LETTER. Preplay of future place cell sequences by hippocampal cellular assemblies

Transcription

1 doi:./nature Preplay of future plae ell sequenes y hippoampal ellular assemlies George Dragoi & Susumu Tonegawa During spatial exploration, hippoampal neurons show a sequential firing pattern in whih individual neurons fire speifially at partiular loations along the animal s trajetory (plae ells, ). Aording to the dominant model of hippoampal ell assemly ativity, plae ell firing order is estalished for the first time during exploration, to enode the spatial experiene, and is susequently replayed during rest or slow-wave sleep for onsolidation of the enoded experiene,. Here we report that temporal sequenes of firing of plae ells expressed during a novel spatial experiene ourred on a signifiant numer of oasions during the resting or sleeping period preeding the experiene. This phenomenon, whih is alled preplay, ourred in disjuntion with sequenes of replay of a familiar experiene. These results suggest that internal neuronal dynamis during resting or sleep organize hippoampal ellular assemlies into temporal sequenes that ontriute to the enoding of a related novel experiene ourring in the future. We reorded neuronal firing sequenes from the CA area of the mouse hippoampus (Supplementary Fig. ) during periods of awake rest (Fam-Rest) alternating with periods of running (Fam-Run) on a familiar trak (Fam session; Supplementary Fig. a) that preeded the exploration of a novel linear arm in ontiguity with the familiar trak (Contig-Run on L-shaped trak; Fig., Supplementary Fig. a and Methods). All the plae ells ative on the novel arm during Contig- Run, whether previously silent (% in oth diretions and % in at least one diretion; Methods and Supplementary Tales ) or ative during Fam-Run (supanels a in Fig. ), fired during Fam-Rest at the ends of the familiar trak (range,.. Hz; Supplementary Fig. ) as part of a numer of spiking events. The spiking events were defined as epohs omposed of multiple individual spikes from at least four different plae ells ative on the novel arm or familiar trak, separated y less than ms and flanked y at least ms of silene,. More signifiantly, the temporal sequene in whih the ells ative on the novel arm fired during Fam-Rest (supanels in Fig. ) was signifiantly orrelated with the spatial sequene in whih they fired later as plae ells on the novel arm during Contig-Run (supanels in Fig. ), despite eing unorrelated with their spatial sequene as plae ells on the familiar trak during Fam-Run. This is illustrated as plae ell sequenes during Contig-Run (supanels in Fig. ) and Fam-Run (supanels a in Fig. ) ompared with the firing sequenes of these ells within individual spiking events oserved during Fam-Rest (supanels in Fig. ). We refer to this proess as preplay of plae ell sequenes eause the temporal sequene of firing during Fam-Rest had ourred efore the atual exploration of the novel arm in the susequent Contig-Run and was not a replay of the plae ell sequenes from the previous Fam-Run. To quantify the signifiane of preplay and to ompare it with replay, we reated plae ell sequene templates aording to the spatial order of the peak firing of plae ells,, on the novel arm during Contig-Run (novel arm templates; supanels in Fig. and Methods) and on the familiar trak during Fam-Run (familiar trak templates) for eah run diretion. The spikes of all the plae ells used to onstrut the two types of template that were emitted during Fam-Rest were sorted y time, and spiking events were determined as explained aove (supanels in Fig. ). For eah spiking event, we alulated a rank-order orrelation etween the novel arm templates and the temporal sequene of firing of the orresponding ells in the spiking events during Fam-Rest. The event orrelation was onsidered signifiant if it exeeded the.th perentile of a distriution of orrelations resulting from randomly shuffling the order of plae ells in the novel arm templates times (P,.). Forward and reverse, preplay refers to the ases in whih the sequene of plae ells during Contig-Run and the firing order of the orresponding ells in Fam-Rest were in the same and opposite diretions, respetively. In % of the preplay ases, the spiking events were orrelated with the novel arm template in one diretion only. The distriution of event orrelation values otained using the original novel arm templates was signifiantly shifted towards higher positive or negative values in omparison with the distriution of orrelation values otained using shuffled templates (Fig. a and Supplementary Fig. ). Figure a also shows the distriution of signifiant preplay events (in red). Of all the spiking events deteted as aove and in whih at least four novel arm plae ells were ative,.% were signifiant preplay events for the plae ell sequene on the novel arm (P,, inomial proaility test ) in the forward or reverse order (Fig. ). The ourrene of signifiant preplay events was orrelated with the ourrene of high-frequeny ripple osillations in CA (Fig. ). The majority of the signifiant preplay events (.%; Fig. d, total, lue) took plae at the juntion etween the familiar and novel arms, and the remaining.% took plae at the free end of the familiar trak (Fig. d, total, purple). The proportion of signifiant preplay events among the total events at eah of the two trak ends was higher at the juntional end (.%, P, ) than at the free end (.