Hippocampal place codes are gated by behavioral engagement

Abstract

As animals explore an environment, the hippocampus is thought to automatically form and maintain a place code by combining sensory and self-motion signals. Instead, we observed an extensive degradation of the place code when mice voluntarily disengaged from a virtual navigation task, remarkably even as they continued to traverse the identical environment. Internal states, therefore, can strongly gate spatial maps and reorganize hippocampal activity even without sensory and self-motion changes.

Fig. 1. The hippocampal spatial code changed despite continuous traversal of the same environment.

a, Mice traversed a 2-m-long virtual linear track with visual cues that repeated in a circular topology. The start of the true reward zone is indicated by a gray vertical line; note that all analyses consider a peri-reward region, including the 10 cm immediately before the true reward zone. For simplicity, we refer to this peri-reward region as the reward zone. b, Three trial types were interleaved. Licking in the true reward zone on standard trials triggered water delivery. Probe trials omitted rewards. Crutch trials delivered rewards regardless of licking. c, Trained mice showed preferential licking in the reward zone on probe and standard trials. Consumption licks were excluded. Error bars: mean ± s.e.m. Raw data points jittered horizontally are shown next to each bar. Two-sided Wilcoxon signed-rank test: reward versus opposite zones on standard trials, P = 8.0 × 10⁻⁷; reward versus opposite zones on probe trials, P = 8.0 × 10⁻⁷. n = 32 sessions and 8 mice. d, Schematic of behavioral and imaging setup. M1 and M2 are mirrors. e, Left: representative imaging field of view. Right: zoomed view. Similar fields of view were obtained from 39 sessions that met inclusion criteria. f, Example jGCaMP8m ΔF/F and deconvolved activity traces from the first and second half of a session. Six cells are shown. g, h, Portion of two example sessions (standard trials only). Top: raster plots of neural activity on each frame, sorted by the location of each neuron’s maximal activity on correct trials. Only neurons whose standard deviation of activity is above the 30th percentile of all neurons’ standard deviation are shown. All neurons’ activities were normalized so that the top 8% pixels of the raster are saturated. Bottom: corresponding linear track position. a.u., arbitrary unit; FOV, field of view.

Fig. 2. The hippocampal spatial code degraded as mice disengaged from the task.

a, Lick raster of an example session with 125 trials. The reward zone (gray shaded area) includes anticipatory licking before reward availability (solid vertical line). The first consumption lick is indicated, whereas subsequent consumption licks inside the reward zone were excluded. b, The trial-wise lick selectivity and number of 5-cm bins with licks are shown for the 39 sessions that met the inclusion criteria. Circle sizes corresponds to the number of trials. The color code corresponds to k-means clustering with two clusters. The ‘engaged’ cluster had higher lick selectivity and number of bins with licks. c, Summary of lick behavior across 32 sessions that had more than ten disengaged trials. Shading represents mean ± s.e.m. d, Distribution of engaged and disengaged trials across 32 sessions. e, Left: distribution of disengaged trials in streaks of different length. Right: transition matrix between trial types averaged across sessions. Mean ± s.e.m. is noted. f, Portion of the standard trials from the same session as Fig. 1g. Decoded positions and decoding error were calculated using decoders trained on sliding windows of 20 trials and tested on the immediate next trial. Chance-level error is 50 cm. Licks, correctness and engagement status (blue: engaged; red: disengaged) are shown at the bottom. g, Sequence plots of cells with significant place fields in correct standard trials pooled across sessions. Cells were sorted by the location of peak activity using the trials indicated. Each cell’s activity was percentile normalized, saturating the top and bottom 2%. h, Decoding performance on held-out trials. The decoder was trained using ten trials of one engagement type at a time and tested on all other trials. The mean trial-wise decoding error was calculated from all iterations of the decoder. Each gray line indicates one session. Chance-level error is 50 cm. Two-sided Wilcoxon signed-rank test: engaged versus disengaged for decoder trained on engaged trials, P = 8.0 × 10⁻⁷; engaged versus disengaged for decoder trained on disengaged trials, P = 0.78. Train/test on engaged versus train/test on disengaged, P = 2.2 × 10⁻⁶. n = 32 sessions and 8 mice. i, Left: mean framewise decoding error by position for the decoder trained on engaged trials. Right: ratio of disengaged to engaged decoding error by position. Shading represents mean ± s.e.m. n = 32 sessions and 8 mice. j, Top: example speed profile (interquartile range) of ten engaged and ten disengaged trials. Bottom: speed profile of the same trials after speed matching. k, Same as h but using speed-matched frames for decoding. Two-sided Wilcoxon signed-rank test: engaged versus disengaged for decoder trained on engaged trials, P = 1.7 × 10⁻⁶; engaged versus disengaged for decoder trained on disengaged trials, P = 0.06. Train/test on engaged versus train/test on disengaged, P = 2.9 × 10⁻⁶. diseng, disengaged; eng, engaged; NS, not significant.

