Abstract
Spatial memories are represented by hippocampal place cells during navigation. This spatial code is dynamic, undergoing changes across time, known as representational drift, and across changes in internal state, even while navigating the same spatial environment with consistent behavior. A dynamic code may provide the hippocampus a means to track distinct epochs of experience that occur at different times or during different internal states and update spatial memories. Changes to the spatial code include place fields (PFs) that remap to new locations and place fields that vanish, while others are stable. However, what determines place field fate across epochs remains unclear. We measured the lap-by-lap properties of place cells in mice during navigation for a block of trials in a rewarded virtual environment. We then determined the position of the place fields in another block of trials in the same spatial environment either separated by a day (a distinct temporal epoch) or during the same session but with reward removed to change reward expectation (a distinct internal state epoch). We found that place cells with remapped place fields across epochs tended to have lower spatial precision during navigation in the initial epoch. Place cells with stable or vanished place fields tended to have higher spatial precision. We conclude that place cells with less precise place fields have greater spatial flexibility, allowing them to respond to, and track, distinct epochs of experience in the same spatial environment, while place cells with precise place fields generally preserve spatial information when their fields reappear.
Significance Statement
Spatial representations in the hippocampus support spatial memory and are comprised of place cells and their corresponding place fields (PFs). Recent work has revealed that spatial representations are not as stable as once thought. Instead, they undergo large changes across days, a phenomenon called representational drift. We present results showing an association between the spatial precision of place fields on 1 d and their drift across subsequent days. We find this association holds true when spatial representations drift in response to changes in reward expectation. We discuss a synaptic-level conceptual model that links place field precision with drift. This study advances our understanding of the mechanisms behind representational drift, providing insight into how the hippocampus updates spatial memories for continual learning.
Introduction
The hippocampus is known to play a role in encoding, consolidating, updating and retrieving episodic memories (Andersen et al., 2006). Within the hippocampus, there are subsets of cells known as place cells, which exhibit spatial activity patterns corresponding to the animal’s location within a specific environment (J. O’Keefe and Dostrovsky, 1971). These locations are referred to as place fields (PFs), and as a population they provide a spatial representation of a given environment. The faithful reinstatement of hippocampal representations is thought to support memory retrieval (J.N. O’Keefe, 1978; Gelbard-Sagiv et al., 2008; Liu et al., 2012; Josselyn et al., 2015; Frankland et al., 2019). However, recent findings show that spatial representations change with time and experience even when animals are navigating the same environment (Hainmueller and Bartos, 2018; J.S. Lee et al., 2020; Mau et al., 2020; Dong et al., 2021; Driscoll et al., 2022; Keinath et al., 2022). This phenomenon is known as representational drift and can occur during navigation of an environment from lap-to-lap, as demonstrated by many PFs shifting backwards (Mehta et al., 2000; Dong et al., 2021; Roth et al., 2012), and across repeated exposures (epochs of experience) to the same environment on different days (Ziv et al., 2013; Dong et al., 2021). Representational drift may track time (Mankin et al., 2012, 2015; Rubin et al., 2015) or amount of experience (Geva et al., 2023; Khatib et al., 2023). Similar changes to the spatial code are observed when animal’s undergo an internal state change during navigation, as demonstrated when attention or reward expectation is altered in an unchanging spatial environment (Krishnan et al., 2022; Pettit et al., 2022). At the single cell level, the fate of preexisting place fields falls into one of three categories. First, place cells can remap their PFs to new locations. Second, PFs can vanish. Third, place fields can remain stable. However, what determines the fate of PFs across time or internal state changes remains unclear.
To investigate this, we reanalyzed previously published data, where two-photon Ca
Materials and Methods
Subjects
All experimental and surgical procedures were in accordance with the Animal Care and Use Committee guidelines of The University of Chicago. For this study, 10- to 12-week-old male C57BL/6J wild-type (WT) mice (23–33 g) were individually housed in a reverse 12/12 h light/dark cycle with an ambient temperature of ∼20°C and ∼50% humidity. Male mice were used over female mice because of the size and weight of the headplates (9.1 × 31.7 mm, ∼2 g), which were difficult to firmly attach to smaller female skulls. All training and experiments were conducted during the animal’s dark cycle.
