Direct Medial Entorhinal Cortex Input to Hippocampal CA1 Is Crucial for Extended Quiet Awake Replay

Transcription

1 Direct Medial Entorhinal Cortex Input to Hippocampal CA1 Is Crucial for Extended Quiet Awake Replay The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters. Citation As Published Publisher Yamamoto, Jun, and Susumu Tonegawa. Direct Medial Entorhinal Cortex Input to Hippocampal CA1 Is Crucial for Extended Quiet Awake Replay. Neuron 96, no. 1 (September 2017): e4. Elsevier BV Version Author's final manuscript Accessed Sun Apr 07 02:55:59 EDT 2019 Citable Link Terms of Use Detailed Terms Creative Commons Attribution-NonCommercial-NoDerivs License

2 Published as: Neuron September 27; 96(1): e4. Direct Medial Entorhinal Cortex Input to Hippocampal CA1 is Crucial for Extended Quiet Awake Replay Jun Yamamoto 1,4,* and Susumu Tonegawa 1,2,3,5,* 1 RIKEN-MIT Center for Neural Circuit Genetics at The Picower Institute for Learning and Memory, Department of Biology and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA 02139, U.S.A 2 Howard Hughes Medical Institute, Massachusetts Institute of Technology, Cambridge, MA 02139, U.S.A 3 RIKEN Brain Science Institute, Wako, Saitama , Japan Summary Hippocampal replays have been demonstrated to play a crucial role in memory. Chains of ripples (ripple bursts) in CA1 have been reported to co-occur with long-range place cell sequence replays during the quiet awake state, but roles of neural inputs to CA1 in ripple bursts and replays are unknown. Here we show that ripple bursts in CA1 and medial entorhinal cortex (MEC) are temporally associated. An inhibition of MECIII input to CA1 during quiet awake reduced ripple bursts in CA1 and restricted the spatial coverage of replays to a shorter distance corresponding to single ripple events. The reduction did not occur with MECIII input inhibition during slow-wave sleep. Inhibition of CA3 activity suppressed ripples and replays in CA1 regardless of behavioral state. Thus, MECIII input to CA1 is crucial for ripple bursts and long-range replays specifically in quiet awake, whereas CA3 input is essential for both, regardless of behavioral state. Introduction When an animal moves through the environment, pyramidal cells in the hippocampus CA1 area fire in a location-specific manner (place cells) (O Keefe and Dostrovsky, 1971). These place cells become reactivated during off-line periods (Pavlides and Winson, 1989; Wilson and McNaughton, 1994; Skaggs and McNaughton, 1996). Moreover, the firing sequences of place cells during running behavior are re-expressed at an accelerated rate during subsequent slow-wave sleep (Kudrimoti et al., 1999; Nadasdy et al., 1999; Lee and Wilson, 2002) or quiet awake pauses in locomotion (Foster and Wilson, 2006; Diba and Buzsaki, 2007). This replay of place cells co-occurs with short-lasting (50~100 ms), high-frequency oscillations * Correspondence: yamajun@mit.edu or Jun.Yamamoto@UTSouthwestern.edu, and tonegawa@mit.edu. 4 Present address: The University of Texas Southwestern Medical Center, Department of Psychiatry, Dallas, TX 75390, U.S.A. 5 Lead Contact Author Contributions Conceptualization, J.Y. and S.T.; Methodology, J.Y. and S.T.; Software, J.Y.; Formal Analysis, J.Y.; Resouces, S.T.; Writing Original Draft, J.Y. and S.T.; Writing Review & Editing, J.Y. and S.T.; Visualization, J.Y. and S.T.; Funding Acquisition, S.T.; Supervision, S.T.;

3 Yamamoto and Tonegawa Page 2 Results (100~200 Hz) called sharp-wave ripples or ripples (Buzsaki et al., 1992) in the local field potential (LFP). These sharp-wave ripple associated replays have been reported during slowwave sleep or non-rem sleep (Lee and Wilson, 2002; Ji and Wilson, 2007; O Neill et al., 2017), and quiet awake (Foster and Wilson, 2006; Jackson et al., 2006; Diba and Buzsaki, 2007; Davidson et al., 2009; Karlsson and Frank, 2009; Singer and Frank, 2009; Gupta et al., 2010; Carr et al., 2011; Carr et al., 2012; Olafsdottir et al., 2016; O Neill et al., 2017; Wu et al., 2017). The replay of hippocampal activity patterns has also been reported in relation to memory tasks (Girardeau et al., 2009; Ego-Stengel and Wilson, 2010; McNamara et al., 2014; de Lavilleon et al., 2015; Maingret et al., 2016). When the linear track explored by an animal is relatively short (around 1 meter), the firing sequence of a set of place cells covering the entire track can be replayed within a single ripple event of ms. However, for a longer track, the replay of an extended place cell sequence of the track during quiet awake spans multiple ripple events called ripple-bursts that span 200~500 ms (Davidson et al., 2009). Anatomically, CA1 primarily receives excitatory input from CA3, and from the medial entorhinal cortex layer III (MECIII) (Witter et al., 2000). The potential roles of these inputs to CA1 in the formation of ripple bursts, and extended replays are unknown. In this study, we addressed these issues by first characterizing ripple bursts in mice when they are exposed to a long (> 6 meters) track. We then investigated the effect of optogenetic or tetanus toxin-mediated inhibition of MECIII or CA3 input on ripples, ripple bursts, and long-distance replays during quiet awake or slowwave sleep in CA1. Our studies revealed differential roles of MECIII and CA3 inputs on these CA1 activities during the animal s different behavioral states. Hippocampal Ripple Bursts within a Long-Track Experience We investigated the features of ripple bursts and ripple-associated replays after mice ran on a long folded linear track (6.3 meters long, Figure 1A). Throughout the behavior and rest periods, we recorded neuronal oscillations and spiking activities from dorsal hippocampal CA1 (Figures 1B). We detected and classified CA1 sharp-wave ripples into singlet, doublets or triplets, using both LFP and multi-unit activities (MUA) (Figures 1C, S1A, S1B, S1C and S1D, see STAR Methods). Each recording session consisted of three parts. The animals were first kept in a sleep box for 30 to 45 minutes (Pre RUN sleep) before they were allowed to run on the track to receive food rewards at both ends. After each running session, they were put back into the sleep box for another minute (Post RUN sleep) session (Figure S1C). To differentiate the animal s quiet awake or slow-wave sleep states from the recorded data, we used delta-to-theta power ratio on hippocampal LFP signals as well as the measurement of animal s head movement (Figure S1C, see STAR Methods). Although there was a tendency of more ripple bursts during slow-wave sleep than quiet awake, there was no significant difference in proportion of ripple burst types (Figure 1D). However, the sharpwave ripple occurrence during slow-wave sleep was significantly more frequent than quiet awake (Buzsaki, 1986) (Figure 1E). Likewise, the distribution of ripple burst occurrence of each type was significantly more frequent during slow-wave sleep than quiet awake (Figure 1F). We did see enhanced ripple bursts (especially in triplets) when animals experienced the

