Organization of hippocampal CA3 into correlated cell assemblies supports a stable spatial code

Abstract

Hippocampal subfield CA3 is thought to stably store memories in assemblies of recurrently connected cells functioning as a collective. However, the collective hippocampal coding properties that are unique to CA3 and how such properties facilitate the stability or precision of the neural code remain unclear. Here, we performed large-scale Ca²⁺ imaging in hippocampal CA1 and CA3 of freely behaving mice that repeatedly explored the same, initially novel environments over weeks. CA3 place cells have more precise and more stable tuning and show a higher statistical dependence with their peers compared with CA1 place cells, uncovering a cell assembly organization in CA3. Surprisingly, although tuning precision and long-term stability are correlated, cells with stronger peer dependence exhibit higher stability but not higher precision. Overall, our results expose the three-way relationship between tuning precision, long-term stability, and peer dependence, suggesting that a cell assembly organization underlies long-term storage of information in the hippocampus.

Keywords: CA1; CA3; CP: Neuroscience; attractor dynamics; calcium imaging; hippocampus; memory; optical imaging; place cells; population coding; representational drift.

Graphical abstract

Figure 1

Ca²⁺ imaging of hippocampal CA1 and CA3 of freely behaving mice during familiarization with novel environments over weeks (A) Representative example of a sagittal section of a mouse implanted with a micro-prism anterior to the hippocampal CA3, showing GCaMP6f expression (green) and DAPI-labeled cell nuclei (blue). The micro-prism location is shown in white. Scale bar, 500 μm. D, dorsal; V, ventral; A, anterior; P, posterior. (B) Ca²⁺ imaging in CA1 or in CA3 during free exploration of initially novel environments (straight and L-shaped linear tracks) every other day. The mice explored environment A on 8 imaging days and then environment B on the following 8 days. Each imaging day consisted of two 10-min sessions separated by a 5-min inter-session interval. (C and D) Five example place cells recorded simultaneously from a mouse imaged in CA1 (C) and a mouse imaged in CA3 (D). Top: position of the mouse at times of estimated neuronal spikes (red and green dots for rightward and leftward running directions, respectively) overlaid on the mouse trajectory (blue curve). Bottom: corresponding spatial tuning curves, shown separately for rightward (red) and leftward (green) running directions.

Figure 2

CA3 place cells exhibit higher spatial tuning precision and short-term stability than CA1 place cells (A) Distributions of average spatial information content carried by the activity of individual hippocampal place cells on the first 3 imaging days in each environment in CA1 (blue) and in CA3 (red). (B) Spatial information of place cells on the first 3 imaging days in each environment (mean ± SEM) was significantly higher in CA3 than in CA1 (Mann-Whitney U test, U = 11, p < 0.05). (C) Spatial information (mean ± SEM) increased over the days of the experiment in CA1 (blue; one-way repeated-measures ANOVA, F₍₁₅₎ = 6.51, p < 10⁻⁶) and CA3 (red; one-way repeated-measures ANOVA, F₍₁₅₎ = 3.74, p < 0.001). The increase in information over time was greater in CA1 than in CA3 (two-way repeated-measures ANOVA, F_(1,15) = 2.56, p < 0.01, interaction between hippocampal subfield and time). (D and E) PV correlation between all pairs of positions across the two running directions for CA1 (D) and CA3 (E). The correlations were averaged across the first 3 imaging days in each environment and over mice. (F) PV correlation between the same positions across the opposite running directions in the track on the first 3 imaging days in each environment (mean ± SEM) was not significantly different between CA1 and CA3 (Mann-Whitney U test, U = 15, p = 0.29). (G) PV correlation between the same positions across the opposite running directions (mean ± SEM) decreased over the days of the experiment for CA1 (blue; one-way repeated-measures ANOVA, F₍₁₅₎ = 4.81, p < 10⁻⁴) and CA3 (red; one-way repeated-measures ANOVA, F₍₁₅₎ = 2.05, p < 0.05). Inset: the across-directions PV correlations on the first 3 imaging days in each environment were significantly higher in the L-shaped than in the straight linear track in CA1 (matched-pairs t test, t₍₃₎ = 7.82, p < 0.01) but not in CA3 (matched-pairs t test, t₍₄₎ = 1.70, p = 0.16). (H) PV correlation between the pairs of sessions imaged 5 min apart on the first 3 imaging days in each environment (mean ± SEM) was significantly higher in CA3 than in CA1 (Mann-Whitney U test, U = 10, p < 0.05). (I) PV correlation between pairs of sessions imaged 5 min apart (mean ± SEM) increased over the days of the experiment for CA1 (blue; one-way repeated-measures ANOVA, F₍₁₅₎ = 7.11, p < 10⁻⁶) and CA3 (red; one-way repeated-measures ANOVA, F₍₁₅₎ = 2.75, p < 0.01). CA1 exhibited a greater increase in PV correlations over time (two-way repeated-measures ANOVA, F_(1,15) = 1.81, p < 0.05, interaction between hippocampal subfield and time). Data were averaged over 4 mice in CA1 and 5 mice in CA3. ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001.