%, P, ) of the familiar trak (P,., Z-test; Fig. d, normalized). We found a relatively high orrelation etween the plae field maps (Fig. A, B and Supplementary Fig. ) of the familiar trak efore and after the novel experiene (median r.; Fig. e, familiar trak, lue); it was signifiantly higher than the orrelations otained when the ell identities were shuffled (median r., P, ; Fig. e, familiar trak, lak). A similar orrelation analysis showed a relatively high staility of the newly formed plae fields on the novel arm from the eginning to the end of Contig-Run (median r. (newly formed) versus median r. (shuffled), P, ; Fig. e, novel arm, lue versus grey). These results suggest that preplay of the novel arm does not our over an entirely new (that is, remapped) representation of the whole L-shaped trak ut rather enefits from the relative staility of the familiar trak representation aross sessions and perhaps failitates the rapid, stale enoding of the novel arm experiene. Using the familiar trak templates and spiking events during Fam- Rest, onstruted as aove, we determined that.% (P, ;data not shown) were signifiant replay events, among the spiking events in whih a minimum of four familiar trak plae ells were ative. All signifiant preplay events ourring during Fam-Rest (n ) were The Piower Institute for Learning and Memory, RIKEN-MIT Center for Neural Ciruit Genetis, Department of Biology and Department of Brain and Cognitive Sienes, Massahusetts Institute of Tehnology, Camridge, Massahusetts, USA. JANUARY VOL NATURE Mamillan Pulishers Limited. All rights reserved

2 RESEARCH LETTER A a Fam-Run Fam-Rest Contig-Run Familiar trak Familiar arm Cell numer B a Cell numer Cell numer Cell numer Fam-Run Familiar trak Fam-Run Familiar trak Fam-Rest tested for possile replay of the familiar trak spatial sequene: these spiking events were more orrelated with the novel arm template (Fig. f, red) than the familiar trak template (Fig. f, lue). Seventytwo perent (n ) of the signifiant events previously onsidered to e preplay had no signifiant orrelation with the familiar trak template. An additional % (n ) of those events were etter orrelated with the novel arm templates (mean asolute r.) than with the familiar trak template (mean asolute r., P, ). Together, these findings rejet the hypothesis that the preplay events simply represent a replay of the familiar trak ativity (see additional ontrols in Supplementary Information). Moreover, we found that the proportion of events exlusively omposed of silent ells that perfetly mathed the novel arm spatial templates was. ( of triplets), whih is signifiantly greater (P,.) than the proportion of y-hane perfet mathes (.). To illustrate the distriution and relative proportions of preplay and replay events among all signifiant spiking events during Fam-Rest, we Contig-Run Familiar arm Position Position C a Fam-Run Fam-Rest Contig-Run Da Fam-Run Fam-Rest Contig-Run Familiar trak Familiar trak Position Position Position Position E a Position ms Position Fam-Rest Contig-Run Position Position Figure Preplay of novel plae ell sequenes. Fam-Run and Fam-Rest respetively denote run and rest sessions on the familiar linear trak efore arrier removal; Contig-Run denotes run sessions on the L-shaped trak after arrier removal. The L-shape trak was linearized for display/analysis. A, B, mouse ; C, D, mouse ; E, mouse. A E, a, Spatial ativity on the familiar trak during Fam-Run of the ells that had plae fields in Contig-Run and preplayed during Fam-Rest (one ell per row); ativity on the novel arm and familiar trak are on the same sale. Horizontal arrows indiate run diretions. Vertial grey ars indiate arrier loations during Fam-Run and Fam-Rest. A E,, Examples of representative spiking events in the forward or reverse N AT U R E V O L J A N U A RY Mamillan Pulishers Limited. All rights reserved diretion during Fam-Rest in -ms time windows ( ms for the seond and fourth panels from left in E, ). Tik marks indiate individual spikes: red, preplay events for plae ell sequenes in the novel arm; lue (in A, and B, ), additional spikes from the familiar trak plae ells partiipating in the spiking event (not shown in C E, ). Numers on the left denote ell numers and orrespond to the plae ell numers in A E, a. Square oxes indiate the ends of the familiar trak where preplay events ourred. Loal field potentials reorded simultaneously with the spikes are shown aove spiking events. A E,, Plae ell sequenes in the novel arm (C E, ; red) or in oth the novel arm (red) and the familiar arm (lue) (A, and B, ) in Contig-Run. alulated a template speifiity index (Fig. g and Methods) for eah event. Pure preplay events (Fig. g, red) and pure replay events (Fig. g, lue) were segregated, and only a minority of events were signifiant for oth preplay and replay (Fig. g, yellow). Consistent with this segregation of preplay and replay events, the novel arm and the orresponding familiar trak templates were not signifiantly orrelated (Fig. h and Methods). The ratio etween the numer of pure replay events (n ) and the numer of pure preplay events (n ) during Fam-Rest was aout. (Fig. g, inset; see Supplementary Information for proportions of events). Preplay and replay events were distriuted in time aross Fam-Rest (Supplementary Fig. a ) and their ourrenes were generally unorrelated (Supplementary Fig. d). The majority (.%) of the spiking events during Fam-Rest did not signifiantly orrelate with either of the two templates (data not shown). We used a Bayesian reonstrution algorithm,,,, (Methods) to deode the animals position from the spiking ativity during Fam- Run (Fig. a) or Fam-Rest (Fig., ). For all original and shuffled