Mouse surgery and virus injection
Mice were anesthetized (∼1–2% isoflurane) and injected with 0.5 ml of saline (intraperitoneal injection) and ∼0.45 ml of meloxicam (1–2 mg/kg, s.c.). For CA1 population imaging, a small (∼0.5–1.0 mm) craniotomy was made over the hippocampus CA1 (1.7 mm lateral, −2.3 mm caudal of bregma). A genetically encoded calcium indicator, AAV1-CamKII-GCaMP6f (Addgene, #100834) was injected into CA1 (∼75 nl) at a depth of 1.25 mm below the surface of the dura using a beveled glass micropipette. Afterwards, the site was covered up using dental cement (Metabond, Parkell Corporation) and a metal head-plate (9.1 × 31.7 mm, Atlas Tool and Die Works) was also attached to the skull with the cement. Mice were separated into individual cages and water restriction began the following day (0.8–1.0 ml/d). Around 7 d later, mice underwent another surgery to implant a hippocampal window as previously described (Dombeck et al., 2010). Following implantation, the head plate was reattached with the addition of a head ring cemented on top of the head plate which was used to house the microscope objective and block out ambient light. Postsurgery, mice were given 2–3 ml of water per day for 3 d to enhance recovery before returning to the reduced water schedule (0.8–1.0 ml/d).
Behavior and calcium imaging
We analyzed previously published data.
For the experiment across days (Dong et al., 2021), mice (n = 5 in total) were trained to run on a treadmill along a 3-m virtual reality (VR) linear track with 4 μm of water reward delivered at the end of the track [the familiar environment (F)]. Mice (n = 3, #1, #2, #3) were then imaged over 2 d in the same environment, without exposure to any novel environment. Calcium activity in CA1 pyramidal neurons (n = 1282 neurons) were extracted using customized MATLAB script (Sheffield et al., 2017), with parameters and procedures detailed in (Dong et al., 2021).
For the experiment involving the novel environment (N) switch, mice were trained to run on a treadmill in F, and then on imaging days, the mice (n = 3, #3, #4, #5) were introduced to N with different 3D visual cues but the same reward location and track length as F. Note that this novel environment N is N2 in Dong et al. (2021). Also note that mouse #3 was also imaged for the experiment in the familiar environment. Again, calcium activity of CA1 pyramidal neurons (n = 1704 neurons) were extracted with the same approach. Table 1 lists the experimental conditions each animal went through and how they were included in (Dong et al., 2021).
Subjects
For multiday imaging datasets, we take an average image at the end of the imaging session on day 1 and use it as a template to find the exact same field of view (FOV) the following day in real time (i.e., before we start collecting data on that day). We match the FOVs within 1 μm of z-plane alignment (that is the limit of our microscope controller). We further check alignment by concatenating the FOVs across days and motion correcting them together as one single time-series movie. This corrects for any differences in X and Y position and allows us to closely inspect any z-differences at the frame transition from day 1 to day 2. If movies from different days are rotated relative to one another, Fiji (ImageJ) is used to correct any rotational displacement between the two movies. We do not use any FOVs that have any noticeable differences in z-planes at the transition. All the data shown in this paper are matched in X, Y, and Z. Example of FOVs across days can be found in Figure 1B. ROIs are extracted after concatenation. To rule out that the results are not because of imaging artifacts, Extended Data Figure 1-4 shows that there is no correlation between anatomic location of place cells in the FOV and their PF fate. If any imaging artifacts were affecting our PF measurements, the three categories would be equally affected.