Figure 3

CA3 exhibits more stable spatial representations over weeks than CA1 (A) Average ensemble rate correlations of place cells across all sessions in the same environment for CA1 (left) and CA3 (right). Correlations were averaged over the two environments and all mice. (B) Ensemble rate correlation of place cells between pairs of sessions as a function of elapsed time between them (mean ± SEM) was not significantly different between CA1 (blue) and CA3 (red; two-way repeated-measures ANOVA, F_(1,7) = 0.29, p = 0.61, main effect of hippocampal subfield). (C) Average PV correlations of place cells across all sessions in the same environment for CA1 (left) and CA3 (right). Correlations were averaged over the two environments and all mice. (D) PV correlation between pairs of sessions as a function of elapsed time between them (mean ± SEM) was significantly higher in CA3 (red) than in CA1 (blue; two-way repeated-measures ANOVA, F_(1,7) = 29.7, p < 0.001, main effect of hippocampal subfield). Data were averaged over 4 mice in CA1 and 5 mice in CA3. ^∗∗∗p < 0.001.

Figure 4

Hippocampal CA3 is organized into functionally related place cell assemblies (A) Illustration of a population of place cells with independent tuning curves versus tuning curves of cells that are organized into a cell assembly. (B) Distribution of pairwise tuning-curve correlations for all pairs of place cells for CA1 (blue) and CA3 (red). Inset: the fraction of cell pairs with a tuning correlation greater than 0.7 was significantly higher in CA3 than in CA1 (Mann-Whitney U test, U = 10, p < 0.05). (C) Illustration of a population of place cells that have the same preferred position with independent shapes of their tuning curves (left) versus a population that is organized into a cell assembly with similar shapes of their tuning curves (right). (D) Distribution of pairwise tuning-curve correlations across all pairs of place cells with the same preferred position for CA1 (blue) and CA3 (red). Inset: fraction of cell pairs with the same preferred position with a tuning correlation greater than 0.7 was significantly higher in CA3 than in CA1 (Mann-Whitney U test, U = 10, p < 0.05). (E) Illustration of a population of place cells that have the same preferred position in either the rightward or leftward running direction. If place cells exhibit a dependence between the shapes of their tuning curves, then the tuning correlations between cells with the same preferred position in the same running direction are expected to be higher than the correlations between pairs of cells with the same preferred position across the opposite running directions. (F) Cumulative distributions of pairwise tuning-curve correlations between cell pairs with the same preferred position (mean ± SEM) for CA1 (blue) and CA3 (red). The within-running-direction distributions (solid curves) are shown against the across-running-direction distributions (dashed curves). (G) Difference between the cumulative within-running-direction and across-running-directions distributions of pairwise tuning-curve correlations (mean ± SEM) for CA1 (blue) and CA3 (red). Inset: the maximal difference between the cumulative distributions was significantly higher in CA3 than in CA1 (Mann-Whitney U test, U = 11, p < 0.05). (H and I) Difference between the cumulative within-running-direction and across-running-direction distributions of pairwise tuning-curve correlations (mean) across all pairs of place cells with the same or with different preferred positions for CA1 (H) and CA3 (I). Note that the x axis in (F)–(I) goes from the high correlations to the low correlations. Data were averaged across the first 3 imaging days in each environment. Data were averaged over 4 mice in CA1 and 5 mice in CA3. ^∗p < 0.05.

