Hippocampal ensemble dynamics timestamp events in long-term memory

Alon Rubin; Nitzan Geva; Liron Sheintuch; Yaniv Ziv

doi:10.7554/eLife.12247

. 2015 Dec 18;4:e12247. doi: 10.7554/eLife.12247

Hippocampal ensemble dynamics timestamp events in long-term memory

Alon Rubin ^1,^†, Nitzan Geva ^1,^†, Liron Sheintuch ¹, Yaniv Ziv ^1,^*

Editor: Howard Eichenbaum²

PMCID: PMC4749549 PMID: 26682652

Abstract

The capacity to remember temporal relationships between different events is essential to episodic memory, but little is currently known about its underlying mechanisms. We performed time-lapse imaging of thousands of neurons over weeks in the hippocampal CA1 of mice as they repeatedly visited two distinct environments. Longitudinal analysis exposed ongoing environment-independent evolution of episodic representations, despite stable place field locations and constant remapping between the two environments. These dynamics time-stamped experienced events via neuronal ensembles that had cellular composition and activity patterns unique to specific points in time. Temporally close episodes shared a common timestamp regardless of the spatial context in which they occurred. Temporally remote episodes had distinct timestamps, even if they occurred within the same spatial context. Our results suggest that days-scale hippocampal ensemble dynamics could support the formation of a mental timeline in which experienced events could be mnemonically associated or dissociated based on their temporal distance.

DOI: http://dx.doi.org/10.7554/eLife.12247.001

Research Organism: Mouse

eLife digest

The ability to recall the timing of events is an important feature of long-term memory. Episodic memory, the mental account of “what” happened, “where” and “when”, depends on a region of a brain called the hippocampus. Certain neurons in the hippocampus, called place-cells, are known to capture information about the locations an animal has visited so that a specific pattern of place cell activity marks each location an animal visits. However, it is not clear how the brain can mark the relationship between the timing of different events.

Some studies have documented gradual changes in the activity patterns of the place cells over time, which could help mark time. If these changes are specific to a particular environment then they would not allow animals to associate in memory events that occurred close in time (for instance, in the same day) if these events occurred in different environments. To do that, a certain component of the changes in the activity patterns would have to be independent of any specific environment or context in which events occur.

Now, Rubin, Geva et al. have captured time-lapse images of the activity of thousands of hippocampal cells in mice as they explored two different environments on repeated occasions over a two-week period. The environments had different shapes, textures, visual cues, and odors. The mice were allowed to explore each environment daily for more than a week prior to the time-lapse filming so that they would be very familiar with the two environments. During the filming portion of the experiments, the mice visited one environment in the morning, and then the other in the afternoon.

The analysis of the images revealed what appeared to be unique patterns of cell activity for specific days, which gradually changed over the course of the experiment. The patterns persisted even when the animals switched to a new environment during the same day, but were different for visits to the same environment on different days. Next, Rubin, Geva et al. used the patterns of activity collected from the mice while they were in one environment to create a timeline of events. From this timeline, it was possible to accurately deduce which day each visit to the other environment occurred based on the patterns of hippocampal cell activity alone. One challenge that stems from this work is to understand the biological mechanisms that drive the patterns in neuronal activity over timescales that are relevant for long-term memory.

DOI: http://dx.doi.org/10.7554/eLife.12247.002

The capacity to remember temporal relationships between different events is essential to our autobiographical memory and sense of self. Autobiographical or episodic memory relies on the hippocampus, whose neurons are thought to encode information about where and when events have occurred (Davachi and DuBrow, 2015; Eichenbaum, 2014; Howard et al., 2014; Rolls, 2010; Shapiro, 2014; Tulving, 2002). Hippocampal place cells encode the spatial location of an animal through localized firing patterns, and have long been considered a substrate for long-term memory of the location in which events occurred (O'Keefe and Dostrovsky, 1971; O'Keefe, 1978). Whereas ample knowledge exists regarding the encoding of location, relatively little is known about the neural mechanisms that enable the encoding of the time in which events occur. Recent work has revealed that in familiar environments hippocampal place cell activity is dynamic over timescales that range from minutes to weeks (Howard and Kahana, 2002; Mankin et al., 2015; Mankin et al., 2012; Manns et al., 2007; Ziv et al., 2013). For timescales that are greater than one day, these dynamics primarily result from ongoing changes in the subsets of place cells that are active during repeated visits to the same fixed environment (Ziv et al., 2013). Such dynamics may contribute information about the temporal relationship between events by providing a unique code that functions as a ‘timestamp’. If such timestamps exist, they would likely aid long-term memory by reducing interference between traces of events that occur at different times at the same place, or that are similar in that they share contextual components such as sensory experience and behavior. Moreover, to support the formation of a mental timeline of experienced events in long-term memory, and the capacity to mentally ‘time-travel’ during memory recall (Kragel et al., 2015; Nyberg et al., 2010), timestamps should change gradually and continuously with time. Such gradual changes in the ensembles of place cells active during similar events on different days have been recently reported, but the extent that these dynamics actually carry temporal information remains unclear (Mankin et al., 2012; Ziv et al., 2013). We consider two alternative hypotheses regarding the possible contribution of the observed dynamics to coding of time. According to one hypothesis, the dynamics in the ensemble activity over days is unique to the environment in which it is observed, and independent from the dynamics in other, dissimilar environments. In this case, the dynamics may contribute ordinal information about different events that occur within a given environment, but will not contribute to associations in memory between events that happen close in time if these events occurred in different or dissimilar environments. An alternative hypothesis asserts that certain aspects of the days-scale dynamics in the ensemble activity are common to different environments. Such environment-nonspecific dynamics could support a linkage in long-term memory between dissimilar events that occur at temporal proximity. If this is the case, we would expect the hippocampal representations of events that occur in different spatial environments but in temporal proximity (e.g. the same day) to share common time-varying components.

To test these alternative hypotheses we investigated hippocampal neuronal representations of different spatial contexts over multiple days and weeks. We combined head-mounted miniaturized fluorescence microscopes (Ghosh et al., 2011; Ziv et al., 2013), chronic microendoscopy (Barretto et al., 2011), and viral-vector based expression of a genetically encoded Ca²⁺ indicator (Chen et al., 2013), to longitudinally image the Ca²⁺ dynamics of large populations (> 1,000 per mouse) of hippocampal CA1 pyramidal cells in freely behaving mice that repeatedly explored two familiar environments (Figure 1A). To avoid circadian effects we alternated the two environments between AM and PM sessions, 4–5 hr apart. Each session consisted of five 3-min trials. To maximize the perceived differences between the environments, we constructed linear tracks (environments A and B) that differed in shape, floor texture, surrounding proximal and distal visual cues, odor, and flavor of the water reward (see Materials and methods). To uncover the time-dependent coding dynamics, while minimizing changes in place codes induced by learning, we familiarized the mice with the two environments before starting the experiment (Figure 1A). During pre-training, the mice ran on each linear track for 15 min per day for 8–11 days, until they performed at least 60 laps in each environment for two consecutive days. Time-lapse imaging began two days after the last pre-training session, and was performed every other day for two weeks (days 1–15, Figure 1B). To confirm that changes in the representations of the two environments are also evident within a single session, we performed on days 16–17 a ‘remapping test’ by shuttling the mice back and forth between environments within the same session, keeping the interval between environments < 10 min. This test confirmed that CA1 neuronal ensembles were differentially and selectively active in the two environments (Figure 1—figure supplement 1), forming place field maps that were markedly different (Figure 1C), in agreement with past studies of place cell remapping (Leutgeb et al., 2004; Muller and Kubie, 1987).

Figure 1. — (A) Mice trained to run back and forth and collect a liquid reward in two different linear tracks. (B) Experimental timeline. After a pre-training period, we trained and imaged mice in the two environments every two days for 15 days. Each session (AM and PM) consisted of five 3-min trials. (C) Remapping tests within the same session performed on day 16 (top panels) and day 17 (bottom panels) confirmed that CA1 representations of the two environments were different. The test consisted of two trials in one environment followed by three trials in the other environment, and then an additional three trials in the first environment (A→B→A, or B→A→B). Shown are place-field maps ordered according to the place field centroid position in the first two trials of the session in environment A (top) and environment B (bottom). (D) Place field maps for the same cells on multiple days of the experiment, ordered by the place fields’ centroid positions in environment A on day 1 (green frame, top panels), or by the place fields’ centroid positions in environment B on day 15 (green frame, bottom panels). The maps depict the changes in place cell activity patterns in different environments and on different days. (**E,F**) Place cells that were active on more than one day typically retained their place fields in subsequent sessions. (E) Examples of a place cell active on all days of the experiment in environment B (top), and a place cell active on some days of the experiment in environment A (bottom). Red dots indicate the location of the mouse along the track during cellular Ca²⁺ excitation. Insets depict the mouse’s trajectory (blue line) and location (red dots) during cellular Ca²⁺ excitation in individual 3-min trials taken from different days. (F) The distributions of centroid shifts (color indicates the interval between sessions, averaged over five mice, mean ± s.e.m) were, centered at zero, and distinct from the null hypothesis that place cells would randomly alter their place fields (Kolmogorov-Smirnov test, p < 10^-14 for all distributions). (G) Average correlations of place cell firing patterns (i.e. the ‘population vector’, PV) during running epochs at different locations along the two linear tracks were low and did not change over time (One-way repeated-measures ANOVA; F<1, p = 0.44). (H) The fraction of place cells out of the population of cells active in both environments on each day also remained stable over time (One-way repeated-measures ANOVA, F_{(2.18, 8.71)}= 1.97, p = 0.19). (I) Average fractions of place cells and non-place cells, and (J) average fractions of Ca²⁺ events generated by place cells and non-place cells when mice were running or stationary.

**DOI:** http://dx.doi.org/10.7554/eLife.12247.003

Figure 1—figure supplement 1. — (A) Mice trained to run back and forth and collect a liquid reward in two different linear tracks. (B) Experimental timeline. After a pre-training period, we trained and imaged mice in the two environments every two days for 15 days. Each session (AM and PM) consisted of five 3-min trials. (C) Remapping tests within the same session performed on day 16 (top panels) and day 17 (bottom panels) confirmed that CA1 representations of the two environments were different. The test consisted of two trials in one environment followed by three trials in the other environment, and then an additional three trials in the first environment (A→B→A, or B→A→B). Shown are place-field maps ordered according to the place field centroid position in the first two trials of the session in environment A (top) and environment B (bottom). (D) Place field maps for the same cells on multiple days of the experiment, ordered by the place fields’ centroid positions in environment A on day 1 (green frame, top panels), or by the place fields’ centroid positions in environment B on day 15 (green frame, bottom panels). The maps depict the changes in place cell activity patterns in different environments and on different days. (**E,F**) Place cells that were active on more than one day typically retained their place fields in subsequent sessions. (E) Examples of a place cell active on all days of the experiment in environment B (top), and a place cell active on some days of the experiment in environment A (bottom). Red dots indicate the location of the mouse along the track during cellular Ca²⁺ excitation. Insets depict the mouse’s trajectory (blue line) and location (red dots) during cellular Ca²⁺ excitation in individual 3-min trials taken from different days. (F) The distributions of centroid shifts (color indicates the interval between sessions, averaged over five mice, mean ± s.e.m) were, centered at zero, and distinct from the null hypothesis that place cells would randomly alter their place fields (Kolmogorov-Smirnov test, p < 10^-14 for all distributions). (G) Average correlations of place cell firing patterns (i.e. the ‘population vector’, PV) during running epochs at different locations along the two linear tracks were low and did not change over time (One-way repeated-measures ANOVA; F<1, p = 0.44). (H) The fraction of place cells out of the population of cells active in both environments on each day also remained stable over time (One-way repeated-measures ANOVA, F_{(2.18, 8.71)}= 1.97, p = 0.19). (I) Average fractions of place cells and non-place cells, and (J) average fractions of Ca²⁺ events generated by place cells and non-place cells when mice were running or stationary.

**DOI:** http://dx.doi.org/10.7554/eLife.12247.003

For longitudinal analysis of the cell activity, we revalidated a previously established routine for aligning cell locations across sessions (Ziv et al., 2013) and applied it to our study (Figure 1—figure supplement 2). Consistent with a recent study using GCaMP3 (Ziv et al., 2013), we found that only a small portion of the cells were active on all days of the experiment (380 and 364 cells, corresponding to 7.4 and 7.2% of the cells for environment A and B respectively; in either environment ~20% of these cells were also place coding on all days). Cells that were place coding on more than one day typically had place fields at the same spatial location (Figure 1D–F).