Place fields that remap across 2 d in a familiar environment have lower spatial precision on day 1. A, Experimental design. Top, The familiar virtual reality (VR) environment. Bottom, Animals trained and recorded in the same VR environment. B, Example field of view showing imaging of the same cells on days 1 and 2. C, Top, Behavior of a single animal showing track location. Blue dots indicate Ca2+ fluorescence peaks from an example cell relative to the animal’s track location. Bottom, Fluorescence trace of the example cell across time. Blue dots indicating the peak of the Fluorescence change. D, Examples of Stable, Remapped, and Vanished place fields across days. Bins near ends of track are excluded. E, Same as D, but place fields are identified by transient peaks. Colored (blue, orange, and green) dots indicate in-field fluorescence peaks. Black dots are out of field peaks. F, Average population backward shift of PF peaks on day 1. PFs are aligned to their onset lap. Line indicates fitted exponential curve:
Extended Data Figure 1-1
Change in PF locations between blocks of trials. A, Change in PF locations across days in the familiar environment. Median of each category shown in figure. B, Change in PF locations across days in the novel environment. Median of each category shown in figure. C, Change in PF locations between R and UR. Median of each category shown in figure. D, Change in PF locations between R and RR. Median of each category shown in figure. Download Figure 1-1, EPS file.
Extended Data Figure 1-2
Following the initial epoch, Remapped and Stable PFs have similar spatial precision during subsequent epochs. A, Comparison of spatial precision of PFs on day 2 in the familiar environment. Right, Bootstrapped median difference between Remapped versus Stable PFs from day 1. 5000 resamples, p = 0.421. B, Comparison of spatial precision of PFs on day 2 in the novel environment. Right, Bootstrapped median difference between Remapped versus Stable PFs from day 1. 5000 resamples, p = 0.811. C, Comparison of spatial precision of PFs in UR condition. Right, Bootstrapped median difference between Remapped versus Stable PFs from R. 5000 resamples, p = 0.869. D, Comparison of spatial precision of PFs in RR condition. Right, Bootstrapped median difference between Remapped versus Stable PFs from R. 5000 resamples, p = 0.848. Download Figure 1-2, EPS file.
Extended Data Figure 1-3
Comparison of median spatial precision of place fields for individual animals. A, Median spatial precision (as measured by the SD of PF peaks) in each category for individual animals in the familiar environment. Significant differences between PF categories are shown using one-way ANOVA (F = 8.814, p = 0.0164), Tukey’s post hoc test shows a significant difference for Stable versus Remapped (p = 0.0218), and for Remapped versus Vanished (p = 0.0300). B, Same as A but for novel environment. Significant differences between PF categories are shown using one-way ANOVA (F = 8.43, p = 0.01810), Tukey’s post hoc test shows a significant difference for Stable versus Remapped (p = 0.0394), and for Remapped versus Vanished (p = 0.0212). C, Same as A but for the Rewarded (R)-Unrewarded (UR) experiment. PFs are categorized by their fate in UR but their spatial precision is measured in R. Significant differences between PF categories is shown using one-way ANOVA (F = 14.24, p = 0.000679), Tukey’s post hoc test shows a significant difference for Stable versus Remapped (p = 4.15E-03), and for Remapped versus Vanished (p = 7.97E-04). D, Same as A but for the Rewarded (R)-re-rewarded (RR) experiment. PFs are categorized by their fate in RR following an epoch of UR, but their spatial precision is measured in R. Significant differences between PF categories is shown using one-way ANOVA (F = 21.05, p = 0.000119), Tukey’s post hoc test shows a significant difference for Stable versus Remapped (p = 2.569E-04), and for Remapped versus Vanished (p = 3.71E-04). Download Figure 1-3, EPS file.
Extended Data Figure 1-4
Anatomical location of place cells color-coded based on their PF fate in three example animals. A, Anatomical location in x and y of place cells [relative to center of field of view (FOV)] in one example animal (#3, imaged across 2 d in familiar environment). Origin represents center of FOV. B, Comparison of anatomical location of place cells from center of FOV grouped by PF fate category (Stable, N = 108; Remapped, N = 32; Vanished, N = 87). One-way ANOVA shows no statistically significant difference between categories (F = 1.160, p = 0.315). C, Same as A but for a different animal (#4, imaged across 2 d in novel environment). D, Same as B but for animal #4 (Stable, N = 176; Remapped, N = 82; Vanished, N = 170). One-way ANOVA shows no statistically significant difference between categories (F = 0.157, p = 0.855). E, Same as A but for a different animal [imaged across blocks of trials with changing reward (R-UR-RR)]. PFs from R are categorized by their fate in UR. F, Same as B but for the animal in E (Stable, N = 68; Remapped, N = 32; Vanished, N = 46). One-way ANOVA shows no statistically significant difference between categories (F = 1.896, p = 0.154). Download Figure 1-4, EPS file.
