing not only involve
structures within the medial temporal lobe but are also
likely to overlap with the mechanisms for event segmen-
tation and episodic memory. However, this approach
does make a critical assumption that the representation
of time used for retrospective timing is present regard-
less of whether animals actually engage in explicit retro-
spective timing. This is in part justified by the fact that
retrospective timing occurs, by definition, without initial
explicit awareness of the importance of tracking time
and is inherently post hoc.
As a starting point for understanding the dynamics
of event trajectories in more detail, a number of studies
have examined how population states in the hippocam-
pus and the entorhinal cortex change over time as ani-
mals explore a constant environment236–238,240,241,248–250.
Establishing the general paradigm for these studies,
Mankin et al. recorded from both CA1 and CA3 while
animals freely foraged within an open field environment.
The temporal structure of the experiment was such that
animals would forage for four 10-min trials, once in the
morning and once in the afternoon six hours later, and
animals would repeat this temporal structure daily for up
to a month. Within CA1, Mankin et al. observed that the
correlation between pairs of population vectors defined
at different time points decreased as a function of tem-
poral distance, reflecting the evolution of CA1 activity
along an event trajectory where events were defined by
the trial structure of the experiment. Given that the same
temporal structure was repeated across days, animals
could potentially have learned this structure implicitly
through temporal expectation processes, generating a
neural trajectory that would similarly exhibit continuous
change over a day. However, when looking across days,
Mankin et al. found that event trajectories continued to
evolve such that population states across days could also
be easily differentiated, suggesting that at least in CA1,
individual days of the same repeated experimental struc-
ture were treated as unique. Other studies using similar
experimental paradigms with fixed temporal structures
and free foraging repeated across days have also observed
event trajectories continue to evolve across days, while
failing to observe repeated stable neural trajectories
corresponding to the repeated temporal structures112,241.
Across the hippocampal subfields, event trajectories
evolve at different rates. The rate of evolution is fastest
in CA2, allowing time points separated by a few hours
to be easily distinguished237, whereas the rate of evolu-
tion is slower in CA1, allowing time points separated
by a day to be easily distinguished237,238. In CA3 and
the dentate gyrus, population states across days remain
largely constant237,238,240,248, suggesting that the repeated
experiments across days may not be clearly differenti-
ated as separate events. However, there is evidence that
evolution of event trajectories in the dentate gyrus may
www.nature.com/nrn

occur on much longer timescales such that time points
separated by several weeks become distinguishable240.
The hippocampus receives most of its cortical input
from the entorhinal cortex, which comprises a medial
subdivision and a lateral subdivision251, and event trajec-
tories have also been observed in the lateral entorhinal
cortex241 (LEC) but not the MEC236,241 (Fig. 4b), which
does not differentiate between repeated experiences
within the same physical environment. The rate of evolu-
tion in the LEC is faster than that in CA2 such that time
points separated by minutes are easily distinguishable241,
and this difference suggests that the entorhinal cortex
and hippocampus may each represent experience on a
different timescale.
Single-cell recordings have also observed several
neural correlates of event segmentation. Event bound-
aries are the defining feature of event segmentation,
and beginning with the initial study of event segmenta-
tion using human functional imaging by Zacks et al.198,
numerous studies have observed increased activity
around event boundaries198,213,252–254. Single-cell record-
ings from both rats241,255 and humans256 have similarly
observed altered population state dynamics follow-
ing event boundaries, as population states change at a
much greater rate immediately following event bound-
aries (Fig. 4b). Whereas in human experiments event
boundaries can be directly reported, event boundaries
in experiments using rats require the assumption that
a sufficiently salient change, such as being physically
transferred between environments, is perceived as an
event boundary. In one study where event boundaries
were defined more subtly by unannounced rule shifts,
hippocampal representations for these event boundaries
were not observed257. Representation of event bounda-
ries extends down to the single-cell level, as cells which
specifically fire following event boundaries have been
observed in both rats255 and humans256. Recording from
the hippocampus, parahippocampal gyrus and amygdala
of human patients, Zheng et al. presented participants
with a number of different video clips which had no
boundaries (one continuous video shot), soft bound-
aries (cuts to a new scene in the same video) or hard
boundaries (cuts to a new scene in a different video) and
found cells which responded immediately after either
hard boundaries alone, or hard and soft boundaries. In
addition to cells with activity tied to event boundaries,
cells potentially representing event identify have also
been observed. Sun et al. trained animals to run four
laps around a track in order to receive a reward at a fixed
reward site258, and found that many place cells in CA1
maintained the same firing location for each lap but
modulated their firing rate depending on the lap num-
ber. Thus, separate events corresponding to separate
laps were encoded by single units through differences
in the firing rate. In addition to the aforementioned cells,
which show transient increases in activity, cells with
ramping activity set to event boundaries have also been
observed241; although these cells exhibit time-varying
activity, they do not appear to be the primary driver
of event trajectories, as clear differentiation of indi-
vidual events is still possible when they are removed
from analysis.
NATuRe RevIeWS | Neuroscience
0123456789();:
Reviews
How neural event trajectories arise
The substantial influence of event structure on retrospec-
tive timing implies that part of understanding the mech-
anisms for retrospective timing involves understanding
the mechanisms for how events are segmented. Existing
work suggests that event boundaries arise either as part
of an active process of perception in which violations of
internal predictive models of the current ongoing event
define event boundaries259,260 or as an inherent property
of the context within which experience occurs211,261,262
(for example, entering a new room generates an event
boundary201). Recent functional imaging experiments
by Baldassano et al. demonstrate that activity patterns
across the brain reflect the segmentation of ongoing
experience into discrete chunks at different timescales213.
In lower-order sensory areas, these chunks were only a
few seconds long, whereas in higher-order areas, such
as the angular gyrus, these chunks could be more than
1 min long. Importantly, the shorter-timescale event
boundaries from lower-order areas aligned with the
longer-timescale event boundaries of higher-order
areas, suggesting that across the cortical processing hier-
archy, shorter-timescale chunks are bound together to
form longer-timescale chunks, which eventually reach
the level of discrete events defining the narrative level
of episodic experience213,214,263. Consistent with this,
Baldassano et al. also observed that human-annotated
event boundaries most closely matched event bound-
aries defined by activity from higher-order areas, while
the degree of matching also increased across the pro-
cessing hierarchy. Earlier in the processing hierarchy,
segmentation of ongoing experience may largely be due
to the temporal structure of the external environment
and the sensory stimuli it provides, while increasing
timescales across the hierarchy of sensory areas may
reflect the formation of bound sensory representations
for external stimuli such as objects and scenes. At the
level of the medial temporal lobe, these representations
across different sensory streams may then be integrated
to generate a multimodal representation of ongoing
experience whose temporal structure may then be
refined in higher-order areas.
The initial generation of a multimodal representation
of experience may occur in the LEC, which has been
proposed to encode the experiential content of epi-
sodic memories264, and both anatomical and functional
data suggest this possibility. Anatomically, the LEC is
one of the major hubs for cortical input across the entire
brain, receiving input from an extremely diverse range
of cortical structures265–268, and is the primary structure
through which integrated multimodal information may
reach the hippocampus251,266,269,270. Functionally, LEC
neurons generally respond to a wide range of stimuli
and associations241,271–282, consistent with a role in bind-
ing multimodal representations, and many LEC cells
encode information about external features, both sen-
sory and non-sensory, within an egocentric reference
frame282, which episodic memory is typically perceived
through. Notably, data from two human functional
imaging studies suggest that entorhinal cortex activity225
and LEC activity specifically247 during the initial encod-
ing of ongoing experience may be particularly important
Reviews
for later retrospective timing. This may be due to the
initial representation of ongoing experience generated
within the LEC being used as a default source of tempo-
ral information. Downstream of the LEC, the functions
of pattern separation and pattern completion carried
out by the hippocampal subfields283 may serve as nat-
ural mechanisms for segmentation of entorhinal input
into episodic events. The fact that hippocampal activ-
ity has often been observed at the end of events may
reflect a combination of this process and consolidation
mechanisms for encoding these events into long-term
memory213,253,284–287. Importantly, all of these processes for
event segmentation and the storage of event representa-
tions in long-term memory define the major source of
temporal information for retrospective timing.
How neural event trajectories are used
Although we know a good deal about the dynamics
of hippocampal population activity during the initial
encoding of experiences, how these dynamics may
subsequently be used to actually estimate retrospective
duration remains essentially unknown, in large part
owing to a lack of experiments directly examining this
question. As a starting point, however, it seems unlikely
that the mechanisms for the actual estimation process
during retrospective timing would be exactly the same
as those observed during prospective timing. The most
salient difference between retrospective timing and pro-
spective timing is that the estimation of duration is a
process separate from the initial encoding of events
defining the duration (outside certain instances of sen-
sory timing which also contain this separation). Because
of this separation, population state dynamics during
the initial experience offer only limited insight into the
subsequent estimation process even though they may
form the basis for subsequent duration estimation —
retrospective estimation of duration cannot be described
simply in terms of the terminal state of a trajectory or the
speed at which a trajectory progresses. Indeed, even if
the trajectory corresponding to an interval were recalled
and replayed in a temporally compressed manner, as has
been observed in the hippocampus in other non-temporal
contexts288, the trajectory’s speed or terminal state alone
would not be sufficient — either of these features would
still need to be converted into a duration value.
If we assume that neural event trajectories during ini-
tial encoding of an experience are recalled253,256,284,286,289–293
and used for subsequent estimation of retrospective
duration, then understanding the estimation process
becomes a question of how event trajectories can be
used to generate a duration magnitude (Fig. 5c). This
process would be straightforward if event trajectories
within the hippocampus progressed in a linear manner,
as suggested by initial observations238: the difference
between population states at the start and the end of
an interval would directly correlate with physical dura-
tion. However, because event trajectories are driven by
representations of ongoing events and can often exhibit
abrupt nonlinear changes due to non-temporal fac-
tors, they are likely to be unsuitable as a linear metric
of physical time. Illustrating this point, the amount by
which hippocampal population states change across two
0123456789();:
different environments experienced within the same day
is larger than the amount by which they change across
2 weeks within the same environment294.
Psychological models of memory for time have pro-
vided a number of different hypotheses about how tem-
poral information may be stored and used158,219,226,295,296.
Among these, a temporal context model based on the
Laplace transform of event history is of particular impor-
tance, offering a potentially simple explanation for how
event trajectories can be used to estimate duration297–299.
In this model, each individual event generates a sequence
of activity within a specific subpopulation of cells
such that a representation of the event persists across
the amount of physical time that the sequence takes.