4 Yamamoto and Tonegawa Page 3 long track compared to a short (1 meter) track or small open field (40 cm 50 cm) (Figure S1E). The distribution of inter-ripple-intervals during ripple bursts showed a monotonous Poisson distribution but there were local peaks observed at around 110, 220, 330, 440 ms respectively, which is a similar time constant to the hippocampal theta oscillation (Figure 1G). The spatial positions of each detected ripple burst type mostly occurred either at both ends of the linear track where animals received food reward or at the corners of the folded track where animals frequently made pauses during RUN behavior (Figures 1H, 1I and 1J). Alternating Ripple Events between MEC and CA1 in Quiet Awake Ripple Bursts We then simultaneously recorded from CA1 and the superficial layers of MEC to further characterize the ripple bursts in light of entorhinal hippocampal network interactions (Witter et al., 2000) (Figure S2A). Strikingly, the ripple bursts detected in CA1 during quiet awake (nearly 50% of detected ripple events: 4837/10274 collected from five subjects) were closely alternating with ripple burst-like activity in MEC (Figures 2A, 2B and S2B) but not during slow-wave sleep (Figures 2C and S2C). To examine the temporal relationship of such burst activities in the two brain areas, we computed the cross-correlation of CA1 ripple-band peak power to MEC ripple-band peak power as well as CA1 MUA to MEC ripple-band LFP. We identified negatively skewed and periodical peaks; MEC ripple-band power increased earlier than that of CA1 during quiet awake ripple bursts (Figure 2D). However, this negative skewness and periodical peaks were not only less clear but also more positively skewed during slow-wave sleep (Figure 2E). The first individual ripple event in superficial MEC layers preceded CA1 ripples by ms (Figure S2D). This was also true for doublet and triplet bursts during quiet awake, respectively, however these peaks were not clear during slow-wave sleep (Figure S2E). Granger causality analysis (Seth et al., 2015) revealed that directional effects from MEC to CA1 and from CA1 to MEC were both significant (~50% of ripple burst-like activity and ~23% of detected ripples: 2385/4837), with the former greater than the latter (model order: 32, lag 256 ms, Figure 2F,). This suggests that MEC ripple bursts are likely the driver of the ripple alternations during quiet awake. This temporal alternation feature of quiet awake ripple bursts was not observed during slow-wave sleep (less than 5%, 436/9348 events, Figure 2C); and the Granger causality analysis was not significant in either direction (Figure 2G). MECIII Input Blockade to CA1 Selectively Reduces Ripple Bursts during Quiet Awake but Not during Slow-Wave Sleep Next, we investigated the role of MECIII pyramidal cell input to CA1 during quiet awake or slow-wave sleep in the generation of ripple bursts and replay using optogenetics. For this purpose, we bilaterally injected Cre-dependent AAVrh8-hSyn1-DIO-eArchT-eYFP into MECIII-specific Cre mice (poxr1-cre) and unilaterally implanted a pair of silicone linear probes paired with an optic fiber into the dorsal hippocampus and MEC (Figure 3A, see STAR Methods). We have previously shown that in these mice the input from MEC layer III to dorsal CA1 is inhibited by activating earcht with yellow-green light (561 nm, 20 mw) through the optic fiber (Yamamoto et al., 2014). Using this system, we placed animals in an open field after exposure to the long-track and monitored the animal s behavior states (motion history and LFP delta/theta power) to identify quiet awake and slow-wave sleep periods (Figure 3B, same criteria as in Figure 1, see STAR Methods). We then performed the

5 Yamamoto and Tonegawa Page 4 light stimulation (20 mw, 1 minute on/off duty cycle, 10 sets) to block MECIII input to CA1 and measured sharp-wave ripples and ripple bursts. The acute blockade did not alter fundamental characteristics of ripples including mean frequency and single ripple duration (Figure 3C), but did reduce ripple burst occurrence (Figure 3D) and the proportion of ripple bursts (Figure 3E) during quiet awake. In contrast, fundamental sharp-wave ripple parameters ripple burst occurrence and proportion of ripple bursts were not affected by MEC inhibition during slow-wave sleep (Figures 3F, 3G and 3H) suggesting that the direct input from MECIII to CA1 may be crucial for ripple bursts depending on the animal s awake/ sleep state. Blockade of MECIII Input Disrupts Long-range CA1 Replay during Quiet Awake We then employed tetanus toxin (TeTX)-mediated inhibition of neurotransmitter release from MECIII pyramidal cells to CA1 by using the triple transgenic mouse, MECIII-TeTX (Suh et al., 2011) (see STAR Methods), referred to hereafter as MT mice. For controls, we used double-transgenic mice lacking a critical component (poxr1 gene) of the triple transgenic mouse system (CT mice, N MT = 3, N CT = 3). For the following replay analysis, we implanted tetrode microdrives (Figure S3A) and recorded CA1 neuronal activities while the mice were running on the long track. We observed no fundamental differences between CT and MT mice in terms of place field size, mean firing rate, spatial information and complex spike index (Figure S3B). Also, no strain-dependent differences were observed in basic ripple parameters, such as mean frequency and single ripple duration (Figure 4A). With around 100 recorded single units from dorsal CA1 in each mouse (Figure S3C, 103±14 single units/mouse per 12 tetrode microdrive), we were able to estimate the animal s position using the Bayesian decoding method (Davidson et al., 2009; Carr et al., 2011; Pfeiffer and Foster, 2013) during the RUN session in both CT and MT (Figures 4B, S3D, S3E, S3F and S3G). Using this decoding approach, we assessed the replay contents within the ripple bursts in terms of replay quality and spatial coverage. We first noticed that a large number of the replay candidates in the MTs were fragmented into multiple of regions on the track (about 43% of replay candidates: 791/1836 events) which we subsequently quantified by introducing a replay fragmentation index (Figure 4C, see STAR Methods) and found that MTs had a significantly higher fragmentation index compared to CTs. Using the replay candidates, we next identified statistically significant replay events with bootstrapping methods (examples Figure 4D) (typically % of replay candidates were significant with this method, see STAR Methods). We observed that statistically significant replays in MTs (225/1836 candidates) covered only % of the length of the track (Figures 4D and 4E). We confirmed that spatial coverage of replay events during quiet awake were reduced by half in MTs compared to that of CTs (Figures 4F, S4A and S4B). The distribution of inter ripple intervals in MT mice were different from that of CT s in terms of the local peaks in the monotonous Poisson distribution (Figure S4C). We also quantified ripple bursts in the full data set and revealed that ripple doublets and triplets were significantly decreased in MTs compared with CTs during quiet awake, and in turn, singlets became more prominent (Figures 4G and 4H), showing that both the spatial and temporal coverage of ripple bursts in the MTs were indeed shorter than those of CTs.

6 Yamamoto and Tonegawa Page 5 MECIII Input is Dispensable for Long-range CA1 Replay during Slow-Wave Sleep Next, we focused on slow-wave sleep during which animals were housed in a high-walled sleep box (Figure 5A), after the RUN period. A total of 9726 replay candidates were detected and tested for statistical significance with the same method as previously described (only 9 10 % of ripple candidates became significant during slow-wave sleep, N CT = 3, N MT = 3). The quality of replay events was assessed by computing a replay score (fidelity) (see STAR Methods), and we confirmed significantly lower replay scores during slow-wave sleep than quiet awake (Karlsson and Frank, 2009) (Figure 5B). Our slow-wave sleep data revealed that both genotypes have comparable spatial coverage of the linear track, unlike quiet awake data (Figures 5C and 5D). The group data analysis revealed no differences in the spatial coverage, with on average ~50% coverage of the track length between the MT and CT genotypes during slow-wave sleep states (Figure 5E). The ripple burst decrease observed in MT mice during quiet awake was also not seen during slow-wave sleep (Figures 5F and 5G). Blockade of CA3 Input Disrupts Ripples and Replays in CA1 Regardless of the Animal s Sleep/Awake State Discussion Comparatively, we also optogenetically blocked CA3 input to CA1 acutely during quiet awake or slow-wave sleep. For this purpose, CA3-Cre (KA1-Cre) mice (Nakazawa et al., 2002) were bilaterally injected with an AAVrh8-hSyn1-DIO-eArchT-eYFP virus and CA1 activity was recorded while mice were in either quiet awake or slow-wave sleep (Figures 6A and 6B, see STAR Methods). Analyses on group data revealed a significant decrease in ripple occurrence during CA3 input inhibition regardless of behavioral state, but not the peak frequency and duration of ripples (Figure 6C right panel, Examples shown in Figures 6D and 6E). Although there was a tendency for the ripple oscillatory frequency to decrease during light stimulation, it was not statistically significant (Figure 6C middle panel). We also performed Bayesian decoding in the CA3-Cre-eArchT mice while they were running on a linear track (1.5 meter). In the light OFF condition (i.e. intact CA3 to CA1 input), we obtained precise position estimates during the RUN session (Figure 6F). We then tested the significance of replays on the residual ripples (amber arrowheads in Figure 6E) during the light ON (i.e. blocked CA3 to CA1 input) condition. Among residual ripples during the light ON condition, only 1.2% (16/1330 significant replays from residual ripples during light ON) of replay candidates passed the significance test. This significant decrease in the amount of statistically significant replays (Figure 6G right panel, *** P = ) suggests that CA3 input is indeed necessary for CA1 to have sharp-wave ripples and replays regardless of the awake/sleep states. The current study demonstrates an awake/sleep state-dependent role of MECIII pyramidal cell input to CA1 on ripples and place cell replays of experience within CA1. We showed that in the quiet awake but not in the slow-wave sleep state, ripple bursts in CA1 were reduced when MECIII pyramidal cells input to CA1 was inhibited. This quiet awakespecific reduction of ripple bursts was demonstrated by two different intervention methods:

