Coordinated memory replay in the visual cortex and hippocampus during sleep

Abstract

Sleep replay of awake experience in the cortex and hippocampus has been proposed to be involved in memory consolidation. However, whether temporally structured replay occurs in the cortex and whether the replay events in the two areas are related are unknown. Here we studied multicell spiking patterns in both the visual cortex and hippocampus during slow-wave sleep in rats. We found that spiking patterns not only in the cortex but also in the hippocampus were organized into frames, defined as periods of stepwise increase in neuronal population activity. The multicell firing sequences evoked by awake experience were replayed during these frames in both regions. Furthermore, replay events in the sensory cortex and hippocampus were coordinated to reflect the same experience. These results imply simultaneous reactivation of coherent memory traces in the cortex and hippocampus during sleep that may contribute to or reflect the result of the memory consolidation process.

Full text

Coordinated memory replay in the visual cortex and
hippocampus during sleep
Daoyun Ji & Matthew A Wilson
Sleep replay of awake experience in the cortex and hippocampus has been proposed to be involved in memory consolidation.
However, whether temporally structured replay occurs in the cortex and whether the replay events in the two areas are related are
unknown. Here we studied multicell spiking patterns in both the visual cortex and hippocampus during slow-wave sleep in rats.
We found that spiking patterns not only in the cortex but also in the hippocampus were organized into frames, defined as periods
of stepwise increase in neuronal population activity. The multicell firing sequences evoked by awake experience were replayed
during these frames in both regions. Furthermore, replay events in the sensory cortex and hippocampus were coordinated to reflect
the same experience. These results imply simultaneous reactivation of coherent memory traces in the cortex and hippocampus
during sleep that may contribute to or reflect the result of the memory consolidation process.
The hippocampus is essential for episodic memory
1,2
. The dominant
theory of system memory consolidation proposes that active commu-
nication between the cortex and hippocampus transforms new mem-
ory in the hippocampus into long-term memory stored in the cortex
3,4
.
Recent studies have provided electrophysiological evidence for the
involvement of the hippocampus and neocortex in memory processing
during sleep, reflecting either active participation in the process of
memory consolidation as proposed in theoretical models
5,6
or reacti-
vation of consolidated memory traces. First, electroencephalogram
(EEG) events between the cortex and hippocampus are correlated
7–11
,
suggesting the two areas are engaged in active interaction during sleep.
Second, cell pairs that are correlated during awake experience are also
correlated during subsequent sleep within the hippocampus
12–14
,
within the cortex
15
, and between the hippocampus and cortex
16
.
These pairwise correlation results and other correlation-based analy-
sis
17
imply that the experience-related neuronal activity is, to some
degree, reactivated during sleep. However, the reactivation in these
studies lacks the specificity presumably required for episodic memory,
which includes a cascade of temporally ordered events encoded by a
unique sequence of activation of different neuronal populations within
the cortex, within the hippocampus, or both
18,19
. If sleep reactivation is
somehow involved in the processing of episodic memory traces, this
sequential structure should be specifically replayed. Indeed, replay of
specific ensemble-level patterns has been utilized in a detailed model of
memory consolidation
6
. Therefore, it is important to experimentally
study the more specific high-order replay, in which a temporally
sequential firing order across multiple cells is recaptured during
sleep. Such high-order replay has been observed in the hippocampus
during slow-wave sleep (SWS)
20,21
and rapid-eye-movement sleep
22
.
However, whether high-order replay exists in the cortex remains
unknown. More importantly, the relationship between replay events in
the cortex and hippocampus has not been studied. The present study
was designed to address these issues by recording spiking activity in
both the visual cortex and the hippocampal CA1 area of rats during
active maze-running and during natural sleep (Fig. 1). As we examined
a primary sensory area that is not explicitly driven by the hippocampus,
any observed replay was more likely to reflect broad cortical reactiva-
tion not limited to directly hippocampus-driven activity. Four rats were
trained to sleep for 1–2 hours (PRE), followed by an awake session
(RUN) during which they alternated between two trajectories (leftright
and rightleft) on a figure-8 maze, followed by another 1–2 hours sleep
session (POST). We found that high-order replay of RUN firing
patterns occurred not only in the hippocampus but also in the visual
cortex during SWS, and the replays in the two areas were coordinated
to represent the same coherent awake experience.
RESULTS
Firing patterns during SWS in the cortex and hippocampus
We first searched for spiking patterns at the population level in the
visual cortex and hippocampus during SWS. In the neocortex, cells
display active depolarized (up) and silent hyperpolarized (down) states
in vitro
23–25
, in anesthetized animals and during SWS
26–29
.Cortical
cells both within and across different cortical regions switch between up
and down states synchronously
9,26,27
. In agreement with these previous
results, we observed that cells across different layers in the visual cortex
displayed synchronized stepwise increases and decreases in multiunit
activity during SWS (Fig. 2a). More specifically, we observed periods of
80–300 ms during which the entire population of the recorded visual
cortical cells were silent. These periods of silence were followed by
increases in activity across the population lasting up to a few seconds.
Received 28 June; accepted 30 November; published online 17 December 2006; doi:10.1038/nn1825
The Picower Institute for Learning and Memory, RIKEN-MIT Neuroscience Research Center, Department of Brain and Cognitive Sciences and Department ofBiology,
Massachusetts Institute of Technology, Building 46, Room 5233, 43 Vassar Street, Cambridge, Massachusetts 02139, USA. Correspondence should be addressed to
M.A.W. (mwilson@mit.edu) or D.J. (dji@mit.edu).
100 VOLUME 10
[
NUMBER 1
[
JANUARY 2007 NATURE NEUROSCIENCE
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience

We refer to these active periods as frames. We are using the term ‘frame’
rather than ‘up state’ because we identified the phenomenon by
changes in multiunit activity rather than EEG rhythms or intracellular
potentials, and because similar structure also exists in the hippocampus
(see below) where no intrinsic up and down states have been reported.
On average, cortical frames occurred at a rate 47.3 ± 2.1 min
–1
(mean ±
s.e.m.) during SWS (n ¼ 20,545 during 20 sleep sessions from four
rats). There was no difference in occurrence rate between PRE and
POST (PRE, 44.3 ± 3.5 min
–1
; POST, 49.9 ± 3.3 min
–1
; P ¼ 0.193,
t-test). The frame durations were distributed widely between 0.1 and 3 s
with a mean 0.96 s and median 0.67 s, whereas the mean and median
durations of the interframe silent periods were 0.17 s and 0.13 s,
respectively (Fig. 2b). Cortical frames during POST had slightly shorter
durations (PRE, mean 1.1 s, median 0.73 s; POST, mean 0.90 s, median
0.65 s; P ¼ 2.2  10
–15
, rank-sum test) and slightly higher within-frame
multiunit firing rates per tetrode (PRE, mean 54.5 Hz, median 48.3 Hz;
POST, mean 58.7 Hz, median 54.1 Hz; P ¼ 1.2  10
–19
, rank-sum test)
than those during PRE. As shown in Figure 2a, the interframe silent
periods were correlated with positive peaks of EEG K-complexes
28
in
layer 5. This observation was confirmed by frame start- and end-time–
triggered EEG averages (Fig. 2c). On average, the cortical frames ended
about 20 ms earlier than the K-complex positive peaks, and they started
about 50 ms earlier than the K-complex negative peaks. Because depth-
positive EEG events are reliably associated with down states
28,29
,the
result imply that the interframe silent periods were produced by
cortical cells’ simultaneous switch to the down state, and that frames
were formed when cells rebounded to the active up state.
