The hippocampal CA2 region discriminates social threat from social safety

Abstract

The dorsal cornu ammonis 2 (dCA2) region of the hippocampus enables the discrimination of novel from familiar conspecifics. However, the neural bases for more complex social-spatial episodic memories are unknown. Here we report that the spatial and social contents of an aversive social experience require distinct hippocampal regions. While dorsal CA1 (dCA1) pyramidal neurons mediate the memory of an aversive location, dCA2 pyramidal neurons enable the discrimination of threat-associated (CS⁺) from safety-associated (CS^-) conspecifics in both female and male mice. Silencing dCA2 during encoding or recall trials disrupted social fear discrimination memory, resulting in fear responses toward both the CS⁺ and CS^- mice. Calcium imaging revealed that the aversive experience strengthened and stabilized dCA2 representations of both the CS⁺ and CS^- mice, with the incorporation of an abstract representation of social valence into representations of social identity. Thus, dCA2 contributes to both social novelty detection and the adaptive discrimination of threat-associated from safety-associated individuals during an aversive social episodic experience.

Extended Data Fig. 1 |. Deep learning pipeline for the automated quantification of behavior.

a, Markerless pose estimation during recall was implemented with DeepLabCut. Training of the ResNet-50 network was based on a total of 9 markers on the body of the subject mouse, 4 markers indicating the corners of the arena and 2 markers indicating the center of the wire cage cups placed on opposite sides of the arena as a proxy of stimulus mouse position. Dotted circles around cups illustrate the social interaction zone (1.5× the radius of the cup). Dark blue lines illustrate the head orientation of the subject mouse (see Methods for further details). b, Markers as predicted with DeepLabCut were employed to obtain features pertaining to posture of the mouse, its velocity and its position and orientation within the arena. c, To categorize behaviors into the five classes, a long short-term memory (LSTM) neural network was used to account for the sequential nature of behaviors, whereby manual annotations served as the ground truth. Illustration of the model for a single time frame, with the LSTM layer comprising 100 memory units (green) and input features described in a. The LSTM layer was followed by a fully connected softmax layer for the prediction of class probabilities. d, Example sequences illustrating behaviors of interest. Behaviors not falling into these categories were labeled as ‘other’. e, F1 score calculated based on softmax output probabilities; shuffled control data based on randomly shifted frames were associated with significantly lower classification accuracies. F1 scores from 10-fold cross-validation. Two-sided Mann–Whitney U = 100, ***P = 1.826 × 10⁻⁴. Box plot: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, decoding accuracies from 10-fold cross-validation.

Extended Data Fig. 2 |. SFC induces freezing and decreases distance traveled during social recall.

a, Freezing duration for mock-SFC (green) and SFC cohorts (purple) during recall trials 24 h after SFC. SFC but not mock-SFC cohort displays elevated freezing duration during the ‘social’ recall trial relative to the non-social ‘cups’ recall trial 24 h after SFC (two-way repeated measures ANOVA: cohort × stage, F(1,100) = 38.82, P = 1.104 × 10⁻⁸). N = 26 C57BL/6J wild-type mice per cohort. b, Freezing duration for Cre⁻ (purple) and Cre⁺ cohorts (orange; dCA2 silenced) during ‘cups’ and ‘social’ recall trials 24 h after SFC. Both cohorts display similar low freezing duration during ‘cups’ recall trials and similarly elevated freezing duration during ‘social’ recall trials (two-way repeated measures ANOVA: cohort × stage, F(1,76) = 1.3846, P = 0.243). N = 10 Avpr1b-Cre⁻, N = 10 Amigo2-Cre⁻, N = 10 Avpr1b-Cre⁺, N = 10 Amigo2-Cre⁺. c, Distance traveled (m) for mock-SFC (green), SFC Cre⁻ (purple) and SFC Cre⁺ (orange) cohorts during ‘cups’ trial in the SFC chamber (before SFC; SFC marked by dashed vertical line) and during ‘cups’ and ‘social’ recall trials in novel recall arena 24 h after SFC. There was no significant difference in distance traveled among cohorts during ‘cups’ trial on day 1 before SFC. On day 2, mock-SFC cohort (green) showed significantly greater mobility during the ‘social’ recall trial than both Cre⁻ (purple) and Cre⁺ (orange) SFC cohorts. Cre⁻ and Cre⁺ cohorts did not differ significantly in terms of their mobility (two-way repeated measures ANOVA: cohort × stage, F(4,171) = 1.35, P = 0.0113). Data pooled and subsampled from Fig. 2; mock-SFC: N = 20 C57BL/6J wild-type mice; SFC Cre⁻: N = 10 Avpr1b-Cre⁻ and N = 10 Amigo2-Cre⁻; SFC Cre⁺: N = 10 Avpr1b-Cre⁺ and N = 10 Amigo2-Cre⁺. Bonferroni post hoc tests: *** P < 0.001, **** P < 0.0001, n.s.: not significant. Box plots: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. Sex is balanced per condition.

Extended Data Fig. 3 |. General anxiety-like behavior and locomotor activity are not affected by SFC, and SFC behaviors do not depend on number of foot shocks administered.

a, No difference in locomotion between mock-SFC and SFC cohorts as quantified by the distance traveled during the non-social ‘cups’ recall trial 24 h after SFC. Two-sided, unpaired t-test: Distance traveled t = 1.019, p = 0.313. N = 26 C57BL/6J wild-type mice per cohort. Data in a and c are from Figs. 2 and 3, respectively. b, No difference in general anxiety behaviors between mock-SFC and SFC cohorts as quantified with the elevated plus maze (EPM) 24 h after SFC. Two-sided, unpaired t-test: open arm entries t = 0.32, p = 0.751, time on open arm: t = 0.934, p = 0.358. N = 15 C57BL/6J wild-type mice per cohort. c, No difference in locomotion between SFC Cre⁻ and Cre⁺ cohorts with CNO injection 30 mins prior to the SFC session is observed for the distance traveled during the ‘cups’ recall trial 24 h after SFC. Two-sided, unpaired t-test: Distance traveled t = 0.767, p = 0.447. SFC Cre⁻: N = 10 Avpr1b-Cre⁻ and N = 10 Amigo2-Cre⁻; SFC Cre⁺: N = 10 Avpr1b-Cre⁺ and N = 10 Amigo2-Cr⁺. d, No difference in general anxiety between SFC Cre⁻ and Cre⁺ cohorts is observed during exploration of the EPM 24 h after SFC, with CNO administration 30 min prior to the EPM assay. Two-sided, unpaired t-test: open arm entries t = 1.092, p = 0.289, time on open arm: t = 0.637, p = 0.531. N = 10 Amigo2-Cre⁻ and N = 10 Amigo2-Cre⁺. e, Discrimination score based on CS⁻ versus CS⁺ interaction times (y-axis) as a function of number of foot shocks received by subject mouse during SFC (x-axis). Data are shown for SFC Cre⁻ control mice (purple outlined boxes) and SFC Cre⁺ mice (orange outlined boxes); both groups received CNO 30 min prior to SFC on day 1. Data for female mice (gray bars) and male mice (open bars) are shown separately. No difference between discrimination scores during social recall 24 h after SFC for Cre⁻ mice receiving 2 versus 3 or 4 foot shocks (two-sided Welch’s t-test: t = 0.213, p = 0.836) and for Cre⁺ mice receiving 2 versus 3 foot shocks (two-sided Welch’s t-test: t = −0.48, p = 0.643). e,f, SFC Cre⁻: N = 31 Avpr1b-Cre⁻ and N = 31 Amigo2-Cre⁻ (N = 53 mice administered 2-foot shocks); SFC Cre⁺: N = 18 Avpr1b-Cre⁺ and N = 18 Amigo2-Cre⁺; 31 mice administered 2-foot shocks). Sex is balanced per condition. f, No difference in freezing duration during social recall trial 24 h after SFC for Cre⁻ mice receiving 2 versus 3 or 4 foot shocks (two-sided Welch’s t-test: t = −1.71, p = 0.121). No difference in freezing duration during the social recall trial 24 h after SFC for Cre⁺ mice receiving 2 versus 3 foot shocks (two-sided Welch’s t-test: t = 0.08, p = 0.938). Box plots: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals.