Despite extensive pre-training in both environments, changes over days in the place codes that are unique to one environment or common to both environments may have resulted from a continuous refinement of the representation due to learning or familiarization (Karlsson and Frank, 2008; Lever et al., 2002). However, this is unlikely to be the case here due to several factors: the correlations between the place field maps of the two environments (Figure 1G), the difference in cell activity rates between the two environments (Figure 1—figure supplement 3A), and the distance between the cell’s peak activity in each environment (Figure 1—figure supplement 3B) all did not change over time, indicating a stable spatial remapping. Furthermore, the fractions of cells that were place coding in both environments (Figure 1H), the average number of Ca²⁺ events in a session, the amplitude of the Ca²⁺ events, and the fraction of cells that were place coding on different days of the experiment in each environment, remained similar between environments throughout the experiment (Figure 1—figure supplement 3C-E). Two-photon imaging of the CA1 before and after the experiment showed no indications of cell morbidity, such as time-dependent differences in the cells’ morphology or Ca²⁺ indicator expression patterns (Tian et al., 2009) (Figure 1—figure supplement 3F). Our findings that the functional properties of CA1 neurons remained stable throughout the experiment, argues against the possibility that the changes in spatial representation we observed result from pathological processes (e.g. due to overexpression of the Ca²⁺ indicator or photo-toxicity). Also, the number of sessions in which cells were active, and the fraction of cells with place fields for left and right directions of the linear tracks remained similar for both environments and GCaMP6 variants (Figure 1—figure supplement 4A,B). Overall, our finding of stable remapping despite fluctuating place cell ensembles on different days of the experiment are consistent with previous reports of place cell remapping between different environments within the same day (Leutgeb et al., 2004; Muller and Kubie, 1987), and within a single fixed environment across days and weeks (Ziv et al., 2013).

In our experiment, place cells constituted 46% of the neurons active in either of the environments, and their activity during locomotion contributed 33% of the Ca²⁺ transients we recorded (Figure 1I,J). Non-place cells generated almost half of the Ca²⁺ transients (46%), mostly during resting or reward consumption at the ends of the linear tracks. Thus, neuronal activity during an episode on the linear track was much richer than the part of the data that is limited to place codes. Furthermore, the same cells can be place coding in one environment but active and not place coding in another, or alternatively place coding on one day and active but not place coding on other days. Thus, we sought in our subsequent analyses to account for the entire episodic representations. We therefore opted to include in the following analyses all the active cells and their Ca²⁺ events, regardless of their spatial tuning properties.

We quantified the correlations between the ensemble cell activity patterns in different sessions, in either the same or different environments for pairs of sessions separated by different intervals (Figure 2A,B). For the same environment, both the degree of overlap between populations of cells active on different days, and correlations in their activity patterns gradually declined as a function of elapsed time (Figure 2B, green data). This gradual decay in the correlations at the ensemble level cannot be explained by monotonic changes in event rates at the single cell level (Figure 2—figure supplement 1A,B). Notably, the correlations in cell activity patterns between environments, and the probability of cells recurring within the subset that was active in both environments declined with elapsed time (Figure 2B, red data). Thus, hippocampal representation of the two environments co-evolved with time.

Figure 2. — (A) Pairwise correlations between the complete activity patterns of all cells in different environments and different days (an average of n = 5 mice). The checkerboard pattern reveals the resemblance between the episodic representations of the same environments at different times. (B) Correlations between cell activity patterns within the same environment (green) or between different environments (red) decay with time. This trend is also visible in A when considering the changes in correlations along a row. Inset, the probability that a cell active on one day will be active on subsequent days in the same environment (green) or in a different environment (red) decayed with time (mean ± s.e.m). (C) Performance of the ordinal time decoder. Red dots indicate perfect performance, as was achieved by using data from both environments together (see Figure 2—figure supplement 2). Sorting of the sessions using data from one of the environments was highly significant in comparison with the results obtained using data in which the day labels of each cell’s activity patterns were randomly shuffled (gray dots, n =10 shuffles per mouse). Red lines indicate p = 0.05. (**D,E**) A ‘within-environment time decoder’ was trained and tested on data from the same environment. (D) Distributions of the decoder’s errors in inferring the day from which the test data (single trials) were taken. The decoder successfully inferred the day from which the test data was taken in the majority of cases, but performed at chance level on shuffled data (gray dots, shuffled data as in C). (E) Successful time decoding (decoder error = 0 days) depended on the amount of test data used by the decoder. Shown are percentages of accurate decoding for test data of different durations. Performance using test data segments as short as 1 sec exceeded chance level (12.5%, gray horizontal line). (**F,G**) An ‘across-environments time decoder’ was trained on data from one environment (A or B) and tested on data from the other environment (B or A, respectively). (F) Distributions of the decoder’s errors in inferring the day from which a test data consisting of single sessions (F) or single trials (G) were taken. The decoder successfully inferred the day from which the test data was taken in the majority of cases, but performed at chance level on shuffled data (shuffled data as in C). Data in B and D-G are mean ± s.e.m.

**DOI:** http://dx.doi.org/10.7554/eLife.12247.008

Figure 2—figure supplement 1. — (A) Pairwise correlations between the complete activity patterns of all cells in different environments and different days (an average of n = 5 mice). The checkerboard pattern reveals the resemblance between the episodic representations of the same environments at different times. (B) Correlations between cell activity patterns within the same environment (green) or between different environments (red) decay with time. This trend is also visible in A when considering the changes in correlations along a row. Inset, the probability that a cell active on one day will be active on subsequent days in the same environment (green) or in a different environment (red) decayed with time (mean ± s.e.m). (C) Performance of the ordinal time decoder. Red dots indicate perfect performance, as was achieved by using data from both environments together (see Figure 2—figure supplement 2). Sorting of the sessions using data from one of the environments was highly significant in comparison with the results obtained using data in which the day labels of each cell’s activity patterns were randomly shuffled (gray dots, n =10 shuffles per mouse). Red lines indicate p = 0.05. (**D,E**) A ‘within-environment time decoder’ was trained and tested on data from the same environment. (D) Distributions of the decoder’s errors in inferring the day from which the test data (single trials) were taken. The decoder successfully inferred the day from which the test data was taken in the majority of cases, but performed at chance level on shuffled data (gray dots, shuffled data as in C). (E) Successful time decoding (decoder error = 0 days) depended on the amount of test data used by the decoder. Shown are percentages of accurate decoding for test data of different durations. Performance using test data segments as short as 1 sec exceeded chance level (12.5%, gray horizontal line). (**F,G**) An ‘across-environments time decoder’ was trained on data from one environment (A or B) and tested on data from the other environment (B or A, respectively). (F) Distributions of the decoder’s errors in inferring the day from which a test data consisting of single sessions (F) or single trials (G) were taken. The decoder successfully inferred the day from which the test data was taken in the majority of cases, but performed at chance level on shuffled data (shuffled data as in C). Data in B and D-G are mean ± s.e.m.

**DOI:** http://dx.doi.org/10.7554/eLife.12247.008

To what extent can the dynamics in the representations of the two environments infer the time in which episodes occurred? To address this question while accounting for the neuronal activity at the ensemble level, we constructed three types of neuronal ‘time decoders’. The decoders are based on the knowledge that the similarity between ensemble activity patterns of different episodes decays as a function of elapsed time (Figure 2A,B). We first quantified whether the coding dynamics are informative about the temporal relationships between different events that occurred days to weeks apart within either one or both of the environments. We used an ‘ordinal-time decoder’ to determine if days-scale changes in CA1 ensemble activity are informative about the order of experienced events. This decoder examines data from all days and orders them by maximizing the mean correlation between cells’ firing patterns on neighboring days (Figure 2C and Figure 2—figure supplement 2A). We tested the decoder’s performance on entire sessions, (each 15 min long) for each environment separately and for both environments together, and contrasted the results with those obtained using data in which the day labels of each cell were randomly shuffled. When data from only one of the environments were included in the analysis, performance was significantly superior compared with chance levels for accurate ordinal sorting and the performance of the decoder on shuffled data (Figure 2C, blue data points). Notably, when data from both environments were included in the analysis, the decoder performed perfectly, accurately ordering all sessions along the experiment in all five mice (Figure 2C, red data points, and Figure 2—figure supplement 2A). Thus, day-to-day differences in the activity patterns of CA1 neurons carry ordinal information. This result argues against the possibilities that the observed differences in neuronal activity between sessions are merely the result of inherently unreliable neuronal responses, or due to fluctuations of cell excitability over minutes to hours. Such alternative explanations are unlikely to support perfect ordinal sorting over multiple days.

Days-scale changes in ensemble activity may facilitate the binding in memory of fractions of events within temporal clusters. Accordingly, we expected that different trials or fractions of trials from the same session would be similarly time-stamped. To determine the consistency of such timestamps within the same environment, we devised a decoder that estimates the day in which an episode was recorded based on the correlations between different trials or fractions thereof (‘within-environment time decoder’). Using single 3-min trials as test data, this decoder accurately inferred the days from which the test data was taken (98 ± 1% of the cases). To determine to what degree temporal information is carried in the changes in neuronal firing rates we performed the analysis again, this time relying only on cells that were active on all days. This decoder was accurate in 80% of the cases, significantly higher than chance (z-test, p < 7 · 10^–48, Figure 2D). The decoder’s performance on shuffled test data was essentially identical to chance levels (12.5% for accurate decoding). Remarkably, even data from very short, several-seconds-long fractions of trials, were sufficient for correct decoding of the day of experiment from which the test data was taken (Figure 2E). Thus, a unique timestamp of events arises from changes in the cell activity levels over time. Our data suggest that such timestamps are not only captured within the CA1 representations of complete episodes but are also highly consistent within short fractions of an episode.

If hippocampal dynamics timestamp different events that occur close in time, as our analysis indicates, then such an aspect of the neural code should also infer the time of an event irrespective of a specific environment or a behavioral context. To test this idea we constructed an ‘across-environment time decoder’ that quantifies the degree to which the time of events in one environment can be inferred by a decoder that is trained on data from another environment. We tested this decoder’s ability to correctly identify from which day of the experiment the test data was taken (the test data consisted of entire sessions or single 3-min-long trials). Despite marked differences in the spatial representation of the two environments (Figure 1) and an inter-session interval of 4–5 hr for sessions taking place on the same day, the across-environment time decoder was able to infer the days from which the test data was taken in 65% and 58% of the cases for sessions and trials respectively, significantly higher than chance (z-test, p < 6 · 10^–7 and 4 · 10⁻²², Figure 2F,G, and Figure 2—figure supplement 2B,C). The decoder’s performance was also essentially identical to chance level when tested on data in which the day labels of the cells were randomly shuffled (Figure 2F,G). Note that the decoder’s errors were mostly in proximal days. Indeed, when excluding the training data from the same day of the tested trials or sessions, the decoders inferred the neighboring days in the vast majority of cases (Figure 2—figure supplement 3). These results are consistent with our finding of a gradual co-evolution of the CA1 ensemble representation of both environments (Figure 2A,B). Furthermore, both place codes and non-place codes carried significant information about time (Figure 2—figure supplement 4), suggesting that time stamping is not intrinsically linked to specific aspects of the episodic representation such as place coding.

Here, we exposed a facet of episodic coding dynamics that has thus far remained hidden: the binding and separation of CA1 episodic codes over timescales of days–weeks and across environments. We capitalized on an imaging approach that allows the tracking of Ca²⁺ dynamics in large populations of hippocampal neurons over timescales of weeks (Ziv et al., 2013). Our results rely on our capability to accurately register the neural data across sessions. Re-analyzing the data using a distance-based (rather than correlation based) registration routine (described in Figure 1—figure supplement 2) yielded highly similar results with regard to the performances of the three time decoders (Figure 2—figure supplement 5A–E). Furthermore, the performance of these decoders was robust to alterations in either the thresholds of correlation or distance used by each of the registration routines (Figure 2—figure supplement 5F,G). Overall, these independent analyses confirm that our findings do not stem from a bias intrinsic to our choice of the across-session registration method or threshold.

The ability to place different experienced events on a personal mental timeline is a central feature of episodic memory. Several recent studies have provided important insights into the hippocampal function with respect to coding of time in memory, suggesting that multiple time-coding mechanisms exist to cover different timescales and mnemonic needs (Mankin et al., 2015; Mankin et al., 2012; Manns et al., 2007; Navratilova and Battaglia, 2015). For instance, ‘time cells’, hippocampal pyramidal cells that fire during a particular moment within a temporally structured episode (Eichenbaum, 2014; Kraus et al., 2013; MacDonald et al., 2011; Modi et al., 2014), may contribute to a representation of sequences of events over timescales of tens of seconds. Due to the seconds to minutes-scale nature of their activity, time cells are less likely to place individual uniquely experienced episodes on a linear temporal axis extending over days. Pyramidal neurons in the hippocampal area CA2 (upstream to CA1) were found to display representations that are similar for different contexts but highly variable upon repeated exposures over timescales of hours. It was suggested that such CA2 dynamics, when combined with spatial information from CA3, could jointly reflect spatial and temporal information in CA1 (Mankin et al., 2012, 2015). However, it was argued that the rapid dynamics in CA2 could support encoding a temporal order between events over timescales that range from hours to one day (Navratilova and Battaglia, 2015), whereas many events in our episodic memory are ordered over much longer timescales.