Individual cells within this sequence receive input from a
set of leaky integrators, and the time at which they fire is
determined by the time constants of these inputs. When
multiple events occur close together in time, they gener-
ate multiple different sequences across the overall popu­
lation which may or may not overlap with each other
depending on the temporal distance between the events.
Altogether then, these sequences provide a readout of
recent events and their temporal history.
Most importantly with regard to retrospective timing,
because a common set of time constants defines how the
sequence of activity for any event progresses in physical
time, a temporal metric is defined by the distribution
of these time constants. Thus, specific population states
(corresponding to specific events) across the overall
popu­lation define specific moments in physical time, and
the magnitude of differences in population state approxi-
mates durations in physical time. Extension of the model
to describe specifically how retrospective timing may be
carried out still remains to be done, but in broad terms,
estimation could be accomplished by counting the num-
ber of changes in population state, similar to evidence
accumulation processes22, by comparison of population
states (for example, through repetition suppression of
match enhancement mechanisms300) or by some means
of converting the output of a pattern separation function
into a magnitude. Notably, the model is able to reproduce
behavioural phenomena observed during free memory
recall (recency and contiguity effects)297,298, bolstering its
relevance to memory for time. Additionally, in terms of
being biologically realistic, it uses representations which
effectively take the form of experimentally observed
time cells69,70 and ramping temporal context cells88,241,301,
although it should be noted that whether time cells actu-
ally encode past event history still remains to be fully ver-
ified, as time cells within existing experiments could also
be explained in terms of prospective timing mechanisms.
Although evidence from human functional imaging
demonstrates that temporal information about duration
may be present within the hippocampus302,303, the actual
computation of duration may ultimately be performed
outside the hippocampus304,305. We have focused on how
duration may be estimated from event trajectories, but
there is also a role within retrospective timing for tem-
poral landmarks and schema, particularly in cases where
events are not contiguous and span multiple experiences.
How these different types of temporal information are
together used in the estimation process still remains to be
www.nature.com/nrn

understood, and overall our understanding of the mech-
anisms for actual estimation of retrospective duration
still remain very limited, but experiments examining
these questions are beginning to be conducted306,307.
Neural event trajectories and order
Although we have focused solely on the estimation of
duration, timing based on memory of past events also
encompasses the temporal order of events, which has
been a larger focus of study211,308–310. Within episodic
memory, it is notably temporal order which plays
the critical role in defining the narrative structure of
experience311 — the relative duration of events and the
temporal distance between them are often secondary —
and evidence from a number of studies210,246,256,309,312–315
suggests that temporal information related to order is
also drawn from event trajectories and is influenced by
the process of event segmentation. The hippocampus in
particular may play an important role in this process, as
the ubiquity with which it generates sequential activity,
including in non-spatial contexts316–320, suggests an intrin-
sic capability for representing order93. In one of the first
studies observing event trajectories, Manns et al. recorded
from the hippocampus of rats engaged in an odour
sequence memory task in which rats were presented in
each trial with a sequences of five odours and had to indi-
cate which of two test odours from the sequence occurred
earlier314. In trials where rats accurately judged order, the
population states for odours which were farther apart in
the initial sequence were less similar (that is, their rep-
resentations were more distinct), while this effect was
not present in trials where rats incorrectly judged order.
Similarly, the event boundary-responsive cells recorded
from human patients by Zheng et al. were also predictive
of subsequent order memory, as they exhibited signif-
icantly reduced activity in trials where temporal order
was incorrectly remembered256. Using a paradigm similar
to that used by Manns et al., Cox et al. presented mice
with a single sequence of five odours and then tested
them with two of the odours, taking advantage of the
animals’ natural instinct to explore the odour presented
earlier in the sequence more and thereby removing the
need for a trained behavioural response315. Inactivation of
CA3 and the LEC but not the MEC impacted encoding
of temporal order, again suggesting that in addition to
the hippocampus, the LEC may also play a key role in the
representation of events and driving the evolution of
event trajectories.
Interaction between timing mechanisms
Although prospective timing and retrospective timing
may arise through different mechanisms, they are not
entirely isolated from each other. Within the context of
learning and memory consolidation, each may influence
the other. First, the learning of temporal structure for
prospective timing may in certain instances reflect the
transformation of temporal information within episodic
memory into semantic knowledge321,322. In this case, the
first experience of a sequence of events with stable tem-
poral structure generates a trajectory through popu­lation
activity space which reflects the salient features of the
experience that drive event segmentation. The encoding
NATuRe RevIeWS | Neuroscience
0123456789();:
Reviews
of subsequent repeated experiences with the same
sequence of events builds on the recalled trajectory of
the initial experience, but the stable temporal structure
allows learning and memory consolidation mechanisms
to shape the neural trajectory to evolve in a reproducible
manner which can support prospective timing. Several
studies provide observations consistent with this sce-
nario. In experiments recording from ensembles of
single neurons in human patients, the population states
observed during initial experience are in part reinstated
during subsequent recall256,289,291. Separately, Tsao et al.
observed that when animals switched between free for-
aging behaviour in an open field and temporally struc-
tured lap running in a maze, the same population of LEC
neurons could generate unique event trajectories dur-
ing free foraging and repeated stable trajectories during
stereotyped lap running241. Human functional imaging
studies provide similar evidence that the entorhinal cor-
tex is capable of generating both event trajectories dur-
ing episodic experience225 and stable trajectories during
repeated experience of a fixed event structure323. Thus,
the activity of a population of entorhinal neurons would,
in principle, be able to transform from an initially unique
trajectory shaped by a single experience into a repeata-
ble trajectory determined by the stable temporal features
of that experience (that is, similar to hippocampal time
cells), although such a transformation still remains to be
directly observed through time-lapse recordings.
Second, the repeated experience of stable temporal
structures can lead to not just the learning of stable
neural trajectories supporting prospective timing but
also temporal schema. Over extended periods in which
multiple experiences are framed within a constant tem-
poral structure, a temporal schema can arise as this
structure is extracted from the different experiences.
The fact that childrens’ ability to estimate duration
using memory of past events develops gradually181,182,
progressing on a timescale from a single day to months
to years as children gain more and more experience
with stable temporal structures of increasing length,
may be related to this process. The use of temporal
schema within retrospective timing may explain how
subjective duration estimates for different groups of
events spanning the same physical duration may vastly
differ324, and human functional imaging studies are now
beginning to examine how temporal schema are formed
and used313,325–329. Temporal schema may also structure
abstract representations of time, thus allowing mental
time travel. In a recent study, human participants first
learned a list of events and their dates, and then on indi-
vidual trials imagined themselves at a specific timepoint
in the past, present or future and were asked whether
a particular event from the learned list was before or
after this imagined position. Magnetoencephalographic
recordings from participants revealed increased activ-
ity in medial temporal lobe structures throughout the
trial, during both the initial imagined self-projection
in time and the subsequent determination of temporal
order330,331. Such mental time travel has been proposed
to require an allocentric-to-egocentric conversion332, and
the allocentric representation of time supporting this
may be built up through temporal schema.
Reviews

Finally, going backwards from prospective timing
to retrospective timing, learned temporal schema may
directly influence the processes underlying event seg-
mentation and the online processing of experience.
Recent work provides evidence that learning temporal
structure through repeated viewings of a video clip can
shift neural activity initially tied to event boundaries for-
wards in time such that event boundaries are anticipated
by neural activity333. In a similar fashion, it may be possi-
ble for existing temporal schema to alter neural activity
online in anticipation of future events.
Conclusions
In this Review, we have focused on how temporal infor-
mation is represented and used during prospective and
retrospective timing. For prospective timing, temporal
information is encoded within stable neural trajectories,
whereas for retrospective timing, temporal information
is encoded within neural event trajectories, which are
a result of the event segmentation process. The sub-
sequent ways in which these two representations of
temporal information are used to generate duration
estimates are relatively distinct. Prospective dura-
tion is directly embedded within the evolution of the
stable trajectories, as features defined by the evolution
of these trajectories are associated with specific dura-
tions, whereas retrospective duration must be sep-
arately computed using reinstated population states
from event trajectories. Although the mechanisms for
prospective and retrospective timing are distinct, the
two separate representations of temporal information
arising from them — stable trajectories and event
trajectories — are likely to interact bidirectionally with
each other.
Published online xx xx xxxx

1. Buhusi, C. V. & Meck, W. H. What makes us tick?
Functional and neural mechanisms of interval timing.
Nat. Rev. Neurosci. 6, 755–765 (2005).
2. Merchant, H., Harrington, D. L. & Meck, W. H.
Neural basis of the perception and estimation of time.
Annu. Rev. Neurosci. 36, 313–336 (2013).
3. Grondin, S. Timing and time perception: a review
of recent behavioral and neuroscience findings and
theoretical directions. Atten. Percept. Psychophys.
72, 561–582 (2010).
4. Gibbon, J., Malapani, C., Dale, C. L. & Gallistel, C. R.
Toward a neurobiology of temporal cognition:
advances and challenges. Curr. Opin. Neurobiol. 7,
170–184 (1997).
5. Ivry, R. B. & Schlerf, J. E. Dedicated and intrinsic
models of time perception. Trends Cogn. Sci. 12,
273–280 (2008).
6. Eichenbaum, H. Time cells in the hippocampus: a new
dimension for mapping memories. Nat. Rev. Neurosci.
15, 732–744 (2014).
7. Gallistel, C. R. The Organization of Learning (The MIT
Press, 1990).
8. Teki, S., Gu, B. M. & Meck, W. H. The persistence
of memory: how the brain encodes time in memory.
Curr. Opin. Behav. Sci. 17, 178–185 (2017).
9. Fraisse, P. The Psychology of Time (Harper & Row,
1963).
10. Fraisse, P. Perception and estimation of time.
Annu. Rev. Psychol. 35, 1–37 (1984).
11. Creelman, C. D. Human discrimination of auditory
duration. J. Acoust. Soc. Am. 34, 582–593 (1962).
12. Treisman, M. Temporal discrimination and the
indifference interval: implications for a model
of the “internal clock”. Psychol. Monogr. Gen. Appl.
77, 1–31 (1963).
13. Gibbon, J., Church, R. M. & Meck, W. H. Scalar timing
in memory. Ann. N. Y. Acad. Sci. 423, 52–77 (1984).
14. Reppert, S. M. & Weaver, D. R. Coordination of circadian
timing in mammals. Nature 418, 935–941 (2002).
15. Welsh, D. K., Logothetis, D. E., Meister, M. &
Reppert, S. M. Individual neurons dissociated from
rat suprachiasmatic nucleus express independently
phased circadian firing rhythms. Neuron 14, 697–706
(1995).