Extended Data Fig. 4 |. No significant effect of Cre⁺ genotype or CNO administration in absence of iDREADD expression.

a, Cre⁻ and Cre⁺ littermates underwent SFC (day 1) and were intraperitoneally injected with CNO on day 2, 30 min prior to social recall. N = 12 Amigo2-Cre⁻ and N = 12 Amigo2-Cre⁺ mice in a–d. Sex is balanced per condition. b, Amigo2-Cre⁻ and Amigo2-Cre⁺ cohorts displayed similar freezing durations 24 h after SFC (two-way repeated measures ANOVA: genotype × stage, F(1,44) = 0.871, p = 0.68). c, Spent a comparable amount of time in the CS⁻ versus CS⁺ arena half (two-way ANOVA: genotype × arena half, F(1,44) = 1.97, p = 0.52; right, discrimination scores, two-sided unpaired t-test t = −0.033, p = 0.973). d, Displayed comparable CS⁻ versus CS⁺ interaction times (two-way ANOVA: genotype × stimulus mouse, F(1,44) = 1.28, p = 0.47; right, discrimination scores, two-sided unpaired t-test: t = 0.292, p = 0.772. e–h, Same paradigm as in a–d but with Avpr1b-Cre⁺ and Avpr1b-Cre⁻ subject mice. N = 8 Avpr1b-Cre⁺, N = 8 Avpr1b-Cre⁻ in e–r. Sex is balanced per condition. f, Avpr1b-Cre⁻ and Avpr1b-Cre⁺ cohorts displayed similar freezing duration 24 h after SFC (two-way repeated measures ANOVA: genotype × stage, F(1,28) = 1.298, p = 0.26) and g, spent comparable times in the CS⁻ and the CS⁺ arena halves, respectively (two-way ANOVA: genotype × arena half, F(1,28) = 1.384, p = 0.24; right, discrimination scores, two-sided unpaired t-test t = −0.832, p = 0.41) and h, displayed comparable CS⁻ versus CS⁺ interaction times (two-way ANOVA: genotype × stimulus mouse, F(1,28) = 0.225, p = 0.638). Right, discrimination scores, two-sided unpaired t-test: t = −0.744, p = 0.46. i, To assess whether effects of SFC differ between CNO-administered (fuchsia) versus saline-administered (blue) Avpr1b-Cre⁻ mice, both cohorts expressed mCherry in dCA2 and underwent SFC on experimental day 1. Twenty-four hours later, mice were intraperitoneally injected with CNO or saline 30 min prior to recall. j–m: N = 8 Avpr1b-Cre⁻ mice per condition (CNO or saline). Sex is balanced per condition. j, CNO- and saline-administered Avpr1b-Cre⁻ cohorts displayed similar freezing durations 24 h after SFC (two-way repeated measures ANOVA: treatment (saline or CNO) × stage, F(1,28) = 0.029, p = 0.865), (k) spent comparable times in the CS⁻ and the CS⁺ arena halves (two-way ANOVA: treatment × arena half, F(1,28) = 4.468, p = 0.051; right, discrimination scores, two-sided, unpaired t-test t = −0.47, p = 0.643), (l) displayed comparable CS⁻ versus CS⁺ interaction times (two-way ANOVA: treatment × stimulus mouse, F(1,28) = 0.243, p = 0.625; right, discrimination scores, two-sided unpaired t-test: t = 0.427, p = 0.676) and (m) traveled comparable distances (two-way repeated measures ANOVA: cohort × stage, F(2,42) = 0.077, p = 0.925). n, As in i–m, with CNO- or saline-administered Avpr1b-Cre⁺ cohorts. o–r: N = 8 Avpr1b-Cre⁺ mice per condition (CNO or saline). Sex is balanced per condition. o, CNO- and saline-administered Avpr1b-Cre⁺ cohorts displayed similar freezing durations 24 h after SFC (two-way repeated measures ANOVA: treatment × stage, F(1,28) = 0.676, p = 0.417), (p) spent comparable times in the CS⁻ and the CS⁺ arena halves (two-way ANOVA: treatment × arena half, F(1,28) = 1.374, p = 0.251; right, discrimination scores, two-sided unpaired t-test t = 0.828, p = 0.42), (q) displayed comparable CS⁻ versus CS⁺ interaction times (two-way ANOVA: treatment × stimulus mouse, F(1,28) = 0.074, p = 0.786; right, discrimination scores, two-sided unpaired t-test: t = −0.185, p = 0.855) and (r) traveled comparable distances (two-way repeated measures ANOVA: cohort × stage, F(2,42) = 0.391, p = 0.678). Bonferroni post hoc tests: *p < 0.05, **p < 0.01, ***p < 0.001. Box plots: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals.