Extended Data Figure 1-5
Other place field metrics are not associated with place field fate across days in a familiar environment. A, Comparison of backward shifting dynamics between the different PF fate categories. Fitting parameters for the exponential
Extended Data Table 1-1
Statistics Download Table 1-1, XLSX file.
For the experiment with change in reward contingencies (Krishnan et al., 2022), mice were trained on a 2-m VR linear track for water reward. Well-trained mice showed preemptive licking before the reward location. On experimental day, the mice (n = 5) ran in the environment with reward (R), then the reward was unexpectedly removed (UR). The reward was then reintroduced (RR). Each condition (R, UR, and RR) lasted 8–10 min. Population activity of CA1 pyramidal neurons (n = 1288) were measured with Ca
Defining PFs
After extracting significant calcium transients, we correlated the transients to the animals’ behavior. We obtained the significant peak of each transient by finding the local maximum of a transient that exceeds the mean
The vast majority of cells had either a single cluster or no cluster in any given epoch. In the limited number of cells that had multiple clusters in either epoch, each cluster was treated as independent PF. For the very few cells with multiple clusters in epoch 1 and epoch 2, the clusters across the epochs were “paired,” such that clusters in epoch 1 were paired with the closet ones in epoch 2 and analyzed as such. In cases with two clusters in epoch 1 and one cluster in epoch 2, we considered the clusters to have merged across epochs If both clusters in the first epoch were within 40 cm of the epoch 2 cluster (if one or both were not, we did not consider them merged and treated each independently). The two clusters in epoch 1 were therefore counted as one Stable PF, and we combined their spatial precision values (see PF properties on how we did this), to avoid overcounting the Stable PFs. This caused the difference in total number of PFs in R for Figure 3C,D, because of cells with multiple clusters in R having different fates in UR and RR.
PF properties
Once clusters were identified as PFs, the time, location, and transient peak
The spatial location of the PF was defined as the median of the locations of the events within the cluster. The spatial precision of the PF was quantified using the SD of the location in the cluster. Therefore, a precise PF means a lower lap-to-lap variation in firing location, and hence a lower SD. To ensure that this measurement was not influenced by the edges of the track, PFs located within 10 cm of the track ends were excluded from the analysis.
In the rare cases where two PFs merged into a single PF across epochs (see above, Defining PFs), we combined their precision measure into a single value and only counted it as a single Stable PF. To obtain a single precision value from the two PFs, we obtained the location of each event from each cluster and then subtracted the mean location of that cluster from each event
The PF firing rate dynamics were investigated by using the
PF categorization
To assess the spatial stability of PFs, we categorized them based on their change in spatial location across days/conditions:
Stable: the PF is present on both days/conditions and any change in PF position is:
Remapped: the PF is present on both days/conditions but changes PF position:
Vanished: The PF is only present on the first day/condition.
The specific choice of 40 cm as the threshold was made to ensure that it was larger than the typical fluctuation observed in peak locations within clusters in the dataset (a measure of PF width). This means that a change in location beyond 40 cm would typically mean a nonoverlapping PF.
PF backward shifting
PFs were aligned by their onset lap, which was defined as the lap number of the first peak in that cluster. Then the spatial positions of the PFs on each lap were obtained with a sliding window of five laps. The five-lap sliding average position of individual PFs were then compared with the median location calculated from laps beyond the 15th lap from the onset lap for that PF. For each lap from the PF onset lap, the average shift over the population of PFs was calculated and plotted. An exponential fit (least square fit, scipy.optimize.curve_fit) was then applied to the trend. The error of the fitting parameters was obtained by the square-root of the diagonal elements of the covariance matrix (returned by curve_fit). The same trend was observed without the smoothing.
PFs were considered to have ceased systematically backward shifting after 2T laps from their onset, as the decay in shift is reduced to
Statistics
For the plots regarding PF backward shifting, the error bars represent mean ± SEM.