16. Miall, C. The storage of time intervals using oscillating
neurons. Neural Comput. 1, 359–371 (1989).
17. Matell, M. S. & Meck, W. H. Cortico-striatal circuits
and interval timing: coincidence detection of oscillatory
processes. Brain Res. Cogn. Brain Res. 21, 139–170
(2004).
18. Gu, B. M., van Rijn, H. & Meck, W. H. Oscillatory
multiplexing of neural population codes for interval
timing and working memory. Neurosci. Biobehav. Rev.
48, 160–185 (2015).
19. Bartolo, R., Prado, L. & Merchant, H. Information
processing in the primate basal ganglia during
sensory-guided and internally driven rhythmic tapping.
J. Neurosci. 34, 3910 (2014).
20. Kononowicz, T. W. & Rijn, H. V. Single trial beta
oscillations index time estimation. Neuropsychologia
75, 381–389 (2015).
21. Kononowicz, T. W., Roger, C. & van Wassenhove, V.
Temporal metacognition as the decoding of
self-generated brain dynamics. Cereb. Cortex 29,
4366–4380 (2018).
22. Balcı, F. & Simen, P. A decision model of timing.
Curr. Opin. Behav. Sci. 8, 94–101 (2016).
23. Buonomano, D. V. & Laje, R. Population clocks: motor
timing with neural dynamics. Trends Cogn. Sci. 14,
520–527 (2010).
24. Buonomano, D. V. & Mauk, M. D. Neural network
model of the cerebellum: temporal discrimination
and the timing of motor responses. Neural Comput.
6, 38–55 (1994).
25. Remington, E. D., Egger, S. W., Narain, D., Wang, J.
& Jazayeri, M. A dynamical systems perspective on
flexible motor timing. Trends Cogn. Sci. 22, 938–952
(2018).
26. Coull, J. T. & Nobre, A. C. Dissociating explicit timing
from temporal expectation with fMRI. Curr. Opin.
Neurobiol. 18, 137–144 (2008).
27. Zelaznik, H. N., Spencer, R. M. C. & Ivry, R. B.
Dissociation of explicit and implicit timing in repetitive
tapping and drawing movements. J. Exp. Psychol.
Hum. Percept. Perform. 28, 575–588 (2002).
28. Ivry, R. B., Spencer, R. M., Zelaznik, H. N. &
Diedrichsen, J. The cerebellum and event timing.
Ann. N. Y. Acad. Sci. 978, 302–317 (2002).
29. Kim, J., Ghim, J.-W., Lee, J. H. & Jung, M. W. Neural
correlates of interval timing in rodent prefrontal
cortex. J. Neurosci. 33, 13834–13847 (2013).
This study demonstrates that sensory timing can
be carried out by evolution of a common neural
trajectory reaching different terminal states for
different physical durations.
30. Gouvêa, T. S. et al. Striatal dynamics explain duration
judgments. Elife 4, e11386 (2015).
31. Mendoza, G., Méndez, J. C., Pérez, O., Prado, L. &
Merchant, H. Neural basis for categorical boundaries
in the primate pre-SMA during relative categorization
of time intervals. Nat. Commun. 9, 1098 (2018).
32. Kim, J., Kim, D. & Jung, M. W. Distinct dynamics
of striatal and prefrontal neural activity during
temporal discrimination. Front. Integr. Neurosci.
12, 34 (2018).
33. Shimbo, A., Izawa, E.-I. & Fujisawa, S. Scalable
representation of time in the hippocampus. Sci. Adv.
7, eabd7013 (2021).
34. Mita, A., Mushiake, H., Shima, K., Matsuzaka, Y.
& Tanji, J. Interval time coding by neurons in the
presupplementary and supplementary motor areas.
Nat. Neurosci. 12, 502–507 (2009).
35. Mello, G. B., Soares, S. & Paton, J. J. A scalable
population code for time in the striatum. Curr. Biol.
25, 1113–1122 (2015).
36. Zhou, S., Masmanidis, S. C. & Buonomano, D. V.
Neural sequences as an optimal dynamical regime for
the readout of time. Neuron https://doi.org/10.1016/
j.neuron.2020.08.020 (2020).
37. Xu, M., Zhang, S. Y., Dan, Y. & Poo, M. M.
Representation of interval timing by temporally
scalable firing patterns in rat prefrontal cortex.
Proc. Natl Acad. Sci. USA 111, 480–485 (2014).
38. Bakhurin, K. I. et al. Differential encoding of time by
prefrontal and striatal network dynamics. J. Neurosci.
37, 854 (2017).
39. Emmons, E. B. et al. Rodent medial frontal control
of temporal processing in the dorsomedial striatum.
J. Neurosci. 37, 8718–8733 (2017).
40. Maimon, G. & Assad, J. A. A cognitive signal for
the proactive timing of action in macaque LIP.
Nat. Neurosci. 9, 948–955 (2006).
41. Churchland, M. M. et al. Neural population dynamics
during reaching. Nature 487, 51–56 (2012).
42. Modi, M. N., Dhawale, A. K. & Bhalla, U. S. CA1 cell
activity sequences emerge after reorganization of
network correlation structure during associative
learning. Elife 3, e01982 (2014).
43. Wang, J., Narain, D., Hosseini, E. A. & Jazayeri, M.
Flexible timing by temporal scaling of cortical
responses. Nat. Neurosci. 21, 102–110 (2018).
This study demonstrates that motor timing can
be carried out by controlling the speed at which
neural trajectories evolve, which is reflected in the
temporal scaling of single-unit responses. Recurrent
neural networks trained to perform the same timing
task also reach the same solution for controlling the
speed at which population activity evolves.
44. Egger, S. W., Remington, E. D., Chang, C.-J. &
Jazayeri, M. Internal models of sensorimotor
integration regulate cortical dynamics. Nat. Neurosci.
22, 1871–1882 (2019).
45. Kim, J., Jung, A. H., Byun, J., Jo, S. & Jung, M. W.
Inactivation of medial prefrontal cortex impairs time
interval discrimination in rats. Front. Behav. Neurosci.
3, 38 (2009).
46. Merchant, H., Zarco, W., Pérez, O., Prado, L. &
Bartolo, R. Measuring time with different neural
chronometers during a synchronization-continuation
task. Proc. Natl Acad. Sci. USA 108, 19784–19789
(2011).
47. Gámez, J., Mendoza, G., Prado, L., Betancourt, A. &
Merchant, H. The amplitude in periodic neural state
trajectories underlies the tempo of rhythmic tapping.
PLoS Biol. 17, e3000054 (2019).
48. Crowe, D. A., Zarco, W., Bartolo, R. & Merchant, H.
Dynamic representation of the temporal and
sequential structure of rhythmic movements in the
primate medial premotor cortex. J. Neurosci. 34,
11972 (2014).
49. Merchant, H. et al. Sensorimotor neural dynamics during
isochronous tapping in the medial premotor cortex of
the macaque. Eur. J. Neurosci. 41, 586–602 (2015).
50. Lewis, P. A. & Miall, R. C. Distinct systems for automatic
and cognitively controlled time measurement: evidence
from neuroimaging. Curr. Opin. Neurobiol. 13,
250–255 (2003).
51. Rammsayer, T. H. Neuropharmacological evidence
for different timing mechanisms in humans. Q. J. Exp.
Psychol. B 52, 273–286 (1999).
52. Karmarkar, U. R. & Buonomano, D. V. Timing in the
absence of clocks: encoding time in neural network
states. Neuron 53, 427–438 (2007).
This study provides the first empirical evidence,
in the form of behavioural results based on
predictions from a state-dependent network, that
timing may be carried out through non-metrical
changes in neural population activity, as opposed
to a dedicated internal clock.

www.nature.com/nrn

0123456789();:

53. Rammsayer, T. H. & Lima, S. D. Duration discrimination
of filled and empty auditory intervals: cognitive
and perceptual factors. Percept. Psychophys. 50,
565–574 (1991).
54. Spencer, R. M. C., Karmarkar, U. & Ivry, R. B.
Evaluating dedicated and intrinsic models of temporal
encoding by varying context. Philos. Trans. R. Soc.
Lond. B Biol. Sci. 364, 1853–1863 (2009).
55. Rammsayer, T. H., Borter, N. & Troche, S. J.
Visual-auditory differences in duration discrimination
of intervals in the subsecond and second range.
Front. Psychol. https://doi.org/10.3389/
fpsyg.2015.01626 (2015).
56. Sohn, H., Narain, D., Meirhaeghe, N. & Jazayeri, M.
Bayesian computation through cortical latent
dynamics. Neuron 103, 934–947.e935 (2019).
57. Shikano, Y., Ikegaya, Y. & Sasaki, T. Minute-encoding
neurons in hippocampal-striatal circuits. Curr. Biol.
https://doi.org/10.1016/j.cub.2021.01.032 (2021).
58. Jacobs, N. S., Allen, T. A., Nguyen, N. & Fortin, N. J.
Critical role of the hippocampus in memory for
elapsed time. J. Neurosci. 33, 13888 (2013).
59. Meck, W. H., Church, R. M. & Matell, M. S.
Hippocampus, time, and memory–a retrospective
analysis. Behav. Neurosci. 127, 642–654 (2013).
60. Palombo, D. J., Keane, M. M. & Verfaellie, M. Does
the hippocampus keep track of time? Hippocampus
26, 372–379 (2016).
61. Sabariego, M. et al. In the temporal organization
of episodic memory, the hippocampus supports the
experience of elapsed time. Hippocampus 31, 46–55
(2021).
62. Narayanan, N. S. Ramping activity is a cortical
mechanism of temporal control of action. Curr. Opin.
Behav. Sci. 8, 226–230 (2016).
63. Komura, Y. et al. Retrospective and prospective coding
for predicted reward in the sensory thalamus. Nature
412, 546–549 (2001).
64. Parker, K. L., Alberico, S. L., Miller, A. D. &
Narayanan, N. S. Prefrontal D1 dopamine signaling
is necessary for temporal expectation during reaction
time performance. Neuroscience 255, 246–254
(2013).
65. Lebedev, M. A., O’Doherty, J. E. & Nicolelis, M. A. L.
Decoding of temporal intervals from cortical
ensemble activity. J. Neurophysiol. 99, 166–186
(2008).
66. Tanaka, M. Cognitive signals in the primate motor
thalamus predict saccade timing. J. Neurosci. 27,
12109 (2007).
67. Jazayeri, M. & Shadlen, M. N. A neural mechanism
for sensing and reproducing a time interval. Curr. Biol.
25, 2599–2609 (2015).