7 Yamamoto and Tonegawa Page 6 acute, targeted optogenetic inhibition of MECIII axon terminals in CA1 (Yamamoto et al., 2014), and chronic TeTX-mediated synaptic inhibition of MECIII cell input (Suh et al., 2011) throughout the recording task. Our MECIII inhibition did not affect decoding results in the downstream CA1, which is potentially inconsistent with the previous report (Brun et al., 2008). This is probably due to the limitation of region specificity of the method used. In the latter intervention method, quiet awake replays, but not slow-wave sleep replays, in a long linear track were fragmented into multiple replay events each covering the distance corresponding to a single ripple event. In contrast to the MEC input inhibition, acute CA3 input inhibition by optogenetics suppressed ripple incidents as well as place cell replays in CA1 regardless of the animal s awake/sleep state, indicating CA3 activity is crucial for ripples and replays; consistent with the conclusions previously formed on the basis of a chronic intervention study (Nakashiba et al., 2009). Mechanism of Ripple Burst Generation Nearly half of the observed ripples during quiet awake are composed of burst trains of multiple ripple events (ripple bursts; Figure 1). Past studies (Chrobak and Buzsaki, 1996; Chrobak et al., 2000) did not focus on how ripple bursts are generated and hence this issue has remained to be investigated. In this study, our simultaneous dual site recording of LFP and MUA of CA1 and MEC during quiet awake periods of long-track runs provided insight to this question. During quiet awake, a ripple burst is initiated by a ripple in the MEC and followed by a CA1 ripple with a long temporal window ( msec). The ripple events alternate between the MEC and CA1, one, two, or sometimes three times with decreasing inter-ripple intervals until they occur nearly simultaneously between the MEC and CA1. It has been reported that the existence of a long time window could facilitate a local computation between MEC and CA1 (Mizuseki et al., 2009; Schomburg et al., 2014; Fernandez-Ruiz et al., 2017). Our cross-correlation analysis on the ripple bursts revealed that MEC ripple activity indeed precedes the CA1 ripple during quiet awake, but this is less clear in slow-wave sleep. The Granger causality index, a measure of causality between a pair of events based on their precise timing of occurrence, suggests that there may be a causal relationship between the individual ripple events in MEC and CA1 with the former playing a more prominent role in driving the alternations. This relationship of ripple bursts in MEC and CA1 might be supported by a reduction of CA1 ripple bursts by inhibition of MEC input to CA1 during quiet awake periods. Two interesting issues emerge from these data. First, the alternating occurrence of the MEC and CA1 ripples may be driven by cyclical activity in the known loop between MEC and CA1: CA1 MEC layer V MEC layer III CA1 (Witter et al., 2000). Second, although the activity within this loop may play a crucial role in the generation of ripple bursts in the quiet awake periods, there must be a separate mechanism for the generation of ripple bursts during slow-wave sleep because MEC input to the CA1 seems to be dispensable for the generation of ripple bursts in this state. During quiet awake, these alternate ripple burst cycles might be triggered by the MEC from external stimuli, such as top-down signal from higher-order cortical areas and/or sensory cue information arriving in

8 Yamamoto and Tonegawa Page 7 the MEC, whereas during slow-wave sleep, they might be internally triggered within the hippocampus (CA3-CA1 network). Our results support the idea that cortical input initiates hippocampal CA1 replays during quiet awake (i.e. on-line state) for the genesis of longrange extended replays for planning for example, while hippocampus itself initiates replay and send them out to cortical areas during slow-wave sleep (i.e. off-line state) for memory consolidation. Coordinated or Uncoordinated Replays between MEC and CA1 Our understanding of whether the MEC and CA1 work dependently or independently to process information has been controversial in the field. Recordings from the deep layers of the MEC reported that there are coordinated replays between the MEC and CA1 during quiet awake (Olafsdottir et al., 2016), whereas recordings from the superficial layers of MEC found replays between MEC and CA1 are not coordinated and work independently during exploration and resting periods in the sleep box (O Neill et al., 2017). The latter report case, especially during sleep period, implies that MEC input is unnecessary for replay in CA1 during sleep, which is consistent with our slow-wave sleep replay results (Figure 5). However, both reports used short track to assess the issue and their replay events (seen in examples) in both cases are relatively short in time (possibly singlets in our definition). Moreover, those studies were not looking at ripple bursts. Our study, which was optimized to do so, suggests that there is an alternating burst activity between superficial MEC and CA1 with a long temporal window and potentially this could be the mechanism for chaining pieces of information into a set of larger information. Causal Relationship Between Ripple Bursts and Long-Distance Replay in Quiet Awake A previous study demonstrated that extended replays are associated with ripple bursts in CA1 during quiet awake (Davidson et al., 2009). This led to the hypothesis that extended hippocampal replay may consist of chains of sub-sequences, each with a spatial extent covered by an individual ripple event. However, the causal relationship between ripple bursts and extended replay has not been demonstrated. The current study provides evidence of such a relationship. Targeted inhibition of MEC input to the CA1 specifically during quiet awake led to a simultaneous decrease of both ripple bursts and extended replay. The presence of this simultaneous down-regulation is unlikely to be due to a general effect of MEC input inhibition because it was not observed during slow-wave sleep. Although the current study was conducted exclusively in a folded linear track, during the exploration of an open field, one of multiple sequences of place cells converging from different angles onto the current location of pose will be replayed. Which set of sequences are replayed is likely determined by the animal s current location of pose and in the case of quiet awake, the sequence set may contain information about where the animal will go next (Pfeiffer and Foster, 2013). MECIII input to CA1 has previously been hypothesized to facilitate on-line processing of information from the current environment during the run state (Colgin et al., 2009; Mizuseki et al., 2009; Zheng et al., 2016) and the present study showed that inhibition of MECIII caused fragmented replay within CA1 during quiet awake but not during slow-wave sleep. This indicates that a potential role of MEC input during quiet awake replay might be to provide current spatial information to the CA1 and thereby