Fig. 3. The disengaged trials had fewer place cells and lower quality place fields.

a, Mean population activity in the engaged and disengaged trials. Each gray line indicates one session. n = 32 sessions and 8 mice. Two-sided Wilcoxon signed-rank test, P = 0.70. b, Left: trial–trial correlation of spatially binned activity averaged across cells of all sessions on 20 trials (at least 80% engaged) immediately before a streak of at least ten disengaged trials. n = 12 sessions and 6 mice. Values along the diagonal were set to 0. The top 1% of correlation values were saturated. Right: mean trial-wise activity of all cells on the corresponding trials. The s.e.m. is shown as gray shading but is contained in the line width. c, Fraction of imaged cells with place fields. Place fields and properties were calculated using a matched number of engaged and disengaged trials that are closest to each other in time. Two-sided Wilcoxon signed-rank test, P = 8.0 × 10⁻⁷. n = 32 sessions and 8 mice. d, Each pair of columns corresponds to one cell. The first two cells have place fields in both engaged and disengaged trials (9,837 cells total showed similar pattern); the third cell lost its place field in disengaged trials (10,792 cells showed similar pattern); and the last cell gained a place field in disengaged trials (3,254 cells showed similar pattern). First row: cell images with cell masks outlined. Second row: trial-wise activity for 20 engaged trials immediately before a streak of 20 disengaged trials. Dark blue represents 0 activity (a.u.). Top 1% values were saturated and indicated as max activity. Third row: mean activity (black traces) across periods. Red lines indicate location of a significant place field. Gray shading indicates 99% confidence bounds for shuffled data. e, Left: reliability of place fields (mean trial-to-trial correlation of spatially binned activity). Two-sided Wilcoxon rank-sum test, P = 0. Right: reliability across position. n = 20,629 place fields in engaged trials; 13,091 place fields in disengaged trials. f, Left: selectivity of place field activity. Two-sided Wilcoxon rank-sum test, P = 9.6 × 10⁻²⁰⁰. Right: selectivity across position. n = 20,629 place fields in engaged trials; 13,091 place fields in disengaged trials. g, Representative trial-wise activity and place field selectivity of four cells around disengaged streak onsets. Dark blue represents 0 activity (a.u.), and the max activity after saturating the top 1% of values is indicated. Green vertical line marks the first disengaged trial. h, Selectivity change around disengaged streak onsets for cells with place fields in engaged trials. Left: shading represents mean ± s.e.m. across cells. Right: shading represents mean ± s.e.m. across sessions. n = 12 sessions and 6 mice. i, Trial-wise decoding error around disengaged streak onsets. The decoder was trained on ten mostly engaged trials and tested on 20 mostly engaged trials immediately before a disengaged streak of at least ten trials. Chance-level error is 50 cm. Left: shading represents mean ± s.e.m. across frames of each trial. Right: shading represents mean ± s.e.m. across sessions. n = 15 sessions and 6 mice. a.u., arbitrary unit; NS, not significant.