68. Kunimatsu, J., Suzuki, T. W., Ohmae, S. & Tanaka, M.
Different contributions of preparatory activity in the
basal ganglia and cerebellum for self-timing. Elife 7,
e35676 (2018).
69. Pastalkova, E., Itskov, V., Amarasingham, A. &
Buzsáki, G. Internally generated cell assembly
sequences in the rat hippocampus. Science 321,
1322–1327 (2008).
This article demonstrates sequential activity in the
hippocampus which spans a delay period in which
animals run in a fixed location.
70. MacDonald, C. J., Lepage, K. Q., Eden, U. T. &
Eichenbaum, H. Hippocampal “time cells” bridge the
gap in memory for discontiguous events. Neuron 71,
737–749 (2011).
71. Jin, D. Z., Fujii, N. & Graybiel, A. M. Neural
representation of time in cortico-basal ganglia
circuits. Proc. Natl Acad. Sci. USA 106, 19156
(2009).
72. Tiganj, Z., Jung, M. W., Kim, J. & Howard, M. W.
Sequential firing codes for time in rodent medial
prefrontal cortex. Cereb. Cortex 27, 5663–5671
(2017).
73. Murakami, M., Vicente, M. I., Costa, G. M. &
Mainen, Z. F. Neural antecedents of self-initiated
actions in secondary motor cortex. Nat. Neurosci.
17, 1574–1582 (2014).
74. Namboodiri, V. M. K., Huertas, M. A., Monk, K. J.,
Shouval, H. Z. & Hussain Shuler, M. G. Visually cued
action timing in the primary visual cortex. Neuron 86,
319–330 (2015).
75. Shuler, M. G. & Bear, M. F. Reward timing in the
primary visual cortex. Science 311, 1606 (2006).
76. Matell, M. S., Meck, W. H. & Nicolelis, M. A.
Interval timing and the encoding of signal duration
by ensembles of cortical and striatal neurons.
Behav. Neurosci. 117, 760–773 (2003).
77. Coull, J. T. & Nobre, A. C. Where and when to pay
attention: the neural systems for directing attention
NATuRe RevIeWS | Neuroscience
to spatial locations and to time intervals as revealed
by both PET and fMRI. J. Neurosci. 18, 7426 (1998).
This study provides the first insights into the
neural mechanisms for temporal expectation,
demonstrating that a network of brain areas is
involved in temporal expectation, and differs from
the network of brain areas involved in spatial
attention.
78. Nobre, A. C. & van Ede, F. Anticipated moments:
temporal structure in attention. Nat. Rev. Neurosci.
19, 34–48 (2018).
79. Nobre, A. C., Correa, A. & Coull, J. T. The hazards of
time. Curr. Opin. Neurobiol. 17, 465–470 (2007).
80. Miniussi, C., Wilding, E. L., Coull, J. T. & Nobre, A. C.
Orienting attention in time. Modulation of brain
potentials. Brain 122, 1507–1518 (1999).
81. Carnevale, F., de Lafuente, V., Romo, R., Barak, O.
& Parga, N. Dynamic control of response criterion in
premotor cortex during perceptual detection under
temporal uncertainty. Neuron 86, 1067–1077 (2015).
This study demonstrates neural trajectories
evolving in a manner reflecting the influence of
temporal expectation, with neural population
activity moving closer to a decision threshold
during the appropriate time window.
82. Ghose, G. M. & Maunsell, J. H. R. Attentional
modulation in visual cortex depends on task timing.
Nature 419, 616–620 (2002).
This study demonstrates that the activity of
individual neurons can be modulated by temporal
expectation, following the temporal structure of the
hazard rate for when a target stimulus might occur.
83. Jaramillo, S. & Zador, A. M. The auditory cortex
mediates the perceptual effects of acoustic temporal
expectation. Nat. Neurosci. 14, 246–251 (2011).
84. Janssen, P. & Shadlen, M. N. A representation of
the hazard rate of elapsed time in macaque area LIP.
Nat. Neurosci. 8, 234–241 (2005).
85. Anderson, B. & Sheinberg, D. L. Effects of temporal
context and temporal expectancy on neural activity
in inferior temporal cortex. Neuropsychologia 46,
947–957 (2008).
86. Toso, A., Reinartz, S., Pulecchi, F. & Diamond, M. E.
Time coding in rat dorsolateral striatum. Neuron 109,
3663–3673.e3666 (2021).
87. Kobak, D. et al. Demixed principal component analysis
of neural population data. Elife https://doi.org/10.7554/
eLife.10989 (2016).
88. Umbach, G. et al. Time cells in the human
hippocampus and entorhinal cortex support episodic
memory. Proc. Natl Acad. Sci. USA 117, 28463
(2020).
89. Kraus, B. J., Robinson, R. J. II, White, J. A.,
Eichenbaum, H. & Hasselmo, M. E. Hippocampal
“time cells”: time versus path integration. Neuron 78,
1090–1101 (2013).
90. MacDonald, C. J., Carrow, S., Place, R. &
Eichenbaum, H. Distinct hippocampal time cell
sequences represent odor memories in immobilized
rats. J. Neurosci. 33, 14607–14616 (2013).
91. Cueva Christopher, J. et al. Low-dimensional dynamics
for working memory and time encoding. Proc. Natl
Acad. Sci. USA 117, 23021–23032 (2020).
92. Koay, S. A., Charles, A. S., Thiberge, S. Y., Brody, C. D.
& Tank, D. W. Sequential and efficient neural-population
coding of complex task information. Neuron 110,
328–349.e311 (2022).
93. Buzsáki, G. & Tingley, D. Space and time: the
hippocampus as a sequence generator. Trends Cogn. Sci.
22, 853–869 (2018).
94. Kawai, R. et al. Motor cortex is required for learning
but not for executing a motor skill. Neuron 86,
800–812 (2015).
95. Heys, J. G. & Dombeck, D. A. Evidence for a subcircuit
in medial entorhinal cortex representing elapsed time
during immobility. Nat. Neurosci. 21, 1574–1582
(2018).
96. Heys, J. G., Wu, Z., Allegra Mascaro, A. L. &
Dombeck, D. A. Inactivation of the medial entorhinal
cortex selectively disrupts learning of interval timing.
Cell Rep. 32, 108163 (2020).
97. Leong, Y. C., Radulescu, A., Daniel, R., DeWoskin, V.
& Niv, Y. Dynamic interaction between reinforcement
learning and attention in multidimensional
environments. Neuron 93, 451–463 (2017).
98. Radulescu, A., Niv, Y. & Ballard, I. Holistic
reinforcement learning: the role of structure and
attention. Trends Cogn. Sci. 23, 278–292 (2019).
99. Suh, J., Rivest, A. J., Nakashiba, T., Tominaga, T. &
Tonegawa, S. Entorhinal cortex layer III input to the
hippocampus is crucial for temporal association
memory. Science 334, 1415–1420 (2011).
0123456789();:
Reviews
100. Kitamura, T. et al. Island cells control temporal
association memory. Science 343, 896–901 (2014).
101. Polti, I., Nau, M., Kaplan, R., van Wassenhove, V. &
Doeller, C. F. Rapid encoding of task regularities in
the human hippocampus guides sensorimotor timing.
bioRxiv https://doi.org/10.1101/2021.08.03.454928
(2022).
102. Taxidis, J. et al. Differential emergence and stability
of sensory and temporal representations in
context-specific hippocampal sequences. Neuron
https://doi.org/10.1016/j.neuron.2020.08.028
(2020).
103. Buzsáki, G. & Llinás, R. Space and time in the brain.
Science 358, 482 (2017).
104. Remington, E. D., Narain, D., Hosseini, E. A. &
Jazayeri, M. Flexible sensorimotor computations
through rapid reconfiguration of cortical dynamics.
Neuron 98, 1005–1019.e1005 (2018).
105. Driscoll, L. N., Golub, M. D. & Sussillo, D. Computation
through cortical dynamics. Neuron 98, 873–875
(2018).
106. Monteiro, T. et al. Using temperature to analyse the
neural basis of a latent temporal decision. bioRxiv
https://doi.org/10.1101/2020.08.24.251827
(2021).
107. Cao, R., Bladon, J. H., Charczynski, S. J.,
Hasselmo, M. E. & Howard, M. W. Internally
generated time in the rodent hippocampus is
logarithmically compressed. bioRxiv https://doi.org/
10.1101/2021.10.25.465750 (2021).
108. Genovesio, A., Tsujimoto, S. & Wise, S. P. Feature-
and order-based timing representations in the frontal
cortex. Neuron 63, 254–266 (2009).
109. Meck, W. H. & Church, R. M. Abstraction of temporal
attributes. J. Exp. Psychol. Anim. Behav. Process. 8,
226–243 (1982).
110. Buonomano, D. V. A learning rule for the emergence
of stable dynamics and timing in recurrent networks.
J. Neurophysiol. 94, 2275–2283 (2005).
111. Liu, J. K. & Buonomano, D. V. Embedding multiple
trajectories in simulated recurrent neural networks
in a self-organizing manner. J. Neurosci. 29, 13172
(2009).
112. Itskov, V., Curto, C., Pastalkova, E. & Buzsáki, G.
Cell assembly sequences arising from spike threshold
adaptation keep track of time in the hippocampus.
J. Neurosci. 31, 2828 (2011).
113. Rajan, K., Harvey, C. D. & Tank, D. W. Recurrent
network models of sequence generation and memory.
Neuron 90, 128–142 (2016).
114. Goudar, V. & Buonomano, D. V. Encoding sensory
and motor patterns as time-invariant trajectories in
recurrent neural networks. Elife https://doi.org/10.7554/
eLife.31134 (2018).
115. Bi, Z. & Zhou, C. Understanding the computation of
time using neural network models. Proc. Natl Acad.
Sci. USA 117, 10530 (2020).
116. Pérez, O. & Merchant, H. The synaptic properties
of cells define the hallmarks of interval timing in
a recurrent neural network. J. Neurosci. 38, 4186
(2018).
117. Hardy, N. F., Goudar, V., Romero-Sosa, J. L. &
Buonomano, D. V. A model of temporal scaling
correctly predicts that motor timing improves with
speed. Nat. Commun. 9, 4732 (2018).
118. Laje, R. & Buonomano, D. V. Robust timing and
motor patterns by taming chaos in recurrent neural
networks. Nat. Neurosci. 16, 925–933 (2013).
This study demonstrates that recurrent neural
networks are able to generate stable trajectories
capable of carrying out motor timing.
119. Goel, A. & Buonomano, D. V. Temporal interval
learning in cortical cultures is encoded in intrinsic
network dynamics. Neuron 91, 320–327 (2016).