3 RESEARCH a Numer of events P < Familiar trak Data Shuffled Signifiant d. Proportion of preplay events Juntional end Opposite end lag (s) Data e f g h Shuffle. P < template Preplay/replay events % Familiar trak template Replay events % % Figure Quantifiation of the preplay phenomenon and omparison with replay. a, Distriution of orrelations etween spiking events in Fam-Rest and spatial templates of the novel arm. Open ars indiate spiking events versus the original (unshuffled) templates; filled ars indiate spiking events versus shuffled templates saled down times; red ars show the distriution of preplay (that is, signifiant) events. Similar distriutions (not shown) of orresponding spiking events were otained when spatial templates were onstruted using all plae ells ative on the L-shaped trak (Figs A,, and B,, ; red and lue)., Proportion of all, forward and reverse preplay events among the spiking events in Fam-Rest. The dotted line indiates the hane level (.%)., Cross-orrelation etween preplay events and ripple epohs. d, Loation of preplay events on the familiar trak: total, proportions of preplay events at ends of the trak; normalized, proportion of preplay events normalized y the numer of spiking events at eah end of trak. represented a trajetory running from the free end of the novel arm to the juntional end (%) or egun near the familiar trak (%); the latter suggests that in some ases preplay events ould e triggered y the ativity of the familiar trak plae ells during Fam-Rest. e, Staility of plae ell spatial tuning aross the novel experiene: familiar trak, staility of the plae fields ative on the familiar trak efore (Fam-Run) versus after (Contig-Run) arrier removal; novel arm, staility of the plae fields ative on the novel arm at the eginning Numer of events Numer of events.. All Forward Reverse Numer of ripples Proportion of preplay events... Total Normalized Template speifiity index Correlation P value (first four laps of run) versus the end (last four laps) of the Contig-Run session. Data (lue), within-ell orrelation of plae ell spatial tuning for the orresponding trak/arm; shuffle (lak), ell identity shuffle (Supplementary Information). Error ars, s.e.m.; asterisks in d and e indiate signifiant differenes. f, Distriution of preplay event orrelations (red) versus distriution of these event orrelations with the familiar trak template (lue). Spiking events were deteted using all plae ells from the familiar trak and novel arm templates (. Hz). Red ars are the same as in a. Correlation is strong with the novel arm template (preplay) and weak with the familiar arm template (replay). The P value orresponds to there eing a signifiant differene etween the two distriutions. g, Disjuntive distriution of pure preplay (red), pure replay (lue) and preplay/replay (yellow) events during Fam-Rest over their template speifiity index (Supplementary Information). Inset, proportions of pure preplay events (red), pure replay events (lue) and preplay/replay events (yellow) among all of the spiking events that were signifiantly orrelated with at least familiar trak templates or novel arm templates. h, Lak of orrelation etween the novel arm template and the orresponding familiar trak template. Eah of the six dots represents either a forward or a reverse run diretion of one of the three mie analysed. Red horizontal line denotes a P value of.. The orrelation values were not signifiant in any of the ases (Supplementary Information). proaility distriutions, a line was fitted to the data using a line-finding algorithm to represent the deoded virtual trajetory (Methods and Supplementary Information). In.% of ases representing trajetories, the reonstruted trajetory during spiking events in Fam- Rest was ontained within the novel arm (Fig., top), a plae the animal had not yet visited (that is, trajetory preplay). Moreover, in.% of the trajetory preplay ases the shuffling proedures resulted in lines that were signifiantly less or not at all ontained within the novel arm (that is, not preplay; Supplementary Information). The remaining trajetories deoded during Fam-Rest represented replay of the familiar trak (.%; Fig., middle) or spanned the joint familiar trak/novel arm spae (.%; Fig., ottom). Means of asolute rank-order orrelations etween spiking ativity and novel arm templates (Fig. a) restrited during epohs of trajetory preplay were signifiantly larger than those etween spiking ativity and familiar trak templates alulated during the same epohs (. versus., P, ). Overall, these results support the existene of the preplay phenomenon. To investigate the possiility that preplay of novel arm plae ell sequenes during Fam-Rest depends on the prior run experiene on the familiar trak, mie with no prior experiene on any linear trak were plaed in a high-walled sleep ox and reorded while resting/ sleeping. The animals were then transferred to a novel isolated linear trak that was in the same room ut ould not e seen from inside the ox, and the reording ontinued during de novo formation of plae ells (Supplementary Fig., de novo session). We found that in a relatively large proportion (.%) of spiking events identified during sleep/rest in the sleep ox, the neuronal firing sequenes were signifiantly orrelated with the plae ell sequenes oserved during the first run session on the novel trak (Fig. A, B and Methods); this was the ase for all four individual mie (Supplementary Fig. ). were assoiated with the ripple ourrene (Fig. C). The plae ells estalished on the novel trak in the de novo session were more dynami (median r.; Fig. D, lue) than in Contig-Run (median r., P,.; Fig. e, right, lue). We have demonstrated that a signifiant numer of temporal firing sequenes of CA ells during resting periods of a familiar trak exploration that preeded a novel trak exploration in the same general environment were orrelated with the plae ell sequenes of the novel trak rather than the familiar trak. This phenomenon, preplay, is temporally opposite to the proess of replay,,, when ativity during rest or sleep periods reapitulates plae ell sequenes that have already ourred during previous explorations. Preplay differs fundamentally from replay eause it ours efore exploration of novel traks. Although our reordings were arried out in CA, we elieve that what we oserved ould e a refletion of the output of the reurrent ellular assemlies from upstream regions (CA or entorhinal ortex). JANUARY VOL NATURE Mamillan Pulishers Limited. All rights reserved