Whereas up and down states have been observed in neocortical cells,
hippocampal cells have not been reported to display such intrinsic
states. Despite this, we observed that the hippocampal neuronal
population also displayed during SWS synchronized periods of
increased and decreased multiunit activity: that is, frame and silent
periods (Fig. 2a). On average, hippocampal frames occurred at a rate of
41.7 ± 2.9 min
–1
during SWS (n ¼ 19,189 during 20 sleep sessions from
four rats). There was no significant difference in occurrence rate
between PRE and POST (PRE, 40.0 ± 4.1 min
–1
; POST, 43.5 ± 4.1
min
–1
; P ¼ 0.35, t-test). Hippocampal frames had shorter duration
(mean 0.78 s, median 0.50 s, P ¼ 0, rank-sum test) than the cortical
frames, and they were separated by longer interframe silent periods
(mean 0.50 s, median 0.22 s, P ¼ 0, rank-sum test) (Fig. 2b). Like
cortical frames, hippocampal frames during POST had slightly (and
insignificantly) shorter durations (PRE, mean 0.81 s, median 0.49 s;
POST, mean 0.76 s, median 0.50 s; P ¼ 0.41, rank-sum test) and slightly
higher multiunit firing rates per tetrode (PRE, mean 63.0 Hz, median
58.1 Hz; POST, mean 67.6 Hz, median 61.5 Hz; P ¼ 0.040, rank-sum
test) than those during PRE. Hippocampal frames were correlated with
CA1
a
b
LR
c
PRE (1–2 h)
RUN (20–40 min) POST (1–2 h)
Visual
1 mm
10 cm
Figure 1 Experimental design. (a) On each recording day, there were three
recording sessions: a 1–2 hour sleep session (PRE), a 20–40 minute maze-
running session (RUN), and another 1–2 hour sleep session (POST) after the
run. (b) During the RUN sessions, rats were trained to run an alternation task
on a figure-8-shaped maze. All the visited position points during a typical
RUN session are plotted to show the shape of the maze. Rats had to alternate
between the red (leftright) and blue (rightleft) running trajectories to receive
a reward at R or L. The arrows mark the running directions. (c) We implanted
tetrodes to record CA1 cells in the hippocampus and cells in the visual
cortex. Histology micrographs show two lesion spots (arrows), which mark the
tetrode tip locations, in the CA1 pyramidal cell layer (‘CA1’), and two in the
deep layers of the primary visual cortex V1 (‘visual’).
b
CTX frame
Count
230
100
200
12301
10
900
1,800
0.5
CTX silent
a
c
d
FS
0
1
e
0
0
0.5
1
End
0
1
EndEndEnd
L5
HP
CTX
Ripple
0.5 s
125
250
1,000
500
Duration (s)
HP frame HP silent
0.1 mV
0.2 s
Start End
Rate (Hz)
0.5
Correlation (× 0.01)
–0.4 –0.2 0.2 0.4 0–0.4 –0.2 0.2 0.4
Time (s) Time (s)
0.5
Correlation (× 0.01)
Start
Duration (s) Duration (s)
10 0.5
Duration (s)
Figure 2 Visual cortical and hippocampal spiking activities were organized as
frames during SWS. (a) Cortical (CTX) and hippocampal (HP) frames during a 5-s
SWS episode. Each tick represents a spike and each row includes all multiunit
spikes recorded from one tetrode. Triangles, frame start times; circles, frame end
times. Cortical EEG in layer 5 (L5, top) and hippocampal EEG within the ripple
band (bottom) are displayed for the same time period. Dotted boxes mark a
K-complex (top) and a ripple event (bottom). Scale bars, 1.5 mV for L5, 0.5 mV
for ripple. (b) Distributions of durations of frames and interframe silent periods in
the cortex and hippocampus. (c) Cortical EEG averages (mean ± s.e.m., s.e.m.
represented by thickness of the curves) triggered by cortical frame start and end times. (d) Occurrence rate (mean ± s.e.m., n ¼ 20 sleep sessions) of
hippocampal ripple events within hippocampal frames (F) and within interframe silent periods (S). (e) Average cross-correlogram (mean ± s.e.m., n ¼ 20 sleep
sessions) between cortical and hippocampal frame start times and between their end times. Here the cortex was the reference, meaning a peak at positivetime
would indicate that the cortex led the hippocampus. Bin size, 10 ms.
NATURE NEUROSCIENCE VOLUME 10
[
NUMBER 1
[
JANUARY 2007 101
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience

ripples (Supplementary Fig. 1 online), which are prominent high-
frequency (80–250 Hz) oscillation events in the hippocampal EEG
30
.
Ripples almost always appeared inside hippocampal frames but not
within interframe silent periods (Fig. 2d). Individual frames could
containnone,oneormultiplerippleevents(Supplementary Fig. 1).
Furthermore, ripple events on average started 30 ms later than the
frame onsets (Supplementary Fig. 1), suggesting ripple events were
triggered by frame activity. These results are consistent with the notion
that ripple events in the hippocampus were modulated and grouped by
the frame structure.
We next studied whether the cortical and hippocampal frames were
related. Frame onset and offset times in the visual cortex and hippo-
campus were significantly correlated (Fig. 2e). On average, the cortical
frames started about 50 ms earlier (P ¼ 2.2  10
–8
, t-test) and ended
about 40 ms earlier (P ¼ 1.0  10
–5
) than the hippocampal ones. There
was no statistically significant difference in the correlation between PRE
and POST, and the temporal relationships were insensitive to para-
meters that define frame boundaries (Supplementary Fig. 2 online).
However, the broad peaks in the cross-correlograms (Fig. 2e)imply
that there was no one-to-one correspondence between cortical and hip-
pocampal frames. Therefore, on average, general activity patterns in the
cortex and hippocampus were correlated, suggesting active interaction
between cortical and hippocampal neuronal ensembles during SWS.
High-order replay in the cortex and hippocampus
To characterize patterned memory reactivation events, we next exam-
ined the contents of the cortical and hippocampal frames in relation to
the activity evoked by maze-running. Unlike pairwise correlation
studies
15,16
(Supplementary Figs. 3 and 4 online), this study addressed
high-order replay by comparing multicell firing sequences generated by
running the two trajectories with the firing sequences of the same cells
in sleep frames during PRE and POST.