Extended Data Fig. 5 |. No significant effect of sex or Amigo2-Cre versus Avpr1b-Cre genotype on SFC behaviors.

a, No significant differences between female (gray fill) and male (white fill) mice with respect to freezing duration during the non-social ‘cups’ and the ‘social’ recall trials 24 h after mock-SFC (green outline) or 24 h after SFC (purple outline; three-way repeated measures ANOVA: cohort × sex × stage, F(1,96) = 0.69, p = 0.4). Data in a–c are from Fig. 2a,b–f: N = 26 C57BL/6J wild-type mice per experimental condition. Sex is balanced per condition. b, Female and male mock-SFC and SFC mice spend a comparable amount of time interacting with the CS⁻ and the CS⁺ stimulus mouse, with no significant difference between sexes (three-way repeated measures ANOVA: cohort × sex × stimulus mouse, F(1,96) = 0.14, p = 0.7). Right, discrimination scores, two-way ANOVA: cohort × sex, F(1,48) = 0.227, p = 0.6357. c, There is no significant difference between female and male mock-SFC and SFC mice in their relative velocities of approach versus retreat toward the CS⁻ and the CS⁺ stimulus mice. Velocity score = (approach velocity − retreat velocity)/sum of velocities, (three-way repeated measures ANOVA: cohort × sex × arena half, F(1,96) = 0.8, p = 0.37). d, Female and male mice display similar freezing durations during the non-social ‘cups’ recall trial and the ‘social’ recall trial 24 h after SFC, for both Cre⁻ control (purple) and Cre⁺ cohorts (orange; three-way repeated measures ANOVA: cohort × sex × stage, F(1,72) = 0.064, p = 0.8). Data in d–i are from Fig. 2g–l; SFC Cre⁻: N = 10 Avpr1b-Cre⁻, N = 10 Amigo2-Cre⁻, SFC Cre⁺: N = 10 Avpr1b-Cre⁺, N = 10 Amigo2-Cre⁺. Sex is balanced per condition. e, No significant differences between female and male Cre⁻ and Cre⁺ mice with respect to social interaction times with the CS⁻ and the CS⁺ stimulus mouse (three-way repeated measures ANOVA: cohort × sex × stimulus mouse, F(1,72) = 0.341, p = 0.56). Right, discrimination scores (two-way ANOVA: cohort × sex, F(1,36) = 0.577, p = 0.452). f, Female and male Cre⁻ and Cre⁺ mice display comparable relative velocities of approach versus retreat toward the CS⁻ and the CS⁺ stimulus mouse (three-way repeated measures ANOVA: cohort × sex × arena half, F(1,72) = 1.15, p = 0.29). g, No significant differences between Cre⁻ (purple) or Cre⁺ (orange) Avpr1b-Cre and Amigo2-Cre mice with respect to freezing duration for the non-social ‘cups’ recall trial and the ‘social’ recall trial 24 h after SFC (three-way repeated measures ANOVA: cohort (Cre⁻ vs. Cre⁺) × genotype (Amigo2- vs. Avpr1b-Cre) × stage, F(1,72) = 0.615, p = 0.44). h, No significant differences between Avpr1b-Cre and Amigo2-Cre Cre⁻ or Cre⁺ mice with respect to social interaction time and no significant difference between sexes (three-way repeated measures ANOVA: cohort × genotype × stimulus mouse, F(1,72) = 0.165, p = 0.685). Right, discrimination scores, two-way ANOVA: cohort × genotype, F(1,36) = 0.004, p = 0.95. i, No significant differences between Avpr1b-Cre and Amigo2-Cre Cre⁻ and SFC Cre⁺ mice for relative approach/retreat velocities (three-way repeated measures ANOVA: cohort × genotype × arena half, F(1,72) = 0.08, p = 0.7758). Box plots: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); Dots, individual animals.

Extended Data Fig. 6 |. Optogenetic silencing of dCA2 during SFC disrupts social fear discrimination memory.

a, An optogenetic approach was used during SFC on day 1 to silence dCA2 selectively during social fear memory encoding: Cre⁺ mice and Cre⁻ littermates were injected in dCA2 with AAV-DIO-eArch3.0-eYFP and implanted with bilateral ferrules for optic fiber probes above dCA2. dCA2 was illuminated in Cre⁺ and Cre⁻ groups with green light pulses continually during the 5-min SFC learning trial. b–f: N = 12 Amigo2-Cre⁻ (purple) mice; N = 12 Amigo2-Cre+(turquoise). Sex is balanced per condition. b, Example confocal image from an Amigo2-Cre⁺ mouse showing optical fiber tract (dashed lines) and eArch3 expression with co-expression of the CA2 marker protein PCP4. c, SFC Cre⁺ and Cre⁻ cohorts display similar freezing duration during recall trials 24 h after SFC (two-way repeated measures ANOVA: cohort × stage, F(1,40) = 0.75, p = 0.72) and significantly elevated freezing levels during the ‘social’ as compared to the ‘cups’ recall trial (Bonferroni post hoc tests). d, Cre⁻ cohorts show a preference for exploring the arena half containing the CS⁻ mouse, whereas Cre⁺ cohorts show generalized avoidance of the CS⁺ and the CS⁻ arena halves (two-way repeated measures ANOVA: cohort × arena half, F(1, 40) = 8.219, p = 6.58 × 10⁻³). Right, discrimination scores: unpaired, two-sided t-test between cohorts: t = −3.256, p = 0.003. e, Cre⁻ cohorts prefer to interact with the CS⁻ mouse, whereas Cre⁺ mice display generalized avoidance toward the CS⁺ and CS⁻ mice (two-way repeated measures ANOVA: cohort × stimulus mouse, F(1,40) = 7.429, p = 9.47 × 10⁻³). Right, discrimination scores: unpaired two-sided t-test between cohorts: t = 2.791, p = 0.01. f, Velocity per head orientation angle plot, illustrating the generalized avoidance behaviors of dCA2-silenced Cre⁺ mice (turquoise) but not of Cre⁻ controls which maintain selective approach-avoidance toward the CS⁻ mouse (purple). Bonferroni post hoc and t-tests, as appropriate: *p < 0.05, **p < 0.01, ***p < 0.001. n.s.: not significant. Box plots: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals.

Extended Data Fig. 7 |. Stimulus mouse experience does not affect social memory or social fear discrimination memory in subject mice.