4 RESEARCH LETTER a Familiar trak Run... Familiar trak Replay familiar trak Preplay novel arm and replay familiar trak (s) Rest (s) (s) Preplay novel arm Max. Novel arm Familiar trak Novel arm Familiar trak Novel arm Familiar trak Figure Bayesian reonstrution of the animal s trajetory in the familiar trak (replay) and novel arm (preplay). a, Position reonstrution of a one-lap run on the familiar trak from the ensemle plae ell ativity during Fam-Run. The heat map displays the reonstruted position of the animal using ensemle plae ell ativity during the run (-ms ins; animal veloity,.ms ). The yellow line indiates the atual trajetory of the animal during Fam-Run., Example of virtual trajetory reonstrution (familiar trak and novel arm) from the ensemle plae ell ativity during Fam-Rest at the ends of the familiar trak (-ms ins; animal veloity,,ms ) efore arrier removal and novel arm exploration. The yellow line reflets the spatial loation of the animal in time: the animal was immoile at the juntion end of the familiar trak. The time-ompressed (,ms ) trajetory reonstrution often jumps over the arrier (top of the figure) into the novel arm area. At around. s, a preplay of the novel arm initiated from the distal (free) end of the novel arm propagates towards the loation of the animal., Examples of preplay of the novel arm (top), replay of the familiar trak (middle) and preplay of the novel arm together with replay of the familiar trak (ottom) during Fam-Rest. All onditions are the same as in. The white line shows the linear fit maximizing the likelihood along the virtual trajetory. Colour ars indiate proaility of trajetory reonstrution. During running on a familiar trak, some of the ells in the postulated upstream ellular assemlies fire sequentially at spatial loations while others, although onneted anatomially to these ells, remain silent. The lak of expression of preplay sequenes during Fam-Run may reflet their state-dependent suppression or suthreshold ativation during these exploratory ehaviours. Owing to inreased net exitation during rest periods predominantly during ripples, some of these silent ells together with some of the familiar trak ells are ativated aove threshold and fire in a ertain sequene. Their sequene of ativation may e determined in part y their funtional onnetivity within the hippoampal formation network. Some of these sequenes may in turn e ativated on a novel trak as plae ell sequenes (Supplementary Fig. ). The ativation of the novel plae ell sequenes during running may strengthen their pre-existing assemly organization manifested during preplay. It ould e argued that during Contig-Run the animals simply onsidered the novel arm to e an extension of the familiar arm and, thus, what we onsidered to e preplay events were replays of the previous runs on the familiar trak. If this was the ase, preplay events would not A a Cell numer B Numer of events P < Pre-Run sleep/rest Data Shuffled Signifiant.. e expeted to e found when the experiene of the familiar trak run is eliminated. This idea was refuted y the demonstration of frequent preplay events in the sleep ox efore the mie were transferred onto a novel linear trak (de novo ondition). Under this ondition, the plae ell sequenes were more dynami and a higher proportion of all spiking events orrelated with the plae ell sequenes in these runs than in the later runs on novel linear traks. These results suggest a shift in the relative ontriution of internal, and external drives in the formation of plae ell sequenes on enounter with a novel trak. In the early phase, internal drives originating in the dynami ellular assemly ativities, whih proaly reflet numerous past experienes distint from the urrent one and expressed as preplay, may have a greater role, whereas in the late phase, external drives that ome from the speifi set of stimuli of the urrent experiene may dominate. Thus, plae ell sequenes on novel traks seem to e produts of a dynami interplay etween the internal and external drives. Several previous studies did not reveal preplay,,,. Although it is diffiult to pinpoint the apparent disrepanies etween these studies and the present one, we suggest that the use of insuffiiently sensitive methods (pairwise orrelations) y some studies,, and small sample sizes y others might have preluded detetion of preplay in previous work (see Supplementary Information for details). Data from the de novo ondition (Fig. ), in whih we oserved an even higher proportion of preplay events, have not een reported previously. Our data showed that novel preplay events oexist in disjuntion with familiar replay events during the rest periods on the familiar trak. This and the finding that these preplay and replay events together make up fewer than one-quarter of all deteted spiking events suggest that they are part of a dynami repertoire of temporal sequenes in the hippoampus that are past-experiene dependent (replay) or future-experiene expetant (preplay). Post-experiene replay of plae ell sequenes during resting or slow-wave sleep has een proposed to have an important role in memory onsolidation,. The temporal preplay of new plae ell sequenes during resting or sleep is onsistent with a preditive funtion for the hippoampal formation and may ontriute to aelerating learning C Numer of ripples.. lag (s) De novo-run Novel trak ms Position D. *.. Data Shuffle Novel trak Figure Preplay of novel plae ell sequenes efore any linear trak experiene. A, Sleep/rest session in the sleep ox (Pre-Run sleep/rest) efore the first run session on a linear trak (De novo-run). Display format is the same as in Fig.. A, a, Representative spiking events in the forward or reverse order during Pre-Run sleep/rest in -ms time windows. A,, Plae ell sequenes on the novel trak (red) during the De novo-run session. Eah row represents one ell in whih the ativity was normalized to the maximum firing rate. One run diretion in one animal is shown. The median numer of plae ells ative on the novel trak partiipating in preplay events is six. B, Distriution of spiking events in Pre-Run sleep/rest as a funtion of the rank-order orrelation with the plae ell sequene template of the novel trak. Display format is the same as in Fig. a. C, Cross-orrelation etween preplay events and ripple epohs during Pre-Run sleep/rest. D, Staility of plae ell spatial tuning aross the novel trak experiene. Display format is the same as in Fig. e (novel arm). Error ars, s.e.m.; asterisk indiates signifiant differene. NATURE VOL JANUARY Mamillan Pulishers Limited. All rights reserved