It is well known that hippocampal cells are active in specific places
(‘place cells’
31
). Place-specific firing has also been reported in the
medial entorhinal cortex
32,33
, but not in sensory cortices. During
RUN, as expected, hippocampal cells fired in their place fields on the
maze. Unexpectedly, many cells in the visual cortex (mostly recorded in
the deep layers in primary visual cortex (V1) and its surrounding
secondary visual cortex), also had localized firing fields (Fig. 3). The
fields were consistent across days (Fig. 3a) as well as across laps within
individual RUN sessions (Fig. 3b). These spatially localized firing
patterns are likely to have resulted from the local visual cues within
the maze which provided consistent patterns of visual input, rather
than any intrinsic ‘place’ specificity as seen in hippocampal cells. Using
spatial information as a measurement, 54 out of 116 cortical cells were
quantified as having localized firing fields on the maze (Fig. 3c). Only
cells with such firing fields were included in the subsequent analysis.
The spatially localized firing fields in the cortex and hippocampus
allowed us to establish repeatable multicell firing sequences in both
areas during the spatial task (Fig. 4a,b, lap). Different sequences
emerged from different trajectories. We extracted these sequences by
assigning numbers (0, 1, etc.) to cells active on a trajectory, and then
arranging them according to the order of the cells’ peak firing times
(Fig. 4a,b, avg). A sequence generated by a RUN trajectory is referred to
as a template sequence. For example, RUN activity patterns in
Figure 4a gave rise to the template sequence 01234567 when the rat
ran the leftright trajectory. We analyzed a total of 12 cortical template
sequences across 10 d and four rats. Among three of the four rats, 15
hippocampal template sequences across 8 d were also constructed. In
the fourth rat, only two hippocampal place cells recorded were active
on the maze, so high-order sequence replay in the hippocampus was
not examined in this individual. For each rat, template sequences on
the same trajectory were extracted from two or three consecutive
recording days. Though these templates may have contained different
number of cells, they were likely to have been drawn from the same cell
population because the tetrodes were not moved during those days.
To determine whether the template sequences were reexpressed
within sleep frames (for example, Fig. 4a,b, frame), we used a
combinatorial method
34
. First, within each frame, a firing sequence
was determined by calculating the relative order of peak firing times
across the same cells as in a template (Fig. 4a,b,seq).Forexample,the
frame in Figure 4a yielded a sequence 0132567 (the number 4 cell in the
template sequence was inactive in this particular frame). We then
defined a matching index I to measure the similarity between the frame
sequence 0132567 and the template sequence 01234567 (see Supple-
mentary Methods online for details). Finally, given a matching index I
we computed the matching probability p that a matching index equal
to or larger than I would be produced by chance, assuming that all
possible orders of the same cells are equally probable. The matching
c
02 4
0
5
Cell count
HP
b
0
60
0
60
100 200
Rate (Hz)
100 200
CTX
Leftright (cm)
0
30
100 200
Rate (Hz)
HP
Rightleft (cm)
0
30
100 200
a
1.2
30
60
0
30
60
0
CTX
HP
2.9
15
30
0
15
30
0
1.1
2.8
Day 1 Day 2
10
0
5
Cell count
10
CTX
Spatial information
(
bits
p
er s
p
ike
)
04
8
Spatial information
(
bits
p
er s
p
ike
)
Figure 3 Visual cortical cells displayed localized firing fields. (a) Firing rate
maps of a cortical cell (CTX) and a hippocampal place cell (HP) on two
consecutive recording days. The maze is shown as blue. Color bars, firing
rates in Hz; bin size, 2 cm  2 cm. The number in each map indicates
spatial information in bits per spike. Scale bar, 20 cm. (b) Consistent firing of
the cortical cell and the hippocampal cell examined lap by lap on day 1 when
the rat was running the leftright and rightleft trajectories. Each trajectory was
linearized and plotted on the x axis. In each panel, black dots represent
spikes fired at the corresponding positions and one row shows all spikes in
one single lap. Laps are arranged top down in increasing temporal order. The
bottom histograms represent the binned firing rate computed from the laps
shown. Bin size, 2 cm. (c) Spatial information from cortical cells and hippo-
campal cells that were active on the maze. The dashed line indicates the
threshold (0.8) used to determine whether a cortical cell had a firing field.
102 VOLUME 10
[
NUMBER 1
[
JANUARY 2007 NATURE NEUROSCIENCE
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience

probability measures the significance of a match between a frame
sequence and a template sequence. Unless otherwise specified, we used
a threshold p o 0.05 to determine whether a sleep frame was a signi-
ficant match. Such a frame is referred to as a replaying frame. Due to
the discrete nature of the matching probability p (see Supplementary
Methods for details), the exact cutoff threshold depended on the
number of cells active in a frame and ranged between 0.028 and
0.049 (Supplementary Table 1 online). A frame with less than four
active cells could not reach this threshold to be considered significant;
thus, a replaying frame necessarily contained at least four active
cells. For example, both the cortical and the hippocampal frames
shown in Figure 4 were replaying frames. The cortical frame con-
tained the sequence 0132567 with I ¼ 0.91 and p ¼ 0.0014. The
hippocampal frame contained the sequence 01235 with I ¼ 1andp ¼
0.0083. More examples of sequence replays are shown in Supplemen-
tary Figure 5 online.
To compute the overall replay effect, we counted the number of
replaying frames out of the total number of candidate frames, defined
as those containing at least four active template cells, during SWS
within last hour in PRE and within first hour in POST. In the cortex,
out of 3,070 PRE and 5,808 POST candidate frames, we identified a
total of 163 PRE and 366 POSTreplaying frames. In the hippocampus,
out of 849 PRE and 1,555 POST candidate frames, we identified a total
of 39 PRE and 121 POSTreplaying frames. The ratio between replaying
and candidate frame numbers, averaged across all the template
sequences, was significantly higher during POST than during PRE in
both the cortex (PRE, 0.052 ± 0.008; POST, 0.073 ± 0.009; P ¼ 0.027,
paired t-test, n ¼ 12 templates) and hippo-
campus (PRE, 0.049 ± 0.011; POST, 0.080 ±
0.007; P ¼ 0.0057, n ¼ 15 templates). There-
fore, in both the cortex and hippocampus,
there were significantly more replaying frames
during POST than PRE, indicating that the
replay was experience dependent. The replay-
ing ratios for every individual template
sequence (trajectory) are listed in Supple-
mentary Table 2 online for cortical templates
and in Supplementary Table 3 online for
hippocampal templates. We then investigated
the properties of these replay events. First, the
ratio in POST decayed back to that of PRE
after about 40 min in the cortex (ratio during
first 20 min, 0.064 ± 0.011; second 20 min,
0.088 ± 0.013; third 20 min, 0.058 ± 0.009;
fourth 20 min, 0.054 ± 0.012), and after about
1 h in the hippocampus (ratio during first
20 min, 0.064 ± 0.006; second 20 min, 0.089 ±
0.012; third 20 min, 0.072 ± 0.017; fourth
20 min, 0.054 ± 0.017). Second, the template
sequences were compressed in these replaying
events in both the cortex and hippocampus by
a similar factor about 5–10 (Supplementary
Fig. 6 online). Third, small differences in
frame properties between PRE and POST
did not contribute to the observed difference
in replaying ratios (S upplementary Fig. 7
online). Fourth, there was no difference in
within-frame multiunit firing rate, within-
frame RUN-active-cell firing rate or frame
duration between replaying and non-
replaying candidate frames (Supplementary
Fig. 8 online). Therefore, the replay identified by the sequence match-
ing method was not biased by differences in these factors between PRE
and POST frames.