Extended Data Fig. 1. Example fluorescence traces and deconvolved activity.

a. Representative jGCaMP8m ∆F/F, deconvolved activity, and reconstructed ∆F/F for four cells, at different temporal scales. Shaded regions in the columns 1 and 2 indicate expanded regions in columns 2 and 3, respectively. b. Representative jRGECO1a traces, plotted as in (a). c. Additional example jGCaMP8m ∆F/F traces. d. Histograms of correlation between ∆F/F and reconstructed ∆F/F from deconvolution across cells (left) and sessions (right).

Extended Data Fig. 2. Stability of the imaging field of view.

a. Example session mean image, image from the engaged trials, and image from the disengaged trials, and local correlation map between them. For each pixel x in the FOV, the local neighborhood used for correlation calculation is a square of 41 by 41 pixels with x in the center. Similar results were obtained from 32 sessions that met inclusion criteria and contained more than 10 engaged and disengaged trials. b. Correlation between the mean session FOV image and FOV images of the engaged and disengaged trials. Two-sided Wilcoxon signed rank test, p = 0.06. The mean correlation between FOVs in the engaged and disengaged trials is 0.992. n = 32 sessions, 8 mice. c. Mean noise in the engaged and disengaged trials. Noise was defined as the activity less than 2 standard deviations from the mean activity of each cell. Two-sided Wilcoxon signed rank test, p = 0.46. n = 32 sessions, 8 mice. d. Standard deviation of noise in the engaged and disengaged trials. Two-sided Wilcoxon signed rank test, p = 0.97. n = 32 sessions, 8 mice.

Extended Data Fig. 3. Session duration and reward volume at the onset of the longest disengaged streak.

a. Histograms of session duration, onset time of the longest disengaged streak, and the fraction of total session duration at the onset of the longest disengaged streak. b. Histograms of cumulative reward volume at the longest disengaged streak onset.

Extended Data Fig. 4. Decoding and place field quality for different calcium indicators, fluorescence trace processing, and decoding methods.

All panels are plotted as in Fig. 2h-i and Fig. 3e-f. a. jRGECO1a sessions. Two-sided Wilcoxon signed rank test: Engaged vs. disengaged for decoder trained on engaged trials, p = 4.4 × 10⁻⁴. Engaged vs. disengaged for decoder trained on disengaged trials, p = 0.03. Train/test on engaged vs. train/test on disengaged, p = 4.4 × 10⁻⁴. Two-sided Wilcoxon rank sum test: reliability p = 0, selectivity p = 9.64 × 10⁻²⁰⁰. n = 16 sessions, 4 mice. Shading represents mean ± SEM. b. jGCaMP8m sessions. Two-sided Wilcoxon signed rank test: Engaged vs. disengaged for decoder trained on engaged trials, p = 4.4 × 10⁻⁴. Engaged vs. disengaged for decoder trained on disengaged trials, p = 0.02. Train/test on engaged vs. train/test on disengaged, p = 2.3 × 10⁻³. Two-sided Wilcoxon rank sum test: reliability p = 0, selectivity p = 9.64 × 10⁻²⁰⁰. n = 16 sessions, 4 mice. Shading represents mean ± SEM. c. ΔF/F instead of deconvolved activity. Two-sided Wilcoxon signed rank test: Engaged vs. disengaged for decoder trained on engaged trials, p = 8.0 × 10⁻⁷. Engaged vs. disengaged for decoder trained on disengaged trials, p = 0.54. Train/test on engaged vs. train/test on disengaged, p = 1.1 × 10⁻⁶. Two-sided Wilcoxon rank sum test: reliability p = 0, selectivity p = 5.7 × 10⁻²⁰⁶. n = 32 sessions, 8 mice. Shading represents mean ± SEM. d. Template matching decoder. Two-sided Wilcoxon signed rank test: Engaged vs. disengaged for decoder trained on engaged trials, p = 8.0 × 10⁻⁷. Engaged vs. disengaged for decoder trained on disengaged trials, p = 0.54. Train/test on engaged vs. train/test on disengaged, p = 1.7 × 10⁻⁶. n = 32 sessions, 8 mice. Shading represents mean ± SEM.

