Matthew L. Shapiro1,*
1Department of Neuroscience, Mount Sinai School of Medicine, New York, NY 10029, USA
In this issue of Neuron, MacDonald et al. describe hippocampal ‘‘time cells’’ that fire during specific delay
periods as rats performed a memory task. Converging results in monkeys suggest that the hippocampus
encodes episodes by signaling events in time.
We remember the events of our lives as
episodes framed by space and time.
Such memories require structures in the
medial temporal lobe (MTL), especially
damage are amnesic, ‘‘lost in time,’’ and
unable to recall the recent past or imagine
the future. Early efforts to model human
amnesia and analyze MTL function in
animals led to the discovery of hippo-
campal place cells and the theory that
the hippocampus supports memory by
constructing cognitive maps that define
spatial contexts (O’Keefe and Nadel,
ronment, ?40% of CA1 cells fire signifi-
cantly in particular locations, and the
same place fields can be recorded for
weeks. Different sets of cells are active
in different environments, and sufficient
changes to stimuli or behavior within an
environment can cause similar ‘‘remap-
ping’’ (Leutgeb et al., 2004). Each set of
place fields is thought to represent a
different spatial context, and amnesia
after hippocampal damage is explained
as an impairment in the representation of
spatial context. Nevertheless it has been
less clear how the hippocampus helps
represent temporal context, i.e., different
events occurring in the same place at
In this issue of Neuron, MacDonald
et al. (2011) describe the activity of dorsal
CA1 pyramidal cells as rats performed
an object-odor delayed association task
in a modified T-maze. In each trial, the
rat was placed in a starting area, pre-
sented briefly with one of two objects,
and allowed to enter a waiting area for
a 6–10 s delay, after which it approached
a scented, sand-filled flowerpot. Each
object-odor pair was associated with
a different response. In ‘‘go’’ trials, the
reward was obtained by digging in the
flowerpot; in ‘‘no-go’’ trials, the reward
could be found in a different place by
not digging. To obtain reward, the rat
had to remember which object had been
presented before the delay (Figure 1A).
Neuronal activity during object presenta-
tion, the delay, and odor presentation
was analyzed with a general linear model
that quantified the extent to which these
variables, together with location, head
direction, movement speed, and time
predicted firing rate.
Consistent with previous reports, the
activity of different neurons was modu-
lated by different task parameters. Thus
many cells had place fields; ?30% of
the neurons distinguished between the
objects, the odors, and the response or
had conjunctive properties, e.g., firing
most when a specific odor was sampled
after a particular object. The authors
discovered that CA1 activity changed
in time so that different populations of
neurons were maximally active through-
out the delay (Figure 1B). One hundred
sixty-seven of the three hundred thirty-
three recorded neurons that were active
during the delay fired in specific periods,
as though the hippocampus coded the
passage of time in an otherwise static
environment. Furthermore, the firing pat-
terns changed smoothly in time, so that
population activity recorded during con-
tiguous intervals was similar, and be-
came more distinct at greater intervals.
Similar patterns of temporal coding were
observed in each of the trial epochs,
showing that the hippocampal code
distinct sequences that linked the object
through the delay and odor presentation
to the response at the end of each trial.
One potential caveat is that hippo-
campal neurons are sensitive to spatial
behavior, especially location, heading di-
rection, and movement speed. If behavior
is stereotyped across trials, then these
variables could masquerade as time cells.
The general linear model quantified the
relative influence of each and found that
most of the ‘‘time cells’’ active during
the delay fired in place fields. However,
the timing signal did not simply reflect
sequentially occupied locations. Rather,
activity within the place fields varied with
the delay, so that while the rat occupied
a given place, a cell could be silent ex-
cept during a particular moment during
the delay. Therefore, the hippocampus
coded both place and time, and 73% of
the cells’ activity was best predicted by
both (Figures 1C–1E). Perhaps the hippo-
campus maps a Minkowski space, in
which all coordinates specify space and
These results add to the growing
evidence that MTL neurons distinguish
temporal context. For example, CA1 and
entorhinal cell activity varies during iden-
tical spatial trajectories depending upon
past or future actions (Ferbinteanu and
Shapiro, 2003;Frank et al., 2000;Smith
and Mizumori, 2006;Wood et al., 2000).
Hippocampal activity changes during the
delay in spatial nonmatching to sample
tasks (Deadwyler et al., 1996), and these
dynamics occur as animals occupy the
same location during the delay (Pastal-
kova et al., 2008). Further, during delayed
eyelid conditioning, hippocampal units
model the acquisition and timing of condi-
tioned responses (Berger et al., 1976).