5 RESEARCH when a new experiene is introdued in multiple steps of inreasing novelty. METHODS SUMMARY We reorded plae ells from the CA area of the hippoampus with six independently movale tetrodes in four mie during sleep/rest sessions in the sleep ox efore any experiene on linear traks and during the first run session on a novel trak. Following familiarization with the linear trak, animals were susequently allowed to explore a ontinuous (L-shaped) trak in whih the now familiar trak and a new novel arm were made ontiguous. To quantify the signifiane of the preplay and replay proesses, spiking events in whih at least four ells were ative were deteted during sleep/rest (speed,,ms ) periods in the sleep ox or awake rest (speed,,ms ) periods at the ends of the familiar trak and novel arm, predominantly during ripple epohs. We alulated statistial signifiane at the P,. level for eah event y omparing the rank-order orrelation etween the event sequene and the plae ell sequene (template) with the distriution of orrelation values from templates otained y shuffling the original order of the plae ells. Proportions of signifiant events were alulated as the ratio etween the numer of signifiant events and the total numer of spiking events. We alulated the overall signifiane of preplay or replay proesses y omparing the distriution of orrelation values of all events with the distriution of orrelation values of shuffled templates (Kolmogorov Smirnov test). The signifiane of the proportion of signifiant events out of the total numer of spiking events was determined as the inomial proaility of oserving the numer of signifiant events (as suesses) from the total numer of spiking events (as independent trials), with a proaility of suess of. in any given trial. We reonstruted the position of the animal from the spiking ativity emitted during resting periods using Bayesian deoding proedures. Full Methods and any assoiated referenes are availale in the online version of the paper at Reeived Deemer ; aepted Otoer. Pulished online Deemer.. O Keefe, J. & Nadel, L. The Hippoampus as a Cognitive Map (Oxford Univ. Press, ).. Wilson, M. A. & MNaughton, B. L. Dynamis of the hippoampal ensemle ode for spae. Siene, ().. Foster, D. J. & Wilson, M. A. Reverse replay of ehavioural sequenes in hippoampal plae ells during the awake state. Nature, ().. Dia, K. & Buzsaki, G. Forward and reverse hippoampal plae-ell sequenes during ripples. Nature Neurosi., ().. Karlsson, M. P. & Frank, L. M. Awake replay of remote experienes in the hippoampus. Nature Neurosi., ().. Davidson, T. J., Kloosterman, F. & Wilson, M. A. Hippoampal replay of extended experiene. Neuron, ().. Wilson, M. A. & MNaughton, B. L. Reativation of hippoampal ensemle memories during sleep. Siene, ().. Skaggs, W. E. & MNaughton, B. L. Replay of neuronal firing sequenes in rat hippoampus during sleep following spatial experiene. Siene, ().. Nádasdy, Z., Hirase, H., Czurko, A., Csisvari, J. & Buzsaki, G. Replay and time ompression of reurring spike sequenes in the hippoampus. J. Neurosi., ().. Lee, A. K. & Wilson, M. A. Memory of sequential experiene in the hippoampus during slow wave sleep. Neuron, ().. Buzsáki, G. Two-stage model of memory trae formation: a role for noisy rain states. Neurosiene, ().. Nakashia, T., Buhl, D. L., MHugh, T. J. & Tonegawa, S. Hippoampal CAoutput is ruial for ripple-assoiatedreativation and onsolidationof memory. Neuron, ().. He, D. O. The Organization of Behavior: A Neuropsyhologial Theory (Wiley, ).. Harris, K. D., Csisvari, J., Hirase, H., Dragoi, G. & Buzsaki, G. Organization of ell assemlies in the hippoampus. Nature, ().. Dragoi, G. & Buzsaki, G. Temporal enoding of plae sequenes y hippoampal ell assemlies. Neuron, ().. Thompson, L. T. & Best, P. J. Plae ells and silent ells in the hippoampus of freely-ehaving rats. J. Neurosi., ().. O Neill, J., Senior, T. & Csisvari, J. Plae-seletive firing of CA pyramidal ells during sharp wave/ripple network patterns in exploratory ehavior. Neuron, ().. Zhang, K., Ginzurg, I., MNaughton, B. L. & Sejnowski, T. J. Interpreting neuronal population ativity y reonstrution: unified framework with appliation to hippoampal plae ells. J. Neurophysiol., ().. Johnson, A. & Redish, A. D. Neural ensemles in CA transiently enode paths forward of the animal at a deision point. J. Neurosi., ().. Kudrimoti, H. S., Barnes, C. A. & MNaughton, B. L. Reativation of hippoampal ell assemlies: effets of ehavioral state, experiene, and EEG dynamis. J. Neurosi., ().. Csisvari, J., Hirase, H., Czurko, A., Mamiya, A. & Buzsaki, G. Osillatory oupling of hippoampal pyramidal ells andinterneurons in theehaving rat. J. Neurosi., ().. Dragoi, G., Harris, K. D. & Buzsaki, G. Plae representation within hippoampal networks is modified y long-term potentiation. Neuron, ().. Pastalkova, E., Itskov, V., Amarasingham, A. & Buzsaki, G. Internally generated ell assemly sequenes in the rat hippoampus. Siene, ().. Blak, J. E. & Greenough, W. T. Advanes in Developmental Psyhology (Lawrene Erlaum, ).. Hassais, D., Kumaran, D., Vann, S. D. & Maguire, E. A. Patients with hippoampal amnesia annot imagine new experienes. Pro. Natl Aad. Si. USA, ().. Tse, D. et al. Shemas and memory onsolidation. Siene, (). Supplementary Information is linked to the online version of