9 Yamamoto and Tonegawa Page 8 STAR Methods Text help concatenate prospective and retrospective sub-sequences for an extended replay (a summary cartoon in Figure S5). Our results may also suggest the existence of functionally diverse cell assemblies in CA1 (van de Ven et al., 2016) that could be dynamically formed into longer episodes by the input from MECIII via ripple bursts, especially during quiet awake state. CONTACT FOR REAGENT AND RESOURCE SHARING Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Susumu Tonegawa (tonegawa@mit.edu). EXPERIMENTAL MODEL AND SUBJECT DETAILS Eight male C57BL/6J WT mice (4 5 months old), four male MECIII-Cre mice (5 6 months old), three MECIII-TeTX-MT mice (5 6 months old), three MECIII-TeTX-CT mice (5 6 months old), four CA3-Cre mice (4 5 months old) were used in this study. All procedures relating to mouse care and treatment conformed to the institutional and NIH guidelines. We previously developed a tetanus toxin (TeTX)-based triple transgenic mouse (MECIII-TeTX MT) that allows for inducible and reversible silencing of synaptic transmission of MEC layer III pyramidal cells with a doxycycline (Dox) containing diet (Suh et al., 2011). We performed in vivo recordings when MT mice were on off-dox period, three weeks after Dox food withdrawal. As controls, littermate double transgenic mice lacking one transgene (poxr1) (Tg2) were used (MECIII-TeTX CT). For acute optogenetic experiments, we used male MECIII-Cre (poxr1-cre) and CA3-Cre (KA1-Cre) mice respectively and Cre specific virus (see Virus Constructs) was bilaterally injected to MEC and CA3 respectively. Where mutant mice (MT or CT, see below) were not used, wild type male C57BL/6J were used. All animals were housed in a 12h/12h light schedule (switching at 7 am and 7 pm). Virus Constructs The AAVrh8-hSyn1-DIO-eArchT-eYFP plasmid was constructed by inserting the earcht-eyfp gene fragment which was obtained from a template, plenti- CaMKIIa-eArchT_3.0eYFP (courtesy of Dr. Karl Deisseroth at Stanford University) (Mattis et al., 2011). This fragment was cloned into AscI (1165bp) and NcoI sites (2719bp) of a linearized and modified AAV vector containing the human Synapsin1 promoter, using the double-floxed inverted construct strategy (Atasoy et al., 2008). Restriction digests were made according to standard protocol and ligations were made using Takara DNA ligation kit version 2.1. The construct was amplified using EndoFree Plasmid Qiagen maxi prep kit. Recombinant AAV vectors were serotyped with AAVrh8 coat proteins and packaged by the viral vector core at the Gene Therapy center and Vector Core at the University of Massachusetts Medical School. The final viral concentration was genome copies ml 1.

10 Yamamoto and Tonegawa Page 9 METHOD DETAILS Stereotactic Injection and Probe Implant Each animal underwent bilateral craniotomies using a 1/4 size drill bit at 4.50 mm anterioposterior (AP), ±3.50 mm mediolateral (ML) for MEC injections. The AAVrh8 virus was injected using a mineral oilfilled glass micropipette joined by a microelectrode holder to a 10 μl Hamilton microsyringe. All mice were injected bilaterally with 250 nl AAV virus at a rate of 100 nl min 1. After recovery from viral injection, two optical fibers were implanted to the left hemisphere of dorsal CA1 and MEC (coordinates: [HPC] AP: 1.80 mm ML: 1.70 mm DV: mm, [MEC] AP: 4.50 mm ML: 3.50 mm DV: mm at 10 degrees anterior-to-posterior). For the right hemisphere, two hybrid silicon probe arrays with an optical fiber (200 μm core) that was attached to the back of the probe was lowered through two right hemisphere holes at the following coordinates: ([HPC] AP: 1.80 mm ML: mm DV: mm, [MEC] AP: 4.50 mm ML: mm DV: mm at 10 degrees anterior-to-posterior). All mice were allowed to recover for seven days before subsequent experiments. All fiber placements and viral injection sites were verified histologically. We only included mice in this study that had fluorophore expression limited to MEC and dorsal CA1 stratum lacunosum moleculare area. Linear Maze Task Linear maze task was conducted with CT, MT and wild type male mice. We used two types of mazes. The short one is made of 1.5 meters long metal gutter (5 centimeter wide, 3 centimeter tall side walls). The 6.3 meters long track was also made of same material but folded into double U-shape to fit into our recording arena. About two weeks after surgery, animals were first trained on the short linear track to receive small food pellets at both ends of track. After 7 to 10 days on the short track, they were then introduced to the long track to receive food rewards. The maze running was conducted when they were in dark light schedule. In Vivo Electrophysiology Two distinct types of electrodes were used in this study. We used silicone probes (linear, ploy and staggered configuration) for Figure 1 (N WT = 4), Figure 2 (N MECIII-Cre = 5), Figure 3 (N MECIII-Cre = 4) and Figure 6 (N CA3-Cre = 4) experiments respectively. For the large-scale single unit recordings, 12 tetrode (4 12.7um, Nicrome) microdrive arrays were used in Figure 4 and Figure 5 (N MT = 3, N CT = 3 respectively) experiments. These electrodes were loaded into a custom designed adjustable miniature microdrives arrays (4.5 g with a protection cone, unpublished) that can take up to two sets of silicone probes (64 channels) or 48 channels microwires. The recorded coordinates were as follows: [HPC] AP: 1.80 mm ML: mm DV: mm (adjustable) [MEC] AP: 4.50 mm ML: mm DV: mm (at 10 degrees anteriorto-posterior). All adjustments were conducted after the mice were fully recovered (at least five days after the surgery). Once all recording sessions were over, the animals were deeply anesthetized and postmortem histology was performed for a subsequent electrode track position confirmation. The fixed brains were cut into 40 μm thick slices and stained with DAPI or fluorescent Nissl staining. In some cases, the electrode track and tip location were identified with help of DiI fluorescence on the silicone probes and tetrodes. We combined the histology information with electrophysiologically unique features (for example, sharpwave-ripple from CA1 cell layer and/or polarities of sharp-waves).

11 Yamamoto and Tonegawa Page 10 Electrophysiology with Optogenetics A yellow-green laser (561 nm, 500 mw; DPSS) with a custom designed 2.5 meters long fiber optics patch cable (dual 200 μm core, NA = 0.22, Doric Lenses) was installed on the same area where the maze and an open arena (50 cm 40 cm 30 cm, copper sink) were situated. A custom designed mechanical shutter using a digital high speed servo motor (S3155, Futaba) was integrated at the laser outlet where the patch cable was connected. The shutter servo motor and the DPSS driver were controlled by custom designed hardware based on an ATmega2560 microcontroller (Mega2560 Rev.3, Arduino). The output power of the DPSS laser was calibrated to 20 mw at the tip with the implanted optical fiber attached to a laser intensity meter (Thorlabs, PM100D with S121C optical sensor). We attached a 200 μm core optic fiber to the back side of silicone linear probes with ~500 μm distance from the first recording channel on the linear probe. For the manipulation, we placed animals in an open field after exposing to the long-track and monitored animal s behavior states (motion history and LFP delta/theta power) to identify quiet awake and slow-wave sleep periods. We then performed the light stimulation (20 mw, 1 minute on/off duty cycle, 10 sets per day) to block MECIII input to CA1 and measured sharp-wave ripples and ripple bursts. During the light stimulation, we did not observe notable rebounds that could confound our BR measurement (however, there were transient photo current potential at the onset and offset of the light that lasted for ~500 ms in the LFP traces but not in the MUA data). We excluded such negligible short time periods for our ripple burst analysis. Behavior Position Tracking All behavior positions were extracted based on the position of light emitting diodes (LEDs) that were mounted on the headstage (preamp, 35mm spacing). The overhead color camera monitored the animal s behavior and the recording system (Digital Lynx) tracked the LED position at 30 Hz. Velocity filters were applied (2 cm/s) to extract valid run segments from the electrophysiological data. QUANTIFICATION AND STATISTICAL ANALYSIS LFP Power Density Estimation In order to estimate the power spectral density (PSD), we used a parametric method, the Modified Covariance Method (MCM), using the autoregressive (AR) model: where a p (k) is AR parameter and F S is the sampling frequency. The power of delta, theta and ripple were obtained by integrating the PSD estimates for 1 4Hz, 6 12 and Hz respectively. For quiet awake states, delta power remained low while theta power had peak at 7Hz. For slow-wave sleep states, delta power increased by two to three folds compared to that of quiet awake states and in turn theta power dropped by three or more folds. Correspondingly, the sharp-wave ripples occurrence increased by two to three folds during slow-wave sleep states compared to quiet awake states.