Moreover, hippocampal neurons fire in
spatiotemporal sequences that reflect
past or future trajectories (place field
‘‘replay and preplay’’) during sharp wave
ripples recorded before or after active
exploration (Davidson et al., 2009;Diba
and Buzsaki, 2007). Indeed, sequential
Neuron 71, August 25, 2011 ª2011 Elsevier Inc.
action potentials recorded at the choice
point of a maze can anticipate the se-
quence of place fields to be occupied
after pending choices (Johnson and Re-
Next, MacDonald et al. (2011) doubled
or tripled the duration of the delay to test
if the cells coded absolute time or inter-
vals relative to the task features. In one
scenario, if the hippocampus codes ab-
solute time, then neuronal activity should
be identical during the initial and familiar
start of the delay (e.g., the first 5 s) and
evolve new codes as the delay is pro-
longed. If the hippocampus codes time
relative to task features, however, then
the same sequential order of activity
should be maintained, corresponding to
the beginning, middle, and end of the
delay, independent of its physical dura-
tion. The authors observed both patterns,
with different cells coding either absolute
or relative time. Nearly 40% of the
neurons fired at the same absolute time
from the start of the trial during different
delay intervals, and a few appeared to
code relative time by scaling, as the
activity was either expanded or com-
pressed. Some cells showed retrospec-
interval after the start of the delay. Other
cells may have been prospective, firing
near the end of the delay as though antic-
ipating the imminent decision. Most cells,
however, ‘‘re-timed’’ and developed new
changed. Just as sufficient changes to
the environment or contingencies cause
place field remapping, altering the delay
items changed time fields. Moreover, the
population as a whole showed ‘‘partial re-
timing.’’ Partial remapping occurs when
subsets of familiar cues are rearranged:
subpopulations of active cells maintain
thesame place fields
develop new ones. Partial remapping
suggests that the hippocampal popula-
tion integrates new information in relation
to prior experience, with the partial over-
lap in activity providing potential links
between new and familiar items. Partial
retiming suggests that the hippocampus
may code the new delay in relation
to the familiar one (Figures 1E and 1F).
Together, the results imply that the hippo-
campus codes event sequences that
link one item to another through space
and time. Even when the outside world
appears static, time and hippocampal
representations continue to evolve.
A new study by Naya and Suzuki (2011)
reports that time is a key feature of hip-
pocampal coding in behaving monkeys.
By recording neuronal activity in four
interconnected MTL regions, the re-
search team used a powerful experi-
mental design to analyze the different
contributions of MTL regions to memory.
As in the study by MacDonald et al.
(2011), animals were trained to perform
a sequence memory task. The monkey
was shown one visual cue and then
another separated by a brief delay; after
another delay, an array of three stimuli
that included the two shown previously
on that trial was presented. The monkey
had to touch the two stimuli in the same
order in which they were previously
presented in the trial to get a reward.
Naya and Suzuki (2011) found that each
MTL region discriminated different task
Figure 1. Coding What, Where, and When
(A) Delayed object-odor association rules. Rats learned to associate objects (a wood block and a green
rubber ball) with odors (cinnamon and basil). In ‘‘GO’’ trials, the rat could dig up food buried in sand in
delay area, and flowerpot were identical in every trial.
(B) Time coding. Different CA1 neurons fired maximally during different temporal periods. The spectrum
depicts CA1 activity dynamics recorded in the sequence task, with each color indicating an active subset
of neurons. Population activity changes smoothly through time.
(C–E) Possible place, time, or episode maps of delay activity in groups of CA1 neurons.
(C) Cells code places independent of time.
(D) Cells code delay periods independent of location.
(E) Cells code episodes in place and time. The actual sequence of neural activity varied across trial types,
and included information about the object, place, delay, and response.
(F) Changing the delay caused ‘‘partial re-timing’’: some cells maintain activity at specific physical times
whereasotherssignaldifferenttaskepochs.Color spectra illustratepopulationactivitychanging smoothly
in space and time.
(G) ‘‘What’’ and ‘‘When’’ is coded differently across the MTL. Hippocampal (‘‘H’’) activity changed
smoothly across delays but did not distinguish visual stimuli. TE cells had the opposite coding properties,
distinguishing cues but not time. Perirhinal (‘‘PR’’) and entorhinal (‘‘ER’’) cells responded to both task
features, as though integrating the ‘‘what’’ and ‘‘when’’ signals from the other regions.