Our present findings of coding dynamics over timescales of days are consistent with a key feature of episodic memory—that each experience is uniquely encoded. This allows for ‘single trial’ learning and events that occur close in time to be linked in long-term memory. This idea is also consistent with previous reports of changes in CA1 neuronal activity over timescales of seconds-hours (Manns et al., 2007), hours-days (Mankin et al., 2012) and days-weeks (Ziv et al., 2013). Such ensemble dynamics may be regarded as a form of ‘remapping’ in the time domain. Spatial remapping is thought to reflect a mechanism for pattern separation, which could help prevent interference between representations of similar places (Colgin et al., 2008). Likewise, we propose that ‘remapping in time’ may be a reflection of a pattern separation mechanism that could mitigate interference between memories of similar events that occurred at different times.

Another aspect of our findings that is key to episodic memory is the co-evolution of the CA1 ensemble representations across spatial environments. While previous work showed that the CA1 place code for a familiar environment changes over timescales of days–weeks (Mankin et al., 2015; Mankin et al., 2012; Ziv et al., 2013) it has remained unclear whether such evolution in the representation of a given environment is independent of the evolution of the representations of other environments. Our findings demonstrate that the representations of two environments were highly distinct, yet sufficiently co-evolved with time to allow us to infer the day in which events occurred in one environment using reference data from the other environment. We propose that such co-evolution may reflect a neural mechanism for binding in long-term memory events that occurred at temporal proximity, even if these events occurred in different contexts and hours-days apart.

Importantly, our data show that such binding/separation of temporally proximal/distal events does not interfere with the ability to distinguish between the two environments; although the spatial representation of each environment changed with time, within environment spatial correlations from sessions that are two weeks apart remained higher than across-environment spatial correlations from the same day (Figure 2—figure supplement 6A,B).

What are the mechanisms that drive ensemble dynamics and time coding over timescales of days? One potential mechanism at the network level is adult neurogenesis in the dentate gyrus (two synapses upstream to CA1) which takes place over timescales of weeks and has been suggested to support time coding via ongoing modification of the hippocampal circuitry by newborn neurons (Aimone et al., 2006, 2009). Another possible mechanism is the recently found days-scale turnover of dendritic spines in basal CA1 dendrites, which could potentially drive coding dynamics at the cellular level (Attardo et al., 2015).

A closely related issue is the role of the interactions between the absolute passage of time and the experience itself on the observed ensemble dynamics. Previous work found no effect for the number of context exposures on the rate of change in CA1 codes over 30 hours (Mankin et al., 2012), but it is possible such an effect could be observed over longer timescales. Our findings are consistent with reports showing that changes in neuronal excitability govern which neurons will be recruited to code and support a memory of a given experience (‘memory allocation’) (Rogerson et al., 2014; Yiu et al., 2014). Repeated training episodes may alter cell excitability via consolidation or reconsolidation processes involving the expression of immediate early genes and their downstream effectors (Dudai, 2012; Lee, 2008; Nader and Hardt, 2009). Such processes may change episodic representations irrespective of the amount of time that has passed between repeated events. Thus, while we have shown in this study that CA1 ensemble dynamics carry temporal information over days, further work is needed to differentiate between the contributions and functional roles of experience-dependent and experience-independent processes that take place in the brain during the passage of time.

Materials and methods

Animals and surgical procedures

All procedures were approved by the Weizmann Institute IACUC. Five male C57BL/6 mice aged 8-12 weeks at the start were used in this study. Mice were housed with 1-4 cage-mates in cages with running wheels, and underwent two surgical procedures under isoflurane anesthesia (1.5-2% volume). First, we injected into the CA1, 400 nL of the viral vector AAV2/5-CaMKIIα-GCaMP6s or AAV2/5-CaMKIIα-GCaMP6f (~2 × 10¹³ particles per ml, packed by University of North Carolina Vector Core)(Chen et al., 2013). Stereotatic coordinates were: -1.9 mm anterio-posterior, -1.4 mm mediolateral, -1.6 mm dorsoventral from bregma. The second surgery, which took place at least one week after the viral injection, was the implantation of a glass guide tube directly above the CA1. We used a trephine drill to remove a circular part of the skull centered posterio-lateral to the viral injection site. We removed the dura and cortex above the CA1 by suction with a 29 gauge blunt needle while constantly washing the exposed tissue with sterile PBS. We then implanted an optical guide tube with its window just dorsal to, but not within, area CA1, and sealed the space between the skull and guide tube using 1.5% agarose in PBS. The exposed areas of the skull were then sealed with Metabond (Parkell, Edgewood, NY) and dental acrylic.

Ca²⁺ imaging and behavioral setup

For time-lapse imaging in freely behaving mice using an integrated miniature fluorescence microscope (nVistaHD, Inscopix), we followed a previously established protocol (Ziv et al., 2013). Briefly, at least three weeks after guide tube implantation, we imaged water restricted mice under isoflurane anesthesia using a two-photon microscope (Ultima IV, Bruker, Germany), equipped with a tunable Ti:Sapphire laser (Insight, Spectra Physics, Santa Clara, CA). We inserted a ‘microendoscope’ consisting of a single gradient refractive index lens (0.44 pitch length, 0.47 NA, GRINtech GmbH, Germany) into the guide tube, and examined Ca²⁺ indicator expression and tissue health. We selected for further imaging only those mice that exhibited homogenous GCaMP6 expression and healthy appearance of the tissue. For the selected mice, we then affixed the microendoscope within the guide tube using ultraviolet-curing adhesive (Norland, NOA81, Edmund Optics, Barrington, NJ). Next, we attached the microscope’s base plate to the dental acrylic cap using light cured acrylic (Flow-It ALC, Pentron, Orange, CA). After a few days, we began training the mice to run back and forth on two elevated linear tracks (Environments A and B). Environment A was a straight 96 cm long track and Environment B was an L-shaped track consisting of two 48 cm long arms. Each environment had distinct sets of visual and tactile cues, overhead lights, flavored liquid rewards, and odor cues. Before the beginning of each pre-training or imaging session we wiped the tracks with differently scented paper towels (0.5% acetic acid for environment A and 10% ethanol for environment B). We trained the mice to run back and forth along the track by giving them a measured amount of water sweetened with commercial fruit juice concentrate, lemon flavored for track A and raspberry flavored for track B, with 2% added sugar by weight. The water reward was dispensed using a custom-made computer controlled device. To record mouse behavior, we used an overhead camera (DFK 33G445, The Imaging Source, Germany), which we synchronized with the integrated microscope. Ca²⁺ imaging was performed at 20Hz. Before beginning with Ca²⁺ imaging, we pre-trained the mice for 8–11 days, until the mice ran at least 60 times the entire length of each track in two consecutive days. Pre-training and imaging sessions consisted of five 3-min-long trials, with an inter-trial interval of 3 min. We imaged a total of 5 mice (two that were injected with AAV2/5-CaMKIIα-GCaMP6f and three that were injected with AAV2/5-CaMKIIα-GCaMP6s) every other day for 15 days, making for 8 recording days. Each day of the experiment consisted of two sessions (AM and PM) separated by 4–5 hr. Remapping tests within a single session were performed on days 16–17. At the end of the experiment we removed the base plate by drilling away the top acrylic cap and re-examined the health of the CA1 neurons by imaging the mice under isoflurane anesthesia using a two-photon microscope as described above.

Processing of Ca²⁺ imaging data

We processed imaging data using commercial software (Mosaic, Inscopix) and custom MATLAB routines as previously described (Ziv et al., 2013). To increase computation speed, we spatially down-sampled the data by a factor of two in each dimension. To correct for non-uniform illumination both in space and time, we normalized the images by dividing each pixel by the corresponding value in a smoothed version. The smoothed version was obtained by applying a Gaussian filter with a radius of 40 pixels on the videos. Normalization enhanced the appearance of the blood vessels, which were later used as stationary fiducial markers for image registration. We used rigid body image registration to correct for lateral displacements of the brain. This procedure was performed on a high contrast subregion of the normalized movies for which the blood vessels were most prominent. The registered movies were transformed to relative changes in fluorescence, $\frac{∆ F' (t)}{F_{0}} = (F' (t) - F'_{0}) / F'_{0}$ , where $F^{'}_{0}$ is the value for each pixel averaged over time. For the purpose of cell identification the movies were downsampled in time by a factor of five. We identified spatial filters corresponding to individual cells using an established cell-sorting algorithm that applies principal and independent component analyses (PCA and ICA) (Mukamel et al., 2009). For each spatial filter, we used a threshold of 50% of the filter’s maximum intensity and each pixel that did not cross the threshold was set to zero. After the cells were identified, further cell sorting was performed to find the spatial filters that follow a typical cellular structure. This was done by measuring the filters’ area and circularity and discarding those whose radius was smaller than 5 μm or larger than 14 μm, or which had a circularity smaller than 0.8. In some cases, the output of the PCA/ICA algorithm included more than one component that corresponded to a single cell. To eliminate such incidents, we examined all cells whose centroids were less than 18 μm apart and whenever their traces had correlation > 0.9, the cell with the lower average peak amplitude was discarded.

Detection of Ca²⁺ events

Ca²⁺activity was extracted by applying the thresholded spatial filters to the full temporal resolution (20Hz) $∆ F' (t) / F_{0}$ videos. Baseline fluctuations were removed by subtracting the median trace (20 s sliding window). The Ca²⁺ traces were smoothed with a low-pass filter with a cutoff frequency of 2Hz. Ca²⁺ candidate events were detected whenever the amplitude crossed a threshold of 4 or 5 median absolute deviations (MAD), for GCaMP6s or GCaMP6f, respectively. Cellular Ca²⁺events are characterized by fast rise and slow decay times. To capture these characteristics in our data we considered for further analysis only candidate Ca²⁺events that followed typical indicator decay time, and decay-to-rise time ratios. In order to avoid the detection of several peaks for a single Ca²⁺ event, only peaks that were 4 or 5 MAD higher than the previous peak (within the same candidate event) and 2 or 2.5 MAD higher than the next peak for GCaMP6s or GCaMP6f, respectively, were regarded as true events. We set the Ca²⁺ event occurrence to the time of the peak fluorescence. To mitigate the effects of crosstalk (i.e., spillover of Ca²⁺ fluorescence from neighboring cells), we adopted a conservative approach, allowing only one cell of a group of neighbors (cells whose centroids are less than 18 μm apart) to register a Ca²⁺ event in a 200 msec time window. If multiple Ca²⁺ events occurred within ~200 msec in neighboring cells, we retained only the events with highest peak $∆ F' (t) / F_{0}$ value. If two neighboring cells had correlation > 0.9 in their events, the cell with the lower average peak amplitude was discarded.

Registration of cells across sessions

For each session we projected centroids of all thresholded filters onto a single image. We computed the spatial cross-correlation among the projections from all sessions to align them according to a reference session. Because changing the reference did not change the alignment output, we chose the first session as the reference. This step corrected slight translations and rotation changes between sessions and yielded each cell’s location in the reference coordinate system. Next, we searched for cells from different sessions that might be the same neuron. This was performed using two separate methods based on either spatial correlations or centroids distances (Figure 1—figure supplement 2). Figures 1 and 2, Figure 1—figure supplement 3, and Figure 2—figure supplements 2–6 show longitudinal data for which we used the spatial correlations-based registration method. Figure 2—figure supplement 5 shows longitudinal data for which we used the centroids distances-based registration method. Within each session, the nearest neighbors spatial correlations were always < 0.6 (Figure 1—figure supplement 2A,C) and the centroids distances were always > 6 μm (Figure 1—figure supplement 2B,D). Between sessions, however, a large amount of cell pairs had spatial correlations > 0.6 and centroid distances < 6 μm. Pairs with spatial correlation > 0.7 or distance < 5 μm were registered as the same neuron. In cases with more than one candidate, the cells with the minimal distance or maximal correlation were assigned to be the same neuron. Analyzing the data using a range of different thresholds demonstrated the robustness of our registration process to the choice of the threshold (Figure 2—figure supplement 5).