Extended Data Fig. 5. Sessions with fewer than ten disengaged trials.

a. Portion of one example session with fewer than ten disengaged trials (standard trials only). Top: Raster plots of deconvolved activity, sorted by the location of each neuron’s maximal activity on correct trials. Only neurons whose standard deviation of activity is above the 30^th percentile of all neurons’ standard deviation are shown. Top 8% pixels were saturated. Decoded positions and decoding error were calculated using a sliding-window decoders as in Fig. 2f. Licks and correctness are shown at the bottom. b. Decoding performance of within-type held-out trials. Each gray line represents one session. Left: mean trial-wise decoding error of sessions without sufficient disengaged trials. The errors on held-out trials from the first half of the engaged trials (trained and tested on the first half) were compared to those from the second half of the engaged trials (trained and tested on the second half). Two-sided Wilcoxon signed rank test, p = 0.81. n = 7 sessions, 4 mice. Right: mean trial-wise decoding error of sessions with more than ten disengaged trials, as in Fig. 2h first and last bar. Two-sided Wilcoxon signed rank test, p = 2.2 × 10⁻⁶. n = 32 sessions, 8 mice.

Extended Data Fig. 6. Decoding and place field characteristics for the non-rewarded half of the track.

a. Similar to Fig. 2h, except for the non-rewarded half of the track. Two-sided Wilcoxon signed rank test: Engaged vs. disengaged for decoder trained on engaged trials, p = 8.0 × 10⁻⁷. Engaged vs. disengaged for decoder trained on disengaged trials, p = 0.91. Train/test on engaged vs. train/test on disengaged, p = 0.04. n = 32 sessions, 8 mice. b. Similar to Fig. 3e,f, except for only the non-rewarded half of the track. Two-sided Wilcoxon rank sum test: reliability p = 0, selectivity p = 1.9 × 10⁻²²². n = 32 sessions, 8 mice.

Extended Data Fig. 7. Decoding error for various trial types.

a. Violin plots showing decoding error distributions for five trial types using the decoder from Fig. 2h trained on engaged trials. Number of trials: engaged rewarded, 3080; engaged probe, 352; disengaged probe, 429; disengaged isolated, 189; disengaged streak, 1855. Only streaks of more than ten trials were considered. Two-sided Wilcoxon rank sum test: Engaged rewarded vs. engaged probe, p = 2.1 × 10⁻¹². Engaged probe vs. disengaged probe, p = 1.9 × 10⁻⁸⁶. Disengaged isolated vs. disengaged streak, p = 1.3 × 10⁻⁵⁰. Disengaged probe vs. disengaged streak, p = 2.3 × 10⁻⁶.

Extended Data Fig. 8. Place field peak locations and properties.

a. Peak location of place fields in engaged and disengaged trials. n = 20,629 place fields in the engaged trials, 13,091 place fields in the disengaged trials. b. Left: Place field status in the disengaged trials for cells with place fields in engaged trials. n = 20,629 cells. Right: peak locations of lost place fields. c. Left: Place field status in the disengaged trials for cells without place fields in the engaged trials. n = 13,091 cells. Right: peak locations of gained place fields. d. Peak locations for cells with place fields in both engaged and disengaged trials. n = 9837 cells. e. Absolute shift in place field peak from engaged to disengaged trials. In the shuffled data (gray), the disengaged place field peak locations were randomly permuted 1000 times, and the absolute shifts were calculated. f. Histogram of true and random shifts in peak locations between engaged to disengaged trials. Same shuffling procedure as in (e).