120. Johnson, H. A., Goel, A. & Buonomano, D. V.
Neural dynamics of in vitro cortical networks reflects
experienced temporal patterns. Nat. Neurosci. 13,
917–919 (2010).
121. Ahrens, M. B. & Sahani, M. Observers exploit
stochastic models of sensory change to help judge
the passage of time. Curr. Biol. 21, 200–206 (2011).
122. Buonomano, D. V. & Maass, W. State-dependent
computations: spatiotemporal processing in
cortical networks. Nat. Rev. Neurosci. 10, 113–125
(2009).
123. Buonomano, D. V. & Merzenich, M. M. Temporal
information transformed into a spatial code by a
neural network with realistic properties. Science 267,
1028–1030 (1995).
124. Wiener, M., Turkeltaub, P. & Coslett, H. B. The image
of time: a voxel-wise meta-analysis. Neuroimage 49,
1728–1740 (2010).
Reviews
125. Coull, J. T., Cheng, R. K. & Meck, W. H.
Neuroanatomical and neurochemical substrates of
timing. Neuropsychopharmacology 36, 3–25 (2011).
126. Rao, S. M. et al. Distributed neural systems underlying
the timing of movements. J. Neurosci. 17, 5528
(1997).
127. Coull, J. T., Vidal, F., Nazarian, B. & Macar, F.
Functional anatomy of the attentional modulation
of time estimation. Science 303, 1506–1508
(2004).
128. Stevens, M. C., Kiehl, K. A., Pearlson, G. &
Calhoun, V. D. Functional neural circuits for mental
timekeeping. Hum. Brain Mapp. 28, 394–408 (2007).
129. Coull, J. T., Nazarian, B. & Vidal, F. Timing, storage,
and comparison of stimulus duration engage discrete
anatomical components of a perceptual timing
network. J. Cogn. Neurosci. 20, 2185–2197 (2008).
130. Harrington, D. L. et al. Neural representation of interval
encoding and decision making. Cogn. Brain Res. 21,
193–205 (2004).
131. Harrington, D. L., Zimbelman, J. L., Hinton, S. C. &
Rao, S. M. Neural modulation of temporal encoding,
maintenance, and decision processes. Cereb. Cortex
20, 1274–1285 (2010).
132. Pouthas, V. et al. Neural network involved in time
perception: an fMRI study comparing long and short
interval estimation. Hum. Brain Mapp. 25, 433–441
(2005).
133. Rao, S. M., Mayer, A. R. & Harrington, D. L.
The evolution of brain activation during temporal
processing. Nat. Neurosci. 4, 317–323 (2001).
134. Emmons, E. et al. Temporal learning among prefrontal
and striatal ensembles. Cereb. Cortex Commun. 1,
tgaa058 (2020).
135. Athalye, V. R., Carmena, J. M. & Costa, R. M.
Neural reinforcement: re-entering and refining
neural dynamics leading to desirable outcomes.
Curr. Opin. Neurobiol. 60, 145–154 (2020).
136. Murray, J. M. & Escola, G. S. Learning multiple
variable-speed sequences in striatum via cortical
tutoring. Elife https://doi.org/10.7554/eLife.26084
(2017).
137. Meck, W. H. Neuropharmacology of timing and time
perception. Brain Res. Cogn. Brain Res. 3, 227–242
(1996).
138. Agostino, P. V. & Cheng, R.-K. Contributions of
dopaminergic signaling to timing accuracy and
precision. Curr. Opin. Behav. Sci. 8, 153–160 (2016).
139. Maricq, A. V., Roberts, S. & Church, R. M.
Methamphetamine and time estimation. J. Exp.
Psychol. Anim. Behav. Process. 7, 18–30 (1981).
140. Meck, W. H. Selective adjustment of the speed of
internal clock and memory processes. J. Exp. Psychol.
Anim. Behav. Process. 9, 171–201 (1983).
141. Meck, W. H. Affinity for the dopamine D2 receptor
predicts neuroleptic potency in decreasing the speed
of an internal clock. Pharmacol. Biochem. Behav. 25,
1185–1189 (1986).
142. Soares, S., Atallah, B. V. & Paton, J. J. Midbrain
dopamine neurons control judgment of time. Science
354, 1273 (2016).
143. Howard, C. D., Li, H., Geddes, C. E. & Jin, X. Dynamic
nigrostriatal dopamine biases action selection. Neuron
93, 1436–1450.e1438 (2017).
144. Daw, N. D., Courville, A. C. & Touretzky, D. S.
Representation and timing in theories of the dopamine
system. Neural Comput. 18, 1637–1677 (2006).
145. Gershman, S. J., Moustafa, A. A. & Ludvig, E. A.
Time representation in reinforcement learning models
of the basal ganglia. Front. Comput. Neurosci. 7, 194
(2014).
146. Mikhael, J. G. & Gershman, S. J. Adapting the
flow of time with dopamine. J. Neurophysiol. 121,
1748–1760 (2019).
147. Petter, E. A., Gershman, S. J. & Meck, W. H. Integrating
models of interval timing and reinforcement learning.
Trends Cogn. Sci. 22, 911–922 (2018).
148. Meck, W. H. & Church, R. M. Cholinergic modulation
of the content of temporal memory. Behav. Neurosci.
101, 457–464 (1987).
149. Chubykin, A. A., Roach, E. B., Bear, M. F. &
Shuler, M. G. H. A cholinergic mechanism for reward
timing within primary visual cortex. Neuron 77,
723–735 (2013).
150. Liu, C.-H. et al. Selective activation of a putative
reinforcement signal conditions cued interval timing
in primary visual cortex. Curr. Biol. 25, 1551–1561
(2015).
151. van Wassenhove, V., Herbst, S. K. & Kononowicz, T. W.
in Magnetoencephalography: From Signals to Dynamic
Cortical Networks (eds Selma S. & Cheryl J. A.)
855–905 (Springer International Publishing, 2019).
152. Paton, J. J. & Buonomano, D. V. The neural basis of
timing: distributed mechanisms for diverse functions.
Neuron 98, 687–705 (2018).
153. van Rijn, H., Gu, B. M. & Meck, W. H. Dedicated
clock/timing-circuit theories of time perception and
timed performance. Adv. Exp. Med. Biol. 829, 75–99
(2014).
154. Wiener, M., Matell, M. & Coslett, H. Multiple
mechanisms for temporal processing. Front. Integr.
Neurosci. https://doi.org/10.3389/fnint.2011.00031
(2011).
155. Issa, J. B., Tocker, G., Hasselmo, M. E., Heys, J. G. &
Dombeck, D. A. Navigating through time: a spatial
navigation perspective on how the brain may encode
time. Annu. Rev. Neurosci. https://doi.org/10.1146/
annurev-neuro-101419-011117 (2020).
156. Bausenhart, K. M., Bratzke, D. & Ulrich, R. Formation
and representation of temporal reference information.
Curr. Opin. Behav. Sci. 8, 46–52 (2016).
157. Savelli, F. & Knierim, J. J. Origin and role of path
integration in the cognitive representations of the
hippocampus: computational insights into open
questions. J. Exp. Biol. 222, jeb188912 (2019).
158. Friedman, W. J. Memory for the time of past events.
Psychol. Bull. 113, 44–66 (1993).
159. Ornstein, R. E. On the Experience of Time (Penguin,
1969).
160. Hicks, R. E., Miller, G. W. & Kinsbourne, M.
Prospective and retrospective judgments of time
as a function of amount of information processed.
Am. J. Psychol. 89, 719–730 (1976).
161. Block, R. A. & Zakay, D. Prospective and retrospective
duration judgments: a meta-analytic review. Psychon.
Bull. Rev. 4, 184–197 (1997).
162. Tobin, S., Bisson, N. & Grondin, S. An ecological
approach to prospective and retrospective timing of
long durations: a study involving gamers. PLoS ONE 5,
e9271 (2010).
163. Block, R. A., Hancock, P. A. & Zakay, D. How cognitive
load affects duration judgments: a meta-analytic
review. Acta Psychol. 134, 330–343 (2010).
164. Block, R. A. in Time, Action and Cognition: Towards
Bridging the Gap: 66 (NATO Science Series D, 66)
(eds. Macar, F., Pouthas, V. & Friedman, W. J.)
141–152 (Springer, 1992).
165. Polti, I., Martin, B. & van Wassenhove, V. The effect
of attention and working memory on the estimation of
elapsed time. Sci. Rep. 8, 6690 (2018).
166. Zakay, D. & Fallach, E. Immediate and remote time
estimation — a comparison. Acta Psychol. 57, 69–81
(1984).
167. Grondin, S., Laflamme, V., Bisson, N. & Désautels, F.
The delay before recall changes the remembered
duration of 15-minute video sequences. Appl. Cogn.
Psychol. 28, 677–684 (2014).
168. Pedri, S. & Hesketh, B. Time perception: effects of task
speed and delay. Percept. Mot. Skills 76, 599–608
(1993).
169. Vitulli, W. F. & Crimmins, K. A. Immediate versus
remote judgements: delay of response and rate
of stimulus presentation in time estimation.
Percept. Mot. Skills 86, 19–22 (1998).
170. Vitulli, W. F. & Shepard, H. A. Time estimation:
effects of cognitive task, presentation rate, and delay.
Percept. Mot. Skills 83, 1387–1394 (1996).
171. Loftus, E. F., Schooler, J. W., Boone, S. M. & Kline, D.
Time went by so slowly: overestimation of event
duration by males and females. Appl. Cogn. Psychol.
1, 3–13 (1987).
172. Brown, N. R. Organization of public events in
long-term memory. J. Exp. Psychol. Gen. 119,
297–314 (1990).
173. Burt, C. D. B. Reconstruction of the duration of
autobiographical events. Mem. Cogn. 20, 124–132
(1992).
174. Burt, C. D. B. & Kemp, S. Retrospective duration
estimation of public events. Mem. Cogn. 19,
252–262 (1991).
175. Gibbons, J. A. & Thompson, C. P. Using a calendar
in event dating. Appl. Cogn. Psychol. 15, 33–44
(2001).
176. Larsen, S. F. & Thompson, C. P. Reconstructive
memory in the dating of personal and public news
events. Mem. Cogn. 23, 780–790 (1995).
177. Loftus, E. F. & Marburger, W. Since the eruption
of Mt. St. Helens, has anyone beaten you up?
Improving the accuracy of retrospective reports
with landmarkevents. Mem. Cogn. 11, 114–120
(1983).
178. Yarmey, A. D. Retrospective duration estimations
for variant and invariant events in field situations.
Appl. Cogn. Psychol. 14, 45–57 (2000).