the paper at Aknowledgements We thank M. A. Wilson for assistane with data aquisition, disussions and omments on an earlier version of the manusript; J. O Keefe, A. Siapas, F. Kloosterman, D. L. Buhl for omments on earlier versions of the manusript; and F. Kloosterman for providing assistane with the line detetion for the Bayesian deoding. This work was supported y NIH grants R-MH and P-MH to S.T., who was an HHMI Investigator in an earlier part of this study. Author Contriutions S.T. and G.D. oneived the projet jointly. G.D. designed and performed the experiments and the analyses. G.D. and S.T. wrote the paper. Author Information Reprints and permissions information is availale at The authors delare no ompeting finanial interests. Readers are welome to omment on the online version of this artile at Correspondene and requests for materials should e addressed to G.D. (gdragoi@mit.edu) or S.T. (tonegawa@mit.edu). JANUARY VOL NATURE Mamillan Pulishers Limited. All rights reserved

6 RESEARCH LETTER METHODS Surgery and experimental design. Eletrophysiologial reordings were performed on four CBL/ mie (strain NR-floxed ) with ages etween and weeks. All animals were implanted under Avertin anaesthesia with six independently movale tetrodes aiming for the CA area of the right hippoampus (. mm posterior to regma and mm lateral to the midline; Supplementary Fig. ). The referene eletrode was implanted posterior to lamda over the ereellum. During the following week of reovery, the eletrodes were advaned daily while animals rested in a small, walled sleeping ox ( m, m high). The animal position was monitored y means of two infrared diodes attahed to the headstage. The experimental apparatus onsisted of a m retangular, walled, linear trak maze. All traks were m wide at the ottom and m wide at the top, and all linear trak walls were m high. Experimental sessions were onduted while the animals explored for hoolate sprinkle rewards plaed always at the ends of the orresponding linear traks (one sprinkle at eah end of the trak on eah lap). Neuronal ativity was reorded in naive animals (four mie) during the sleep/rest session in the sleep ox immediately preeding the first experiene on linear traks, and ontinued (Fig. ) during the first run session on a novel trak. After familiarization with the linear trak, the animals went through a reording session of min (Fam session), and the reordings ontinued for the next min (Contig session) while the animals explored an L-shaped trak for the first time. In this trak, the familiar arm and the novel arm were made ontiguous y removing the arrier that had separated them (Fig. ). For the purpose of analysing the reording data, the Fam session was further divided into Fam-Run, in whih the animals ran through the trak (veloity of animal s movement was higher than m s ), and Fam-Rest, where the animals took awake rests at the ends of the trak (veloity of animal s movement was less than m s ). During resting periods, the animals onsumed the hoolate sprinkle and groomed, ut mostly they were still until they self-initiated the next lap of run on the linear trak. After ompletion of the experiments, the rains of all mie were perfused, fixed, setioned and stained using nulear fast red (Supplementary Fig. ) or resyl violet for eletrode trak reonstrution. Reordings and single-unit analysis. A total of neurons were reorded from the CA area of the hippoampus in four mie during the Fam and Contig sessions (Supplementary Tales ). A total of CA neurons were reorded from the four mie in the de novo ondition (,, and ells, respetively). Single ells were identified and isolated using the manual lustering method Xlust and the appliation of luster quality measurements. Pyramidal ells were distinguished from interneurons on the asis of spike width, average rate and autoorrelations. Plae fields were omputed as the ratio etween the numer of spikes and the time spent in -m ins along the trak, smoothed with a Gaussian kernel with a standard deviation of m. Bins where the animal spent a total of less than. s and periods during whih the animal s veloity was elow m s were exluded. Plae field length and peak rate were alulated after separating the diretion of movement and linearizing the trajetory of the animal. Linearized plae fields were defined as areas with a loalized inrease in firing rate aove Hz for at least five ontiguous ins ( m). The plae field peak rate and loation were given y the rate and loation of the in with the highest ratio etween spike ounts and time spent. Plae field orders were defined as the points where the firing rate eame less than % of the peak firing rate or Hz (whihever was igger) for at least m. Loal field potential analysis. Ripple osillations were deteted during sleep/rest periods in the sleep ox and during rest periods at the ends of the traks. The eletroenephalography signal was filtered ( Hz) and ripple-and amplitude was omputed using the Hilert transform. Ripple epohs with maximal amplitude more than s.d. aove the mean, eginning and ending at s.d. were deteted. The time of ripple ourrene (Figs and C) was the time of its maximal amplitude. The proportion of ripples with whih ells with plae fields on the novel arm of the L-shaped trak fired in the preeding session (Supplementary Fig. ) was alulated for eah qualifying ell as the ratio etween the numer of ripples during whih the ell fired at least one spike and the total numer of ripples during the orresponding exploratory session. Preplay and replay analyses. To analyse the preplay and replay proesses, spiking events were deteted during Pre-Run sleep/rest periods in the sleep ox (de novo ondition; veloity,,ms ) or during awake rest periods at the ends of the running traks (Contig ondition; veloity,,ms ). A spiking event was defined as a transient inrease in the firing ativity of a population of at least four different plae