Place fields

We analyzed mouse behavior videos using a custom MATLAB (Mathworks) routine that detected the mouse’s center of mass in each frame, calculated its velocity and applied a rectangular smoothing window of 250 msec. For place field analysis, we considered periods when the mouse ran >1 cm s⁻¹. We divided each track into 24 bins (4 cm each) and excluded the last 2 bins at both ends of the tracks where water rewards were consumed and the mouse was generally stationary (Ziv et al., 2013). We computed the time spent in each bin, and the number of Ca²⁺ events per bin, and smoothed these two maps (‘occupancy’ and ‘Ca²⁺ event number’) using a truncated Gaussian kernel (σ = 1.5 bins, size = 5 bins). We then computed the place field map for each neuron by dividing the two smoothed maps of Ca²⁺ event number and occupancy. We separately considered place fields for left and right running directions and normalized each place field by its maximum value. We defined each place field's position at its peak value. For each place field with >5 events for a given session, we computed the spatial information (in bits per event) using the unsmoothed events-rate map of each cell, as previously described (Markus et al., 1994):

S p a t i a l I n f o r m a t i o n = \sum_{i} p_{i} (r_{i} / \bar{r}) l o g_{2} (r_{i} / \bar{r})

Where r_i is the Ca²⁺ event rate of the neuron in the i^th bin; p_i is the probability of the mouse being in the i^th bin (time spent in i^th bin/total session time); r̄ is the overall mean Ca²⁺ event rate; and i running over all the bins. We then performed 1000 distinct shuffles of animal locations during Ca²⁺ events, accounting for the spatial coverage statistics at the relevant session and direction, and calculated the spatial information for each shuffle. This yielded the p value of the measured information relative to the shuffles. Place fields with p ≤ 0.05 were considered significant.

Population vector correlation

To determine the level of similarity between representations of the different environments, we calculated the mean population vector correlation between them (Leutgeb et al., 2005). For each spatial bin (excluding the last 2 bins at both ends of the tracks) we defined the population vector as the mean event rate for each neuron given that bin’s occupancy. We computed the correlation between the population vector in one environment with that of the matching location in the other environment, and averaged the scores over all positions. Since there are two edges to each of the two linear tracks there are two possible transformations between them. Therefore, we used the one that resulted in higher global population vector correlation.

Statistical analysis

We generated the null hypothesis for place fields' displacements between a pair of days by taking the measured centers of place fields in the same environment on the two days and shuffling cells' identities on each of the days. We calculated the distribution of all displacements and averaged them over 10,000 distinct pairs of shuffles. Figure 1F shows the mean null hypothesis for the displacement curve found by averaging over all pairs of days for a given elapsed time. For the analyses shown in Figure 1G,H, and Figure 1—figure supplement 3A-E we used analysis of variance (ANOVA) with repeated measures. Greenhouse-Geisser estimates of sphericity were used for degrees of freedom adjustment.

Time decoders

To capture the temporal information encoded in the hippocampal neural representations of different episodes we constructed three types of time decoders: (1) ordinal time decoder, (2) within-environment time decoder, and (3) across environments time decoder. The time decoders estimated the true order of the recording days from sets of eight episodes (sessions or trials) from the different days in the experiment. Decoding analyses were performed separately for five mice. Vectors of ensemble activity patterns were constructed where each element corresponded to the total number of events of one neuron within an episode. We notated the full-session ensemble activity pattern in day d in environment E as $V_{d}^{E}$ , and the ensemble activity pattern in the t^th trial on day d in environment E as $v_{d, t}^{E}$ . To quantify the similarities between ensemble activity patterns of different episodes we calculated the Pearson correlation between the activity vectors. For the within and between environments time decoders, we normalized each correlation value between a test-data pattern and a training-data pattern by subtracting the average correlation of the training-data vector over all the vectors.

Ordinal time decoder

To sort shuffled, unlabeled activity patterns according to their ordinal positions (chronological order), we sought to maximize the average of all correlations between activity patterns taken on neighboring days in the shuffled test-data. This was achieved by calculating the average correlation between activity patterns for all 20,160 possible permutations of the eight recording days (8!/2) (see Figure 2—figure supplement 2A and Figure 2—figure supplement 5B). The decoder output the day-ordering that maximized the average correlation:

\hat{O} ({V_{i}^{E}}_{i = 1 : 8}) = \underset{< d_{1}, d_{2}, d_{3} \dots d_{8} >}{argmax} \{\frac{1}{7} \sum_{j = 1}^{7} c o r r (V_{d_{j}}^{E}, V_{d_{j + 1}}^{E})\}

Where $\hat{O} ({V_{i}^{E}}_{i = 1 : 8})$ is the inferred days order for a set of eight ensemble activity patterns in environment E, <d₁,d₂,d₃...d₈> is a possible ordering of the eight patterns, and $V_{i}^{E}$ is the ensemble activity pattern of the i^th full-session in environment E (environment A, environment B, or both environments together).

To obtain the significance of the ordinal time decoder performance versus chance level, we divided the number of permutations with a mean correlation that is equal or greater than the mean correlation for the correct order by the total number of possible days’ permutations.

Within environment time decoder

Within environment time decoder inferred the time in which episodes (single trials) in one environment were recorded by comparing their ensemble activity patterns to the activity patterns in all the sessions from the same environment. Specifically, we calculated the normalized correlations between single-trial ensemble activity patterns in one environment (test-data) and each of the full-session activity patterns in all the sessions from the same environment (training-data). To evade bias, in addition to the exclusion of test trials from their sessions in the training-data, we also excluded the corresponding trials from the rest of the sessions in the training-data. Then, the decoder output the time of the session that maximized the correlation with the test-data:

\hat{d} (v_{i, j}^{E}) = \underset{d}{argmax} \{c o r r (v_{i, j}^{E}, V_{d}^{E} - v_{d, j}^{E}) - \underset{d'}{E} [c o r r (v_{d', j}^{E}, V_{d}^{E} - v_{d, j}^{E})]\}

Where $\hat{d} (x)$ is the inferred day for ensemble activity pattern $x$ , $\underset{d^{'}}{E} [\cdot]$ is the average over all $d^{'}$ , $v_{i, j}^{E}$ is the ensemble activity pattern in the j^th trial of the i^th day in environment E and $V_{i}^{E}$ is the ensemble activity pattern in the i^th full-session in environment E. For the results presented in Figure 2—figure supplement 3 and 4 we applied the same decoder while excluding from the training-data the session from the day of the test-data.

Across environments time decoder

Across environments time decoder inferred the time in which episodes (trials or sessions) in one environment were recorded, by comparing their ensemble activity patterns to the activity patterns in all the sessions in the other environment. Decoding was done at the trial level as for the within environment time decoder. For across environments time decoder, decoding was done at the session level as well. Specifically, we calculated the normalized correlation between the full-session ensemble activity pattern in one environment (test-data) and the full-session activity patterns in all the sessions in the other environment (training-data). Then, the decoder output the time of the session that maximized the correlation with the test-data:

Trial-based:

\hat{d} (v_{i, j}^{E_{1}}) = \underset{d}{argmax} \{c o r r (v_{i, j'}^{E_{1}} V_{d}^{E_{2}} - v_{d, j}^{E_{2}}) - \underset{d'}{E} [c o r r (v_{d', j'}^{E_{1}} V_{d}^{E_{2}} - v_{d, j}^{E_{2}})]\}

Session-based:

\hat{d} (V_{i}^{E_{1}}) = \underset{d}{argmax} \{c o r r (V_{i}^{E_{1}}, V_{d}^{E_{2}}) - \underset{d'}{E} [c o r r (V_{d'}^{E_{1}}, V_{d}^{E_{2}})]\}

Where $\hat{d} (x)$ is the inferred day for ensemble activity pattern $x$ , $\underset{d'}{E} [\cdot]$ is the average over all $d'$ , $v_{i, j}^{E}$ is the ensemble activity pattern in the j^th trial of the i^th day in environment E and $V_{i}^{E}$ is the ensemble activity pattern in the i^th full-session in environment E. For the results presented in Figure 2—figure supplement 3 we applied the same decoder while excluding from the training-data the session from the day of the test-data.

Measures of divergence

We used two measures of divergence between the representations of the two environments: ‘activity divergence’ and ‘peak displacement’.

Activity divergence

The ratio of the average absolute difference in event rate between trials in different environments divided by the average absolute difference in average event rate between trials in the same environment:

A c t i v i t y D i v e r g e n c e (d) = \frac{< | r_{n, d, t}^{E} - r_{n, d, t'}^{E'} | {>_{n, t, t'}}^{D}}{< | r_{n, d, t}^{E} - r_{n, d, t'}^{E} | {>_{n, t, t'}}^{S}}

where $r_{n, d, t}^{E}$ is the event rate of the n^th cell in the t^th trial of the d^th day in environment E, $< \cdot {>^{D}}_{n, t, t'}$ denotes the average over all cells and over all pairs of trials $(t and t')$ from different environments $(E and E')$ , $< \cdot {>^{S}}_{n, t, t^{'}}$ denotes the average over all cells and over all pairs of trials from the same environment and |x| denotes the absolute value of x. We modified this analysis from (Lever et al., 2002) to be suited for Ca²⁺ imaging data.

Peak displacement

The average distance between locations of peak activity in different environments:

P e a k D i s p l a c e m e n t (d) = < | p_{n, d}^{E} - p_{n, d}^{E^{'}} | >_{n}

where $p_{n, d}^{E}$ is the location of the peak event rate of the $n^{t h}$ cell in the $d^{t h}$ day in environment $E$ , $< \cdot >_{n}$ denotes the average over all place cells from different environments $(E and E')$ , and |x| denotes the absolute value of x (distance for 1-dimnesional environments). Note that there are two possible transformations between the two environments and here we used the one that resulted in higher global population vector correlation.

Cell-level dynamics

To investigate cell-level dynamics we analyzed the changes in event rates over time for each cell. We applied these analyses separately for each of the two environments.

Maximal monotonic sequence

For each cell we calculated the difference in event rates between consecutive recording days, resulting in a sequence of seven difference values per cell. We then found the longest sub-sequence with the same trend (either monotonic increase or decrease in event rates) and compared the distribution of these lengths obtained from all cells to the distribution of lengths obtained from shuffled data (per-cell random permutation of the order of eight recording days).

Monotonicity score

In order to check whether single cells exhibit monotonic behavior that is obscured by noise we derived a more relaxed definition of monotonicity. For each cell we calculated the difference between the number of day-pairs in which the later had a higher event rate, and the number of day-pairs in which the earlier had a higher event rate, normalized by the total number of day-pairs (28). This resulted in a measure that ranges from -1 for monotonically decreasing cells and +1 for monotonically increasing cells.

Coefficient of variation (CV)

For the cells that were active in all the sessions (in one environment) we calculated the average and the standard deviation of event rates over all eight sessions in that environment. We then compared the distribution of event rates’ CV for those cells to the CV distribution of a population of simulated cells with equivalent average event rates that follow a stationary Poisson model of activity.

Population monotonicity

We aligned each cell’s activity to the day of its maximal event rate and normalized the activity of all days by the maximum event rate. Consequently, the obtained activity level in day-0 is 1 for all cells. Shuffled data was obtained by a per-cell random permutation of the order of eight recording days (Figure 2— figure supplement 1D). To separate the effect of event rate dynamics from cell recruitment dynamics we repeated this procedure for subgroups of cells according to the maximal number of consecutive days in which they were active (2–8 days). For each subgroup we extracted the maximal segment and computed the normalized event rates only for this segment (Figure 2— figure supplement 1E).

Acknowledgements

Yaniv Ziv is incumbent of the Daniel E. Koshland SR. career development chair. We thank Aia Haruvi, Linor Balilti-Turgeman, Noa Brande-Eilat, Yadin Dudai, Ofer Yizhar, and Inbal Goshen for help, advice and comments on the manuscript.

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Funding Information

This paper was supported by the following grants:

Israel Science Foundation 2184/14 to Yaniv Ziv.
European Research Council 638644 to Yaniv Ziv.
Marie Curie Actions CIG 630852 to Yaniv Ziv.
Weizmann Institute of Science Hymen T. Milgrom Trust to Yaniv Ziv.
Weizmann Institute of Science Mr. and Mrs. Mike Kahn Fund for Alzheimer’s related research to Yaniv Ziv.
Weizmann Institute of Science Abraham and Sonia Rochlin Foundation to Yaniv Ziv.
Weizmann Institute of Science Harold and Faye Liss Foundation for Brain Related Research to Yaniv Ziv.
Weizmann Institute of Science Lulu P and David J Levidow Fund for Alzheimer's Diseases and Neuroscience Research to Yaniv Ziv.
Weizmann Institute of Science Mr and Mrs Mike Kahn Fund for Alzheimer's Related Research to Yaniv Ziv.

Additional information

Competing interests

The other authors declare that no competing interests exist.

YZ: Has ownership interests at Inscopix Inc.

Author contributions

AR, Conception and design, Analysis and interpretation of data, Drafting or revising the article.

NG, Conception and design, Acquisition of data, Drafting or revising the article.

LS, Analysis and interpretation of data, Drafting or revising the article.

YZ, Conception and design, Analysis and interpretation of data, Drafting or revising the article.

Ethics

Animal experimentation: All animal work was approved by the Weizmann Institute institutional animal care and use committee (IACUC protocol 18030515-3).