a,b, Social memory assay with dCA2-silenced male Amigo2-Cre stimulus mice. a, After 3–4 weeks of iDREADD injections, CNO was administered to stimulus mice 30 min prior to the first social memory encoding session (‘social 1’), which was immediately followed by a second encoding session (‘social 2’) with the position of stimulus mice swapped. Thirty minutes later, a stimulus mouse, chosen at random, was replaced by a novel mouse (‘recall’). One week later (week 2), the procedure is repeated with saline injection administered to the same stimulus mice and a new cohort of subject mice. This was repeated with a different pair of stimulus mice and two cohorts of subject mice, resulting in a total of N = 20 C57BL/6J wild-type male subject mice (N = 5 per cohort). b, Left, subject mice display comparable interaction times when stimulus mice were administered CNO versus saline, two-way ANOVA: injection type × stimulus mouse (novel or familiar), F(1,36) = 1.14, p = 0.293. Right, discrimination scores; unpaired, two-sided t-test: t = −0.879, p = 0.39). One-sided, one-sample t-tests against zero: stimulus mouse + saline, p = 0.0057, t = 3.603; stimulus mouse + CNO, t = 3.882, p = 0.0037. c,d, SFC assay with dCA2-silenced male Amigo2-Cre stimulus mice. c, After 3–4 weeks of iDREADD injections, CS⁺ and CS⁻ stimulus mice received CNO 30 min before SFC. The same stimulus mice were used for SFC 1 week later with saline injections and a new cohort of subject mice. This was repeated with a different pair of stimulus mice and two different cohorts of subject mice, resulting in a total of N = 20 male subject mice (N = 5 per cohort). d, Left, subject mice display comparable interaction times toward CNO- versus saline-administered CS⁺ versus CS⁻ stimulus mice. Two-way ANOVA: injection type × stimulus mouse (CS⁺ or CS⁻), F(1,36) = 0.98, p = 0.328. Right, discrimination scores; unpaired, two-sided t-test: t = 0.342, p = 0.736. One-sided, one-sample t-tests against zero: stimulus mouse + saline, t = 2.806, p = 0.02; Stimulus mouse + CNO, t = 3.403, p = 0.007. Bonferroni post hoc tests: *p < 0.05, ** p < 0.01. e–g, Effect of prior SFC experience by CS⁺ and CS⁻ stimulus mice on subject mouse behavior for mock-SFC (green; N = 6 C57BL/6J wild-type mice per condition), SFC Cre⁻ (purple; N = 3 Avpr1b-Cre⁻, N = 3 Amigo2-Cre⁻ per condition) or SFC Cre⁺ (orange; N = 3 Avpr1b-Cre⁺, N = 3 Amigo2-Cre⁺ per condition) subject mice during recall trial on day 2. Sex is balanced per condition. e, Freezing duration of mock-SFC (green), SFC Cre⁻ (purple) or SFC Cre⁺ (orange) subject mice during recall trial on day 2 does not significantly differ when stimulus mice are ‘naive’ (gray-filled bars; no prior SFC experience) versus ‘experienced’ (open bars; used in prior SFC experiments with N = 5 subjects; three-way repeated measures ANOVA: cohort × stage × stim. mouse experience, F(2,40) = 0.016, p = 0.984). f, No significant difference in relative velocities toward naive or experienced CS⁻ or CS⁺ mice, three-way ANOVA: cohort × arena half × stim. mouse experience, F(2,40) = 0.2, p = 0.82. g, Comparable interaction times of subjects with naive versus experienced stimulus mice (three-way repeated measures ANOVA: cohort × stimulus mouse (CS⁻ vs. CS⁺) × stim. mouse experience, F(2,40) = 0.126, p = 0.88). Right, discrimination scores, two-way ANOVA: cohort × stim. mouse experience, F(2,20) = 1.04, p = 0.37. Data pooled from Figs. 2 and 3 and Extended Data Fig. 2a,b, with corresponding sex and genotype. Box plots: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. Bonferroni post hoc and t-tests, as appropriate: *p < 0.05, **p < 0.01. n.s.: not significant.

Extended Data Fig. 8 |. Comparison of dCA2 neuron stimulus selectivity and stability before and after mock-SFC and SFC for individual mice.

a, No change in the CS⁻ or CS⁺ selective fraction of cells was observed across mock-SFC (two-sided χ²(6) = 4.943, p = 0.551). Dots, individual mice. SE: social exploration trials before mock-SFC; SR: social recall trials after mock-SFC. Data corresponding to Fig. 5e. a–d: N = 5 male Amigo2-Cre⁺ and 5 male Avpr1b-Cre⁺ mice. b, SFC significantly increased the fraction of cells that responded selectively to the CS⁺ or the CS⁻ (chi-square test; two-sided χ²(6) = 34.382, p = 5.675 × 10⁻⁶). Dots, individual mice. Data corresponding to Fig. 5e. SE: social exploration trials before SFC; SR: social recall trials after SFC. c, Percentage of CS⁺ (left) and CS⁻ (right) selective dCA2 pyramidal neurons (y-axis; see a and b), plotted as a function of social exploration or social recall trials (x-axis): for mock-SFC (top) and SFC cohorts (bottom). Mock-SFC: one-way repeated measures ANOVA CS⁺: F(2,18) = 0.68, p = 0.519; CS⁻: F(2,18) = 1.497, p = 0.2503. SFC: one-way repeated measures ANOVA CS⁺: F(2,18) = 7.177, p = 0.00512; CS⁻: F(2,18) = 8.996, p = 0.00195. Thin lines: individual animals. Filled symbols connected by thick lines and error bars in a and c: mean ± SEM averaged across animals. Data are from Fig. 5e. d, Individual animal data showing the stability of dCA2 neuron response profiles for cells that initially responded on day 1 selectively to CS⁺ or CS⁻ (labeled above graphs) before mock-SFC (left) or SFC (right). Graphs show response profiles for CS⁺ and CS⁻ selective cells identified on day 1 when measured during the recall trial on day 2 after mock-SFC or SFC. For the SFC cohort, cells selective for the CS⁺ or CS⁻ on day 1 show a greater probability of responding, respectively, to the CS⁺ or the CS⁻ in the recall trial after SFC (day 2). Data are from Fig. 5h. Total number of cells (mean ± SEM across mice): mock-SFC day 1: 672 (69 ± 14); mock-SFC day 2: 658 (64 ± 10); SFC day 1: 647 (62 ± 7); SFC day 2: 630 (58 ± 6). Cells that responded to both CS⁺ and CS⁻ were classified as ‘mixed selective’. Box plots: central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals.

Extended Data Fig. 9 |. CS-selective cells enhance identity decoding after SFC, but spatial decoding performance is not enhanced after SFC.