12 Yamamoto and Tonegawa Page 11 Candidate Ripple and Replay Event Detection In order to determine candidate events, we used both ripple band power ( Hz) and multi-unit activities. We first identified periods in which ripple band LFP power of selected recording channel exceeded the 3 S.D. level of the baseline. The sum of MUA was computed in two ways in this manuscript as we used two different types of electrodes (ie, silicone linear probes and tetrodes) to measure neuronal activities. For the linear probes, we first identified positional range of recording sites along the probes and summed spike counts within the region of interest at 10 millisecond time bins as spiking activities. For the tetrodes, we first performed standard spike sorting based on spike amplitudes and binned them into 10 millisecond bins. Sum was then computed across tetrodes that are located in the putative pyramidal cell layer. We then identified instantaneous multi-unit activity peak firing rates that exceeded the threshold (i.e., 3 S.D. of the baseline level) and then searched a time-window around the instantaneous multi-unit activity peak until the power reached the cut-off (1 S.D.) level at both ends. We defined the extracted time window as candidate events and assign identification numbers that was similar algorithm to the previous report (Davidson et al., 2009). Sleep Classification We employed delta versus theta power ratio that has been used to measure sleep states in rodent (Haggerty and Ji, 2014). To enhance detectability of pronounced delta power in deeper sleep, we used delta to theta power ratio (Figure S1C). In order to further classify quiet awake versus slow-wave sleep, we also used temporal history of animal s head position. We defined slow-wave sleep as following: first monitor animal s head position movement and compute amount of movement in chunks of every 15 seconds. The 15 second epochs that have movement of less than 2 cm and delta/theta power ratio that is greater than 5 S.D. were classified as slow-wave sleep. We defined this elevated delta power and overall stable head position every 15 seconds as slow-wave sleep period. Similar algorithm is used to define quiet awake but with different thresholds (more than 2 cm movement over 15 seconds and delta/theta power ratio of 2 SD to 5 SD period. Ripple Burst Analysis After detecting candidate ripple events (see Candidate Ripple and Replay Event Detection), time lags between peaks of ripple event were computed. Singlet ripples were defined as ripple events that are temporally separated by 200 milliseconds or more to adjacent ripple events. For the doublets and triplets (ripple bursts), the temporal lags were set to less than 200 milliseconds but greater than 70 milliseconds. These parameters were determined by computing the single ripple duration (~80 ms) and inter-ripple intervals (~125 ms) when ripple bursts occurred. In the event that adjacent ripple closer than 70 milliseconds, these were categorized as single ripple. Number of adjacent ripple peaks was determined and assigned to doublets, triplets and others (more peaks than three). We performed this analysis to both quiet awake and slow-wave sleep periods that are identified by elevated delta wave power (1 4 Hz), lowered theta wave power Granger Causality Analysis We used time series domain Multivariate Granger causality (MVGC) analysis in this study (Seth, 2010; Barnett and Seth, 2014; Seth et al., 2015) to assess temporal dependencies of neuronal burst, namely the ripples in both MEC and CA1.

13 Yamamoto and Tonegawa Page 12 This analysis is performed by fitting a Vector Autoregressive model (VAR) to a set of time series: where U t is split into two jointly distributed multivariate time series processes (in this study, MEC MUA burst and CA1 MUA burst time series), p is model order of vector autoregressive (VAR), A k are regression coefficients, U t-k are time lag from U t and ε t are regression residuals. The Granger causality from Y to X is defined to be the log-likelyhood ratio: where Σ xx =cov(ε x, t ) and Σ xx =cov(ε x, t ) are the residual covariance matrices of the VAR model. We have normalized the firing rates to maximum in each field before we performed the test. To statistically test the GC value, we performed two different types of shuffling: (1) random time lags were add to shuffle original data for 1000 times and (2) spike theta phase preserved while temporal association between MEC and CA1 randomized The p values were computed using Monte-Carlo bootstrap method. Replay Fragmentation Analysis To assess fragmentation of replay events, we introduced Replay Fragmentation Index. Within a replay candidate, probability density functions that are higher than 3 S.D. of the mean PDF value were detected. For each high PDF, spatial lag between adjacent time bins are computed and contiguous high PDF points were identified if the adjacent high PDFs were within 20% of the track length. If the lag was greater than 20% of the track length, it was categorized as discontinuous PDF. Then temporal length of continuous PDF events was computed and tested if the events have minimum temporal length (4 bins). Next, replay scores were computed using a linear fitting algorithm that was used previously (Davidson et al., 2009). The fragmentation index is a product of replay scores and the number of fragmentation within a replay candidate. where n denotes number of isolated replay events. See Replay Fidelity Analysis for Rscore value. Bayesian Decoding Analysis We used a Bayesian decoding algorithm to decode population data from CA1. The likelihood of animal location can be computed according to Bayes rule:

14 Yamamoto and Tonegawa Page 13 where we assume a uniform prior probability over position P r (pos) so that only Poisson firing probability Pr(spike pos) is needed. For the CA3 optogenetic inhibition experiment, we used unsorted version of Bayesian decoder due to the limitation of clustering, which is demonstrated to provide comparable result with standard sorted version (Kloosterman et al., 2014). To test statistical significance, we used a bootstrapping method. Within a replay candidate, spatial column of PDF were randomly shuffled (column shuffle) for 2000 times. Replay scores were determined on the randomized replay patterns and significance P values were computed using Monte-Carlo method. Replay Fidelity Analysis We computed replay score for each significant replay events: where V is animal s velocity, L is starting location, Δt is temporal bin size. This scoring assumes that each replay trajectory is linear and is not considering the curve of the trajectory. The value ranges 0 1 and 1 means perfect fit to a line with highest PDF value (which is highest replay fidelity) (Davidson et al., 2009). All statistical analyses were performed in MATLAB. We used Student s t-tests and ANOVA for normally distributed data (after testing for normality). We used Wilcoxon rank-sum tests for non-normally distributed data sets. All tests were two-sided. DATA AND SOFTWARE AVAILABILITY The data and code that support the findings of this study are available from the corresponding author upon reasonable request. Supplementary Material Acknowledgments References Refer to Web version on PubMed Central for supplementary material. We thank C. Sun, and M. Morrissey for comments and discussions on the manuscript; S. Huang for the viral construction; C. Lovett for help with the viral injections and histology, and other members of the Tonegawa lab for their support. This work was supported by the RIKEN Brain Science Institute, the Howard Hughes Medical Institute, the JPB Foundation (to S.T.). Atasoy D, Aponte Y, Su HH, Sternson SM. A FLEX switch targets Channelrhodopsin-2 to multiple cell types for imaging and long-range circuit mapping. J Neurosci. 2008; 28: [PubMed: ]

