Susceptible vs resilient: neural correlates of the CSDS phenotype

Background & rationale

Only a subset of mice subjected to CSDS develop lasting social avoidance (susceptible); the rest remain resilient (Krishnan et al. 2007; Golden et al. 2011). Resilience is an active process with its own neural signatures. Pre-stress nucleus accumbens D1-MSN activity predicts later resilience (Muir et al. 2018), a brain-wide electrical network signature tracks vulnerability (Hultman et al. 2018), and susceptible and resilient mice differ in large-scale functional connectivity after CSDS (Mol Psychiatry 2025). Hippocampal social coding also differs between the phenotypes (Biol Psychiatry Glob Open Sci 2025). Here the lab classified 3 mice as resilient (4703, 6597, 6669) and 3 as susceptible (6445, 6672, 6675), so we can ask which single-neuron network measures, measured during the defeat period, differ between the groups.

Methodology

Per-animal features (days 1–9 without day 7). * Behavior (S01): defeat-response index (fraction of SD time submissive + freezing + escape) and the annotation-based SI proxy. * Activity and networks (S02, S04): for event rate, mean pairwise correlation, significant-edge density, clustering, global efficiency and modularity, the animal's baseline (BL) mean and the defeat response (SD − BL) and sensory-stress response (SS − BL), computed over days on which the animal has all of BL, SD and SS. * Reorganisation (S06): FC context specificity and the drift slope of SD networks (similarity per day of separation, from the pairwise session similarities). * Coding (S08): context-decoding accuracy and behavior-decoding accuracy above each animal's own null.

Group comparison. For each feature, the susceptible − resilient mean difference was tested with an exact permutation test over all 20 ways of labelling 3 of the 6 animals as susceptible (two-sided). With 3 vs 3 the smallest attainable p is 0.10, which happens only when the groups are completely separated. The study therefore reports Hedges' g and complete separation (every susceptible value above, or below, every resilient value) as the primary descriptors. Under the null, any given feature shows complete separation with probability 0.10. BH-FDR q-values are reported for completeness.

Trajectories. Session-level measures were averaged per phenotype for each recording day and condition to show whether the groups diverge as defeats accumulate.

Continuous coupling. Spearman ρ (exact, 720 permutations) between each neural feature and the defeat-response index.

Parameters

Results

Group separation. Of 25 per-animal features, 2 completely separated the 3 susceptible from the 3 resilient mice, against ≈2.5 expected by chance: event rate (Hz) · SS − BL (g = -3.22, susceptible < resilient); event rate (Hz) · SD − BL (g = -1.90, susceptible < resilient). The chance of seeing at least this many separating features is p = 0.73 (binomial, treating features as independent; they are correlated, so this is optimistic). The count is compatible with chance, so no single feature should be read as a phenotype marker. Each such feature has the minimum attainable exact p of 0.10. Table 1 lists all features ranked by separation and |Hedges' g|, and Figure 1 shows the strongest ones per animal.

Largest effects. event rate (Hz) · SS − BL: susceptible -0.107 vs resilient 0.0151 (g = -3.22); event rate (Hz) · SD − BL: susceptible -0.31 vs resilient -0.0536 (g = -1.90); SD network drift (Δsim/day): susceptible -0.0125 vs resilient -0.0372 (g = +1.35); SI proxy (annotations): susceptible 1.41 vs resilient 3.13 (g = -0.83); modularity Q · SD − BL: susceptible 0.0464 vs resilient 0.0149 (g = +0.80).

Trajectories. Figure 2 shows activity and network measures per recording day for each phenotype in BL, SD and SS sessions. Figure 3 shows the BL → SD → SS profile of each group.

Interpretation. With three animals per group, this analysis cannot establish a phenotype difference. Features that separate the groups and agree with prior literature are the natural pre-registered hypotheses for a larger cohort.

Limitations

• n = 3 per group. No feature can reach p < 0.05, and with ~30 features about 3 would be expected to separate the groups by chance. Results identify candidate markers for a confirmatory cohort and are not evidence of a phenotype difference.
• Mouse 6597 has no sensory-stress block on day 9, so that animal's network features use days 1, 2, 3 and 5. Mouse 6669 lacks the SIT imaging block; day-1–9 features for that animal are complete. Day 7 is excluded for every mouse.
• No non-defeated control group, so "resilient" differences cannot be separated from "less affected by the assay".

References

1. Krishnan V. et al. (2007) Molecular adaptations underlying susceptibility and resistance to social defeat in brain reward regions. Cell 131:391–404.
2. Golden S.A. et al. (2011) A standardized protocol for repeated social defeat stress in mice. Nat Protoc 6:1183–1191.
3. Muir J. et al. (2018) In vivo fiber photometry reveals signature of future stress susceptibility in nucleus accumbens. Neuropsychopharmacology 43:255–263.
4. Hultman R. et al. (2018) Brain-wide electrical spatiotemporal dynamics encode depression vulnerability. Cell 173:166–180.
5. Resting-state fMRI reveals altered functional connectivity associated with resilience and susceptibility to CSDS (2025) Mol Psychiatry. doi:10.1038/s41380-025-02897-2
6. Diminished social memory and hippocampal correlates of social interactions in chronic social defeat stress susceptibility (2025) Biol Psychiatry Glob Open Sci. https://www.bpsgos.org/article/S2667-1743(25)00009-6/fulltext
7. Button K.S. et al. (2013) Power failure: why small sample size undermines the reliability of neuroscience. Nat Rev Neurosci 14:365–376.