To generate the plots comparing the spatial precision, we used the package DABEST (data analysis with bootstrap-coupled estimation; Ho et al., 2019). As explained in PF properties, this gives a measure of the precision of PFs in each category. To compare the difference in the population, the median difference between the distributions and its confidence level were obtained with bootstrapping (5000 resamples); p-values of the nonparametric two-sided approximate permutation t test were reported. See Extended Data Table 1-1 for the specific statistical tests used in each figure.
Schematic figures (Figs. 1A, 2A, 3A) were created with BioRender.
Place fields that remap across 2 d in a novel environment have lower spatial precision on day 1. A, Experimental design. Top, The familiar environment (F) and the novel environment (N). Bottom, The animals were trained in a familiar environment, then switched to a novel environment and imaged for 2 d. B, Example field of view for days 1 and 2 showing the same imaged cells. C, Top, Behavior of a single animal showing track location. Blue dots indicate the animal’s location when the example cell’s calcium transient is at its peak. Bottom, Time-series fluorescent trace for an example cell. Blue dots indicating the transient peaks. D, Average population backward shifting of PF peaks within the session on day 1. PFs are aligned to their onset lap. Line indicates fitted exponential curve:
Extended Data Figure 2-1
Other place field metrics are not associated with place field fate across days in a novel environment. A, Comparison of backward shifting dynamics between the different categories of PF fate. Fitting parameters for the exponential
Place fields that remap in response to changed reward expectation tend to have lower spatial precision before the change. A, Experimental design. Top, The familiar VR environment. Bottom, The Animals were trained and recorded in the same VR environment during changes in reward. B, Example of Stable (top), Remapped (middle), and Vanished (bottom) place fields across changes in reward expectation. Colored (blue, orange, and green) dots indicate in-field transient peaks. The place field can undergo remapping in UR and then again in RR. C, Comparison of spatial precision of PFs (561 PFs in R; see Materials and Methods, Defining PFs) from the three categories by measuring the SD of the lap-by-lap peak locations of Stable (277/561, 49.4%), Remapped (106/561, 18.9%), and Vanished (178/561, 31.7%) PFs between R and UR. Medium with inter-quartile range are shown by the broken vertical line next to the each dataset. Bottom, Bootstrapped median difference between the three groups. 5000 resamples, **p = 0.0016 for Remapped versus Stable, p = 0.480 for Vanished versus Stable. D, Same as C, but comparison between R and RR. (568 PFs in R, see Materials and Methods, Defining PFs; Stable PFs: 284/568, 50.0%; Remapped PFs: 55/568, 9.7%; Vanished PFs: 229/568, 40.3%) 5000 resamples, **p = 0.0022 for Remapped versus Stable, p = 0.683 for Vanished versus Stable. R, Rewarded; UR, Unrewarded; RR, Re-rewarded; n.s., not significant.
Extended Data Figure 3-1
Other place field metrics in R are not associated with place field fate in UR. A, Comparison of backward shifting dynamics between the different categories of PF fate. Fitting parameters for the exponential
Extended Data Figure 3-2
Other place field metrics in R are not associated with place field fate in RR. A, Comparison of backward shifting dynamics between the different categories of PF fate. Fitting parameters for the exponential
Results
Place fields that remap across 2 d in a familiar environment have lower spatial precision on day 1
To address whether place field characteristics during a single epoch of navigation in an environment were associated with their fate during a second epoch of the environment, head-fixed mice (n = 3) were trained to navigate a familiar VR environment (Fig. 1A) while the same populations of place cells were imaged in CA1 during two blocks of trials (distinct temporal epochs) separated by a day (Fig. 1B). The peak Ca2+ fluorescence on each lap traversal was used as a proxy for maximum spatial firing position on each lap. Peaks were then treated as events in 2D parameter space (time and spatial position). Clusters of events with consistent spatial position were identified as PFs (see Materials and Methods). The mean spatial position was then calculated and the same analysis was done the following day. PFs calculated on day 1 were then defined as either Stable, Remapped, or Vanished based on their mean activity on day 2 (see PF categorization and Extended Data Fig. 1-1). PFs were considered Remapped if the change in PF location across days was >40 cm (for why this threshold was chosen, see Materials and Methods). Note that in this paper we distinguish between PFs that change spatial position (referred as “Remapped”) and PFs that disappear (referred as “Vanished”; see examples in Fig. 1D). The median change in spatial position across days in the Stable PF group was 2.6 versus 179.8 cm in the Remapped group (Extended Data Fig. 1-1A).