We then examined whether the observed numbers of replaying
frames significantly deviated from those expected by chance, using
two methods to evaluate the significance. First, we computed the
theoretical distribution of replaying frame numbers by assuming a
binomial process in which every frame independently matches a
template sequence at the same probability as the cutoff threshold.
This distribution is referred to as chance distribution. We compared
the observed numbers of replaying frames with those expected from the
chance distribution (Fig. 5a,b). For all the trajectories combined, the
observed numbers in the visual cortex were statistically significant
in both POST (n ¼ 366, P o 1  10
–38
)andPRE(n ¼ 163, P ¼ 1.4 
10
–6
). In the hippocampus, the observed numbers were significant in
POST (n ¼ 121, P ¼ 8.1  10
–12
), but not in PRE (n ¼ 39, P ¼ 0.16).
We repeated the analysis for each individual rat. In the cortex, the
observed replaying frame numbers during POSTwere significant for all
four rats (rat 1, P ¼ 5.0  10
–11
;rat2,P ¼ 2.8  10
–11
;rat3,P ¼
0.00031; rat 4, P ¼ 0.0028), whereas the numbers during PRE were
significant for two rats (rat 1, P ¼ 1.4  10
–5
;rat4,P ¼ 0.00041), close
to being significant for another (rat 3, P ¼ 0.067) and not significant for
thelast(rat2,P ¼ 0.50). In the hippocampus, the numbers for all three
rats were significant in POST (rat 1, P ¼ 0.0017; rat 2, P ¼ 1.5  10
–7
;
rat 3, P ¼ 0.00019), but not in PRE (rat 1, P ¼ 0.17; rat 2, P ¼ 0.52; rat
3, P ¼ 0.12). The second method tested the null hypothesis that a RUN
template sequence is replayed with the same probability as any of its
1 s
Avg 012345
Lap
Frame
0.2 s
Seq 01235
0
5
0
5
Cell number
HP
a
RUN
1 s
Avg 01234567
Lap
0.5 s
Seq 0132567
Cell number
0
7
0
7
b
Lap
CTX
POST RUN POST
Frame
Figure 4 Sleep frames replayed multicell firing sequences during RUN in both the visual cortex and the
hippocampus. (a) Cortical firing sequence during RUN and in a POST sleep frame. Lap, firing pattern
during a single running lap on the leftright trajectory. Each row represents a cell and each tick represents
a spike. Avg, template firing sequence obtained by averaging over all laps on the trajectory. Each curve
represents the average firing rate of a cell. Cells were assigned to numbers 0, 1, etc. and then arranged
(01234567) from bottom to top according to the order of their firing peaks (vertical lines). Frame, the
same cells’ firing patterns in a POST sleep frame. Triangles and circles, frame start and end times,
respectively. Seq, firing sequence in the frame. Spike trains were convolved with a gaussian window and
cells were ordered (0132567) according to the peaks (vertical lines) of the resulted curves. (b)Sameas
a, but for cells in the hippocampus on the same trajectory.
a
POST
80 350
0
215
Number of replaying frames
Prob. density
PRE
60 160110
b
PRE
15 35
0
0.08
Prob. density
55
POST
40 120
0
0.04
0.08
80
0.04
0.02
0
0.04
0.02
0.04
Number of replaying frames
Figure 5 Frame replays occurred significantly more often than chance in POST in both the visual
cortex and hippocampus. (a) Chance (dotted line) and shuffle (solid line) distributions of
the number of replaying frames that were randomly generated for the visual cortex during PRE
and POST. Vertical lines, the actual observed numbers of replaying frames. (b) Same as a, but for
the hippocampus.
NATURE NEUROSCIENCE VOLUME 10
[
NUMBER 1
[
JANUARY 2007 103
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience

random shuffles. From the null hypothesis, a shuffle distribution of
replaying frame number was obtained. Against this shuffle distribution,
the observed numbers of replaying frames in the visual cortex were also
significant (Fig. 5a,b)inbothPOST(P o 0.001) and PRE (P ¼ 0.009),
whereas in the hippocampus the numbers were only significant in
POST (P o 0.001), not in PRE (P ¼ 0.19). These analyses verify that
replaying frames in POST did not arise from chance. Thus, the
sequence-matching analysis demonstrates that a significant number
of sleep frames replayed running-evoked firing sequences in both the
visual cortex and hippocampus, providing the first direct evidence for
high-order replay in the neocortex.
Interaction between cortical and hippocampal replays
To study the interaction between the cortical and hippocampal replays,
we next asked whether the replaying frames in the two areas were
independent of each other. As we identified only a relatively small
number of frames as replaying among a large number of total sleep
frames (see numbers above), replaying frames were sparsely distributed
during SWS. As a result, the chance that a cortical replaying frame and a
hippocampal replaying frame would occur together would be very small
if replaying frames in the two areas were not temporally related. We
identified replaying cortical and hippocampal frame pairs that matched
thesametrajectoryandoverlappedintime(‘same-trajectory’).An
example of such a pair is shown in Figure 6a.Thecorticalframehad
a sequence 023489567 with a matching probability p ¼ 0.0063, and
the overlapping hippocampal frame had a sequence 012345 with
p ¼ 0.0014. From the three rats in which both cortical and hippocampal
templates were available on the same trajectory, a total of nine such pairs
were observed in POST (rat 1, three; rat 2, two; rat 3, four) whereas only
one was observed in PRE. As a control comparison, we also counted
overlapping frame pairs in which the cortical frame replayed one
trajectory while the hippocampal frame replayed the other on the
same day (‘different-trajectory’). In this case, we observed only three
pairs in POST and none in PRE (rat 1, zero; rat 2, two; rat 3, one). We
then evaluated the significance of the observed overlapping pairs by
comparing the numbers with those expected from the null hypothesis
that the replaying frames in the two areas are independent. For this
purpose, we applied a shuffling procedure in which replaying frames in
the cortex and hippocampus were randomly and independently redis-
tributed among all the candidate frames (Supplementary Fig. 9 online).
We compared the actual observed numbers with distribution of the
shuffling-produced overlapping pair numbers (Fig. 6b,c). The signifi-
cance level (P value) was defined as the number of shuffles that yielded
the same or more overlapping pairs than the actual observed pairs
divided by the total number of shuffles. In the case of same-trajectory,
the observed number of pairs was significant in POST (P ¼ 0.01) but
not in PRE (P ¼ 0.75). For different-trajectory, the observed numbers
were not significant in either POST (P ¼ 0.59) or PRE (P 4 0.99). This
result indicates that frames in the visual cortex and hippocampus that
replayed the same trajectories overlapped more than chance.
The observed number of overlapping replaying pairs was low.