References

Aimone JB, Wiles J, Gage FH. Potential role for adult neurogenesis in the encoding of time in new memories. Nature Neuroscience. 2006;9:723–727. doi: 10.1038/nn1707. [DOI] [PubMed] [Google Scholar]
Aimone JB, Wiles J, Gage FH. Computational influence of adult neurogenesis on memory encoding. Neuron. 2009;61:187–202. doi: 10.1016/j.neuron.2008.11.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
Attardo A, Fitzgerald JE, Schnitzer MJ. Impermanence of dendritic spines in live adult CA1 hippocampus. Nature. 2015;523:592–596. doi: 10.1038/nature14467. [DOI] [PMC free article] [PubMed] [Google Scholar]
Barretto RPJ, Ko TH, Jung JC, Wang TJ, Capps G, Waters AC, Ziv Y, Attardo A, Recht L, Schnitzer MJ. Time-lapse imaging of disease progression in deep brain areas using fluorescence microendoscopy. Nature Medicine. 2011;17:223–228. doi: 10.1038/nm.2292. [DOI] [PMC free article] [PubMed] [Google Scholar]
Chen T-W, Wardill TJ, Sun Y, Pulver SR, Renninger SL, Baohan A, Schreiter ER, Kerr RA, Orger MB, Jayaraman V, Looger LL, Svoboda K, Kim DS. Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature. 2013;499:295–300. doi: 10.1038/nature12354. [DOI] [PMC free article] [PubMed] [Google Scholar]
Colgin LL, Moser EI, Moser M-B. Understanding memory through hippocampal remapping. Trends in Neurosciences. 2008;31:469–477. doi: 10.1016/j.tins.2008.06.008. [DOI] [PubMed] [Google Scholar]
Davachi L, DuBrow S. How the hippocampus preserves order: the role of prediction and context. Trends in Cognitive Sciences. 2015;19:92–99. doi: 10.1016/j.tics.2014.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
Dudai Y. The restless engram: consolidations never end. Annual Review of Neuroscience. 2012;35:227–247. doi: 10.1146/annurev-neuro-062111-150500. [DOI] [PubMed] [Google Scholar]
Eichenbaum H. Time cells in the hippocampus: a new dimension for mapping memories. Nature Reviews Neuroscience. 2014;15:732–744. doi: 10.1038/nrn3827. [DOI] [PMC free article] [PubMed] [Google Scholar]
Ghosh KK, Burns LD, Cocker ED, Nimmerjahn A, Ziv Y, Gamal AE, Schnitzer MJ. Miniaturized integration of a fluorescence microscope. Nature Methods. 2011;8:871–878. doi: 10.1038/nmeth.1694. [DOI] [PMC free article] [PubMed] [Google Scholar]
Howard MW, Kahana MJ. A distributed representation of temporal context. Journal of Mathematical Psychology. 2002;46:269–299. doi: 10.1006/jmps.2001.1388. [DOI] [Google Scholar]
Howard MW, MacDonald CJ, Tiganj Z, Shankar KH, Du Q, Hasselmo ME, Eichenbaum H. A unified mathematical framework for coding time, space, and sequences in the hippocampal region. Journal of Neuroscience. 2014;34:4692–4707. doi: 10.1523/JNEUROSCI.5808-12.2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
Karlsson MP, Frank LM. Network dynamics underlying the formation of sparse, informative representations in the hippocampus. Journal of Neuroscience. 2008;28:14271–14281. doi: 10.1523/JNEUROSCI.4261-08.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
Kragel JE, Morton NW, Polyn SM. Neural activity in the medial temporal lobe reveals the fidelity of mental time travel. Journal of Neuroscience. 2015;35:2914–2926. doi: 10.1523/JNEUROSCI.3378-14.2015. [DOI] [PMC free article] [PubMed] [Google Scholar]
Kraus BJ, Robinson RJ, White JA, Eichenbaum H, Hasselmo ME. Hippocampal “time cells”: time versus path integration. Neuron. 2013;78:1090–1101. doi: 10.1016/j.neuron.2013.04.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
Lee JLC. Memory reconsolidation mediates the strengthening of memories by additional learning. Nature Neuroscience. 2008;11:1264–1266. doi: 10.1038/nn.2205. [DOI] [PubMed] [Google Scholar]
Leutgeb S. Distinct ensemble codes in hippocampal areas CA3 and CA1. Science. 2004;305:1295–1298. doi: 10.1126/science.1100265. [DOI] [PubMed] [Google Scholar]
Leutgeb S. Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science. 2005;309:619–623. doi: 10.1126/science.1114037. [DOI] [PubMed] [Google Scholar]
Lever C, Wills T, Cacucci F, Burgess N, O'Keefe J. Long-term plasticity in hippocampal place-cell representation of environmental geometry. Nature. 2002;416:90–94. doi: 10.1038/416090a. [DOI] [PubMed] [Google Scholar]
MacDonald CJ, Lepage KQ, Eden UT, Eichenbaum H. Hippocampal “time cells” bridge the gap in memory for discontiguous events. Neuron. 2011;71:737–749. doi: 10.1016/j.neuron.2011.07.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
Mankin EA, Sparks FT, Slayyeh B, Sutherland RJ, Leutgeb S, Leutgeb JK. Neuronal code for extended time in the hippocampus. Proceedings of the National Academy of Sciences of the United States of America. 2012;109:19462–19467. doi: 10.1073/pnas.1214107109. [DOI] [PMC free article] [PubMed] [Google Scholar]
Mankin EA, Diehl GW, Sparks FT, Leutgeb S, Leutgeb JK. Hippocampal CA2 activity patterns change over time to a larger extent than between spatial contexts. Neuron. 2015;85:190–201. doi: 10.1016/j.neuron.2014.12.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
Manns JR, Howard MW, Eichenbaum H. Gradual changes in hippocampal activity support remembering the order of events. Neuron. 2007;56:530–540. doi: 10.1016/j.neuron.2007.08.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
Markus EJ, Barnes CA, McNaughton BL, Gladden VL, Skaggs WE. Spatial information content and reliability of hippocampal CA1 neurons: effects of visual input. Hippocampus. 1994;4:410–421. doi: 10.1002/hipo.450040404. [DOI] [PubMed] [Google Scholar]
Modi MN, Dhawale AK, Bhalla US. CA1 cell activity sequences emerge after reorganization of network correlation structure during associative learning. eLife. 2013;3:e12247. doi: 10.7554/eLife.01982. [DOI] [PMC free article] [PubMed] [Google Scholar]
Mukamel EA, Nimmerjahn A, Schnitzer MJ. Automated analysis of cellular signals from large-scale calcium imaging data. Neuron. 2009;63:747–760. doi: 10.1016/j.neuron.2009.08.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
Muller RU, Kubie JL. The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. The Journal of Neuroscience. 1987;7:1951–1968. doi: 10.1523/JNEUROSCI.07-07-01951.1987. [DOI] [PMC free article] [PubMed] [Google Scholar]
Nader K, Hardt O. A single standard for memory: the case for reconsolidation. Nature Reviews Neuroscience. 2009;10:224–234. doi: 10.1038/nrn2590. [DOI] [PubMed] [Google Scholar]
Navratilova Z, Battaglia FP. CA2: it’s about time—and episodes. Neuron. 2015;85:8–10. doi: 10.1016/j.neuron.2014.12.044. [DOI] [PubMed] [Google Scholar]
Nyberg L, Kim ASN, Habib R, Levine B, Tulving E. Consciousness of subjective time in the brain. Proceedings of the National Academy of Sciences of the United States of America. 2010;107:22356–22359. doi: 10.1073/pnas.1016823108. [DOI] [PMC free article] [PubMed] [Google Scholar]
O'Keefe J, Dostrovsky J. The hippocampus as a spatial map. preliminary evidence from unit activity in the freely-moving rat. Brain Research. 1971;34:171–175. doi: 10.1016/0006-8993(71)90358-1. [DOI] [PubMed] [Google Scholar]
O'Keefe JN. The Hippocampus as a Cognitive Map. Oxford: Clarendon Oxford; 1978. [Google Scholar]
Rogerson T, Cai DJ, Frank A, Sano Y, Shobe J, Lopez-Aranda MF, Silva AJ. Synaptic tagging during memory allocation. Nature Reviews Neuroscience. 2014;15:157–169. doi: 10.1038/nrn3667. [DOI] [PMC free article] [PubMed] [Google Scholar]
Rolls ET. A computational theory of episodic memory formation in the hippocampus. Behavioural Brain Research. 2010;215:180–196. doi: 10.1016/j.bbr.2010.03.027. [DOI] [PubMed] [Google Scholar]
Shapiro ML. Time and again. Neuron. 2014;81:964–966. doi: 10.1016/j.neuron.2014.02.034. [DOI] [PubMed] [Google Scholar]
Tian L, Hires SA, Mao T, Huber D, Chiappe ME, Chalasani SH, Petreanu L, Akerboom J, McKinney SA, Schreiter ER, Bargmann CI, Jayaraman V, Svoboda K, Looger LL. Imaging neural activity in worms, flies and mice with improved GCaMP calcium indicators. Nature Methods. 2009;6:875–881. doi: 10.1038/nmeth.1398. [DOI] [PMC free article] [PubMed] [Google Scholar]
Tulving E. Episodic memory: from mind to brain. Annual Review of Psychology. 2002;53:1–25. doi: 10.1146/annurev.psych.53.100901.135114. [DOI] [PubMed] [Google Scholar]
Yiu AP, Mercaldo V, Yan C, Richards B, Rashid AJ, Hsiang H-LL, Pressey J, Mahadevan V, Tran MM, Kushner SA, Woodin MA, Frankland PW, Josselyn SA. Neurons are recruited to a memory trace based on relative neuronal excitability immediately before training. Neuron. 2014;83:722–735. doi: 10.1016/j.neuron.2014.07.017. [DOI] [PubMed] [Google Scholar]
Ziv Y, Burns LD, Cocker ED, Hamel EO, Ghosh KK, Kitch LJ, Gamal AE, Schnitzer MJ. Long-term dynamics of CA1 hippocampal place codes. Nature Neuroscience. 2013;16:264–266. doi: 10.1038/nn.3329. [DOI] [PMC free article] [PubMed] [Google Scholar]

eLife. 2015 Dec 18;4:e12247. doi: 10.7554/eLife.12247.015

Decision letter

Editor: Howard Eichenbaum¹

eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.

Thank you for submitting your work entitled "Hippocampal ensemble dynamics timestamp events in long-term memory" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor (Howard Eichenbaum) and Timothy Behrens as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Stephan Leutgeb and Albert Lee (peer reviewers).

The reviewers have discussed the reviews with one another and the Reviewing editor has drafted this decision to help you prepare a revised submission.

The reviewers were generally enthusiastic about the study and the promise of the optical recording approach highlighted here. At the same time, the reviewers were concerned about the novelty of the results compared to published work by the author and others. The reviewers were clear that substantial novelty existed, but thought that the writing of the manuscript made it unclear which results were new and which could be inferred from previous data. It is essential that the authors clearly lay out what is confirmatory of existing findings with the current approach and what is qualitatively new here. The reviewers had several other major concerns and recommendations outlined below that should be addressed.

Reviewer #1:

The use of calcium imaging of large numbers of neurons over many days in awake and locomoting mice is potentially a very important technological contribution to the study of how populations of hippocampal neurons change over time and contribute to memory. The question though is whether or not the technological advancement evident in the current study resulted in substantive advancement of the field above and beyond the methodology. One complication in evaluating this latter question is the extent to which the authors focused on only the most recent literature and thus provided a narrow perspective of how their study fit into existing knowledge. Numerous earlier studies and models of how gradually changing neural representations could add a temporal dimension to memory are ignored. In addition, of the more recent work that is cited (e.g., Mankin et al., 2012), the authors do not clearly show how the present results differ from and thus add to the prior results (aside from the imaging technique used to obtain the data). As a result, one gets the sense that the current study reveals more about the promise of the technique rather than contributing new findings.

Reviewer #2:

The report by Rubin et al. finds that hippocampal CA1 ensembles gradually change over time such that ensemble that are recorded several days apart do not resemble each other. In contrast, they find that CA1 ensembles that are recorded on the same day in two different environments resemble each other. While the number of cells that can be recorded with calcium imaging is certainly much higher than with standard electrophysiological techniques, the reported results largely resemble those that were obtained in previous recording studies which have already reported and proposed many of the key conceptual points. Notably, closely related publications (e.g., Manns et al., 2007; Howard and Kahana, 2002, and others) are not included here and standard analysis methods (detailed comparison of firing during immobility and across the two environments) are not performed.

1) The Introduction begins with mentioning the work of others, but in the second half, predominantly cites the work from their own group (Ziv et al., 2013). Furthermore, it is not mentioned that Mankin et al. (2012 and 2014) recorded in two environments and reported correlations between the two environments, which show a similar pattern as those reported here.

2) Paragraph four. Non-place cells are presented as if it were not a well-known fact that many cells that are not place cells are active during sharp-wave ripples and in other phases of the task. In fact, many well performed electrophysiology studies report the fraction of cells that are place cells compared to those that are only observed during immobility. A much more detailed analysis of the firing patterns of non-place cells within and across environments is needed before any conclusions about their contribution to stability over time can be drawn.