0123456789();:
179. Friedman, W. J. Time in autobiographical memory.
Soc. Cogn. 22, 591–605 (2004).
180. Shum, M. S. The role of temporal landmarks in
autobiographical memory processes. Psychol. Bull.
124, 423–442 (1998).
181. Friedman, W. J. The development of children’s
memory for the time of past events. Child. Dev. 62,
139–155 (1991).
182. Friedman, W. J. in Time, Action and Cognition: Towards
Bridging the Gap: 66 (NATO Science Series D, 66)
(eds. Macar, F., Pouthas, V. & Friedman, W. J.)
165–172 (Springer, 1992).
183. Zacks, J. M. & Tversky, B. Event structure in perception
and conception. Psychol. Bull. 127, 3–21 (2001).
184. Zakay, D., Tsal, Y., Moses, M. & Shahar, I. The role of
segmentation in prospective and retrospective time
estimation processes. Mem. Cogn. 22, 344–351
(1994).
185. Poynter, D. in Time and Human Cognition: A Life-Span
Perspective (eds. Levin, I. & Zakay, D.) 305–331
(North-Holland, 1989).
186. Poynter, W. D. Duration judgment and the segmentation
of experience. Mem. Cogn. 11, 77–82 (1983).
187. Poynter, W. D. & Homa, D. Duration judgment and
the experience of change. Percept. Psychophys. 33,
548–560 (1983).
188. Boltz, M. G. Effects of event structure on retrospective
duration judgments. Percept. Psychophys. 57,
1080–1096 (1995).
189. Predebon, J. Organization of stimulus events and
remembered apparent duration. Aust. J. Psychol. 36,
161–169 (1984).
190. Block, R. A. Temporal judgments and contextual
change. J. Exp. Psychol. Learn. Mem. Cogn. 8,
530–544 (1982).
191. Block, R. A. Remembered duration: imagery processes
and contextual encoding. Acta Psychol. 62, 103–122
(1986).
192. Faber, M. & Gennari, S. P. In search of lost time:
reconstructing the unfolding of events from memory.
Cognition 143, 193–202 (2015).
193. Block, R. A. & Reed, M. A. Remembered duration:
evidence for a contextual-change hypothesis.
J. Exp. Psychol. Hum. Learn. Mem. 4, 656–665
(1978).
194. Kellaris, J. J. & Kent, R. J. The influence of music on
consumers’ temporal perceptions: Does time fly when
you’re having fun. J. Consum. Psychol. 1, 365–376
(1992).
195. Pollatos, O., Laubrock, J. & Wittmann, M. Interoceptive
focus shapes the experience of time. PLoS ONE 9,
e86934 (2014).
196. Newtson, D. & Engquist, G. The perceptual
organization of ongoing behavior. J. Exp. Soc. Psychol.
12, 436–450 (1976).
197. Newtson, D., Engquist, G. A. & Bois, J. The objective
basis of behavior units. J. Pers. Soc. Psychol. 35,
847–862 (1977).
198. Zacks, J. M. et al. Human brain activity time-locked
to perceptual event boundaries. Nat. Neurosci. 4,
651–655 (2001).
This study provides the first demonstration
of neural activity changes related to event
segmentation.
199. Kurby, C. A. & Zacks, J. M. Segmentation in the
perception and memory of events. Trends Cogn. Sci.
12, 72–79 (2008).
200. DuBrow, S. & Davachi, L. The influence of context
boundaries on memory for the sequential order
of events. J. Exp. Psychol. Gen. 142, 1277–1286
(2013).
201. Horner, A. J., Bisby, J. A., Wang, A., Bogus, K. &
Burgess, N. The role of spatial boundaries in shaping
long-term event representations. Cognition 154,
151–164 (2016).
202. Sargent, J. Q. et al. Event segmentation ability
uniquely predicts event memory. Cognition 129,
241–255 (2013).
203. Swallow, K. M., Zacks, J. M. & Abrams, R. A. Event
boundaries in perception affect memory encoding
and updating. J. Exp. Psychol. Gen. 138, 236–257
(2009).
204. Gold, D. A., Zacks, J. M. & Flores, S. Effects of cues
to event segmentation on subsequent memory.
Cogn. Res. Princ. Implic. 2, 1 (2017).
205. Flores, S., Bailey, H. R., Eisenberg, M. L. & Zacks, J. M.
Event segmentation improves event memory up to one
month later. J. Exp. Psychol. Learn. Mem. Cogn. 43,
1183–1202 (2017).
206. Schwan, S. & Garsoffky, B. The cognitive representation
of filmic event summaries. Appl. Cogn. Psychol. 18,
37–55 (2004).
www.nature.com/nrn

207. DuBrow, S. & Davachi, L. Temporal binding within and
across events. Neurobiol. Learn. Mem. 134, 107–114
(2016).
208. Ezzyat, Y. & Davachi, L. What constitutes an episode
in episodic memory? Psychol. Sci. 22, 243–252
(2011).
209. Ezzyat, Y. & Davachi, L. Similarity breeds proximity:
pattern similarity within and across contexts is related
to later mnemonic judgments of temporal proximity.
Neuron 81, 1179–1189 (2014).
210. Hsieh, L. T., Gruber, M. J., Jenkins, L. J. &
Ranganath, C. Hippocampal activity patterns carry
information about objects in temporal context.
Neuron 81, 1165–1178 (2014).
211. Clewett, D., DuBrow, S. & Davachi, L. Transcending
time in the brain: how event memories are constructed
from experience. Hippocampus 29, 162–183
(2019).
212. Radvansky, G. A. & Zacks, J. M. Event boundaries
in memory and cognition. Curr. Opin. Behav. Sci. 17,
133–140 (2017).
213. Baldassano, C. et al. Discovering event structure in
continuous narrative perception and memory. Neuron
95, 709–721.e705 (2017).
This study demonstrates that event segmentation
occurs in a hierarchical manner across the brain,
with early sensory areas segmenting events on
short timescales and higher-order areas such as
the hippocampus segmenting events on longer
timescales which approximately correspond to
the timescale of the perceived narrative structure
of events.
214. Hasson, U., Yang, E., Vallines, I., Heeger, D. J. &
Rubin, N. A hierarchy of temporal receptive windows
in human cortex. J. Neurosci. 28, 2539 (2008).
215. Lerner, Y., Honey, C. J., Silbert, L. J. & Hasson, U.
Topographic mapping of a hierarchy of temporal
receptive windows using a narrated story. J. Neurosci.
31, 2906 (2011).
216. Zacks, J. M., Tversky, B. & Iyer, G. Perceiving,
remembering, and communicating structure in events.
J. Exp. Psychol. Gen. 130, 29–58 (2001).
217. Bower, G. H. Stimulus-sampling theory of encoding
variability. Coding Process. Hum. Mem. 3, 85–123
(1972).
218. Estes, W. K. Statistical theory of distributional
phenomena in learning. Psychol. Rev. 62, 369–377
(1955).
219. Howard, M. W. & Kahana, M. J. A distributed
representation of temporal context. J. Math. Psychol.
46, 269–299 (2002).
220. Mensink, G.-J. & Raaijmakers, J. G. A model
for interference and forgetting. Psychol. Rev. 95,
434–455 (1988).
221. DuBrow, S., Rouhani, N., Niv, Y. & Norman, K. A.
Does mental context drift or shift. Curr. Opin.
Behav. Sci. 17, 141–146 (2017).
222. Rule, M. E., O’Leary, T. & Harvey, C. D. Causes and
consequences of representational drift. Curr. Opin.
Neurobiol. 58, 141–147 (2019).
223. Jenkins, L. J. & Ranganath, C. Prefrontal and medial
temporal lobe activity at encoding predicts temporal
context memory. J. Neurosci. 30, 15558–15565
(2010).
This study provides the first link between evolution
of event trajectories and estimation of temporal
distance in humans, demonstrating that when the
neural activity patterns for different objects within
an overall sequence of objects were more distinct
from each other, estimates for when a specific
object was shown were more accurate.
224. Nielson, D. M., Smith, T. A., Sreekumar, V., Dennis, S.
& Sederberg, P. B. Human hippocampus represents
space and time during retrieval of real-world memories.
Proc. Natl Acad. Sci. USA 112, 11078 (2015).
225. Lositsky, O. et al. Neural pattern change during
encoding of a narrative predicts retrospective duration
estimates. Elife 5, e16070 (2016).
This study experimentally demonstrates that
changes in neural activity patterns measured using
functional MRI — particularly in the entorhinal
cortex — are correlated with retrospective duration
estimates.
226. Block, R. A. & Zakay, D. in Time and Mind
(ed. Helfrich, H.) Ch. 9, 171–195 (Hogrefe & Huber,
1996).
227. Fountas, Z. et al. A predictive processing model of
episodic memory and time perception. Neural Comput.
34, 1501–1544 (2022).
228. Roseboom, W. et al. Activity in perceptual classification
networks as a basis for human subjective time
perception. Nat. Commun. 10, 267 (2019).
NATuRe RevIeWS | Neuroscience
229. Clayton, N. S. & Dickinson, A. Episodic-like memory
during cache recovery by scrub jays. Nature 395,
272–274 (1998).
230. Fetterman, J. G. & Killeen, P. R. Prospective and
retrospective timing by pigeons. Learn. Behav. 38,
119–125 (2010).
231. Babb, S. J. & Crystal, J. D. Episodic-like memory in the
rat. Curr. Biol. 16, 1317–1321 (2006).
232. Roberts, W. A. & Feeney, M. C. The comparative study
of mental time travel. Trends Cogn. Sci. 13, 271–277
(2009).
233. Wang, L. et al. Fallacious reversal of event-order
during recall reveals memory reconstruction in
rhesus monkeys. Behav. Brain Res. 394, 112830
(2020).
234. Roberts, W. A. et al. Episodic-like memory in rats:
is it based on when or how long ago? Science 320,
113 (2008).
235. Zhou, W. & Crystal, J. D. Evidence for remembering
when events occurred in a rodent model of episodic
memory. Proc. Natl Acad. Sci. USA 106, 9525
(2009).
236. Diehl, G. W., Hon, O. J., Leutgeb, S. & Leutgeb, J. K.
Stability of medial entorhinal cortex representations
over time. Hippocampus 29, 284–302 (2019).
237. Mankin, E. A., Diehl, G. W., Sparks, F. T., Leutgeb, S.
& Leutgeb, J. K. Hippocampal CA2 activity patterns
change over time to a larger extent than between
spatial contexts. Neuron 85, 190–201 (2015).
238. Mankin, E. A. et al. Neuronal code for extended time
in the hippocampus. Proc. Natl Acad. Sci. USA 109,
19462 (2012).