ells within a temporal window preeded and followed y at least ms of silene. Overall, similar results were otained using -, -, - and -ms time windows. The spikes of all the plae ells ative on the novel trak that were emitted during the Pre-Run sleep/rest in the ox for the de novo ondition as well as the spikes of all the plae ells ative on the familiar trak or the novel arm that were emitted during Fam-Rest session at the two ends of the familiar trak for Contig ondition were respetively sorted y time and further used for the detetion of the spiking events. All four animals exhiited a signifiant numer of spiking events in the Pre-Run session of the de novo ondition. Three of the four animals (mie ) exhiited a signifiant numer of spiking events in the Contig ondition, the remaining animal (mouse ) having a elow-threshold numer of simultaneously ative CA plae ells. The time of the spiking event used to ompute the ross-orrelation with ripple epoh ourrene (Figs and C) was the average time of all spikes omprising the spiking event. The plae ell sequenes (templates) were alulated for eah diretion of the animal s movement and for eah run session (De novo-run, Fam-Run and Contig-Run) y ordering the spatial loation of the plae field peaks that were aove Hz. For plae ells with multiple plae fields aove Hz on a partiular arm or trak in the Contig ondition (six of plae ells ative on the novel arm in the two diretions, or %: two for eah diretion in mouse, one in mouse and one in mouse ), only the plae field orresponding to the peak firing rate of the plae ell on that arm or trak was onsidered for the onstrution of the template of that partiular arm or trak, to e onsistent with all the previous studies that used spatial templates to demonstrate replay during sleep or awake rest,,. Plae ells with fields on oth the novel arm in the Contig-Run session and the familiar trak in the Fam-Run session partiipated in the onstrution of oth the novel arm and familiar trak templates. Statistial signifiane was alulated for eah event y omparing the rankorder orrelation etween the sequene of ells firing during the event (that is, event sequene) and the plae ell sequene (template), on the one hand, and the distriution of orrelation values etween the event sequene and surrogate templates otained y shuffling the order of plae ells, on the other (Fig. a). The signifiane level was set at. to ontrol for multiple omparisons (two diretions of run). The proportions of signifiant events (preplay novel trak, preplay novel arm (Fig. ), replay novel arm and replay familiar trak) were eah alulated as the ratio etween the numer of signifiant events and the total numer of spiking events in whih at least four orresponding plae ells were ative. Corresponding familiar trak templates (Fig. h) were onstruted y ordering the loation of peak firing on the familiar trak during Fam-Run (no minimum threshold of firing) of all plae ells that susequently fired on the novel arm. Cells omprising the orresponding familiar trak templates are the same as those omprising the novel arm templates. We note that these orresponding familiar trak templates are different from the ones used in Figs and a g, whih were onstruted y ordering the peak firing of all plae ells ative on the familiar trak. Hz. The overall signifiane of the preplay (Fig. a) or replay proess was alulated y omparing the distriution of orrelation values of all events relative to the original template with the distriution of orrelation values relative to the shuffled surrogate templates, using the Kolmogorov Smirnov test. Quantifiation of the replay versus preplay events during the Fam-Run session (Fig. f, g) was performed as desried aove using different spatial templates for the familiar trak and the novel arm. All spiking events were orrelated with oth the novel arm and the familiar trak templates. Events signifiantly orrelated only with familiar trak or with novel arm templates were onsidered pure replay and pure preplay, respetively. The template speifiity index was alulated for eah event as the differene etween the asolute value of the event s orrelation with the novel arm template (preplay, high positive index) and the event s orrelation with the familiar trak template (replay, high negative index). For the purpose of displaying the template speifiity index, events orrelated with the novel arm ut not with the familiar trak templates were onsidered preplay and events orrelated with the familiar trak ut not with the novel arm templates were onsidered replay (Fig. g). Additionally, events orrelated with oth the familiar trak and the novel arm templates formed a third group, preplay/replay events, displayed in yellow in the inset of Fig. g. Correlations etween pairs of familiar trak and novel arm templates (Fig. h) were performed using modified familiar trak templates that were onstruted using the loation of peak firing (. Hz) of only those ells that had plae fields on the novel arm (peak rate,. Hz). The lak of signifiant orrelation in this ase demonstrates that the novel arm plae ell sequene is not simply a transposition of a familiar trak plae ell sequene on the novel arm. We also identified neurons that did not fire during Fam-Run, that ativated during Fam-Rest events and that orresponded to trajetories on the novel arm during Contig-Run (silent ells). We alulated the orrelation etween the order in whih they fired during Fam-Rest events and their spatial sequene as new plae ells on the novel arm during Contig-Run, as previously explained. Owing to the low asolute numer of silent neurons, only triplets of ells were availale for further analysis (n ). The proportion of events perfetly mathing the spatial template was ompared with the proportion of y-hane perfet mathing (.). Mamillan Pulishers Limited. All rights reserved