3) Paragraph five. A much more detailed analysis of the correlation across environments for place cells and non-place cells is needed. It is of course expected that correlations within environments are higher than between environments but the key question is the extent of the neuronal coding difference for the two environments. From the information that is currently provided it appears that A and B are partially correlated and that each environment is more correlated with itself from at temporally proximal compared to temporally distant time points. If A and B and A and A' are correlated, is appears trivial that B and A' are also correlated, and all classifiers and conclusions seem to be based on this finding.

4) Paragraphs six and nine. These results have been shown before by others.

5) Paragraphs eleven and thirteen. Previous similar findings are not discussed.

6) Paragraph twelve. The largest fraction of change is found within 1-2 days, which is not consistent with the time course of neurogenesis. In contrast to what is reported here (i.e., constant remapping between environments), Rangel et al. (2014) found a change in the degree of remapping over the time course of neurogenesis and when manipulating neurogenesis.

7) Paragraph thirteen. While the experimental design in Mankin et al. (2012) could distinguished between effects of elapsed time and intervening experience, no such results are reported here. If this is discussed, the text should refer to previously published data on this topic.

Reviewer #3:

In this manuscript the authors build on previous work (Ziv et al., 2013) using the same technique of freely moving 1-photon imaging of hippocampal CA1 populations employed there. In the previous study they showed significant changes to the place cell representation of a single environment over ~35 days primarily due to continual changes to the subset of cells that were active (using GCaMP3 as the indicator). In particular, a cell that was active on a given day was less and less likely to be active in the same environment over several weeks of time. However, they also showed that if a cell was active at different time points within this ~35 day period that their place fields were in the same location, indicating stability of the spatial representation in that sense. Given the amount of change they observed in the representation, it is important to know to what extent this amount of change may affect how spatial memories can be recalled over the long term. One way to investigate this is to consider how the representations of two different environments could be distinguished over time, which is the experiment presented here. Here they show that while the activity levels (a graded version of the active/inactive distinction that Ziv et al. (2013) focused on) of individual cells changed over time, the representations of the environments were distinguishable from each other with a similar level of contrast at any given time point, indicating a similar magnitude of global remapping over time. They also showed that the activity levels of both environments changed gradually over time so that the amount of similarity in activity levels at different time points could be used to distinguish how far apart in time those experiences were. An important aspect of this last point was that the activity levels of the cells changed in a similar fashion for both environments, allowing them to determine when an experience occurred in a given environment by using the activity levels over time from either that or the other environment.

Long-term tracking of the activity patterns of the same set of cells is an important and understudied area. The extent to which a representation stays the same versus changes – and the manner in which it changes – over long time periods is critical for understanding memory storage and recall. There are several points in the manuscript that I believe should be addressed, but overall this is a valuable contribution to the field of hippocampal research which is also potentially relevant for memory representations in other brain areas (e.g. the amygdala).

Major comments:

1) The Abstract exclusively emphasizes the time evolution of CA1 representations independent of spatial context. The absence of a clear statement in the Abstract that important aspects of the representations of spatial contexts remain constant over time – the preserved locations of place fields (already in the Ziv et al., 2013 Abstract) and the preserved global remapping between spatial contexts (new here) – could lead to misunderstanding by readers. Therefore, the authors should modify the Abstract so that it is clear which important aspects of the representation of spatial contexts remain stable over long time periods.

2) The ability to order representations of each environment across days could be inferred from Ziv et al. (2013). What would have been the alternative to this, given the findings in this previous work? (On the other hand, the stability of global remapping across time and the co-evolution of activity levels across multiple environments is new.) The authors should delve into the details of the time evolution of single-environment representations more. For example, is the gradual evolution of activity levels over time due to half of the neurons gradually increasing their activity levels over the 2 weeks and half decreasing? Or do many neurons increase then decrease, or decrease then increase, activity over that period? Or does some substantial fraction of cells have activity levels that remain basically the same over time? What are the percentages of each kind of pattern?

3) For the within-environment decoder, it seems not surprising that the activity pattern in a trial of a day would best correlate with the activity pattern on the rest of that day. The important issue is whether the pattern evolves gradually over time versus just being noisy over time. Therefore, what the authors should do is remove the session from the day of the test trial and see if the best match from the remaining days came from a day that was closer to the actual day than chance. Similarly, this calculation should be done for the across-environments decoder (that is, eliminating the same day's data). (However, it is useful to also have the answer when the same day is included, which is the calculation they already did.)

4) An important part of the memory function of the hippocampus would be the ability to distinguish between the two spatial contexts and correctly identify them with previous experience in each of those contexts. While the authors showed intact global remapping between environments within each day, they should analyze if they can distinguish which previous environment they are recalling across days. Specifically, they should compare how the spatial population vectors of the representation of environment A on day n for n = 3, 5, […] correlate with the spatial population vectors for environment A and separately for environment B on day n = 1. Then they should repeat it for day 5, 7, […] versus day 3, and so on. (This should also be done for environment B versus environment B or A on different days.) This analysis will essentially be an across-days combination of Figure 1G and Figure 2A-B.

5) The authors write "Thus, having shown that remapping between familiar environments is stable over time, we tried in our subsequent analyses to account for the entire episodic representations. We therefore opted to include in the following analyses all the active cells and their Ca²⁺ events, regardless of their spatial running properties." They describe in the Materials and methods that they excluded the 2 spatial bins on each end of the tracks when defining place fields, but did they then include these bins when determining the spatial population vector correlation between representations in different environments to show remapping? And I assume based on the quoted text above that they did not exclude the 2 spatial bins on each end of the tracks when defining the "ensemble activity patterns" for each time period, which was used to do their time evolution analysis? This should be clarified in the Materials and methods and main text.

6) Related to (5): They appear to have excluded the reward-related activity at the track ends in the global remapping analysis. They appear to have included this activity in the "ensemble activity pattern" vectors – which contain the relative activity of each cell in a given environment, independent of the locations at which that activity for each cell occurred – for the time ordering analysis. Assuming this is the case, the authors should re-run the analysis for the time ordering results using only the subset of cells that show place activity and, separately, all cells excluding the track ends where reward is consumed because the remaining activity would represent the theta-state associated activity patterns. This analysis would also reveal to what extent the reward-related and other activity at the track ends contributes to their time ordering results.

7) One important difference between this study and the previous one (Ziv et al., 2013) is the use of GCaMP6s and GCaMP6f here versus the previous use of GCaMP3. Both studies point out that there is significant change in the hippocampal representation over time. However, the GCaMP6 variants are more sensitive detectors of neural activity than GCaMP3, which detects strong bursts more than single spikes. Because models of memory can depend on parameter values such as how stable activity is over time, it would be very valuable for the hippocampal field to have a quantitative comparison between the key numbers from the previous GCaMP3-based study and the GCaMP6-based values here (especially since other differences in methodology are likely to be low due to overlap in the author lists). Therefore, the authors should quantitatively compare their data to Ziv et al. (2013) by adding to the current manuscript's supplementary figures the same plots corresponding to Figure 2E, Figure 3B, and Figure 3C of Ziv et al. (2013). It looks like one of these (the fraction of cells that are place cells) is already shown in both (Ziv et al., 2013 Figure 2E that shows a place cell fraction of ~10-15% and this manuscript's Figure 1H shows a value of ~50%), but it's not clear that it has been computed in exactly the same way. As for Figure 3C in Ziv et al. (2013), the previous study used a binary distinction between active and inactive cells. The current manuscript's Figure 2B uses a more graded activity level measure for cells. In the inset of Figure 2B, does the "overlap" plot use exactly the same definition of active and inactive cells as Ziv et al. (2013) Figure 3C? Finally, the authors should show separate plots for GCaMP6s and for 6F, which could have different detectability for events.

8) How does the data relate to Lever et al. (2002)? In that paper, it was shown that the representations of the two environments diverged over time. Here, did the representations of the two environments separate more with time in certain aspects? While the authors show that their global remapping score is stable over time, it looks like Figure 2A shows some degree of increasing divergence even though Day 1 already represents ~2 weeks of familiarization with both environments. The authors should check for this and report any change or lack of change.

eLife. 2015 Dec 18;4:e12247. doi: 10.7554/eLife.12247.016

Author response

Reviewers were generally enthusiastic about the study and the promise of the optical recording approach highlighted here. At the same time, the reviewers were concerned about the novelty of the results compared to published work by the author and others. The reviewers were clear that substantial novelty existed, but thought that the writing of the manuscript made it unclear which results were new and which could be inferred from previous data. It is essential that the authors clearly lay out what is confirmatory of existing findings with the current approach and what is qualitatively new here. The reviewers had several other major concerns and recommendations outlined below that should be addressed.

We thank the reviewers for their comments and suggestions for improvement of the manuscript, and appreciate that they were enthusiastic about the study. In the revised manuscript we have clarified which results are confirmatory of previous findings and which ones constitute a conceptual advance. To address specific points raised by the reviewers, we have revised two figures from the original submission and added five new Supplementary Figures, overall containing 27 new display items. We believe that these additions have substantially improved the manuscript, and we thank the reviewers for the constructive criticism that prompted these additional analyses. We summarize here the main novel contributions of our paper, as well as the new material added in the revised submission:

• We now emphasize the key novelty of our findings – that the CA1 ensemble representations of two distinct familiar environments co-evolve with time. This co-evolution allowed us to infer the day in which events occurred in one environment using reference data from the other environment, pointing to a potential mechanism for binding in long-term memory events that occurred at temporal proximity, even if these events occurred in different contexts and hours-days apart.

• Our ensemble-level analysis of the data, done for individual mice and days, and particularly our decoding approaches, provide the first quantitative measure regarding the degree to which the ensemble dynamics carry information about time.

• In the revised manuscript, we separately analyze the performance of the time decoders on place code and non-place code data, and show that both categories of activity carry information about the temporal relationship between experienced events. This finding suggests that time stamping is not intrinsically linked to specific aspects of the episodic representation such as place coding.

• We now show that the population vector correlation between activity patterns in two sessions that took place in the same environment was higher than the correlation for two sessions that took place in different environment regardless of the elapsed time between the sessions, suggesting that binding or separation of temporally proximal or distal events does not interfere with the ability to distinguish between the two environments – as expected for intact long-term memory.

• We now demonstrate that the performance of the within-environment time decoder does not solely rely on within-day similarity in activity patterns, further supporting the notion that ensemble dynamics over multiple days could associate episodes that occurred close in time.

• We now added detailed analysis, done separately for GCaMP6s and GCaMP6f, which confirms previous findings with an inferior Ca²⁺ indicator, GCaMP3.

• Finally, we show that the gradual decay in the correlations at the ensemble level between spatial representations cannot be explained by monotonic changes in event rates at the single cell level.

Reviewer #1:

Thank you for these remarks which helped point out that we did not adequately present the conceptual novelty of this work. We hope the revised manuscript now clarifies this matter. In this work we tested the hypothesis that CA1 ensemble dynamics over timescales of days carry information about the temporal relationship between experienced events, both within and across spatial contexts. We wish to elucidate that unlike recent studies that introduced the imaging technique that we have used here (Ziv et al., 2013, and Ghosh et al., 2011), the current work is not about the technique. Several conceptual aspects of our work prove it to be a substantive advancement beyond the methodology:

1) Distinct CA1 representations co-evolve with time.

We agree that the idea that changes in neuronal activity could infer time has been investigated in multiple studies. For example, Manns et al., 2007 showed population evolution at a time scale of minutes and Mankin et al., 2015 showed dynamics in CA2 that allowed the distinction of representations for up to 6 hours (CA2 was shown to reach a plateau at longer time intervals). However, unlike previous studies, in this work we focused on the co-evolution of CA1 representations of different contexts. Until now it has not been explicitly demonstrated that CA1 neural codes can be both substantially different between contexts and yet co-evolve with time in a manner that allows the binding in memory of events that occurred at different contexts. In the revised manuscript we further substantiate that the CA1 representations of the two environments are highly distinct (Figure 1C, G, Figure 1—figure supplement 3A,B and Figure 2— figure supplement 6). (This is an important point because it would not be surprising if representations that were similar to each other were found to co-evolve with time). We also show that our time decoders could infer the day from which the test data was taken – both within and across environments – even if we use data from only place cells or from only non-place cells, suggesting that time stamping is not intrinsically linked to specific aspects of the episodic representation such as place coding. Overall, the results support our hypothesis, and suggest that CA1 ensemble dynamics can both associate or dissociate distinct experienced events in long-term memory based on their temporal distance.

2) Days-scale dynamics are relevant to long-term episodic memory.