Neuron 71, August 25, 2011 ª2011 Elsevier Inc.
features, as if coding different types of
abstract representations. Most hippo-
campal neurons (88%) distinguished the
order of events, e.g., firing most during
the delay after the first cue was removed
and continuing during the presentation
of the second cue, or vice versa. As in
the study by MacDonald et al. (2011),
the activity of hippocampal neurons
changed gradually during the delay, so
that population activity recorded during
contiguous intervals was similar and be-
came more distinct at greater intervals
(Manns et al., 2007). Few hippocampal
neurons signaled unique stimulus items.
In stark contrast, most TE neurons (94%)
encoded the cues, but not presentation
order or time. Subpopulations of entorhi-
nal and perirhinal cortical neurons sig-
naled both item and time in different
ways. Entorhinal activity patterns shifted
gradually away from the response to the
first cue during the delay, but responded
abruptly to presentation of the second
cue, as though the initial representation
was sensitive to or fading in time. Entorhi-
nal cells also showed a strong interaction
between the items and their presentation
order, distinguishing items during the first
or the second cue period, but not both.
Perirhinal neurons encoded items more
strongly, but their responses to preferred
items were modulated by presentation
order. The results suggest a mechanism
by which different MTL structures code
that together represent the content and
sequential flow of episodes (Figure 1G).
Hence, the identity of viewed objects
appears to be coded by TE neurons inde-
pendent of time and place, whereas
events defined in time and place are
coded by the hippocampus independent
of object identity. These signals are
combined in the entorhinal cortex to
represent subsets of objects in sequence
and in the perirhinal cortex to distinguish
the behavioral context in which identical
The two studies agree that hippo-
campal representations evolve in time
independent of other external variables
ing history of experience. These results
break new ground and raise fundamental
questions. What mechanisms drive time
cells? Computational models suggest
that instantaneous activity in the hippo-
campus is determined in part by its prior
Both CA3 and the dentate gyrus include
powerful recurrent connections thatcould
maintain similar activity patterns during
contiguous intervals yet drive continuous
shifts in activity as time proceeds. When
and why are hippocampal neurons sensi-
tive to discriminative stimuli? Naya and
Suzuki (2011) found that most hippo-
campal cells coded time, but very few
discriminated the visual cues. Perhaps
the monkeys were so familiar with the
sequences that the hippocampus repre-
sented the stimuli only as steps in a
visual cues could be incorporated to
disrupt the expected routines and engage
hippocampal processing to encode the
new cues as distinct episodes in memory.
campus should be prominent during
probe tests. Finally, are hippocampal
time fields needed for event memory?
The ‘‘retiming’’ described by MacDonald
et al. (2011) suggest that the hippo-
campus is not merely counting time, but
includes duration and temporal contiguity
among task epochs as an intrinsic coding
feature. Nonetheless, memory perfor-
mance did not require memory for time,
and time codes did not predict perfor-
mance levels. Future recording experi-
ments that require animals to compare
different durations are needed to test
whether time fields contribute to memory
Berger, T.W., Alger, B., and Thompson, R.F.
(1976). Science 192, 483–485.
Davidson, T.J., Kloosterman, F., and Wilson, M.A.
(2009). Neuron 63, 497–507.
Deadwyler, S.A., Bunn, T., and Hampson, R.E.
(1996). J. Neurosci. 16, 354–372.
Diba, K., and Buzsaki, G. (2007). Nat. Neurosci. 10,
Ferbinteanu, J., and Shapiro, M.L. (2003). Neuron
Frank, L.M., Brown, E.N., and Wilson, M. (2000).
Neuron 27, 169–178.
Howard, M.W., and Kahana, M.J. (2002). J. Math.
Psychol. 46, 269–299.
Johnson, A., and Redish, A.D. (2007). J. Neurosci.
Leutgeb, S., Leutgeb, J.K., Treves, A., Moser,
M.B., and Moser, E.I. (2004). Science 305, 1295–
MacDonald, C.J., Lepage, K.Q., Eden, U.T., and
Eichenbaum, H. (2011). Neuron 71, this issue,
Manns, J.R., Howard, M.W., and Eichenbaum, H.
(2007). Neuron 56, 530–540.
Naya, Y., and Suzuki, W.A. (2011). Science 333,
O’Keefe, J., and Nadel, L. (1978). (Oxford: Oxford
Pastalkova, E., Itskov, V., Amarasingham, A., and
Buzsaki, G. (2008). Science 321, 1322–1327.
Wood, E.R., Dudchenko, P.A., Robitsek, R.J., and
Eichenbaum, H. (2000). Neuron 27, 623–633.
Neuron 71, August 25, 2011 ª2011 Elsevier Inc.