a, CS⁺ versus CS⁻ decoding accuracies for two social exploration trials (SE) on day 1 before SFC and social recall trials on day 2 after SFC (SR 1 and 2; SR 3 and 4). Decoding is based on the original dataset (purple; Fig. 6) or when a randomly selected subset of 11% of cells were removed from (ochre). Thin lines show individual animals; circles and thick lines show mean ± SEM. Decoding accuracies for the two datasets are comparable throughout the experiment: two-way repeated measures ANOVA: cohort × session, F(2,54) = 0.018, p = 0.98. For both datasets, decoding of CS⁺ versus CS⁻ is significantly greater after SFC (one-sided, paired t-tests: all cells SR12 vs. SE12: t = −4.42, p = 0.00083; random subset removed SR12 vs. SE12: t = −2.82, p = 0.01). a–e: N = 5 male Amigo2-Cre⁺ and 5 male Avpr1b-Cre⁺ mice per cohort. b, Decoding based on the original dataset (purple; Fig. 6) or with the 11% of CS⁺ and CS⁻ selective cells newly recruited after SFC was removed from the population for all time points (blue). Removal of the newly recruited selective cells caused a significant decrease in decoding accuracy after SFC but not before SFC. Decoding accuracies for the two datasets are significantly different for the different SFC stages (two-way repeated measures ANOVA: cohort × session, F(2,54) = 3.623, p = 0.0334). When CS⁺ and CS⁻ selective cells newly recruited by SFC have been removed, no enhancement of decoding accuracy can be observed (one-sided, paired t-tests, CS-selective cells removed SR12 vs. SE12: t = −1.81, p = 0.051). Bonferroni post hoc tests: *p < 0.05, n.s.: not significant. Decoding accuracy, mean ± SEM: all cells (purple; SE 1 and 2: 57.1 ± 1.15, SR 1 and 2: 69.24 ± 2.62, SR 3 and 4: 66.17 ± 2.25); random subset (11% cells) removed (ochre; SE 1 and 2: 58.74 ± 1.32, SR 1 and 2: 66.39 ± 2.42, SR 3 and 4: 64.03 ± 1.3); SFC-recruited CS⁺ and CS⁻ selective cells removed (blue; SE 1 and 2: 57.27 ± 1.76, SR 1 and 2: 61.76 ± 2.13, SR 3 and 4: 59.53 ± 1.6). Unpaired, two-sided t-test, comparison of decoding accuracies when random subset removed versus when CS⁺ and CS⁻ selective cells removed for SR 1 and 2: t = −2.34, p = 0.031. Thin lines show individual animals; circles and thick lines show mean ± SEM. c, To assess how SFC impacts dCA2 representations of spatial variables, dCA2 calcium activity data were acquired during social–spatial interactions with the CS⁺ and CS⁻ in front and back right cups before and after SFC. Arena and spatial location of cups differ from SFC learning trial so only social information is present in the recall trial relevant to SFC. dCA2 activity data were pooled from interactions around the same cup location (i.e., back or front cup) from two consecutive recall trials, irrespective of stimulus mouse identity. d, r, Pearson’s correlation coefficient. Changes in spatial decoding are not correlated with changes in freezing behavior (top graphs) or behavioral discrimination of CS⁺ from CS⁻ (bottom graphs) for SFC (left graphs) or mock-SFC cohorts (right graphs). e, Spatial decoding accuracy is similar before (SE trial) and after (SR trials) SFC (purple). SFC and mock-SFC (green) cohorts show similar spatial decoding accuracy throughout the paradigm, suggesting that the past threat-associated experience associated with a conspecific does not affect the representation of position in the recall arena. Two-way ANOVA: cohort × session, F(2,54) = 1.284, p = 0.285. Bonferroni post hoc tests: n.s.: not significant. Thin lines show individual animals; circles and thick lines show mean ± SEM.

Extended Data Fig. 10 |. OFC does not enhance dCA2 single-cell or population-level responses to safety- versus threat-associated objects.

a, OFC paradigm. On day 1, a subject mouse was allowed to explore an open arena (the same used for SFC social exploration and recall trials) for 5 min (‘Hab.’), followed by two 5-min trials in which the subject explored the prospective CS⁺ and CS⁻ objects (OE, ‘object exploration’ trials), with positions swapped across trials. In week 1, OE trials were followed 2 hours later by a mock-OFC trial. On day 2, the subject was re-exposed to the open arena (‘Hab.’), followed by four consecutive ‘object recall’ (OR) trials, with positions of stimulus objects swapped between each trial. In week 2, the same procedure was repeated with two novel objects and with the mock-OFC trial replaced by an OFC trial. b, dCA2 cells with indicated response selectivity profiles; cells that responded to both CS⁺ and CS⁻ objects were classified as ‘mixed selective’. b–d, N = 7 mice (N = 3 male Amigo2-Cre⁺; N = 4 male Avpr1b-Cre⁺ mice). OFC did not significantly change the fraction of CS⁺ or CS⁻ selective, mixed-selective or spatially selective cells (two-sided Chi-square tests; χ²(6) = 6.26, p = 0.395), no change of the responsive cell populations was observed for the mock-SFC cohort (two-sided χ²(6) = 3.678, p = 0.72). c, OFC did not alter population-level decoding of CS⁺ versus CS⁻ objects. Top left, calcium activity was measured from dCA2 pyramidal neurons during exploration of CS⁺ and CS⁻ objects in two object exploration (OE) trials on day 1 and four object recall (OR) trials on day 2 (after mock-OFC or OFC), with positions of objects swapped after each trial. Bottom left, linear SVM decoding of CS⁺ versus CS⁻ objects from calcium data recorded with object positions swapped across two successive trials. Right graphs: top, CS⁺ versus CS⁻ object decoding accuracy during two object exploration trials (OE) before OFC (purple) or before mock-OFC (green) and during the first two or the second two object recall trials (OR 1 and 2; OR 3 and 4) 24 h after OFC (purple) or 24 h after mock-OFC (green). Thin lines show individual animals; circles and thick lines are mean ± SEM. Bottom, behavioral discrimination scores before and after OFC and mock-OFC (averaged across two consecutive trials). OFC enhances behavioral discrimination performance but not dCA2-based decoding accuracy of objects (two-way repeated measures ANOVA: cohort × session, F(2,36) = 0.1933, p = 0.825 for decoding and F(2,36) = 5.017, p = 0.0119 for behavior). Paired, one-sided t-test against chance-level accuracies from null model. Bonferroni post hoc and t-tests, as appropriate: *p < 0.05, ***p < 0.001, n.s.: not significant. d, r, Pearson’s correlation coefficient. Changes in freezing behavior (top graphs) and behavioral discrimination of CS⁺ from CS⁻ objects (bottom graphs) plotted versus change in decoding accuracy of CS⁺ from CS⁻ objects after OFC (left graphs; purple) or mock-OFC (right graphs; green) relative to values before OFC or mock-OFC. Filled symbols and solid lines; data from OR1 and OR2 recall trials; open symbols and dashed lines, data from OR3 and OR4 recall trials. No significant correlation between behavior and decoding performance. Circles represent individual animals. All mice were administered 2-foot shocks during OFC. Total number of cells (mean ± SEM per mouse): mock-OFC—object exploration trial, n = 456 cells (65 ± 11 per animal); object recall trial, n = 418 (64 ± 9 per animal); OFC—object exploration trials, n = 402 (57 ± 8 per animal); OFC object recall trials, n = 388 (55 ± 8 per animal).