15 Yamamoto and Tonegawa Page 14 Barnett L, Seth AK. The MVGC multivariate Granger causality toolbox: a new approach to Grangercausal inference. J Neurosci Methods. 2014; 223: [PubMed: ] Brun VH, Leutgeb S, Wu HQ, Schwarcz R, Witter MP, Moser EI, Moser MB. Impaired spatial representation in CA1 after lesion of direct input from entorhinal cortex. Neuron. 2008; 57: [PubMed: ] Buzsaki G. Hippocampal sharp waves: their origin and significance. Brain Res. 1986; 398: [PubMed: ] Buzsaki G, Horvath Z, Urioste R, Hetke J, Wise K. High-frequency network oscillation in the hippocampus. Science. 1992; 256: [PubMed: ] Carr MF, Jadhav SP, Frank LM. Hippocampal replay in the awake state: a potential substrate for memory consolidation and retrieval. Nat Neurosci. 2011; 14: [PubMed: ] Carr MF, Karlsson MP, Frank LM. Transient slow gamma synchrony underlies hippocampal memory replay. Neuron. 2012; 75: [PubMed: ] Chrobak JJ, Buzsaki G. High-frequency oscillations in the output networks of the hippocampalentorhinal axis of the freely behaving rat. J Neurosci. 1996; 16: [PubMed: ] Chrobak JJ, Lorincz A, Buzsaki G. Physiological patterns in the hippocampo-entorhinal cortex system. Hippocampus. 2000; 10: [PubMed: ] Colgin LL, Denninger T, Fyhn M, Hafting T, Bonnevie T, Jensen O, Moser MB, Moser EI. Frequency of gamma oscillations routes flow of information in the hippocampus. Nature. 2009; 462: [PubMed: ] Davidson TJ, Kloosterman F, Wilson MA. Hippocampal replay of extended experience. Neuron. 2009; 63: [PubMed: ] de Lavilleon G, Lacroix MM, Rondi-Reig L, Benchenane K. Explicit memory creation during sleep demonstrates a causal role of place cells in navigation. Nat Neurosci. 2015; 18: [PubMed: ] Diba K, Buzsaki G. Forward and reverse hippocampal place-cell sequences during ripples. Nat Neurosci. 2007; 10: [PubMed: ] Ego-Stengel V, Wilson MA. Disruption of ripple-associated hippocampal activity during rest impairs spatial learning in the rat. Hippocampus. 2010; 20:1 10. [PubMed: ] Fernandez-Ruiz A, Oliva A, Nagy GA, Maurer AP, Berenyi A, Buzsaki G. Entorhinal-CA3 Dual-Input Control of Spike Timing in the Hippocampus by Theta-Gamma Coupling. Neuron. 2017; 93: e1215. [PubMed: ] Foster DJ, Wilson MA. Reverse replay of behavioural sequences in hippocampal place cells during the awake state. Nature. 2006; 440: [PubMed: ] Girardeau G, Benchenane K, Wiener SI, Buzsaki G, Zugaro MB. Selective suppression of hippocampal ripples impairs spatial memory. Nat Neurosci. 2009; 12: [PubMed: ] Gupta AS, van der Meer MA, Touretzky DS, Redish AD. Hippocampal replay is not a simple function of experience. Neuron. 2010; 65: [PubMed: ] Haggerty DC, Ji D. Initiation of sleep-dependent cortical-hippocampal correlations at wakefulnesssleep transition. J Neurophysiol. 2014; 112: [PubMed: ] Jackson JC, Johnson A, Redish AD. Hippocampal sharp waves and reactivation during awake states depend on repeated sequential experience. J Neurosci. 2006; 26: [PubMed: ] Ji D, Wilson MA. Coordinated memory replay in the visual cortex and hippocampus during sleep. Nat Neurosci. 2007; 10: [PubMed: ] Karlsson MP, Frank LM. Awake replay of remote experiences in the hippocampus. Nat Neurosci. 2009; 12: [PubMed: ] Kloosterman F, Layton SP, Chen Z, Wilson MA. Bayesian decoding using unsorted spikes in the rat hippocampus. J Neurophysiol. 2014; 111: [PubMed: ] Kudrimoti HS, Barnes CA, McNaughton BL. Reactivation of hippocampal cell assemblies: effects of behavioral state, experience, and EEG dynamics. J Neurosci. 1999; 19: [PubMed: ]

16 Yamamoto and Tonegawa Page 15 Lee AK, Wilson MA. Memory of sequential experience in the hippocampus during slow wave sleep. Neuron. 2002; 36: [PubMed: ] Maingret N, Girardeau G, Todorova R, Goutierre M, Zugaro M. Hippocampo-cortical coupling mediates memory consolidation during sleep. Nat Neurosci. 2016; 19: [PubMed: ] Mattis J, Tye KM, Ferenczi EA, Ramakrishnan C, O Shea DJ, Prakash R, Gunaydin LA, Hyun M, Fenno LE, Gradinaru V, et al. Principles for applying optogenetic tools derived from direct comparative analysis of microbial opsins. Nat Methods. 2011; 9: [PubMed: ] McNamara CG, Tejero-Cantero A, Trouche S, Campo-Urriza N, Dupret D. Dopaminergic neurons promote hippocampal reactivation and spatial memory persistence. Nat Neurosci. 2014; 17: [PubMed: ] Mizuseki K, Sirota A, Pastalkova E, Buzsaki G. Theta oscillations provide temporal windows for local circuit computation in the entorhinal-hippocampal loop. Neuron. 2009; 64: [PubMed: ] Nadasdy Z, Hirase H, Czurko A, Csicsvari J, Buzsaki G. Replay and time compression of recurring spike sequences in the hippocampus. J Neurosci. 1999; 19: [PubMed: ] Nakashiba T, Buhl DL, McHugh TJ, Tonegawa S. Hippocampal CA3 output is crucial for rippleassociated reactivation and consolidation of memory. Neuron. 2009; 62: [PubMed: ] Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S. Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science. 2002; 297: [PubMed: ] O Keefe J, Dostrovsky J. The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res. 1971; 34: [PubMed: ] O Neill J, Boccara CN, Stella F, Schoenenberger P, Csicsvari J. Superficial layers of the medial entorhinal cortex replay independently of the hippocampus. Science. 2017; 355: [PubMed: ] Olafsdottir HF, Carpenter F, Barry C. Coordinated grid and place cell replay during rest. Nat Neurosci. 2016; 19: [PubMed: ] Pavlides C, Winson J. Influences of hippocampal place cell firing in the awake state on the activity of these cells during subsequent sleep episodes. J Neurosci. 1989; 9: [PubMed: ] Pfeiffer BE, Foster DJ. Hippocampal place-cell sequences depict future paths to remembered goals. Nature. 2013; 497: [PubMed: ] Schomburg EW, Fernandez-Ruiz A, Mizuseki K, Berenyi A, Anastassiou CA, Koch C, Buzsaki G. Theta phase segregation of input-specific gamma patterns in entorhinal-hippocampal networks. Neuron. 2014; 84: [PubMed: ] Seth AK. A MATLAB toolbox for Granger causal connectivity analysis. J Neurosci Methods. 2010; 186: [PubMed: ] Seth AK, Barrett AB, Barnett L. Granger causality analysis in neuroscience and neuroimaging. J Neurosci. 2015; 35: [PubMed: ] Singer AC, Frank LM. Rewarded outcomes enhance reactivation of experience in the hippocampus. Neuron. 2009; 64: [PubMed: ] Skaggs WE, McNaughton BL. Replay of neuronal firing sequences in rat hippocampus during sleep following spatial experience. Science. 1996; 271: [PubMed: ] Suh J, Rivest AJ, Nakashiba T, Tominaga T, Tonegawa S. Entorhinal cortex layer III input to the hippocampus is crucial for temporal association memory. Science. 2011; 334: [PubMed: ] van de Ven GM, Trouche S, McNamara CG, Allen K, Dupret D. Hippocampal Offline Reactivation Consolidates Recently Formed Cell Assembly Patterns during Sharp Wave-Ripples. Neuron. 2016; 92: [PubMed: ] Wilson MA, McNaughton BL. Reactivation of hippocampal ensemble memories during sleep. Science. 1994; 265: [PubMed: ]

17 Yamamoto and Tonegawa Page 16 Witter MP, Wouterlood FG, Naber PA, Van Haeften T. Anatomical organization of the parahippocampal-hippocampal network. Ann N Y Acad Sci. 2000; 911:1 24. [PubMed: ] Wu CT, Haggerty D, Kemere C, Ji D. Hippocampal awake replay in fear memory retrieval. Nat Neurosci. 2017; 20: [PubMed: ] Yamamoto J, Suh J, Takeuchi D, Tonegawa S. Successful execution of working memory linked to synchronized high-frequency gamma oscillations. Cell. 2014; 157: [PubMed: ] Zheng C, Bieri KW, Hsiao YT, Colgin LL. Spatial Sequence Coding Differs during Slow and Fast Gamma Rhythms in the Hippocampus. Neuron. 2016; 89: [PubMed: ]