This study demonstrates that the evolution of event
trajectories can occur over the timescale of days
and at different rates within the hippocampal
subfields (CA1 and CA3).
239. Mau, W. et al. The same hippocampal CA1 population
simultaneously codes temporal information over
multiple timescales. Curr. Biol. 28, 1499–1508.e1494
(2018).
240. Rangel, L. M. et al. Temporally selective contextual
encoding in the dentate gyrus of the hippocampus.
Nat. Commun. 5, 3181 (2014).
241. Tsao, A. et al. Integrating time from experience in the
lateral entorhinal cortex. Nature 561, 57–62 (2018).
This study demonstrates that event trajectories in
the LEC evolve in a manner reflecting the content
of ongoing experience, including learned temporal
structures, and suggests that the LEC may play a
central role in the generation of event trajectories.
242. Eichenbaum, H. On the integration of space, time, and
memory. Neuron 95, 1007–1018 (2017).
243. Moser, M.-B., Rowland, D. C. & Moser, E. I. Place cells,
grid cells, and memory. Cold Spring Harb. Perspect. Biol.
7, a021808 (2015).
244. Buzsáki, G. & Moser, E. I. Memory, navigation and
theta rhythm in the hippocampal-entorhinal system.
Nat. Neurosci. 16, 130–138 (2013).
245. Josselyn, S. A. & Tonegawa, S. Memory engrams:
recalling the past and imagining the future. Science
367, eaaw4325 (2020).
246. Jenkins, L. J. & Ranganath, C. Distinct neural
mechanisms for remembering when an event
occurred. Hippocampus 26, 554–559 (2016).
247. Montchal, M. E., Reagh, Z. M. & Yassa, M. A. Precise
temporal memories are supported by the lateral
entorhinal cortex in humans. Nat. Neurosci. 22,
284–288 (2019).
248. Hainmueller, T. & Bartos, M. Parallel emergence
of stable and dynamic memory engrams in the
hippocampus. Nature 558, 292–296 (2018).
249. Hyman, J. M., Ma, L., Balaguer-Ballester, E.,
Durstewitz, D. & Seamans, J. K. Contextual encoding
by ensembles of medial prefrontal cortex neurons.
Proc. Natl Acad. Sci. USA 109, 5086 (2012).
250. Ziv, Y. et al. Long-term dynamics of CA1 hippocampal
place codes. Nat. Neurosci. 16, 264–266 (2013).
251. Cappaert, N. L. M., van Strien, N. M. & Witter, M. P.
Hippocampal formation. in The Rat Nervous System,
511–573 (Academic Press, 2015).
252. Speer, N. K., Zacks, J. M. & Reynolds, J. R. Human
brain activity time-locked to narrative event
boundaries. Psychol. Sci. 18, 449–455 (2007).
253. Ben-Yakov, A., Eshel, N. & Dudai, Y. Hippocampal
immediate poststimulus activity in the encoding of
consecutive naturalistic episodes. J. Exp. Psychol. Gen.
142, 1255–1263 (2013).
254. Magliano, J. P. & Zacks, J. M. The impact of continuity
editing in narrative film on event segmentation.
Cogn. Sci. 35, 1489–1517 (2011).
255. Bulkin, D. A., Sinclair, D. G., Law, L. M. & Smith, D. M.
Hippocampal state transitions at the boundaries
0123456789();:
Reviews
between trial epochs. Hippocampus 30, 582–595
(2020).
256. Zheng, J. et al. Neurons detect cognitive boundaries
to structure episodic memories in humans.
Nat. Neurosci. 25, 358–368 (2022).
This study demonstrates in humans that single
cells can encode event boundaries, that the
evolution of event trajectories is influenced by
event boundaries, and that both the activity of
cells encoding event boundaries and the dynamics
of event trajectories are each correlated with
discrimination of temporal order.
257. Bladon, J. H., Sheehan, D. J., De Freitas, C. S. &
Howard, M. W. In a temporally segmented experience
hippocampal neurons represent temporally drifting
context but not discrete segments. J. Neurosci. 39,
6936 (2019).
258. Sun, C., Yang, W., Martin, J. & Tonegawa, S.
Hippocampal neurons represent events as
transferable units of experience. Nat. Neurosci. 23,
651–663 (2020).
259. Reynolds, J. R., Zacks, J. M. & Braver, T. S.
A computational model of event segmentation from
perceptual prediction. Cogn. Sci. 31, 613–643
(2007).
260. Zacks, J. M., Speer, N. K., Swallow, K. M., Braver, T. S.
& Reynolds, J. R. Event perception: a mind-brain
perspective. Psychol. Bull. 133, 273–293 (2007).
261. Schapiro, A. C., Rogers, T. T., Cordova, N. I.,
Turk-Browne, N. B. & Botvinick, M. M. Neural
representations of events arise from temporal
community structure. Nat. Neurosci. 16, 486–492
(2013).
262. Clewett, D. & Davachi, L. The ebb and flow of
experience determines the temporal structure
of memory. Curr. Opin. Behav. Sci. 17, 186–193
(2017).
263. Hasson, U., Chen, J. & Honey, C. J. Hierarchical
process memory: memory as an integral component
of information processing. Trends Cogn. Sci. 19,
304–313 (2015).
264. Knierim, J. J., Neunuebel, J. P. & Deshmukh, S. S.
Functional correlates of the lateral and medial
entorhinal cortex: objects, path integration and
local-global reference frames. Philos. Trans. R. Soc.
Lond. B Biol. Sci. 369, 20130369 (2013).
265. Bota, M., Sporns, O. & Swanson, L. W. Architecture
of the cerebral cortical association connectome
underlying cognition. Proc. Natl Acad. Sci. Usa. 112,
E2093–E2101 (2015).
266. Burwell, R. D. & Amaral, D. G. Cortical afferents of
the perirhinal, postrhinal, and entorhinal cortices
of the rat. J. Comp. Neurol. 398, 179–205 (1998).
267. Doan, T. P., Lagartos-Donate, M. J., Nilssen, E. S.,
Ohara, S. & Witter, M. P. Convergent projections
from perirhinal and postrhinal cortices suggest
a multisensory nature of lateral, but not medial,
entorhinal cortex. Cell Rep. 29, 617–627.e617 (2019).
268. Zingg, B. et al. Neural networks of the mouse
neocortex. Cell 156, 1096–1111 (2014).
269. Burwell, R. D. The parahippocampal region:
corticocortical connectivity. Ann. N. Y. Acad. Sci. 911,
25–42 (2000).
270. Nilssen, E. S., Doan, T. P., Nigro, M. J., Ohara, S. &
Witter, M. P. Neurons and networks in the entorhinal
cortex: A reappraisal of the lateral and medial
entorhinal subdivisions mediating parallel cortical
pathways. Hippocampus 29, 1238–1254 (2019).
271. Deshmukh, S. S. & Knierim, J. J. Representation
of non-spatial and spatial information in the lateral
entorhinal cortex. Front. Behav. Neurosci. 5, 69–69
(2011).
272. Hargreaves, E. L., Rao, G., Lee, I. & Knierim, J. J.
Major dissociation between medial and lateral
entorhinal input to dorsal hippocampus. Science 308,
1792–1794 (2005).
273. Igarashi, K. M., Lu, L., Colgin, L. L., Moser, M.-B. &
Moser, E. I. Coordination of entorhinal–hippocampal
ensemble activity during associative learning. Nature
510, 143–147 (2014).
274. Keene, C. S. et al. Complementary functional
organization of neuronal activity patterns in the
perirhinal, lateral entorhinal, and medial entorhinal
cortices. J. Neurosci. 36, 3660–3675 (2016).
275. Leitner, F. C. et al. Spatially segregated feedforward
and feedback neurons support differential
odor processing in the lateral entorhinal cortex.
Nat. Neurosci. 19, 935–944 (2016).
276. Pilkiw, M. et al. Phasic and tonic neuron ensemble
codes for stimulus-environment conjunctions in
the lateral entorhinal cortex. Elife https://doi.org/
10.7554/eLife.28611 (2017).
Reviews
277. Tsao, A., Moser, M. B. & Moser, E. I. Traces of
experience in the lateral entorhinal cortex. Curr. Biol.
23, 399–405 (2013).
278. Yoganarasimha, D., Rao, G. & Knierim, J. J. Lateral
entorhinal neurons are not spatially selective in
cue-rich environments. Hippocampus 21, 1363–1374
(2011).
279. Bitzenhofer, S. H., Westeinde, E. A., Zhang, H. B.
& Isaacson, J. S. Rapid odor processing by layer 2
subcircuits in lateral entorhinal cortex. Elife
https://doi.org/10.7554/eLife.75065 (2022).
280. Lee, J. Y. et al. Dopamine facilitates associative
memory encoding in the entorhinal cortex. Nature
598, 321–326 (2021).
281. Suter, E. E., Weiss, C. & Disterhoft, J. F. Differential
responsivity of neurons in perirhinal cortex,
lateral entorhinal cortex, and dentate gyrus during
time-bridging learning. Hippocampus 29, 511–526
(2019).
282. Wang, C. et al. Egocentric coding of external items in
the lateral entorhinal cortex. Science 362, 945 (2018).
283. Knierim, J. J. & Neunuebel, J. P. Tracking the flow
of hippocampal computation: pattern separation,
pattern completion, and attractor dynamics.
Neurobiol. Learn. Mem. 129, 38–49 (2016).
284. Ben-Yakov, A. & Henson, R. N. The hippocampal
film editor: sensitivity and specificity to event
boundaries in continuous experience. J. Neurosci.
38, 10057–10068 (2018).
285. Sols, I., DuBrow, S., Davachi, L. & Fuentemilla, L.
Event boundaries trigger rapid memory reinstatement
of the prior events to promote their representation in
long-term memory. Curr. Biol. 27, 3499–3504.e3494
(2017).
286. Silva, M., Baldassano, C. & Fuentemilla, L. Rapid
memory reactivation at movie event boundaries
promotes episodic encoding. J. Neurosci. 39, 8538
(2019).
287. Tambini, A. & Davachi, L. Awake reactivation of prior
experiences consolidates memories and biases
cognition. Trends Cogn. Sci. 23, 876–890 (2019).
288. Carr, M. F., Jadhav, S. P. & Frank, L. M. Hippocampal
replay in the awake state: a potential substrate for
memory consolidation and retrieval. Nat. Neurosci.
14, 147–153 (2011).
289. Folkerts, S., Rutishauser, U. & Howard, M. W.
Human episodic memory retrieval is accompanied by
a neural contiguity effect. J. Neurosci. 38, 4200–4211
(2018).