Various properties of individual PFs in each category were then analyzed on day 1. First, lap-wise spatial precision was quantified using the SD of spatial locations of the fluorescent peaks. We compared the spatial precision of PFs across the three categories (Fig. 1F). The PFs that remapped on day 2 exhibited significantly lower spatial precision on day 1, as revealed by a higher variation in lap-to-lap PF position (Median (Mdn) SD = 6.9 cm, Inter-quartile range (IQR) = 2.1–18.8 cm) than Stable PFs (Mdn SD = 1.6 cm, IQR = 0.8–3.3 cm; p = 4.135E-07). Vanished PFs, however, exhibited similar spatial precision to Stable PFs (Mdn SD = 2.2 cm, IQR = 1.1–3.9 cm; p = 0.14).
Studies have reported a type of drift on a lap-by-lap basis that occurs during navigation, also known as PF backward shifting (Mehta et al., 2000; I. Lee and Knierim, 2007; Roth et al., 2012; Dong et al., 2021; Geva et al., 2023; Khatib et al., 2023). Backward shifting could reduce the lap-wise precision of PFs as we measured it here. To determine whether backward shifting contributed to our measure of precision and its association with PF fate, we measured the extent of backward shifting on day 1 (Fig. 1E). Further, because not all PFs emerge immediately when mice start navigating a familiar environment on any particular day, we first defined the PF onset lap for each PF (Sheffield et al., 2017; Dong et al., 2021). Aligning PFs to their onset lap, we found that backward shifting ceased after a finite number of laps, and the decay of shifting could be well fitted to an exponential (Fig. 1E). We estimated the time constant T of the decay. We considered PFs to have ceased backward shifting after 2T laps from their onset, as this is the point at which 90% of the shifting had decayed (see PF backward shifting). Then, the same comparison was performed as Figure 1F, but restricted to PF activity following the backward shifting. We found that the association between PF precision and PF fate across days was maintained even when backward shifting on day 1 was excluded from the analysis Figure 1G (Stable PFs: Mdn SD = 1.4 cm, IQR = 0.5–3.0 cm; Remapped PFs: Mdn SD = 5.7 cm, IQR = 1.6–16.2 cm; Vanished PFs: Mdn SD = 1.7 cm, IQR = 0.8–3.5 cm). This was also true for individual animals, shown in Extended Data Figure 1-3A.
It is possible that cells closer to the center of the imaging field of view (FOV) are of higher image quality, i.e., are less sensitive to imaging artifacts which could introduce noise. Cells further from the center of the FOV could therefore produce noisier signals that could make their PFs appear less precise and also make them more difficult to detect across days, making them appear less stable. To test whether this potential artifact is driving the association between PF precision and PF fate, we correlated the anatomic location of place cells in the FOV with their PF fate, which is shown in Extended Data Figure 1-4 for three example animals. We found no correlation between the location of place cells in the FOV and the fate of their PFs across days, ruling out any imaging artifacts driving the association between PF precision and fate.
We next checked whether the Remapped and Stable PFs continued to have differences in spatial precision on day 2 (Extended Data Fig. 1-2A). We found no such difference, showing that on day 2, the PFs that had remapped and the PFs that had stabilized have the same median precision. This suggests that PFs can switch from one category to the other. In other words, an imprecise PF that remaps across a day can then become precise and therefore stable across a subsequent day, and vice versa. While the spatial precision on day 1 is relevant to PF fate across a day, we next tested whether other PF properties on day 1 were associated with PF fate. We first asked whether the extent of backward shifting of PFs was associated with their fate. In Extended Data Figure 1-5A, we show the lap-by-lap shifting of all the PFs, and separately, the PFs from each category fitted to an exponential. Remapped, Stable, and Vanished PFs showed similar backward shifting dynamics. Comparing both the amplitudes and the time constants for the exponential fits along with their uncertainty values for all the categories demonstrated no differences between the groups (Extended Data Fig. 1-5A).