However, because only a small fraction of cells that would actually be
participating in replay events were recorded, many more frames
could be replaying but not detected because of the limited number of
cells available. To investigate how robust the overlapping effect was,
we varied the matching probability (p) threshold that defines re-
playing frames. As the threshold increased, we found more overlapp-
ing replaying pairs in POST and the number of pairs in POST became
statistically significant for a large range of p threshold in the
case of same-trajectory (Fig. 6d). For example, at the more relaxed
No. of pairs No. of pairs
b
POST
PRE
0105
0
0.2
0.4
a
POSTPOST
0.3
0.5
0 0.1 0.2
PRE
P value
0
1
PRE
d
Probability
0
0.2
0.4
Probability
0
0.2
0.4
Probability
0
0.2
0.4
Probability
c
POST
PRE
0105
No. of pairs No. of pairs
01050105
CTX: 023489567
HP: 012345
0.5 s
Prob. threshold
0.3
0.5
0 0.1 0.2
P value
0
1
Prob. threshold
Prob. density
a
Sleep interval (s)
024–2–4
0
2
4
–2
–4
b
0
0
0.03
0.06
c
0510
0
0.04
0.08
P value
0510
0
P value
d
Run interval (s)
–0.2 0.2
Correlation coefficient
Tra
j
ector
y
number
0.08
0.04
Trajectory number
Figure 7 Cortical and hippocampal frames co-replayed the same running trajectory as revealed by interval analysis. (a) Time intervals between cortical and
hippocampal cell pairs based on cortical replaying frames, compared with their corresponding RUN intervals on a trajectory. Solid line, linear regression
between the sleep and RUN intervals. (b) Distribution of shuffling-produced correlation. Vertical line, actual observed correlation. (c) P values of the actual
observed correlations based on cortical replaying frames for all trajectories. Trajectories represented by the same shape were from the same rat. Horizontal
lines, significance level P ¼ 0.05. (d)Sameasc, but based on hippocampal replaying frames.
Figure 6 Visual cortical and hippocampal frames that replayed the same
trajectories tended to occur at the same time. (a) A cortical (CTX) and a
hippocampal (HP) replaying frame that overlapped in time. Each row
represents a cell and each tick represents a spike. Triangles and circles,
frame start and end times, respectively. The two frames replayed the same
rightleft trajectory. (b,c) Distributions of pair numbers produced by shuffling
for overlapping cortical-hippocampal frame pairs that replayed the same (b)
and different (c) trajectories in PRE and POST. Vertical gray lines, actual
observed numbers. (d) Dependence of the significance P values of the actual
observed numbers on the matching probability threshold in PRE and POST.
Lines with filled triangles, same-trajectory; lines with filled circles, different-
trajectory; dotted horizontal lines, significance level P ¼ 0.05.
104 VOLUME 10
[
NUMBER 1
[
JANUARY 2007 NATURE NEUROSCIENCE
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience

threshold P o 0.12, we found 25 same-trajectory pairs in POST
(P ¼ 0.004) and only 3 in PRE (P ¼ 0.91), whereas for different-
trajectory pairs we found 11 in POST (P ¼ 0.35) and 5 in
PRE (P ¼ 0.24). When the threshold became too large (40.18),
the differences between PRE and POST and between same- and
different-trajectory were eventually lost. This analysis demonstrates
that the overlapping effect was robust and did not depend on a
particular choice of matching probability threshold.
To provide further evidence that the replays in the hippocampus and
the cortex were coordinated, we applied an interval analysis as follows.
For a cell in a replaying frame in one area and a cell in one of its
overlapping frames in the other area, we computed the temporal
interval between their peak firing times in their corresponding frames,
and compared it with the temporal interval between their peak firing
times on a RUN trajectory. Based on all cortical replaying frames for a
trajectory (as results were similar if replaying frames in PRE and POST
were computed separately (data not shown), we combined all the
replaying frames in PRE and POST), we first collected sleep intervals of
all cell pairs with one cell in a cortical replaying frame and the other in
one of its overlapping hippocampal frames (not necessarily replaying),
and their corresponding RUN intervals on the trajectory. We then
examined whether the sleep intervals and the RUN intervals were
correlated. For 11 out of 11 trajectories from four rats, sleep intervals
based on cortical replaying frames were significantly correlated with
their RUN intervals (P r 0.033, Pearson’s r)(forexample,Fig. 7a:
r ¼ 0.23, P ¼ 1.7  10
–17
). Significant correlation could be a result of
systematic temporal bias of hippocampal cells or cortical cells on a
trajectory or of overall relationship between hippocampal frames and
cortical frames (as shown in Fig. 2e). To control for this possibility, we
shuffled cell identities in the template sequence for the replaying frames
and obtained a distribution of correlation from the shuffled templates.
We then compared the actual observed correlation with the distribu-
tion. For the trajectory shown in Figure 7a, the actual observed
correlation (0.23) was significantly higher than those produced by
the shuffling (P o 0.001, Fig. 7b). The same was true for all 11
trajectories examined (P r 0.035, Fig. 7c). Similarly, we also per-
formed the analysis based on all the hippocampal replaying frames. In
this case, for nine out of ten trajectories from three rats, the correlations
between sleep and RUN intervals were significantly higher than those
produced by the shuffling (P r 0.048, Fig. 7d). This interval analysis
result indicates that, if a frame in the cortex or hippocampus replayed a
trajectory, cells in its overlapping frames in the other area fired at the
relative temporal interval predicted from the RUN template. Together
with the result that frames replaying the same trajectory in the two
areas tended to appear simultaneously, the data provide evidence that
the fine details of replaying events seem to match coherently the same
awake experience in the two areas.
DISCUSSION
Current theory proposes that active interaction between the cortex and
hippocampus during offline periods, such as sleep, plays an important
role in memory consolidation
5,6
. Here we have described two types of
interaction between the neocortical and hippocampal spiking activities
during SWS. First, both visual cortical and hippocampal activity
patterns seem to be organized into periods of elevated activity referred
to as frames. These frames tend to start and end together at a fine time
scale, with hippocampal frames briefly lagging cortical frames. Second,
at the level of detailed activity pattern, both visual cortical and
hippocampal frames replay the multicell firing sequences evoked by
awake experience, and the replay in the two areas tends to reflect the
same experience (in this case the same trajectory).
Cortical cells switch between up and down state in a synchronized
manner during SWS
9,26,27
. This has been described as the cortical slow
oscillation in intracellular membrane potential
35,36
, and is also seen in
the EEG
10,11,37
. We have characterized the extracellular multiunit
activity pattern (frame) that presumably arises from such intracellular
events. In measurements of similar alternating active and silent periods
of individual cortical cells
27
, the silent period duration is comparable
with that in our data, whereas the active period length is shorter
than the frame duration that we measured. This is consistent with
the observation that cortical cells are not perfectly synchronized in
switching to up state
25–27
. The correlation of cortical frames with
K-complexes, a major component of the slow oscillation
28,36
,andcon-
currence of the cortical frame occurrence rate (0.8 Hz) with the slow
oscillation frequency range further imply a direct relationship between
cortical frames and the slow oscillation. Although these findings
indicate that frames may be equivalent to the slow oscillation of cortical
cells, the frame structure is also observed in the hippocampus, even
though general EEG events are distinctly different. The fact that cortical
frames led hippocampal frames by about 50 ms indicates that hippo-
campal frames may be the result of cortical drive rather than intrinsic
state change. Recently, hippocampal interneurons have been found to
be phase-locked to cortical up and down state transitions
38
, indicating
that the frame structure in the hippocampus may be primarily driven
or shaped by the interneuron activity. It has also recently been found
that slow oscillation in the EEG can be seen in the hippocampus and
that the cortical slow EEG oscillation leads the hippocampal one by a
similar interval (56 ms)
11
. Thus, it is possible that the slow oscillation
reflects or underlies the emergence of the frame structure in both areas.