290. Gelbard-Sagiv, H., Mukamel, R., Harel, M., Malach, R.
& Fried, I. Internally generated reactivation of single
neurons in human hippocampus during free recall.
Science 322, 96 (2008).
291. Howard, M. W., Viskontas, I. V., Shankar, K. H. &
Fried, I. Ensembles of human MTL neurons “jump
back in time” in response to a repeated stimulus.
Hippocampus 22, 1833–1847 (2012).
292. Manning, J. R., Polyn, S. M., Baltuch, G. H., Litt, B.
& Kahana, M. J. Oscillatory patterns in temporal lobe
reveal context reinstatement during memory search.
Proc. Natl Acad. Sci. USA 108, 12893 (2011).
293. Miller, J. F. et al. Neural activity in human
hippocampal formation reveals the spatial context
of retrieved memories. Science 342, 1111 (2013).
294. Rubin, A., Geva, N., Sheintuch, L. & Ziv, Y.
Hippocampal ensemble dynamics timestamp events
in long-term memory. Elife 4, e12247 (2015).
295. Block, R. Models of psychological time. in Cognitive
Models of Psychological Time (Ed. Block, R. A.) 1–35
(Lawrence Erlbaum Associates, 1990).
296. Howard, M. W. in The New Handbook of Mathematical
Psychology (eds. Ashby,F. G., Colonius,H. & Dzhafarov, E.)
Vol. 3 (Cambridge Univ. Press, in the press).
297. Howard, M. W., Shankar, K. H., Aue, W. R. &
Criss, A. H. A distributed representation of internal
time. Psychol. Rev. 122, 24–53 (2015).
298. Shankar, K. H. & Howard, M. W. A scale-invariant
internal representation of time. Neural Comput. 24,
134–193 (2012).
299. Howard, M. W. et al. A unified mathematical
framework for coding time, space, and sequences in
the hippocampal region. J. Neurosci. 34, 4692–4707
(2014).
This study describes a biologically plausible
computational model which in principle is capable
of carrying out retrospective timing.
300. Engel, T. A. & Wang, X.-J. Same or different? A neural
circuit mechanism of similarity-based pattern match
decision making. J. Neurosci. 31, 6982 (2011).
301. Bright, I. M. et al. A temporal record of the past with
a spectrum of time constants in the monkey entorhinal
cortex. Proc. Natl Acad. Sci. USA 117, 20274 (2020).
302. Thavabalasingam, S., O’Neil, E. B. & Lee, A. C. H.
Multivoxel pattern similarity suggests the integration
of temporal duration in hippocampal event sequence
representations. Neuroimage 178, 136–146 (2018).
303. Thavabalasingam, S., O’Neil, E. B., Tay, J., Nestor, A.
& Lee, A. C. H. Evidence for the incorporation of
temporal duration information in human hippocampal
long-term memory sequence representations.
Proc. Natl Acad. Sci. USA 116, 6407–6414 (2019).
304. Foudil, S.-A., Kwok, S. C. & Macaluso, E.
Context-dependent coding of temporal distance
between cinematic events in the human precuneus.
J. Neurosci. 40, 2129 (2020).
305. Kwok, S. C., Shallice, T. & Macaluso, E. Functional
anatomy of temporal organisation and domain-specificity
of episodic memory retrieval. Neuropsychologia 50,
2943–2955 (2012).
306. Cohn-Sheehy, B. I. et al. The hippocampus constructs
narrative memories across distant events. Curr. Biol.
31, 4935–4945.e4937 (2021).
307. Azizi, L., Polti, I. & van Wassenhove, V. Episodic timing:
how spontaneous alpha clocks, retrospectively. bioRxiv
https://doi.org/10.1101/2021.10.01.462732 (2021).
308. Cohn-Sheehy, B. I. & Ranganath, C. Time regained:
how the human brain constructs memory for time.
Curr. Opin. Behav. Sci. 17, 169–177 (2017).
309. Davachi, L. & DuBrow, S. How the hippocampus
preserves order: the role of prediction and context.
Trends Cogn. Sci. 19, 92–99 (2015).
310. Ranganath, C. & Hsieh, L.-T. The hippocampus:
a special place for time. Ann. N. Y. Acad. Sci. 1369,
93–110 (2016).
311. Tulving, E. Elements of Episodic Memory (Oxford
University Press, 1983).
312. DuBrow, S. & Davachi, L. Temporal memory is
shaped by encoding stability and intervening item
reactivation. J. Neurosci. 34, 13998 (2014).
313. Hsieh, L.-T. & Ranganath, C. Cortical and subcortical
contributions to sequence retrieval: Schematic coding
of temporal context in the neocortical recollection
network. NeuroImage 121, 78–90 (2015).
314. Manns, J. R., Howard, M. W. & Eichenbaum, H.
Gradual changes in hippocampal activity support
remembering the order of events. Neuron 56,
530–540 (2007).
This study provides the first experimental
observation of evolving event trajectories in the
hippocampus, and demonstrates a correlation
between the rate of drift and accuracy in
determining temporal order of events.
315. Cox, B. M. et al. Acquisition of temporal order requires
an intact CA3 commissural/associational (C/A) feedback
system in mice. Commun. Biol. 2, 251 (2019).
316. Aronov, D., Nevers, R. & Tank, D. W. Mapping of a
non-spatial dimension by the hippocampal–entorhinal
circuit. Nature 543, 719–722 (2017).
317. Terada, S., Sakurai, Y., Nakahara, H. & Fujisawa, S.
Temporal and rate coding for discrete event sequences
in the hippocampus. Neuron 94, 1248–1262.e1244
(2017).
318. Allen, T. A., Salz, D. M., McKenzie, S. & Fortin, N. J.
Nonspatial sequence coding in CA1 neurons.
J. Neurosci. 36, 1547–1563 (2016).
319. Shahbaba, B. et al. Hippocampal ensembles represent
sequential relationships among an extended sequence
of nonspatial events. Nat. Commun. 13, 787 (2022).
320. Schuck, N. W. & Niv, Y. Sequential replay of nonspatial
task states in the human hippocampus. Science 364,
eaaw5181 (2019).
321. Coull, J. T. & Droit-Volet, S. Explicit understanding
of duration develops implicitly through action.
Trends Cogn. Sci. 22, 923–937 (2018).
322. Faber, M. & Gennari, S. P. Effects of learned episodic
event structure on prospective duration judgments.
J. Exp. Psychol. Learn. Mem. Cogn. 43, 1203 (2017).
323. Bellmund, J. L., Deuker, L. & Doeller, C. F. Mapping
sequence structure in the human lateral entorhinal
cortex. Elife 8, e45333 (2019).
324. Zauberman, G., Levav, J., Diehl, K. & Bhargave, R.
1995 feels so close yet so far: the effect of event
0123456789();:
markers on subjective feelings of elapsed time.
Psychol. Sci. 21, 133–139 (2010).
325. Baldassano, C., Hasson, U. & Norman, K. A.
Representation of real-world event schemas during
narrative perception. J. Neurosci. https://doi.org/
10.1523/JNEUROSCI.0251-18.2018 (2018).
326. Morton, N. W., Schlichting, M. L. & Preston, A. R.
Representations of common event structure in medial
temporal lobe and frontoparietal cortex support
efficient inference. Proc. Natl Acad. Sci. USA
https://doi.org/10.1073/pnas.1912338117 (2020).
327. Pudhiyidath, A., Roome, H. E., Coughlin, C., Nguyen, K. V.
& Preston, A. R. Developmental differences in
temporal schema acquisition impact reasoning
decisions. Cogn. Neuropsychol. 37, 25–45 (2020).
328. Roach, N. W., McGraw, P. V., Whitaker, D. J. &
Heron, J. Generalization of prior information for rapid
Bayesian time estimation. Proc. Natl Acad. Sci. USA
114, 412 (2017).
329. Bellmund, J. L. S., Deuker, L., Montijn, N. D. &
Doeller, C. F. Structuring time: The hippocampus
constructs sequence memories that generalize
temporal relations across experiences. bioRxiv
https://doi.org/10.1101/2021.04.23.440002 (2021).
330. Gauthier, B., Pestke, K. & van Wassenhove, V. Building
the arrow of time… over time: a sequence of brain
activity mapping imagined events in time and space.
Cereb. Cortex 29, 4398–4414 (2019).
331. Gauthier, B., Prabhu, P., Kotegar, K. A. &
van Wassenhove, V. Hippocampal contribution
to ordinal psychological time in the human brain.
J. Cogn. Neurosci. 32, 2071–2086 (2020).
332. Gauthier, B. & van Wassenhove, V. Cognitive mapping
in mental time travel and mental space navigation.
Cognition 154, 55–68 (2016).
333. Lee, C. S., Aly, M. & Baldassano, C. Anticipation
of temporally structured events in the brain. Elife
https://doi.org/10.7554/eLife.64972 (2021).
334. Block, R. A., Grondin, S. & Zakay, D. in Timing
and Time Perception: Procedures, Measures &
Applications (eds Vatakis, A., Balcı, F., Di Luca, M.
& Correa, Á.) 32–51 (Brill, 2018).
335. Block, R. A. Memory and the experience of duration
in retrospect. Mem. Cogn. 2, 153–160 (1974).
Acknowledgements
The authors dedicate this Review to the memory of Warren H.
Meck. Sadly, Warren passed away in 2020 before they could
complete the writing of this Review. However, it was first pro-
posed by him as an opportunity to bring together two fields
of time research which have long been isolated from each
other, and the authors are honoured and grateful to continue
this work in his memory. The authors thank V. van
Wassenhove for discussion and comments on the manuscript.
The work was supported by a Synergy Grant from the
European Research Council to E.I.M. (‘KILONEURONS’, grant
agreement number 951319), an RCN FRIPRO grant to E.I.M.
(grant number 286225), a Centre of Excellence scheme grant
to M.-B.M. and E.I.M. from the Research Council of Norway
(Centre for Neural Computation, grant number 223262), the
Kavli Foundation (M.-B.M. and E.I.M.), and a direct contribu-
tion to M.-B.M. and E.I.M. from the Ministry of Education and
Research of Norway.
Author contributions
All authors contributed to the manuscript.
Competing interests
The authors declare no competing financial interests.
Peer review information
Nature Reviews Neuroscience thanks Y.-J. Lin and the other
anonymous reviewers for their contribution to the peer review
of this work.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
Springer Nature or its licensor holds exclusive rights to this
article under a publishing agreement with the author(s) or
other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the
terms of such publishing agreement and applicable law.
© Springer Nature Limited 2022
www.nature.com/nrn