Next, we asked whether PF firing rates were associated with PF fate. Using peak amplitudes of calcium transients as a proxy for max firing rate on each PF traversal, we quantified the median and deviation in amplitudes from lap-to-lap. Comparing this measure between the three categories of PF fate revealed no significant difference (Extended Data Fig. 1-5B,C).
Not only do PFs emerge on different laps in a familiar environment, place cells can stop firing in their PF before the session ends. The PF onset lap and PF end lap, as well as the total laps in between onset and end (PF duration), can therefore be quantified for each PF. Extended Data Figure 1-5D shows the histograms of PF onset laps, end laps and total laps for the three PF fate categories. We found no differences between the PF fate categories.
Together, our investigation into place field properties and PF fate across days in a familiar environment suggests that it is randomly varying lap-by-lap spatial dynamics on day 1 that is related to the across-day fate of PFs, and other PF properties are unrelated.
Place fields that remap across 2 d in a novel environment have lower spatial precision on day 1
When mice are introduced to a novel environment, global remapping occurs in CA1 in which a new map forms (Colgin et al., 2008; Sheffield et al., 2017; Dong et al., 2021). Once the PFs that comprise the new map emerge, they typically are less precise than in familiar environments (Frank et al., 2004). We therefore tested whether the relationship between lap-wise precision and across-day PF fate that we observed in a familiar environment also occurred in a novel environment during familiarization. We therefore switched mice (n = 3) to a novel VR environment while imaging CA1 (Fig. 2A,B) and identified PFs (Fig. 2C). We first wanted to determine how the newly-formed PF map backward shifted from lap-to-lap on day 1 (Fig. 2D). We found backward shifting was prolonged compared with the familiar environment (
Place fields that remap in response to changed reward expectation tend to have lower spatial precision before the change
A recent study showed that some PFs remap when the internal state of reward expectation changes in an unchanging spatial environment (Krishnan et al., 2022). We therefore asked whether lap-by-lap spatial precision of PFs was associated with remapping under these conditions of altered internal state. To do this, mice were trained and then imaged in the same familiar rewarded environment (Fig. 3A). Trained mice (n = 5) were first water rewarded for a block of trials and then reward was removed for a subsequent block of trials [unrewarded condition (UR)]. After a few laps, mice stopped preemptively licking for reward, demonstrating a loss of reward expectation (Krishnan et al., 2022). Then, reward was reintroduced [rerewarded condition (RR)].
Figure 3B shows example place cells; one with a Stable PF across the R-UR-RR conditions (top), one with a Remapped PF across all conditions (middle), and one PF that vanished in UR but reappeared in RR (bottom). We then investigated whether the lap-by-lap spatial precision of PFs in R determined their fate in UR or RR (Fig. 3C,D). Similar to the fate of PFs across days, we found that the Remapped PFs in UR exhibited lower spatial precision in R (Fig. 3C; Remapped PFs: Mdn SD = 1.9 cm, IQR = 1.2–3.0 cm; Stable PFs: Mdn SD = 1.4 cm, IQR = 0.8–2.2 cm; Vanished PFs: Mdn SD = 1.3 cm, IQR = 0.4–2.1 cm) and RR (Fig. 3D; Remapped PFs: Mdn SD = 2.0 cm, IQR = 1.2–3.4 cm; Stable PFs: Mdn SD = 1.3 cm, IQR = 0.8–2.2 cm; Vanished PFs: Mdn SD = 1.2 cm, IQR = 0.5–1.9 cm). Again, other PF properties had no association with remapping or stability (Extended Data Figs. 3-1 and 3-2). This indicates that across the population, Remapped PFs caused by internal state changes tend to have lower precision than Stable/Vanished PFs, behaving similarly to Remapped PFs across days.
Discussion
We investigated hippocampal CA1 spatial code dynamics occurring across epochs in unchanging spatial environments to determine whether firing characteristics during a single epoch was associated with how cells encode future epochs. Our findings show that PFs that remapped across epochs separated in time by a day, or across epochs distinguished by differences in reward expectation, had a statistically lower lap-by-lap spatial precision during the initial epoch, compared with Stable and Vanished PFs. This held true across epochs in novel environments as mice underwent familiarization (a form of learning). Other lap-by-lap characteristics of PFs, such as firing rate variability, backward shifting dynamics, PF onset, PF offset, and PF duration were not associated with the fate of PFs across epochs. This indicates that the spatial firing precision of PFs during navigation is related to their tendency to remap or stabilize/vanish across distinct epochs of experience.