18 Yamamoto and Tonegawa Page 17 Figure 1. Ripple Bursts under Long-track Exposure (A) Long-track setup of the 6.3 meter linear track in a visual cue surrounded room. The folded track was set up on a table and elevated by 10 cm from the surface of the table. Three explicitly large visual cues were on the white surrounding walls. The experimenter and recording setups were isolated with a thick black curtain. Bottom panel shows an example of an animal trajectory during run. Two reward sites were in the left bottom corner in the panel. Animals were able to freely run and turn corners of the track. (B) Recording configuration in dorsal CA1 with a silicone linear probe. A solid line on the top panel and a dotted line on the bottom panel show electrode track in CA1. (C) Examples of identified single and ripple bursts in CA1. Top panel: Wideband LFP ranging from striatum-oriens to striatum-radiatum of the dorsal CA1, Middle panel: same recording position as top panel: color-coded LFP ripple band amplitude, Bottom panel: Multi-unit activities and sum of MUA (see STAR Methods). (D) Proportion of different types of ripple bursts during quiet awake (QAW) and slow-wave sleep (SWS) [QAW: Singlet: 54.1±8.2 (%)/Doublet: 28.6±6.2 (%)/Triplet:

19 Yamamoto and Tonegawa Page ±4.2 (%)/Other: 3.1±1.8 (%)], [SWS: Singlet: 47.6±10.3 (%)/Doublet: 32.1±6.2 (%)/ Triplet: 18.2±4.2 (%)/Other: 2.7±1.4 (%)], paired test, t singlet (1235) = 1.462, P singlet = 0.106/t doublet (638) = 1.176, P doublet = 0.228/t triplet (331) = 0.562, P triplet = 0.559/t other (63) = 1.091, P other = respectively, Bonferroni corrected. (E) Ripple occurrence per behavioral session. QAW: 2.4±0.37 (10s 1 ), SWS: 6.5±0.81 (10s 1 ), paired test, t(2267) = 3.291, ***P = (F) Distribution of ripple occurrence per ripple type. [QAW: Singlet: 1.22±0.39 (10s 1 )/Doublet: 0.75±0.29 (10s 1 )/Triplet: 0.32±0.16 (10s 1 )/Other: 0.11±0.04 (10s 1 )], [SWS: Singlet: 3.21±0.37 (10s 1 )/Doublet: 1.62±0.31 (10s 1 )/Triplet: 0.88±0.19 (10s 1 )/Other: 0.32±0.12 (10s 1 )], paired test, t singlet (1235) = 3.293, ***P singlet = /t doublet (638) = 2.498, **P doublet = 0.006/t triplet (331) = 2.502, **P triplet = 0.008/ t other (63) = 2.263, *P other = respectively, Bonferroni corrected. (G) Histogram of Interripple interval. Red arrowheads denote local peaks (around 110, 220, 330, 440 ms) in the monotonous Poisson distribution. (H) Examples of detected singlet ripple location on the long track (blue dots: singlet start position). (I) Examples of detected doublet ripple location on the long track (green dots: doublet start position). (J) Examples of detected triplet ripple location on the long track (red dots: triplet start position). Data are represented as mean ± SD.

20 Yamamoto and Tonegawa Page 19 Figure 2. Quiet Awake Specific Ripple Burst Associated Alternating Burst Activities in MEC- CA1 Network (A) An example of singlet sharp-wave ripple. Top panel: color-coded ripple band LFPs of superficial MEC and dorsal CA1 cell layer. Middle panel: Associated superficial MEC MUA with elevated burst activities (red downward arrow heads). Green dotted trace is smoothed sum of MEC MUA (see STAR Methods), Bottom panel: Detected CA1 MUA and its smoothed sum (blue dotted traces) with elevated ripple activities (a red upward arrow head). (B) Same as in (A), Examples of doublets and a triplet, inter-regional burst interaction during quiet awake. (C) Same as in (A). Examples of non-interregional ripple bursts during slow-wave sleep. (D) Cross-correlation between MEC and CA1 using ripple-band LFP peak

21 Yamamoto and Tonegawa Page 20 power times (top) and LFP and spikes (bottom) during QAW. Blue trace: 10 ms bin. Red trace: smoothed trend line (Gaussian kernel: σ= 20ms). (E) Cross-correlation between MEC and CA1 using MEC ripple-band LFP peak power times and CA1 MUA spike times (top) and LFP and spikes (bottom) during SWS. Blue trace: 10 ms bin. Red trace: smoothed trend line (Gaussian kernel: σ= 20ms). (F) Granger causal directional effect of inter-regional bursts between MEC and CA1 during quiet awake, MEC to CA1: 0.57±0.09, MEC to CA1: 0.32±0.08, z = 2.674, **P = 0.004, Wilcoxson rank-sum test. Dotted line denotes 95% significance level. Red dotted line: ripple power peak times between MEC and CA1 shuffled, Blue dotted line: spike theta phase preserved while temporal association between MEC and CA1 randomized. (G) No Directional effect between MEC and CA1 during slowwave sleep, MEC to CA1: 0.12±0.05, MEC to CA1: 0.17±0.06, z = 0.552, P = 0.681, Wilcoxson rank-sum test. Dotted lines denote 95% significance level. Red dotted line: ripple power peak times between MEC and CA1 shuffled, Blue dotted line: spike theta phase preserved while temporal association between MEC and CA1 randomized. Data are represented as mean ± SD.

22 Yamamoto and Tonegawa Page 21 Figure 3. Effects of Optogenetic Intervention of MECIII Input on Ripple Bursts (A) Sagittal sections of MEC and CA1 area. MECIII-Cre mice were bilaterally infected with the Cre specific AAVrh8-hSyn1-DIO-eArchT-eYFP virus and silicone linear probes with an optical fiber were implanted in both areas. (B) Examples of light stimulation during quiet awake when animals were in an open arena. Top panel: Light OFF condition. Bottom panel: Light ON condition. For each panel: traces from top: light timing, detected MUA ripples, ripple-band LFP, wide-band LFP traces and classified ripple bursts. (C) Fundamental sharpwave ripples parameters during quiet awake including ripple frequency (Light OFF: 149.3±4.4 Hz, Light ON: 148.7±4.9 Hz, paired test, t(302) = 1.003, P = 0.302), single

23 Yamamoto and Tonegawa Page 22 ripple duration (Light OFF: 91.2±6.3 ms, Light ON: 90.8±5.8 Hz, paired test, t(302) = 0.575, P = 0.553) during optical stimulation. (D) Distribution of ripple occurrence during quiet awake by ripple burst type. Singlet: Light OFF: 1.32±0.23 (10s 1 )/Light ON: 2.29±0.21 (10s 1 ), t single (638) = 3.093, **P single = 0.002, Doublet: Light OFF: 0.72±0.12 (10s 1 )/Light ON: 0.42±0.11 (10s 1 ), t double (304) = 2.681, **P double = 0.007, Triplet: Light OFF: 0.30±0.06 (10s 1 )/Light ON: 0.12±0.05 (10s 1 ), t triple (138) = 2.235, **P triple = 0.009, Other: Light OFF: 0.10±0.03 (10s 1 )/Light ON: 0.06±0.02 (10s 1 ), paired test, t other (34) = 1.018, P other = 0.342, paired test, respectively, Bonferroni corrected.(e) Distribution of ripple bursts with optogenetic stimulation during quiet awake, Singlet (Light-OFF: 57.5±9.3, Light-ON: 74.2±10.2, t single (638) = 1.841, P single = 0.061; Doublet (Light-OFF: 27.2±4.2, Light-ON: 17.4±2.7, t double (304) = 2.301, *P double = 0.032), Triplet (Light OFF: 13.2±3.5, Light ON: 7.3±2.1, t triple (138) = 2.129, *P triple = 0.025), Other (Light OFF: 2.1±0.6, Light ON: 1.1±0.8, t other (34) = 1.903, P other = 0.061), paired test, Bonferroni corrected. (F) Same as in (C) but during slow-wave sleep. Ripple frequency (Light OFF: 150.2±4.7 Hz, Light ON: 151.4±6.3 Hz, paired test, t(428) = 0.755, P = 0.462), single ripple duration (Light OFF: 95.3±7.2 ms, Light ON: 92.5±5.8 Hz, paired test, t(428) = 0.225, P = 0.824) during optical stimulation. (G) Same as in (D) but during slow-wave sleep. Singlet: Light OFF: 3.13±0.39 (10s 1 )/Light ON: singlet: 3.19±0.37 (10s 1 ), t single (948) = 0.552, P single = 0.693, Doublet: Light OFF: 1.49±0.29 (10s 1 )/Light ON: 1.54±0.31 (10s 1 ), t double (428) = 0.248, P double = 0.835, Triplet: Light OFF: 0.77±0.16 (10s 1 )/Light ON: 0.79±0.19 (10s 1 ), t triple (231) = 0.325, P triple = 0.769, Other: Light OFF: 0.25±0.04 (10s 1 )/Light ON: 0.21±0.12 (10s 1 ), paired test, t other (63) = 0.221, P other = 0.895, paired test, respectively (mean±s.d.), Bonferroni corrected. (H) Distribution of ripple bursts with optogenetic manipulation during slow-wave sleep, Singlet (Light OFF: 54.6±9.3, Light ON: 55.7±11.5, t single (948) = 0.635, P single = 0.524), Doublet (Light OFF: 32.2±6.1, Light ON: 31.5±5.2, t double (428) = 0.921, P double = 0.382), Triplet (Light OFF: 13.1±4.0, Light ON: 12.5±3.1, t triple (231) = 1.748, P triple = 0.081), Other (Light OFF: 0.5±0.2, Light ON: 0.7±0.2, t other (63) = 1.349, P other = 0.226), paired test, Bonferroni corrected. Data are represented as mean ± SD.