Whereas previous electrophysiological studies have attributed a role in time coding for changes in hippocampal neuronal activity over timescales of seconds to ~1.5 days, our study investigated CA1 coding dynamics over much longer timescales of days–weeks. Dynamics over such timescales are relevant for long-term episodic and autobiographical memory, and specifically for the ability to form a mental timeline of events that were experienced days or weeks apart. By analogy, just as recordings of place cells in simple small-sized environments can only inform us to limited extent about the way larger real-life environments are represented, recordings over minutes–hours may not reveal all aspects of the dynamics that are relevant to coding of time in long-term memory. Our results show that the hippocampus could enable the formation of a timeline of memories over extended days-weeks timescales while maintaining coding specificity for different environments, consistent with its hypothesized role of encoding ‘where’ and ‘when’ into long-term memory.

3) “Time decoders” quantify time information carried by the ensemble dynamics.

Whereas previous studies have averaged the data over mice and/or recording days, the large-scale recordings in our study allowed us to longitudinally analyze the data at the ensemble level, and for each mouse individually. To quantify the degree to which the ensemble dynamics is informative about time, we devised three different approaches for neuronal “time decoding”. Using these analyses we demonstrate that events that occurred at temporal proximity (e.g. the same day) shared a common timestamp, even if these events have occurred in two distinct spatial contexts. We also found that information about time is embedded in the ensemble activity even in short, seconds-long fractions of a session. These are new results, which could not have been derived from previous studies that characterized the dynamics of modest numbers of neurons per animal. Furthermore, thanks to suggestions by the Reviewers, we now provide a detailed analysis of the dynamics of individual cells, and of different subsets of cells. The new analyses show that weak monotonic changes in firing rates at the single cell level can translate into strong monotonic changes at the ensemble level (Figure 2—figure supplement 1), and that both place cells and non-place cells contribute information about time (Figure 2—figure supplement 4).

Reviewer #2:

We agree that our findings are consistent with previous electrophysiological studies in the field, and recognize that our writing did not sufficiently emphasize the novelty of our results with respect to these studies. We hope the revised manuscript now clarifies this issue. Thank you also for drawing our attention to the studies by Manns et al. (2007) and Howard & Kahana (2002) which are now cited in the Introduction.

We now cite the studies by Manns et al. (2007) and Howard & Kahana (2002) in the Introduction, and in the Discussion mention that Mankin et al. (2012 and 2015) recorded in two environments. The revised manuscript now states:

“Pyramidal neurons in the hippocampal area CA2 (upstream to CA1) were found to display representations that are similar for different contexts but highly variable upon repeated exposures over timescales of hours. It was suggested that such CA2 dynamics, when combined with spatial information from CA3, could jointly reflect spatial and temporal information in CA1 (Mankin et al., 2012, 2015).”

Our intention was not to imply that it was not well known that non-place cells are also active during different phases of the exploration task. The aim was to show that in our data non-place cells generated nearly half of the Ca²⁺ transients; but by no means do we claim this to be a new finding. We omitted the words “typically analyzed” from the text to avoid giving this impression. The revised manuscript now states:

“Thus, neuronal activity during a training episode on the linear track was much richer than the part of the data that is limited to place codes.”

A much more detailed analysis of the firing patterns of non-place cells within and across environments is needed before any conclusions about their contribution to stability over time can be drawn.

Thank you for this suggestion. We now provide a new supplementary figure (Figure 2—figure supplement 4) with a detailed analysis of the temporal information carried by place cells and non-place cells. First, we compared the performance of the time decoders on Ca²⁺-event data from epochs in which the animal was either moving or stationary (mostly at the ends of the tracks). The performance of the time decoder on these two parts of the data was similar. Next, we compared the performance of time decoders that were trained on data from either place cells or non-place cells. Both subsets of cells supported a highly significant performance of the decoders, and were not significantly different from each other.

Thank you for suggesting this. We now added Figure 2—figure supplement 6, which presents the correlations of the place cell population vector within and between the two environments. The correlation values between environments were low throughout the experiment. Thus, at any point along the course of the experiment the place cell representation of the two environments was different. As mentioned in the previous point, Figure 2—figure supplement 4 shows that both place cells and non-place cells carried information about time.

From the information that is currently provided is appears that A and B are partially correlated and that each environment is more correlated with itself from at temporally proximal compared to temporally distant time points. If A and B and A and A' are correlated, it appears trivial that B and A' are also correlated, and all classifiers and conclusions seem to be based on this finding.

If we understand correctly the notation you use (i.e. A and B are the representations of the two environments, and B’ is the representation of the same environment as in B at a different time), our data suggest that the correlation between A and B is higher than the correlation between A and B’. This result could not have been derived from the correlations between A and A’, or B and B’ alone. To place this logic in perspective, consider that previous work showed that the place code for a fixed environment changes over timescales of days–weeks. But what has remained unknown until our current study is the degree to which such days-scale evolution in the hippocampal CA1 representation of a given environment is independent from the evolution of the representations of other environments. We think this point is far from trivial: if the dynamics of the representations of different environments were independent, then different visits to the same environment would still be encoded as different traces; however, with such a coding scheme events that occurred at different environments but at temporal proximity will not be associated in long-term memory. Contrary to this scenario, our current work revealed that the representations of two different environments co-evolved with time, sufficing to infer the time in which events occurred in one environment using reference data from the other environment.

4) Paragraphs six and nine. These results have been shown before by others.

The reviewer refers to the results of the ordinal time decoder, which indicate that day-to-day differences in activity patterns of CA1 neurons carry ordinal information. We accept that our findings are consistent with previous work that showed changes in CA1 neuronal activity over seconds-hours (Manns et al., 2007), hours-days (Mankin et al., 2012) and days-weeks (Ziv et al., 2013), and now state this explicitly in the revised manuscript. However, to the best of our knowledge, none of our ensemble-level time decoding approaches regarding the degree to which days-scale ensemble dynamics carry information about time have been published before. On this point the revised manuscript now states:

“Our present findings of coding dynamics over timescales of days are consistent with a key feature of episodic memory— that each experience is uniquely encoded. […] This idea is also consistent with previous reports of changes in CA1 neuronal activity over timescales of seconds-hours (Manns et al., 2007), hours-days (Mankin et al., 2012) and days-weeks (Ziv et al., 2013).”

5) Paragraphs eleven and thirteen. Previous similar findings are not discussed.

We have revised the text and now cite and discuss previous similar work (see points #4 above, and #7 below).

6) Paragraph twelve. The largest fraction of change is found within 1-2 days, which is not consistent with the time course of neurogenesis.

Indeed, the large fraction of the change is within 1-2 days, but results from the current study show a gradual change over the course of weeks. Thanks to the reviewers’ suggestions, we now show that this gradual change is sufficient for the performance of the time decoders (Figure 2—figure supplement 3). Furthermore, there is currently no evidence at the neural code level regarding the degree to which neurogenesis in the DG affects representational stability in the downstream CA1. However, there is relevant evidence at the behavioral level: neurogenesis rates were found to be inversely correlated with the strength of long-term episodic-like memories, or the duration of such memories were hippocampus dependent (Akers et al., Science 2014, and Kitamura et al., Cell 2009). Given that the CA1 is the main output from the hippocampus, it is plausible that such behavioral effects were associated with representational differences in CA1.

In contrast to what is reported here (i.e., constant remapping between environments), Rangel et al. (2014) found a change in the degree of remapping over the time course of neurogenesis and when manipulating neurogenesis.

We do not think our results are inconsistent with those reported by Rangel et al. In their work, cells exhibited higher context selectivity if the rats were introduced to the different contexts gradually over weeks (first context 1, then context 2, and finally context 3), than if exposed to all contexts in the same week. Our experimental design is very different: we pre-trained the mice for 8-11 days in both environments, before starting the two-week long imaging experiment, but did not examine the effects of gradual versus simultaneous introduction of the two environments on their representations or dynamics. Beyond this basic conceptual difference in study design, there are multiple additional differences between the studies that make a direct comparison between them difficult. For example, in Rangel et al., they recorded from the DG of rats whereas we recorded from the CA1 of mice. The DG and CA1 have different qualitative and functional differences. Rangel et al. pooled data from four days of recordings at the end of the training phase, whereas we recorded longitudinally over two weeks. They did not compare neuronal activity from the same animals at different intervals along the course of the experiment, whereas we did. In their experiments rats explored environments of different sizes, whereas in our experiments the environments were the same size. In their experiments distal external cues were the same for the different environments, whereas in our experiment they were different.

We agree with this comment and have now added a sentence to the Discussion that makes reference to this previous work. The revised manuscript now states:

”Previous work found no effect for the number context exposures on the rate of change in CA1 codes over 30 hours (Mankin et al., 2012), but it is possible such an effect could be observed over longer timescales.”

Reviewer #3:

Done. We have revised the Abstract, and added the information about the important aspects of the representation that remains stable over time. The new sentence states:

“Longitudinal analysis exposed ongoing environment-independent evolution of episodic representations, despite stable place field locations and constant remapping between the two environments.”

The results of previous work indeed hinted that we should be able, given similar data, to order the representations of each environment. However, given that the analyses in Ziv et al. (2013) are based on data averaged over four mice, and in most cases also over day-pairs, it has been unclear to what degree ordering the representations will be accurate at the individual mouse and day levels. The gradual changes that were observed when averaged over all mice could originate from abrupt changes in the representations in individual mice. Such abrupt changes in the representations of each environment would not support perfect ordinal sorting of the data. Furthermore, taken together with our findings of the co-evolution of the representation of the two environments, we have now ruled out the possibility that the previously reported results regarding time evolution of a representation of a given environment are only due to dynamics in the representation that are specific to that environment.

The authors should delve into the details of the time evolution of single-environment representations more. For example, is the gradual evolution of activity levels over time due to half of the neurons gradually increasing their activity levels over the 2 weeks and half decreasing? Or do many neurons increase then decrease, or decrease then increase, activity over that period? Or does some substantial fraction of cells have activity levels that remain basically the same over time? What are the percentages of each kind of pattern?

We now added a new supplementary figure (Figure 2—figure supplement 1) that presents a detailed analysis of the cell-level time evolution in a single environment. Specifically, we found that:

1) There were no cells that exhibit monotonic increase or decrease in activity throughout the two-week period of the experiment.

2) Nevertheless, the single-cell dynamics were slightly more monotonic than the shuffled data. Such weak monotonic behavior at the single cell level seems to underlie monotonic dynamics at the ensemble level.

3) Even neurons that were active throughout all eight days of the experiment did not tend to demonstrate a constant firing rate.

Thank you for this suggestion – we agree that it is not surprising that the activity patterns in a trial of a session would best correlate with the activity pattern of the rest of the session. We have now added these analyses in a new supplementary figure (Figure 2—figure supplement 3). The results confirm that the ensemble activity evolves gradually over time.

We have now added these analyses in a new supplementary figure (Figure 2—figure supplement 6). The results confirm that at any point throughout the experiment the ensemble representation of the two environments was different.

Yes, the last two bins of the tracks were excluded from the analysis of the population vector correlations (Figure 1G), but included in the analysis of the ensemble activity patterns (Figure 2). This is now clarified in revised manuscript.

The caption for Figure 1 now states:

“(G) Average correlations of place cell firing patterns (i.e. the ‘population vector’, PV) during running epochs at different locations along the two linear tracks were low and did not change over time”

The Materials and methods section now states:

“For each spatial bin (excluding the last 2 bins at both ends of the tracks) we defined the population vector as the mean event rate for each neuron given that bin’s occupancy.”

Thank you for this suggestion. To quantify whether non-place codes contained temporal information we analyzed the performance of the different time decoders, separately for periods in which the mouse was active versus stationary (Figure 2—figure supplement 4A-C), and for place cells versus non-place cells (Figure 2—figure supplement 4D-F). Active periods were defined as times when the animal was at velocity $\geq 1 cm / sec$ and located more than 8cm from either end of the track. Place cells were defined as cells with significant place coding in at least one session. Figure 2—figure supplement 4 now shows that: (1) Ca²⁺ events taken from epochs in which the mouse was either active or stationary, and (2) place cells and non-place cells all carried significant temporal information. Thus, time stamping is not intrinsically linked to specific aspects of the episodic representation such as place coding.

We have now added a new supplementary figure (Figure 1—figure supplement 4) that addresses these issues, and separately presents for GCaMP6s and GCaMP6f the analyses done in Ziv et al. (2013) for: (1) number of days that the cells were active in each environment, (2) fraction of place cells active during rightward and leftward movement in each environment, and (3) the changes in the ensemble activity correlations, or overlap, within and between environments. Please note that while we present here data that was analyzed as in Ziv et al. (2013), several other aspects of the current experiment (beyond the use of a more sensitive indicator) were different than the previous studies. For example, the interval between imaging days was five days in the previous study and two days in ours, which may account for some of the differences seen in the overlap between the place cell ensemble that were active on different days; in the previous work mice explored a single environment, whereas in the current study mice explored two different environments in each day of recording; both the length of the linear tracks and their distance from distal visual cues are longer in the current study, which may have contributed to some of the differences in the fraction of place cells, or the degree to which they are direction selective.