Drift across time has been observed in different parts of the brain (Driscoll et al., 2017; Deitch et al., 2021; Marks and Goard, 2021; Schoonover et al., 2021). In the hippocampus, an accurate representation of the spatial environment is preserved during drift (Ziv et al., 2013; Keinath et al., 2022), suggesting drift may encode nonspatial factors of the context such as time (Mankin et al., 2012, 2015), and experience (Geva et al., 2023; Khatib et al., 2023). Our data suggest that the dynamics of drift across epochs is related to the cellular activity within an epoch. This also holds true for epochs that are separated by internal state changes. The dynamics of the hippocampal spatial code across epochs may therefore not be random and may instead be predictable. However, the extent of predictability remains to be directly tested.
A possible explanation for the relationship between lap-by-lap dynamics and the tendency to remap is that those PFs with less spatial precision simply receive higher variability in the activation of the CA3 inputs they receive (Davoudi and Foster, 2019; Devalle and Roxin, 2022; Zutshi et al., 2022). Alternatively, evidence suggests that all CA1 pyramidal cells may receive synaptic input regarding all locations in an environment from CA3 (Grienberger et al., 2017). What determines whether a cell fires at a given location may therefore be the strength of synapses activated at particular locations. Indeed, dendritic spikes in CA1 place cells, which is a reflection of strong synaptic input to a dendritic branch, is associated with PF stability across days (Sheffield and Dombeck, 2015). One idea is that these strong synapses may have undergone Hebbian potentiation, and together with the resultant somatic firing may induce homeostatic mechanisms that lower overall cellular excitability (through synaptic or intrinsic excitability renormalization; Miller, 1996) and make other sets of synapses too weak to cause somatic firing in a winner-takes-all manner (Barry and Burgess, 2007; Sheffield and Dombeck, 2015). This process would result in a precise PF as the cell would fire only in response to those specific inputs. The dendritic spikes associated with this strong input may further serve to maintain synaptic strength to stabilize the PF across days (Sheffield and Dombeck, 2015). On the other hand, PFs with more lap-wise fluctuations in spatial firing may reflect differences in the sets of synapses activated from lap-to-lap. Such variations may not engage Hebbian potentiation and thus avoid the homeostatic winner-takes-all process described above. This would both cause the cell to be less precise from lap-to-lap but also allow the cell more flexibility to respond to new sets of synaptic activation that may occur across distinct epochs of experience, allowing for continuous encoding of new information in the hippocampus.
Our results also show that the PFs that vanished across epochs are indistinguishable from the stable ones in terms of spatial precision. This aligns with previous literature (Ziv et al., 2013), that place cells enter and exit an active subset of an underlying stable map. When these cells are active again, their PFs retain their locations (Ziv et al., 2013). The PFs that vanished across epochs may actually be part of this stable map, they are just not participating in the active subset during a particular epoch. This is likely because of the CA3 inputs that could drive them to fire not being activated.
Overall, our study presents how lap-by-lap dynamics of PFs during an epoch of experience relate to spatial code dynamics across epochs in the same environment. This work provides insight into why some cells remap and others remain stable/vanish. It also provides insight into the synaptic mechanisms which may facilitate these cellular dynamics to support episodic memory encoding.
Acknowledgments
Acknowledgments: We thank D. Goodsmith and A. Madar for manuscript comments.
Footnotes
The authors declare no competing financial interests.
This work was supported by The Whitehall Foundation, The Searle Scholars Program, The Sloan Foundation, The University of Chicago Institute for Neuroscience start-up funds, the National Institutes of Health Grants 1DP2NS111657-01 and 1RF1NS127123-01 (to M.E.J.S.), and the National Institute on Drug Abuse T32 Training Grant T32DA043469 (to S.K.).
This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International license, which permits unrestricted use, distribution and reproduction in any medium provided that the original work is properly attributed.
References
In this issue