Fig. 1 |. Differential roles for dCA2 and dCA1 in spatial and social aspects of social episodic memory.

a, SSFC paradigm. Colored dots show pose estimation markers. b, Experimental timeline. c, Top row, coronal section showing hM4Di-mCherry in dCA1 pyramidal neurons from an Lypd1-Cre mouse. Bottom row, coronal section showing co-expression of hM4Di-mCherry and CA2 marker RGS14 in dCA2 pyramidal neurons from an Amigo2-Cre mouse. d, Example mouse position heatmaps during SSFC spatial (top panels) and social (bottom panels) recall sessions for mice expressing: left, mCherry in dCA1 (control); middle, hM4Di-mCherry in dCA1; and right, hM4Di-mCherry in dCA2. e–h, n = 13 Lypd1-Cre⁺ (mCherry in dCA1), n = 13 Lypd1-Cre⁺ (hM4Di-mCherry in dCA1), n = 13 Amigo2-Cre⁺ (hm4Di-mCherry in dCA2). e, Left, mCherry and dCA2-silenced mice but not dCA1-silenced mice spent during spatial recall more time in chamber half where the CS⁻ mouse was present in the SSFC learning trial (two-way repeated measures ANOVA—cohort × arena half, F(2,72) = 17.7, P = 5.8 × 10⁻⁷). Right, discrimination scores (Methods), two-tailed one-sample t-tests against zero. f, Time spent freezing during spatial recall (day 2) was significantly greater than that during the ‘Cups’ trial before SFC (day 1) for all three groups (two-way repeated measures ANOVA—cohort × stage, F(2,72) = 4.34, P = 0.017). g, Left, mCherry and dCA1-silenced mice but not CA2-silenced mice spent more time during social recall interacting with the CS⁻ compared to CS⁺ mouse during the social recall in the novel arena (two-way repeated measures ANOVA—cohort × arena half, F(2,72) = 28.45, P = 7.8 × 10⁻¹⁰). Right, discrimination scores, two-tailed one-sample t-tests against zero. h, Cohorts displayed similarly elevated freezing durations during social recall versus either Cups trial before SSFC (Cups 1) or cup recall trial after SSFC (Cups 2; three-way repeated measures ANOVA—Cups (Cups 1 or Cups 2) × social recall × cohort, F(3,108) = 8.124, P = 0.014). Box plots—central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. Bonferroni post hoc and t-tests, as appropriate—*P < 0.5, **P < 0.01 and ***P < 0.001. Sex was balanced per condition, no sex differences were observed (for statistical comparison between females and males, see Supplementary Table 1). NS, not significant.

Fig. 2 |. Social fear conditioning induces robust social fear discrimination memory that requires the activity of dCA2.

a, SFC paradigm. b, Left, CS⁻ and CS⁺ interaction times after mock-SFC or SFC (two-way repeated measures ANOVA—cohort × stimulus mouse, F(1,100) = 10.91, P = 1.32 × 10⁻³). Right, discrimination scores. b–f, n = 26 C57BL/6J wild-type mice per cohort. Discrimination scores for b and h—unpaired, two-sided t-tests between cohorts; two-sided one-sample t-tests against zero. c, Illustration of approach-avoidance behaviors. d, Velocity versus head orientation angle plots toward CS⁻ (top) and CS⁺ (bottom) for mock-SFC (green) and SFC (purple) groups. Approach, 0° and retreat, 180°. e, Relative approach/retreat velocity after mock-SFC and SFC. Note faster approach than retreat to CS⁻ and faster retreat than approach to CS⁺ after SFC (Methods; two-way repeated measures ANOVA—cohort × arena half, F(1,100) = 117.05, P = 1.6 × 10⁻¹⁸). f, Frequency of approach-avoidance behaviors toward CS⁺ (open bars, solid lines) and CS⁻ (filled bars, dashed lines) after mock-SFC or SFC. Two-sided Kolmogorov–Smirnov test, ****P = 8.1 × 10⁻⁵. g, Timeline of chemogenetic experiment. h, Left, Cre⁻ but not Cre⁺ (dCA2silenced) cohort interacted more with CS⁻ than CS⁺ (two-way repeated measures ANOVA—cohort × stimulus mouse, F(1,76) = 32.92, P = 1.85 × 10⁻⁷). CS⁺ interaction times are similar between groups (P = 0.071; Bonferroni post hoc test). Right, discrimination scores. h–l, n = 10 Avpr1b-Cre⁻, n = 10 Amigo2-Cre⁻, n = 10 Avpr1b-Cre⁺ and n = 10 Amigo2-Cre⁺. i, Cartoon illustrating CS⁺ selective approach-avoidance behaviors for Cre⁻ but not Cre⁺ group. j, Velocity versus head orientation angle plots show Cre⁻ mice (purple) but not Cre⁺ mice (orange) selectively retreated from CS⁺ after SFC. k, Relative approach/retreat velocity scores. Two-way repeated measures ANOVA—cohort × arena half, F(1,76) = 200, P = 5.5 × 10⁻²³. l, Frequency of approach-avoidance events toward CS⁺ was greater than toward CS⁻ for Cre⁻ but not Cre⁺ group after SFC. Two-sided Kolmogorov–Smirnov test, ****P = 2.4 × 10⁻⁵. Box plots—central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. Bonferroni post hoc and t-tests, as appropriate—*P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Sex was approximately balanced across conditions, and no sex differences were observed (for statistical comparison between females and males, see Supplementary Table 1 and Extended Data Fig. 5a–f).

Fig. 3 |. dCA2 pyramidal neuron activity is required for social fear discrimination memory during recall.

a, Experimental timeline of chemogenetic silencing of dCA2 pyramidal neurons during SFC recall (day 2). b, Cre⁺ mice and Cre⁻ mice displayed comparable freezing durations (two-way repeated measures ANOVA—cohort × stage, F(1,60) = 0.0001, P = 0.99) that were significantly greater during social than cup recall trials. b–g, n = 8 Amigo2-Cre⁻ and n = 8 Avpr1b-Cre⁻; n = 8 Amigo2-Cre⁺ and n = 8 Avpr1b-Cre⁺. c, Left, Cre⁻ but not Cre⁺ mice selectively avoided the CS⁺ arena half (left, two-way repeated measures ANOVA—cohort × arena half, F(1,60) = 168.02, P = 4.48 × 10⁻¹⁹). Right, discrimination scores. Discrimination scores for c and d—unpaired, two-sided t-tests between cohorts, two-sided one-sample t-tests against zero. d, Left, Cre⁻ mice preferentially interacted with CS⁻ relative to CS⁺; Cre⁺ mice showed little interaction with either (two-way repeated measures ANOVA—cohort × stimulus mouse, F(1,60) = 15.46, P = 2.221 × 10⁻⁴). CS⁺ interaction times did not significantly differ between the Cre⁻ and Cre⁺ cohorts (P = 0.068, Bonferroni post hoc test). Right, discrimination scores. e, Velocity versus head orientation angle plots showing Cre⁻ (purple) but not Cre⁺ (orange) mice selectively retreat from CS⁺. f, Relative approach/retreat velocity score (two-way repeated measures ANOVA—cohort × arena half, F(1,60) = 37.45, P = 7.726 × 10⁻⁸). g, Density plots show Cre⁻ but not Cre⁺ cohort had significantly more approach-avoidance events to the CS⁺ than CS⁻. Two-sided Kolmogorov–Smirnov test, ****P = 3.1 × 10⁻⁵. Box plots—central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. Bonferroni post hoc and t-tests, as appropriate—*P < 0.05, ***P < 0.001 and ****P < 0.0001. Sex was balanced across conditions, and no sex differences were observed (for statistical comparison between females and males, see Supplementary Table 1 and Extended Data Fig.5a–f).