7 RESEARCH Staility of plae ell maps. Stailities of plae ell firing on the familiar trak efore and after arrier removal as well as on the novel trak (de novo ondition) and the novel arm (Contig ondition) in the eginning versus the end of the run session were assessed y alulating, for eah plae ell and eah diretion, a orrelation etween the spatial firing in the orresponding paired situations (efore versus after arrier removal for the familiar trak or the first four laps versus the last four laps of the De novo-run or Contig-Run session for the novel trak or arm, respetively). The plae ell ativity was not partitioned in plae fields; rather, the whole ativity on the partiular trak or arm was onsidered separately for eah ell and diretion (average orrelations are shown in Figs e and D, lue ars). In addition, we alulated the same type of orrelation after shuffling the identity of the ell in one memer of the orrelation (one for eah different ell; average orrelations are in Figs e and D, lak ars). Shuffle results (Figs e and D, lak ars) were omputed as orrelation etween spatial tuning of ells on the familiar trak during Fam-Run and spatial tuning of all other simultaneously reorded ells on the familiar arm during Contig-Run (familiar trak group; Fig. e, left), or orrelation etween spatial tuning of ells on the novel arm (or novel trak) during the eginning of Contig-Run (or De novo-run) and spatial tuning of all the other simultaneously reorded ells on the novel arm (or novel trak) during the end of Contig-Run (novel arm group; Fig. e, right) or De novo-run (Fig. D). Original and shuffled orrelations were ompared using the rank-sum test. The average numer of laps (traversal of the novel trak in oth diretions) per session was. in De novo-run (,, and in the four mie) and. in Contig-Run (, and in the three mie). Bayesian reonstrution of atual and virtual trajetories. For eah ell, we alulated a linearized spatial tuning urve on the familiar trak during the Fam-Run session and a linearized spatial tuning urve on the novel arm during the Contig-Run session. The tuning urves were onstruted in -m ins from spikes emitted in oth run diretions at veloities higher than m s, and were smoothed with a Gaussian kernel with a standard deviation of m. We onstruted a joint spatial tuning urve for eah ell y juxtaposing the spatial tuning urve on the familiar trak during the Fam-Run session and the spatial tuning urve on the novel arm during the Contig-Run session. We also deteted for eah ell all the spiking ativity emitted at veloities elow m s during the Fam- Rest session, where replay and preplay events where shown to our using the rank-order orrelation method. We used a Bayesian reonstrution algorithm, to deode the virtual position of the animal from the spiking ativity during Fam- Rest (Fig. ) in non-overlapping, -ms ins using the joint spatial tuning urves. We then extrated epohs of reonstruted trajetory mathing the time of the spiking events as deteted using multiunit ativity of plae ells from the familiar trak and novel arm (rank-order orrelation method; see Preplay and replay analyses, aove). We used two shuffling proedures to measure the quality of the Bayesian deoding. In the first shuffling proedure, for eah event, the original time-in olumns of the proaility distriution funtion (PDF) were replaed with an equal numer of time-in olumns randomly extrated from a pool ontaining the time-in olumns of all PDFs of all deteted events. The shuffling proedure was repeated times. In the seond shuffling proedure, the identity of the plae ells was randomly shuffled times and new PDFs were alulated for all events. For all original and shuffled PDFs, a line was fitted to the data using a previously desried line-finding algorithm. Lines fitted to the original and shuffled data were ompared using slope, spatial extent, loation on the trak and proaility sore. We defined replay and preplay as the epohs of Fam-Rest in whih the reonstruted trajetory was loated on the familiar trak or the novel arm, respetively. The trajetory was defined aross a set of position estimates during the orresponding epoh (Fig. ). Only epohs that lasted at least ms (three ins) and whih ontained reonstruted trajetories spanning at least m were onsidered for further analysis. Trajetories for whih % or more of their length was loated on the familiar trak were onsidered to represent replay of an animal s trajetory on the familiar trak (Fig., middle), and trajetories for whih % or more of their length was loated in the novel arm were onsidered to represent preplay of the animal s future trajetory on the novel arm (Fig., top). The remaining events were onsidered preplay replay (Fig., ottom). An epoh was onsidered signifiant if the new line was less than % ontained in the familiar trak for replay or novel arm for preplay in at least % of the shuffled ases. For eah epoh that was signifiant for replay or preplay using the reonstrution method, we retrieved the value of the rank-order orrelation etween the neuronal firing sequenes and the familiar trak and novel arm spatial templates as alulated using the rank-order orrelation method. We ompared the asolute orrelation values etween the epoh s firing sequenes and familiar trak templates with the asolute orrelation values etween the same epoh s firing sequenes and novel arm templates. We also reonstruted the trajetory of the animal on the familiar trak from the spiking ativity during the Fam-Run session at veloities aove m s in -ms ins using the spatial tuning urves on the familiar trak, (Fig. a) to validate the deoding proedure.. Tsien, J. Z., Huerta, P. T. & Tonegawa, S. The essential role of hippoampal CA NMDA reeptor-dependent synapti plastiity in spatial memory. Cell, ().. Shmitzer-Torert, N., Jakson, J., Henze, D., Harris, K. & Redish, A. D. Quantitative measures of luster quality for use in extraellular reordings. Neurosiene, (). Mamillan Pulishers Limited. All rights reserved