24 Yamamoto and Tonegawa Page 23 Figure 4. MECIII Input is Indispensable for Long-Range CA1 Replay in Quiet Awake (A) Comparison of the fundamental parameters of individual ripple: peak frequency: CT: 152.5±4.1 (Hz), MT: 151.4±5.0 (Hz), paired test, t(336) = 0.738, P = and duration: CT: 89.3±6.9 (ms), MT: 92.4±5.2 (ms), t(336) = 0.633, P = 0.528, respectively. (B) Examples of ripple-burst associated replays in CT and MT. Top panel: decoded replay trajectory. Current location is shown as a solid green line. Bottom panel: Detected replay candidates (red) using MUA (blue). (C) Examples of fragmented replay trajectories in MECIII input blocked mice (MT). Blue solid lines are detected fragmented replays. Fragmentation index: CT: 1.3±0.4, MT: 3.3±0.6, paired test, t(142) = 3.022, **P = (D) Examples of positive and negative correlated replays. Current locations at the time of replay indicated as green solid lines. (E) Examples of replay trajectories of MECIII blocked MT and CT. Each line represents single replay events and color-coded with single ripple (blue) or doublets (green) or triplets (red). Replay start and end locations were used to draw each line. (F) Group data of the spatial coverage. (CT: 57.3±9.3 %, MT: 25.7±8.3 %, z = 3.298, ***P = , Wilcoxson rank-sum test. (G) Distribution of ripple occurrence by ripple burst type. Singlet: CT: 1.29±0.18 (10s 1 )/MT: 2.30±0.16 (10s 1 ), t single (328) = 3.295, ***P single = , Doublet: CT: 0.72±0.12 (10s 1 )/MT: 0.42±0.11 (10s 1 ), t double (147) = 2.817, **P double = 0.006, Triplet: CT: 0.30±0.06 (10s 1 )/MT: 0.12±0.05 (10s 1 ), t triple (63) = 2.856, **P triple = 0.008, Other: CT: 0.10±0.03 (10s 1 )/MT: 0.06±0.02 (10s 1 ), paired test, t other (21) = 1.829, P other = 0.072, paired test, respectively, Bonferroni corrected. (H) Distribution of ripple bursts during quiet awake, Singlet (CT: 54.0±8.3, MT: 79.3±10.4, t single (328) = 2.327, *P single = 0.021), Doublet (CT: 28.6±6.2, MT: 13.9±4.6, t double (147) =

25 Yamamoto and Tonegawa Page , **P double = 0.006), Triplet (CT: 13.2±4.2, MT: 7.3±2.5, t triple (63) = 3.243, **P triple = 0.002), Other (CT: 3.1±2.1, MT: 2.7±1.2, t other (21) = 2.063, P other = 0.053), paired test, Bonferroni corrected. Data are represented as mean ± SD.

26 Yamamoto and Tonegawa Page 25 Figure 5. MECIII Input is Dispensable for Both Ripple Bursts and for Long-Range CA1 Replay during Slow-Wave Sleep (A) An example of sleep period after RUN session expressing elevated sharp-wave ripples during prolonged immobility. (B) Replay fidelity measurement by replay score. quiet awake: 0.63±0.12, slow-wave sleep: 0.41±0.10, z = 2.844, **P = 0.004, Wilcoxson rank-sum test. (C) Same as in Figure 4D. Examples of statistically significant slow-wave sleep replay episodes. (D) Same as in Figure 4E. Spatial coverage plot of slow-wave sleep replays. (E) Group data of spatial coverage of sleep replays, CT: 53.2±25.3 %, MT: 49.7±22.9 %, P = 0.092, Wilcoxson rank-sum test. (F) Distribution of ripple occurrence by ripple burst type. Singlet: CT: 3.24±0.39 (10s 1 )/MT: 3.19±0.37 (10s 1 ), t single (544) = 0.539, P single = 0.731, Doublet: CT: 1.53±0.29 (10s 1 )/MT: 1.54±0.19 (10s 1 ), t double (221) = 1.088, P double = 0.268, Triplet: CT: 0.84±0.16 (10s 1 )/MT: 0.79±0.19 (10s 1 ), t triple (102) = 1.295, P triple = 0.291, Other: CT: 0.24±0.04 (10s 1 )/MT: 0.21±0.12 (10s 1 ), paired test, t other (42) = 0.717, P other = 0.482, paired test, respectively, Bonferroni corrected. (G) Distribution of ripple bursts during slow-wave sleep, Singlet [CT: 51.0±10.3, MT: 54.3±12.1, t single (544) = 1.206, P single = 0.203), Doublet [CT: 24.3±6.7, MT: 27.2±4.8, t double (221) = 0.602, P double = 0.552), Triplet [CT: 14.3±4.4, MT: 10.5±2.2, t triple (102) = 1.611, P triple = 0.128), Other [CT: 10.4±2.4, MT: 8.2±1.1, t other (42) = 1.310, P other = 0.098) respectively, paired test, Bonferroni corrected. Data are represented as mean ± SD.

27 Yamamoto and Tonegawa Page 26 Figure 6. CA3 Input is Crucial for CA1 Ripples and Replays Regardless of Behavioral State (A) Schematic of CA3-Cre-eArchT mouse, N mice = 4. CA3 input is optogenetically blocked by axon terminal stimulation in CA1 area. Bilaterally injected and stimulated. (B) Viral expression pattern in CA3-Cre (KA1-Cre) mouse. Unilaterally recorded spiking activities and LFPs. (C) Group data of CA3 input blockade in CA1 area. Ripple peak frequency: (light OFF 155.4±4.4 Hz; light ON: 143.2±4.9 Hz, paired test, t(739) = 0.387, P = 0.681), single ripple duration: (light OFF 97.3±6.2 ms; light ON: 75.2±10.3 ms, paired test, t(739) = 1.726, P = 0.063), and occurrence: (light OFF 0.81±0.051 Hz; light ON: 0.051±0.05 Hz, paired test, t(395) = 3.294, *** P = ). (D) An example of optogenetic blockade of ripples during quiet awake. Top panel: green line indicates animal s position. Bottom panel: MUA as blue lines. Detected ripples as red lines. (E) An example of optogenetic blockade of ripples during slow-wave sleep. Amber arrowheads highlight residual ripple events during blockade. (F) An example of position reconstruction in a CA3-Cre-eArchT mouse. (G) Examples of residual replay (three left top panels) and bootstrap results (three left bottom panels) that correspond to panel (D) arrowheads. No significant replays during CA3