As suggested, we adopted a similar approach to that presented in Lever et al. (2002) and added two new analyses of the degree to which the representational difference between the two environments changes with time (Figure 1—figure supplement 4A,B). Due to the nature of the Ca²⁺ imaging data, in which bursts of action potentials are often detected as single Ca²⁺ excitation events and not as trains of spikes (as in electrophysiological recordings), it is hard to reliably evaluate a spatial activity map of single neurons based on a single 3-minute trial. By comparison, Lever et al. typically analyzed 8-10 minute trials. We therefore devised two scores: ‘activity divergence’ and ‘peak displacement’, which resemble Lever’s ‘rate divergence’ and ‘peak divergence’ scores, but different in that they do not require the evaluation of firing map based on data from a single trial. Importantly, according to our new analyses the dynamics of neither of these scores changed over the time scale of the experiment.

[bib1] Aimone JB, Wiles J, Gage FH. Potential role for adult neurogenesis in the encoding of time in new memories. Nature Neuroscience. 2006;9:723–727. doi: 10.1038/nn1707. [DOI] [PubMed] [Google Scholar]

[bib2] Aimone JB, Wiles J, Gage FH. Computational influence of adult neurogenesis on memory encoding. Neuron. 2009;61:187–202. doi: 10.1016/j.neuron.2008.11.026. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib3] Attardo A, Fitzgerald JE, Schnitzer MJ. Impermanence of dendritic spines in live adult CA1 hippocampus. Nature. 2015;523:592–596. doi: 10.1038/nature14467. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib4] Barretto RPJ, Ko TH, Jung JC, Wang TJ, Capps G, Waters AC, Ziv Y, Attardo A, Recht L, Schnitzer MJ. Time-lapse imaging of disease progression in deep brain areas using fluorescence microendoscopy. Nature Medicine. 2011;17:223–228. doi: 10.1038/nm.2292. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib5] Chen T-W, Wardill TJ, Sun Y, Pulver SR, Renninger SL, Baohan A, Schreiter ER, Kerr RA, Orger MB, Jayaraman V, Looger LL, Svoboda K, Kim DS. Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature. 2013;499:295–300. doi: 10.1038/nature12354. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib6] Colgin LL, Moser EI, Moser M-B. Understanding memory through hippocampal remapping. Trends in Neurosciences. 2008;31:469–477. doi: 10.1016/j.tins.2008.06.008. [DOI] [PubMed] [Google Scholar]

[bib7] Davachi L, DuBrow S. How the hippocampus preserves order: the role of prediction and context. Trends in Cognitive Sciences. 2015;19:92–99. doi: 10.1016/j.tics.2014.12.004. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib8] Dudai Y. The restless engram: consolidations never end. Annual Review of Neuroscience. 2012;35:227–247. doi: 10.1146/annurev-neuro-062111-150500. [DOI] [PubMed] [Google Scholar]

[bib9] Eichenbaum H. Time cells in the hippocampus: a new dimension for mapping memories. Nature Reviews Neuroscience. 2014;15:732–744. doi: 10.1038/nrn3827. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib10] Ghosh KK, Burns LD, Cocker ED, Nimmerjahn A, Ziv Y, Gamal AE, Schnitzer MJ. Miniaturized integration of a fluorescence microscope. Nature Methods. 2011;8:871–878. doi: 10.1038/nmeth.1694. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib11] Howard MW, Kahana MJ. A distributed representation of temporal context. Journal of Mathematical Psychology. 2002;46:269–299. doi: 10.1006/jmps.2001.1388. [DOI] [Google Scholar]

[bib12] Howard MW, MacDonald CJ, Tiganj Z, Shankar KH, Du Q, Hasselmo ME, Eichenbaum H. A unified mathematical framework for coding time, space, and sequences in the hippocampal region. Journal of Neuroscience. 2014;34:4692–4707. doi: 10.1523/JNEUROSCI.5808-12.2014. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib13] Karlsson MP, Frank LM. Network dynamics underlying the formation of sparse, informative representations in the hippocampus. Journal of Neuroscience. 2008;28:14271–14281. doi: 10.1523/JNEUROSCI.4261-08.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib14] Kragel JE, Morton NW, Polyn SM. Neural activity in the medial temporal lobe reveals the fidelity of mental time travel. Journal of Neuroscience. 2015;35:2914–2926. doi: 10.1523/JNEUROSCI.3378-14.2015. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib15] Kraus BJ, Robinson RJ, White JA, Eichenbaum H, Hasselmo ME. Hippocampal “time cells”: time versus path integration. Neuron. 2013;78:1090–1101. doi: 10.1016/j.neuron.2013.04.015. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib16] Lee JLC. Memory reconsolidation mediates the strengthening of memories by additional learning. Nature Neuroscience. 2008;11:1264–1266. doi: 10.1038/nn.2205. [DOI] [PubMed] [Google Scholar]

[bib17] Leutgeb S. Distinct ensemble codes in hippocampal areas CA3 and CA1. Science. 2004;305:1295–1298. doi: 10.1126/science.1100265. [DOI] [PubMed] [Google Scholar]

[bib18] Leutgeb S. Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science. 2005;309:619–623. doi: 10.1126/science.1114037. [DOI] [PubMed] [Google Scholar]

[bib19] Lever C, Wills T, Cacucci F, Burgess N, O'Keefe J. Long-term plasticity in hippocampal place-cell representation of environmental geometry. Nature. 2002;416:90–94. doi: 10.1038/416090a. [DOI] [PubMed] [Google Scholar]

[bib20] MacDonald CJ, Lepage KQ, Eden UT, Eichenbaum H. Hippocampal “time cells” bridge the gap in memory for discontiguous events. Neuron. 2011;71:737–749. doi: 10.1016/j.neuron.2011.07.012. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib21] Mankin EA, Sparks FT, Slayyeh B, Sutherland RJ, Leutgeb S, Leutgeb JK. Neuronal code for extended time in the hippocampus. Proceedings of the National Academy of Sciences of the United States of America. 2012;109:19462–19467. doi: 10.1073/pnas.1214107109. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib22] Mankin EA, Diehl GW, Sparks FT, Leutgeb S, Leutgeb JK. Hippocampal CA2 activity patterns change over time to a larger extent than between spatial contexts. Neuron. 2015;85:190–201. doi: 10.1016/j.neuron.2014.12.001. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib23] Manns JR, Howard MW, Eichenbaum H. Gradual changes in hippocampal activity support remembering the order of events. Neuron. 2007;56:530–540. doi: 10.1016/j.neuron.2007.08.017. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib24] Markus EJ, Barnes CA, McNaughton BL, Gladden VL, Skaggs WE. Spatial information content and reliability of hippocampal CA1 neurons: effects of visual input. Hippocampus. 1994;4:410–421. doi: 10.1002/hipo.450040404. [DOI] [PubMed] [Google Scholar]

[bib25] Modi MN, Dhawale AK, Bhalla US. CA1 cell activity sequences emerge after reorganization of network correlation structure during associative learning. eLife. 2013;3:e12247. doi: 10.7554/eLife.01982. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib26] Mukamel EA, Nimmerjahn A, Schnitzer MJ. Automated analysis of cellular signals from large-scale calcium imaging data. Neuron. 2009;63:747–760. doi: 10.1016/j.neuron.2009.08.009. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib27] Muller RU, Kubie JL. The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. The Journal of Neuroscience. 1987;7:1951–1968. doi: 10.1523/JNEUROSCI.07-07-01951.1987. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib28] Nader K, Hardt O. A single standard for memory: the case for reconsolidation. Nature Reviews Neuroscience. 2009;10:224–234. doi: 10.1038/nrn2590. [DOI] [PubMed] [Google Scholar]

[bib29] Navratilova Z, Battaglia FP. CA2: it’s about time—and episodes. Neuron. 2015;85:8–10. doi: 10.1016/j.neuron.2014.12.044. [DOI] [PubMed] [Google Scholar]

[bib30] Nyberg L, Kim ASN, Habib R, Levine B, Tulving E. Consciousness of subjective time in the brain. Proceedings of the National Academy of Sciences of the United States of America. 2010;107:22356–22359. doi: 10.1073/pnas.1016823108. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib31] O'Keefe J, Dostrovsky J. The hippocampus as a spatial map. preliminary evidence from unit activity in the freely-moving rat. Brain Research. 1971;34:171–175. doi: 10.1016/0006-8993(71)90358-1. [DOI] [PubMed] [Google Scholar]

[bib32] O'Keefe JN. The Hippocampus as a Cognitive Map. Oxford: Clarendon Oxford; 1978. [Google Scholar]

[bib33] Rogerson T, Cai DJ, Frank A, Sano Y, Shobe J, Lopez-Aranda MF, Silva AJ. Synaptic tagging during memory allocation. Nature Reviews Neuroscience. 2014;15:157–169. doi: 10.1038/nrn3667. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib34] Rolls ET. A computational theory of episodic memory formation in the hippocampus. Behavioural Brain Research. 2010;215:180–196. doi: 10.1016/j.bbr.2010.03.027. [DOI] [PubMed] [Google Scholar]

[bib35] Shapiro ML. Time and again. Neuron. 2014;81:964–966. doi: 10.1016/j.neuron.2014.02.034. [DOI] [PubMed] [Google Scholar]

[bib36] Tian L, Hires SA, Mao T, Huber D, Chiappe ME, Chalasani SH, Petreanu L, Akerboom J, McKinney SA, Schreiter ER, Bargmann CI, Jayaraman V, Svoboda K, Looger LL. Imaging neural activity in worms, flies and mice with improved GCaMP calcium indicators. Nature Methods. 2009;6:875–881. doi: 10.1038/nmeth.1398. [DOI] [PMC free article] [PubMed] [Google Scholar]

[bib37] Tulving E. Episodic memory: from mind to brain. Annual Review of Psychology. 2002;53:1–25. doi: 10.1146/annurev.psych.53.100901.135114. [DOI] [PubMed] [Google Scholar]

[bib38] Yiu AP, Mercaldo V, Yan C, Richards B, Rashid AJ, Hsiang H-LL, Pressey J, Mahadevan V, Tran MM, Kushner SA, Woodin MA, Frankland PW, Josselyn SA. Neurons are recruited to a memory trace based on relative neuronal excitability immediately before training. Neuron. 2014;83:722–735. doi: 10.1016/j.neuron.2014.07.017. [DOI] [PubMed] [Google Scholar]

[bib39] Ziv Y, Burns LD, Cocker ED, Hamel EO, Ghosh KK, Kitch LJ, Gamal AE, Schnitzer MJ. Long-term dynamics of CA1 hippocampal place codes. Nature Neuroscience. 2013;16:264–266. doi: 10.1038/nn.3329. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Hippocampal ensemble dynamics timestamp events in long-term memory

Alon Rubin

Nitzan Geva

Liron Sheintuch

Yaniv Ziv

Roles

Abstract

eLife digest

Figure 1. Stable spatial remapping between two familiar environments despite evolving place codes.

Figure 1—figure supplement 1. CA1 neurons exhibit selective Ca2+ responses in the two environments.

Figure 1—figure supplement 2. Accurate across-session cell identification based on correlation or distance between candidate cells.

Figure 1—figure supplement 3. Stable neuronal function and environment-specific activity.

Figure 1—figure supplement 4. Activity and long-term dynamics of GCaMP6s- and GCaMP6f expressing cells.

Figure 2. CA1 ensemble codes are informative about the temporal relationship between experienced events.

Figure 2—figure supplement 1. Within-environment cell-level dynamics (A–E).

Figure 2—figure supplement 2. Time decoders accurately expose temporal information in all individual mice.

Figure 2—figure supplement 3. Time decoders accurately infer the time of the episode even when the tested session is excluded from the training data (A–D).

Figure 2—figure supplement 4. CA1 ensemble place codes and non-place codes are informative about the temporal relationship between experienced events.

Figure 2—figure supplement 5. Time decoding is robust to changes in the across-session cell identification method.

Figure 2—figure supplement 6. The two environments can be distinguished according to CA1 place cell firing patterns regardless of recording day (A,B).

Materials and methods

Animals and surgical procedures

Ca2+ imaging and behavioral setup

Processing of Ca2+ imaging data

Detection of Ca2+ events

Registration of cells across sessions

Place fields

Population vector correlation

Statistical analysis

Time decoders

Ordinal time decoder

Within environment time decoder

Across environments time decoder

Measures of divergence

Activity divergence

Peak displacement

Cell-level dynamics

Maximal monotonic sequence

Monotonicity score

Coefficient of variation (CV)

Population monotonicity

Acknowledgements

Funding Statement

Funding Information

Additional information

Competing interests

Author contributions

Ethics

References

Decision letter

Roles

Author response

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases

Ca²⁺ imaging and behavioral setup

Processing of Ca²⁺ imaging data

Detection of Ca²⁺ events