Therefore, frames may serve as basic functional processing units during
SWS in many brain areas, and may provide a framework for studying
cortical-hippocampal interactions involved in memory consolidation.
Consolidation of episodic memory presumably requires or results in
replay of specific neuronal patterns that encode the temporally sequen-
tial events in an episode. We have demonstrated that such high-order
replay not only occur in the hippocampus but also in the primary visual
cortex. The replay implies that specific activity patterns of those cells
involved in visual perception (during maze-running) are reactivated
during sleep, even if no visual stimuli are present. This is consistent
with imaging studies showing that early visual cortices are activated
during mental imagery
39
and memory recall
40
in the absence of visual
input. Furthermore, the replay also raises the possibility that even early
sensory cortices may be involved in memory consolidation, long-term
memory storage or both. It has been proposed that episodic memory
may be stored distributedly with components involving a particular
sensory modality stored in that sensory cortex
41
. Our study is con-
sistent with this hypothesis.
Our data provide evidence that there may be a difference between the
hippocampal and cortical replays. In this experiment, we studied
memory reactivation only after the memory was well-formed. There-
fore, it is quite possible that we observed events that resulted from
earlier consolidation
42
. But even for well-trained rats, replays were
enhanced by the running experience between PRE and POST sleep in
both the cortex and the hippocampus, demonstrating the experience
dependence of both cortical and hippocampal replays. During PRE,
however, replaying frames already seemed to occur significantly more
often than chance in the cortex, but not in the hippocampus. By
contrast, the interval analysis result showed that when a cortical
replaying frame occurred, during either PRE or POST, hippocampal
cells were biased to fire at the same time at locations consistent with
those in RUN, meaning that on average there was some degree of
reactivation in hippocampal frames that overlapped with cortical
NATURE NEUROSCIENCE VOLUME 10
[
NUMBER 1
[
JANUARY 2007 105
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience

replaying frames; however, the robustness of high-order hippocampal
replay was reduced during PRE. In contrast, PRE cortical frames
showed a more robust high-order replay, indicating that in well-trained
rats cortical memory traces expressed during SWS may be more likely
to reflect past RUN experience than hippocampal traces are. This
observation is consistent with the theoretical hypothesis that the cortex
and hippocampus play complementary roles in memory formation and
storage
43,44
, with the cortex reflecting long-term memory and the
hippocampus reflecting new short-term memory.
We found that the cortical and hippocampal replays were coordi-
nated to match the same awake experience during SWS. The coordina-
tion is likely to require active communication between the cortex and
hippocampus. The observation that cortical frame onset times precede
hippocampal ones (Fig. 2e) implies an initial feed-forward interaction
from the cortex to hippocampus. However, it remains unclear which
area is responsible for initiating individual replay events after frame
onsets. Although our data revealed a trend toward hippocampal replays
leading those in the neocortex (Supplementary Fig. 10 online), we
were unable to definitively establish the direction of interaction.
Overall, these findings are consistent with a bidirectional interaction
model. First, cortical frame activation during SWS biases hippocampal
activity and triggers the start of hippocampal frames through cortical-
hippocampal projections
45
. This could establish the context or initial
conditions for subsequent replay within hippocampal frames. Sequence
memories are then reactivated during ripple events that occur within
hippocampal frames. The replayed sequence memories are sent back to
the associational and then primary sensory cortices through hippo-
campal-cortical back projections
46
, and this biases the cortical activity
toward simultaneous cortical frame replay which gradually strengthens
cortical-cortical synapses for long-term memory storage. In this model,
the two-way interaction and memory trace transfer occur within
individual hippocampal and cortical frames. Indeed, there is evidence
that neuronal activity propagates among cortical layers
25
and among
cortical areas
26,27
under broad synchrony of up state activation. The
expression of these reactivated memory traces in sensory cortex may
directly relate to the perceptual imagery experienced during sleep
and dream states.
METHODS
Rats and experimental procedures. Four Long-Evans rats (5–8 months old)
were trained to sleep in a sleep box and run an alternation task on a figure-8-
shaped maze (Fig. 1). The daily training procedure was exactly same as in later
recording days. Intra-maze cues, such as black and white stripes with different
orientations and simple geometric shapes, were added to the maze floors and
inner walls. The entire maze was surrounded by a black curtain without obvious
distal cues except for the irregular wrinkles of the curtain. The rats were trained
to alternate between two running trajectories (leftright and rightleft) to get food
at two reward sites. The training and later recording protocol was approved by
the Committee on Animal Care at Massachusetts Institute of Technology and
followed US National Institutes of Health guidelines.
After about 2–3 weeks’ training, we implanted on the rat’s skull a micro-
electrode array containing 18 independently adjustable tetrodes. Six to eight
tetrodes were assigned to the hippocampus (anteroposterior –3.9, mediolateral
2.2, relative to bregma) and 10–12 tetrodes aimed at primary visual cortex
(anteroposterior –7.1, mediolateral 3.5). We inserted a bipolar electrode into
the rat’s neck muscle to record the electromyogram (EMG). We reintroduced
rats to the maze one week after the surgery and retrained them for about 10–15
d before the recording. Recording began once units were stable and rats ran
each trajectory at least 20 times. This study only includes data taken from well-
trained rats (alternation with at least 80% accuracy). Spikes from tetrodes with
any of the four channels crossing a preset triggering threshold were acquired at
32 kHz. EMG and EEG signals were filtered at 0.1–475 Hz and recorded
continuously at 2 kHz. Two infrared diodes were used to track the rat’s position
during a RUN session. Diode positions were sampled at 30 Hz with a resolution
of approximately 0.67 cm. On some days, diodes were mounted not directly
over but on one side of the rat’s head, causing one loop of the maze to appear
slightly smaller than the other.
Data analysis. We used ten datasets (two or three consecutive days per rat), each
of which contained at least ten RUN-active visual cortical cells and ten RUN-
active hippocampal cells, in this analysis. In total, we recorded 116 cortical cells
and 294 CA1 cells. Among them, 97 cortical cells (RUN mean rate Z 0.5 Hz)
and 129 CA1 place cells (RUN mean rate Z 0.2 Hz and o 4 Hz) were active on
the maze. Most of the cortical cells were located in the deep layers (5 or 6) in
primary visual cortex (V1). A few cells were recorded from layers 4 and 3 in V1
and some other cells from deep layers of the visual cortical area immediately
lateral to V1. Tetrode locations were identified according to ref. 47.
Sleep stage classification. EMG, hippocampal and cortical EEGs were used to
classify sleep states at 1-s resolution into four stages: wake state, SWS, rapid-
eye-movement sleep and an unspecified intermediate state (Supplementary
Fig. 11 online). SWS was characterized as having low EMG, high hippocampal
ripple, low hippocampal theta and high cortical delta power
48
.