Figure 5

Place cells with peer-dependent tuning exhibit higher long-term tuning stability but not higher tuning precision (A–C) Relationship between long-term tuning stability, tuning precision, and tuning peer dependence at the single-cell level. (A) Tuning-curve correlations between sessions imaged 2 days apart increase as a function of the spatial information (mean ± SEM) in CA1 (blue; multiple regression t = 22.4, p < 10⁻¹⁰⁰) and CA3 (red; multiple regression t = 19.1, p < 10⁻⁷⁹). (B) Spatial information does not change with the level of tuning peer dependence (mean ± SEM) in CA1 (blue; multiple regression t = 0.38, p = 0.70) or CA3 (red; multiple regression t = 1.82, p = 0.07). Inset: tuning-curve correlation between sessions imaged 5 min apart does not change with the level of tuning peer dependence (mean ± SEM) in CA1 (blue; multiple regression t = −0.07, p = 0.94) or CA3 (red; multiple regression t = −1.70, p = 0.09). (C) Tuning-curve correlations between sessions imaged 2 days apart increase as a function of the level of tuning peer dependence (mean ± SEM) in CA1 (blue; multiple regression t = 12.0, p < 10⁻³²) and CA3 (red; multiple regression t = 8.8, p < 10⁻¹⁷). (A–C) A linear multiple regression model was separately fitted to the data of each hippocampal subfield between the tuning-curve correlation, spatial information, tuning peer dependence, average estimated firing rate, and mouse identity of each cell. (D–I) Comparison of the precision and stability of spatial tuning between place cells that were highly correlated with their peers (peer-dependent cells) and place cells that were not (peer-independent cells). (D) Spatial information over the days of the experiment (mean ± SEM) did not differ between the peer-dependent (solid) and peer-independent (dashed) cells in CA1 (blue; Friedman’s test, χ²₍₁₎ = 0.13, p = 0.72) or CA3 (red; Friedman’s test, χ²₍₁₎ = 0.39, p = 0.53). (E) PV correlation between two sessions imaged 5 min apart over the days of the experiment (mean ± SEM) was not different between the peer-dependent cells (solid) and the peer-independent cells (dashed) in CA1 (blue; Friedman’s test, χ²₍₁₎ = 0.08, p = 0.77) or CA3 (red; Friedman’s test, χ²₍₁₎ = 1.51, p = 0.22). (F) PV correlation between pairs of sessions as a function of elapsed time between them (mean ± SEM) was significantly higher for the peer-dependent cells (solid) than for the peer-independent cells (dashed) in CA1 (blue; Friedman’s test, χ²₍₁₎ = 6.51, p < 0.05) and CA3 (red; Friedman’s test, χ²₍₁₎ = 17.72, p < 10⁻⁴). (G) Average spatial information across all imaging days (mean ± SEM) did not differ between the peer-dependent and peer-independent cells in CA1 (matched pairs t test, t₍₃₎ = −0.33, p = 0.76) or CA3 (matched pairs t test, t₍₄₎ = −0.26, p = 0.81). (H) PV correlation (mean ± SEM) between sessions imaged 5 min apart did not differ between the peer-dependent and peer-independent cells in CA1 (matched pairs t test, t₍₃₎ = −0.32, p = 0.77) or CA3 (matched pairs t test, t₍₄₎ = −2.36, p = 0.08). (I) PV correlation (mean ± SEM) between sessions imaged 2 days apart was significantly higher for the peer-dependent cells than for the peer-independent cells in CA1 (matched pairs t test, t₍₃₎ = 7.06, p < 0.01) and CA3 (matched pairs t test, t₍₄₎ = 5.27, p < 0.01). (E, F, H, and I) Subsets of peer-independent cells were randomly chosen to match the peer-dependent cells’ population size. Data were averaged over 4 mice in CA1 and 5 mice in CA3. ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001.