Fig. 4 |. dCA2 pyramidal neurons are not required for object fear discrimination memory.

a, OFC paradigm. b, OFC but not mock-OFC cohort displayed increased freezing during object recall relative to the habituation trial (two-way repeated measures ANOVA—cohort × stage, F(1,44) = 18.797, P = 8.335 × 10⁻⁵). b–g, n = 12 C57BL/6J wild-type mice per cohort. c, Left, OFC but not mock-OFC mice selectively avoided the CS⁺ arena half (two-way repeated measures ANOVA—cohort × arena half, F(1,44) = 19.36, P = 6.8 × 10⁻⁵). Right, discrimination scores. Discrimination scores for c, d, j and k—unpaired, two-sided t-tests between cohorts; two-sided one-sample t-tests against zero. d, Left, OFC but not mock-OFC cohort interacted more with the CS⁻ than CS⁺ object (two-way repeated measures ANOVA—cohort × stimulus object, F(1,44) = 6.1, P = 0.017). Right, discrimination scores. e, Speed versus head orientation angle plot. f, Relative approach/retreat velocity score for mock-OFC and OFC groups (two-way repeated measures ANOVA—cohort × arena half, F(1,44) = 51.87, P = 5.781 × 10⁻⁹). g, Frequency of approach-avoidance events, two-sided Kolmogorov–Smirnov test, ****P = 9 × 10⁻⁵. h, Experimental timeline of chemogenetic experiment. i, Cre⁺ (dCA2-silenced) and Cre⁻ controls injected with CNO 30 min prior to OFC displayed comparably elevated freezing during object recall relative to the habituation trial (two-way repeated measures ANOVA—cohort × stage, F(1,44) = 0.694, P = 0.409). i–n, n = 12 Amigo2-Cre⁻ and n = 12 Amigo2-Cre⁺ mice. j, Left, Cre⁻ and Cre⁺ cohorts both avoided the CS⁺ arena half (two-way repeated measures ANOVA—cohort × arena half, F(1,44) = 2.096, P = 0.154). Right, discrimination scores. k, Left, comparable CS⁻ and CS⁺ object interaction times between cohorts (two-way repeated measures ANOVA—cohort × stimulus object, F(1,44) = 0.235, P = 0.629). Right, discrimination scores. l, Cre⁻ and Cre⁺ mice exhibit similar selective retreat from CS⁺. m, Relative approach/retreat velocity to CS⁻ and CS⁺ for Cre⁻ and Cre⁺ cohorts (two-way repeated measures ANOVA—cohort × arena half, F(1,44) = 0.078, P = 0.78). n, Frequency of approach-avoidance. Two-sided Kolmogorov–Smirnov test for CS⁻ relative to CS⁺, ****P = 9.14 × 10⁻⁵ and ****P = 8 × 10⁻⁵ for OFC Cre⁺ and OFC Cre⁻, respectively. Box plots—central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. Bonferroni post hoc and t-tests, as appropriate—*P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Sex was balanced across conditions, and no sex differences were observed (for statistical comparison between females and males, see Supplementary Table 1). Hab., habituation trial.

Fig. 5 |. SFC enhanced and stabilized dCA2 pyramidal neuron representations of social identity.

a, dCA2 pyramidal neuron microendoscopic calcium imaging. GCaMP8m is expressed in n = 5 male Avpr1b-Cre⁺ and n = 5 male Amigo2-Cre⁺ mice. b, Example GCaMP8m expression. Dashed lines, GRIN lens tract. c, Example field of view and cell detection with CNMF-E. d, Calcium imaging SFC paradigm. SE, social exploration trials 2 h before SFC or mock-SFC; SR, social recall trials 24 h after SFC or mock-SFC. e, Pie charts showing the fraction of dCA2 cells with indicated response selectivity during two SE trials before mock-SFC and SFC, and during four SR trials after mock-SFC or SFC. Fraction of CS⁺ or CS⁻ selective increased after SFC (two-sided chi-square test—χ²(4) = 30.237, P = 4.38 × 10⁻⁶) but not after mock-SFC (two-sided χ²(4) = 3.364, P = 0.338). f, Representative GCaMP8m z-scored activity traces as the subject mouse interacted with empty cups (left column) or a stimulus mouse during indicated SR trials after SFC (middle and right columns). Top row, CS⁺ selective cell; middle, CS⁻ selective cell; bottom, spatially selective cell. Shaded regions and labels indicate periods of interaction with an indicated cup (‘back’ or ‘front’ of the arena) or mouse. g, Average z-scored responses for CS⁻ and CS⁺ selective dCA2 cells during SE trials before and SR trials after SFC (mean ± s.e.m.; across mice) aligned to start of social interactions (t = 0; dashed line). h, CS⁺ and CS⁻ selective cells identified on day 1 (labels under the charts) showed an increased probability of maintaining their selectivity on day 2 after SFC (right) but not after mock-SFC (left). Two-sided chi-square tests, SFC for CS⁺ cells—χ²(4) = 24.307, P = 6.9 × 10⁻⁵; for CS⁻ cells—χ²(4) = 21.409, P = 2.62 × 10⁻⁴; mock-SFC for CS⁺ cells—χ²(4) = 0.8378, P = 0.933; for CS⁻ cells—χ²(4) = 1.9028, P = 0.754. Total number of cells (mean ± s.e.m. across mice)—mock-SFC SE, 672 (69 ± 14); mock-SFC SR, 658 (64 ± 10); SFC SE, 647 (62 ± 7) and SFC SR, 630 (58 ± 6). Stability analysis (h) based on tracking the same cell population across both days (mock-SFC, 658 cells; SFC, 630 cells total).