Frame definition. All multiunit spikes (not necessarily sorted single-unit
spikes) from all tetrodes within the same recording area were used to determine
frame boundaries (see Supplementary Fig. 12 online for details). Spikes from a
recording area were combined and counted in 10 ms time bins. Spike counts
were then smoothed using a gaussian window with s ¼ 30 ms. Interframe
silent periods were defined as periods with spike counts below a preset
threshold, and frames as periods in between. Furthermore, consecutive frames
with a gap shorter than a threshold were combined.
Frame-triggered EEG and ripple detection. Broadband (0.1–475 Hz) EEGs
recorded in layer 5 were used for cortical frame-triggered averages. For the
hippocampus, EEGs recorded from the CA1 pyramidal cell layer were first
filtered for ripple band (80–250 Hz), and then ripple power was calculated as
squared EEG value at each time point. For a selected time point (start or end
time) of a frame, a 5-s EEG (or EEG power) segment centered at the time was
selected. All the segments triggered by all frames in consideration were then
averaged to obtain the mean trace. Ripple events were detected using a
threshold-crossing method on the filtered hippocampal EEG at ripple band
7,30
.
Two thresholds were defined. If S is the standard deviation of an EEG trace, 3S
was set as cross-threshold and 7S as peak-threshold. All the time points with
absolute EEG values larger than the cross-threshold were identified. Time
points separated by gaps smaller than 50 ms were grouped as a single event.
Furthermore, only events with a peak absolute value larger than the peak-
threshold were taken as ripple events and the peak time was considered to be
the ripple event time. The method also determined the start and end times for
every ripple event.
Frame cross-correlation. Frame start (or end) times were treated as discrete
events. We first converted the events to occurrence rates with a bin size 10 ms.
Given two event rates f
1
(t)andf
2
(t), where t ¼ 1,2,y,n, the cross-correlation
coefficient at time lag Dt between the two events was computed as
C
12
ðDtÞ¼
P
n
t ¼ 1
ðf
1
ðtÞf
1
Þðf
2
ðt + DtÞf
2
Þ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
n
t ¼ 1
ðf
1
ðtÞf
1
Þ
2
s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
n
t ¼ 1
ðf
2
ðtÞf
2
Þ
2
s
;
where
f
i
¼
1
n
X
n
t ¼ 1
f
i
ðtÞ; for i ¼ 1; 2:
As the correlation coefficient is normally distributed if we assume a null
hypothesis that two events are independent Poisson processes
49
,weusedat-test
to test the dependence between two event trains at a time lag.
Firing rate map and spatial information. Position points on the maze were
binned into 2-cm  2-cm grids. A firing rate map was obtained by simply
counting a cell’s spikes in a grid divided by the rat’s total occupancy time in it.
106 VOLUME 10
[
NUMBER 1
[
JANUARY 2007 NATURE NEUROSCIENCE
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience

Only position points and spikes during trajectory running were included.
Spatial information was computed using one-dimensional linearized trajec-
tories instead of the two-dimensional maze. The two trajectories (leftright and
rightleft) were linearized separately, binned with 2-cm bins, and then com-
bined. The cell’s firing rate in each bin of the two linearized trajectories was
computed similarly to that of the two-dimensional maze by counting spikes
divided by occupancy time. If f
i
, t
i
(i ¼ 1,2,y,n) are the firing rate and
occupancy time for the i
th
bin, spatial information is given by
50
SpI ¼
X
n
i¼1
p
i
f
i
f
log
2
f
i
f
;
where
p
i
¼ t
i
=
X
i
t
i
is the occupancy probability and
f ¼
X
i
p
i
f
i
is the mean firing rate.
Sequence matching and interval analysis. Sequence construction, sequence
similarity, sequence matching probability, overall replay significance, over-
lapping frame pairs and overlapping significance, and interval analysis
are briefly described in the Results section. See Supplementary Methods for
more details.
Note: Supplementary information is available on the Nature Neuroscience website.
ACKNOWLEDGMENTS
We thank E. Miller, C. Moore, J. Fisher and F.-M. Zhou for critical readings
on the manuscript, and Wilson laboratory members for technical help and
suggestions and comments on the project and manuscript. Supported by grants
to M.A.W. from the Brain Science Institute at the Institute of Physical and
Chemical Research (RIKEN) in Japan and the US National Institutes of Health.
COMPETING INTERESTS STATEMENT
The authors declare that they have no competing financial interests.
Published online at http://www.nature.com/natureneuroscience
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions/
1. Squire, L.R. Memory and the hippocampus: a synthesis from findings with rats,
monkeys, and humans. Psychol. Rev. 99, 195–231 (1992).
2. Fortin, N.J., Agster, K.L. & Eichenbaum, H.B. Critical role of the hippocampus in
memory for sequence of events. Nat. Neurosci. 5, 458–462 (2002).
3. Squire, L.R., Stark, C.E. & Clark, R.E. The medial temporal lobe. Annu. Rev. Neurosci.
27, 279–306 (2004).
4. Hasselmo, M.E. & McClelland, J.L. Neural models of memory. Curr. Opin. Neurobiol. 9,
184–188 (1999).
5. Alvarez, P. & Squire, L.R. Memory consolidation and the medial temporal lobe: a simple
network model. Proc. Natl. Acad. Sci. USA 91, 7041–7045 (1994).
6. Ka
´
li, S. & Dayan, P. Off-line replay maintains declarative memories in a model of
hippocampal-neocortical interactions. Nat. Neurosci. 7, 286–294 (2004).
7. Siapas, A.G. & Wilson, M.A. Coordinated interactions between hippocampal ripples and
cortical spindles during slow-wave sleep. Neuron 21, 1123–1128 (1998).
8. Sirota, A., Csicsvari, J., Buhl, D. & Buzsa
´
ki, G. Communication between neocortex and
hippocampus during sleep in rodents. Proc. Natl. Acad. Sci. USA 100, 2065–2069
(2003).
9. Battaglia, F.P., Sutherland, G.R. & McNaughton, B.L. Hippocampal sharp wave bursts
coincide with neocortical up-state transitions. Learn. Mem. 11, 697–704 (2004).
10. Mo
¨
lle, M., Yeshenko, O., Marshall, L., Sara, S.J. & Born, J. Hippocampal sharp wave-
ripples linked to slow oscillations in rat slow-wave sleep. J. Neurophysiol. 96,62–70
(2006).
11. Wolansky, T., Clement, E.A., Peters, S.R., Palczak, M.A. & Dickson, C.T. Hippocampal
slow oscillation: a novel EEG state and its coordination with ongoing neocortical activity.
J. Neurosci. 26, 6213–6229 (2006).
12. Wilson, M.A. & McNaughton, B.L. Reactivation of hippocampal ensemble memories
during sleep. Science 265, 676–679 (1994).
13. Skaggs, W.E. & McNaughton, B.L. Replay of neuronal firing sequences in rat
hippocampus during sleep following spatial experience. Science 271, 1870–1873
(1996).