Fig. 6 |. Enhanced decoding of CS⁺ and CS⁻ by dCA2 population activity after SFC.

a, Top left, dCA2 pyramidal neuron activity based on calcium signals measured during interactions with stimulus mice. Bottom left, binary classes formed by grouping calcium signals for each neuron during exploration of CS⁺ or CS⁻ mice in front and back cups across two successive SE or SR trials. Top right, the accuracy of decoding interactions with CS⁺ versus CS⁻ during the two SE (SE 1 and 2) trials before mock-SFC (green) or SFC (purple), and during the two pairs of SR trials (SR 1 and 2, SR 3 and 4) after mock-SFC or SFC. Bottom right, behavioral discrimination scores for CS⁻ versus CS⁺ interaction times. Two-way repeated measures ANOVA—cohort × session. Decoding accuracy—F(2,54) = 8.4, P = 6.67 × 10⁻⁴ and discrimination score—F(2,54) = 6.65 × 10⁻⁴. Thin lines, individual animals. Filled symbols connected by thick lines and error bars in a and c, mean ± s.e.m. averaged from different animals for SFC, mock-SFC and null model as indicated. a–c, n = 5 male Avpr1b-Cre⁺ and n = 5 male Amigo2-Cre⁺ mice, 514 cells total. Bonferroni post hoc tests between SFC and mock-SFC, **P < 0.01 and ****P < 0.0001 (see Supplementary Table 2 for P values of CS-decoding against the null model). b, Correlation between change in freezing time (top graphs) or behavioral discrimination score (bottom graphs) with change in decoding accuracy before and after SFC (left graphs) or mock-SFC (right graphs). Dots, individual animals. Filled symbols and solid lines, data from SR trials 1 and 2; open symbols and dashed lines, data from SR trials 3 and 4. r, Pearson’s correlation coefficient. For SFC, both correlations were significant with SR trials 1 and 2 data; neither was significant for trials 3 and 4 data. For mock-SFC, no correlations were significant. c, Accuracy of decoding CS⁺ (solid lines) or CS⁻ (dashed lines) versus empty cup (same arena position) was enhanced after SFC but not mock-SFC (two-way repeated measures ANOVA—cohort × session, F(5,108) = 4.717, P = 6.18 × 10⁻⁴). Symbols and error bars show mean ± s.e.m. averaged from different animals. Significantly greater than chance decoding accuracies for SFC and mock-SFC for all trials. Significantly greater decoding accuracies after SFC versus mock-SFC on day 2 (one-sided, paired t-tests; see Supplementary Table 2 for P values across all trials).

Fig. 7 |. Enhanced decoding of stimulus mouse during free interactions after SFC.

a, Subject and stimulus mice during free interactions in an open arena. b, Trajectories of the subject (top) and indicated stimulus mouse (bottom) during SE before SFC (left) and during two successive recall sessions with CS⁻ and CS⁺ 24 h after SFC (right) with reduced social interactions during the latter. c, Left, subject mouse shows greater interaction time with CS⁻ relative to CS⁺ after SFC; no difference before SFC (two-way repeated measures ANOVA—cohort × stimulus mouse, F(1,60) = 15.46, P = 2.21 × 10⁻⁴). Right, discrimination scores for CS⁻ versus CS⁺ interaction times. Unpaired, two-sided t-tests between cohorts and two-sided one-sample t-tests against zero. c,d, n = 5 male Amigo2-Cre⁺ mice, 382 cells total. d, CS⁺ (solid lines) or CS− (dashed lines) decoding accuracy as a function of distance to subject mouse relative to no interaction at the same distance. After SFC (purple), CS⁺ interactions decoded over a greater range of distances and with higher accuracy than CS⁻ interactions (dashed line; ANOVA—cohort × distance, F(30,176) = 3.215, P = 6.13 × 10⁻⁶). After SFC, CS⁺ decoding accuracy was above the chance level at all distances. CS⁻ only was decoded above chance for separations ≤1.5 cm. CS⁺ accuracy was significantly greater than CS⁻ accuracy at all distances. After mock-SFC (green), decoding accuracies for interactions with the two stimulus mice did not differ and were above chance only for separations ≤0.5 cm. Bonferroni post hoc tests and one-sided, paired t-tests against decoding accuracies obtained from the null model (see Supplementary Table 3 for P values across all distances). Bold lines and bars in d, mean ± s.e.m. across cohorts. Box plots—central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. Bonferroni post hoc tests (c, left) and t-test (c, right), *P < 0.05 and **P < 0.01.

Fig. 8 |. dCA2 population activity encodes an abstract representation of social valence and position.

a, Protocol to measure CCGP. Two SFC sessions, separated by 1 week, with stimulus mice CS1⁺ and CS1⁻ in week 1 and CS2⁺ and CS2⁻ in week 2. CS⁻ mice, blue; CS⁺ mice, red. Week 1 mice, darker shades; week 2 mice, lighter shades. Back cup interactions, solid color; front cup interactions, cross-hatched. b, Assignment of binary classes for training and testing data to measure the following CCGPs: (1) cross-identity CCGP for social valence, (2) cross-identity and cross-position CCGP for social valence and (3) cross-identity CCGP for position. c, (1) Cross-identity social valence; CCGP = 70.6% ± 1.4% (mean ± s.e.m. throughout figure) after SFC (***P = 3.11 × 10⁻⁶ compared to null model, one-sided paired t-test throughout). CCGP = 51.5% ± 2.1% after mock-SFC (NS, P = 0.97). (2) Cross-identity and cross-position social valence; CCGP = 63.5% ± 1.3% after SFC (***P = 1.2 × 10⁻⁵). CCGP = 51.7% ± 1.8% after mock-SFC (NS, P = 0.94). (3) Position; CCGP = 67.2% ± 1.3% after SFC (***P = 8.43 × 10⁻⁶). CCGP = 63.8% ± 1.8% after mock-SFC (***P = 2.17 × 10⁻⁵). c,d, n = 5 male Amigo2-Cre⁺ mice with 309 cells tracked across both SFC sessions. Box plots—central line, median; bottom and top edges, 25th and 75th percentiles; whiskers, most extreme data points (excluding outliers); dots, individual animals. d, Response profiles of dCA2 cells during week 2 mock-SFC or SFC recall sessions that were originally responsive to CS⁺ or CS⁻ (labeled below pie charts) during corresponding week 1 recall sessions. Two-sided chi-square tests against cell fractions expected by chance: mock-SFC CS⁺—χ²(4) = 5.2103, P = 0.266; mock-SFC CS⁻—χ²(4) = 4.7123, P = 0.318; SFC CS⁺—χ²(4) = 28.353, P = 1.1 × 10⁻⁵ and SFC CS⁻—χ²(4) = 24.711, P = 5.8 × 10⁻⁵. e, Hypothetical decoding plane from a linear classifier trained to decode CS1⁺ from CS1⁻ based on three-dimensional trapezoidal-like representational geometry before SFC (left) fails to separate (decode) CS2⁺ from CS2⁻. Classifier plane trained to decode CS1⁺ from CS1⁻ based on two-dimensional planar representations after SFC (right) will decode CS2⁺ from CS2⁻. Distances between points are drawn roughly to scale based on pairwise decoding results from Supplementary Table 4.