14. Kudrimoti, H.S., Barnes, C.A. & McNaughton, B.L. Reactivation of hippocampal cell
assemblies: effects of behavioral state, experience, and EEG dynamics. J. Neurosci. 19,
4090–4101 (1999).
15. Hoffman, K.L. & McNaughton, B.L. Coordinated reactivation of distributed memory
traces in primate neocortex. Science 297, 2070–2073 (2002).
16. Qin, Y.L., McNaughton, B.L., Skaggs, W.E. & Barnes, C.A. Memory reprocessing in
corticocortical and hippocampocortical neuronal ensembles. Phil. T rans. R. Soc. Lond.
B 352, 1525–1533 (1997).
17. Ribeiro, S. et al. Long-lasting novelty-induced neuronal reverberation during slow-wave
sleep in multiple forebrain areas. PLoS Biol. 2,24(2004).
18. Eichenbaum, H., Dudchunko, P., Wood, E., Shapiro, M. & Tanila, H. The hippocampus,
memory, and place cells: is it spatial memory or a memory space? Neuron 23, 209–226
(1999).
19. Jensen, O. & Lisman, J.E. Hippocampal sequence-encoding driven by a cortical multi-
item working memory buffer. Trends Neurosci. 28, 67–72 (2005).
20. Na
´
dasdy, Z., Hirase, H., Czurko
´
, A., Csicsvari, J. & Buzsa
´
ki, G. Replay and time
compression of recurring spike sequences in the hippocampus. J. Neurosci. 19,
9497–9507 (1999).
21. Lee, A.K. & Wilson, M.A. Memory of sequential experience in the hippocampus during
slow wave sleep. Neuron 36, 1183–1194 (2002).
22. Louie, K. & Wilson, M.A. Temporally structural replay of awake hippocampal ensemble
activity during rapid eye movement sleep. Neuron 29, 145–156 (2001).
23. Cossart, R., Aronov, D. & Yuste, R. Attractor dynamics of network up states in the
neocortex. Nature 423, 283–288 (2003).
24. Shu, Y., Hasenstaub, A. & McCormick, D.A. Turning on and off recurrent balanced
cortical activity. Nature 423, 288–293 (2003).
25. Sanchez-Vives, M.V. & McCormick, D.A. Cellular and network mechanisms of rhythmic
recurrent activity in neocortex. Nat. Neurosci. 3, 1027–1034 (2000).
26. Petersen, C.C., Hahn, T.T.G., Metha, M., Grinvald, A. & Sakmann, B. Interaction of
sensory responses with spontaneous depolarization in layer 2/3 barrel cortex. Proc. Natl.
Acad. Sci. USA 100, 13638–13643 (2003).
27. Volgushev, M., Chauvette, S., Mukovski, M. & Timofeev, I. Precise long-range synchro-
nization of activity and silence in neocortical neurons during slow-wave sleep.
J. Neurosci. 26, 5665–5672 (2006).
28. Amzica, F. & Steriade, M. Cellular substrates and laminar profile of sleep K-complex.
Neuroscience 82, 671–686 (1998).
29. Steriade, M., Timofeev, I. & Grenier, F. Natural waking and sleep states: a view from
inside neocortical neurons. J. Neurophysiol. 85, 1969–1985 (2001).
30. Csicsvari, J., Hirase, H., Mamiya, A. & Buzsa
´
ki, G. Ensemble patterns of hippocampal
CA3-CA1 neurons during sharp wave-associated population events. Neuron 28,
585–594 (2000).
31. O’Keefe, J. & Dostrovsky, J. The hippocampus as a spatial map: preliminary evidence
from unit activity in the freely-moving rat. Brain Res. 34, 171–175 (1971).
32. Hafting, T., Fyhn, M., Molden, S., Moser, M.B. & Moser, E.I. Microstructure of a spatial
map in the entorhinal cortex. Nature 436, 801–806 (2005).
33. 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).
34. Lee, A.K. & Wilson, M.A. A combinatorial method for analyzing sequential firing patterns
involving an arbitrary number of neurons based on relative time order. J. Neurophysiol.
92, 2555–2573 (2004).
35. Steriade, M., Nubez, A. & Amzica, F. A novel slow (o1 Hz) oscillation of neocortical
neurons in vivo: depolarizing and hyperpolarizing components. J. Neurosci. 13, 3252–
3265 (1993).
36. Steriade, M. & Amzica, F. Slow sleep oscillation, rhythmic K-complexes, and their
paroxysmal developments. J. Sleep Res. 7 (suppl. 1): 30–35 (1998).
37. Achermann, P. & Borbely, A.A. Low-frequency (o1 Hz) oscillations in the human sleep
EEG. Neuroscience 81, 213–222 (1997).
38. Hahn, T.T.G., Sakmann, B. & Mehta, M.R. Phase-locking of hippocampal interneurons’
membrane potential to neocortical up-down states. Nat. Neurosci. 9, 1359–1361
(2006).
39. Kosslyn, S.M. et al. The role of area 17 in visual imagery: convergent evidence from PET
and rTMS. Science 284, 167–170 (1999).
40. Wheeler, M.E., Petersen, S.E. & Buckner, R.L. Memory’s echo: vivid remembering reac-
tivates sensory-specific cortex. Proc. Natl. Acad. Sci. USA 97, 11125–11129 (2000).
41. Harris, J.A., Petersen, R.S. & Diamond, M.E. The cortical distribution of sensory
memories. Neuron 30, 315–318 (2001).
42. Suzuki, W.A. Encoding new episodes and making them stick. Neuron 50, 19–21 (2006).
43. McClelland, J.L. & Goddard, N.H. Considerations arising from a complementary learning
systems perspective on hippocampus and neocortex. Hippocampus 6, 654–665
(1996).
44. O’Reilly, R.C. & Rudy, J.W. Computational principals of learning in the neocortex and
hippocampus. Hippocampus 10, 389–397 (2000).
45. Lavenex, P. & Amaral, D.G. Hippocampal-neocortical interaction: a hierarchy of asso-
ciativity. Hippocampus 10, 420–430 (2000).
46. Rolls, E.T. Hippocampal-cortical and cortico-cortical backprojections. Hippocampus
10, 380–388 (2000).
47. Paxinos, G. & Watson, C. The Rat Brain in Stereotaxic Coordinates 4th edn. (Academic,
New York, 1998).
48. Robert, C., Guilpin, C. & Limoge, A. Automated sleep staging systems in rats.
J. Neurosci. Methods 88, 111–122 (1999).
49. Siapas, A.G., Lubenov, E.V. & Wilson, M.A. Prefrontal phase locking to hippocampal
theta oscillations. Neuron 46, 141–145 (2005).
50. Skaggs, W.E., McNaughton, B.L., Gothard, K.M. & Markus, E.J. An information-
theoretic approach to deciphering the hippocampal code. In Advances in Neural
Information Processing Systems Vol. 5 (eds. Hanson, S.J., Cowan, J.D. & Giles, C.J.)
1030–1037 (Morgan Kaufmann, San Mateo, California, USA, 1993).
NATURE NEUROSCIENCE VOLUME 10
[
NUMBER 1
[
JANUARY 2007 107
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/natureneuroscience