What was recorded, and how it is analysed

Experimental design

Six male C57BL/6N mice (6–7 weeks at arrival) carried a Doric miniscope during chronic social defeat stress (CSDS). Imaging was in prelimbic mPFC (AP +1.80 mm, ML +0.35 mm, DV −2.30 mm, no angle) with AAV2/9-Syn-GCaMP6f through a 0.5 mm Doric lens (SICL_D_500_80). The lab classified 4703, 6597 and 6669 as resilient and 6445, 6672 and 6675 as susceptible. Imaging was done on CSDS days 1, 2, 3, 5, 7 and 9 at 6.666 frames/s. Each day was a ~15-min recording divided into baseline (BL) → social defeat (SD) → sensory stress (SS, videos labelled "PS"). Day 7 is left out of the primary analyses so that every mouse is compared on the same days: it was not recorded for 6597, and it is usable for the other five. Mouse 6597 is also missing SS on day 9. The files also contain day 11: a baseline, a social interaction test (SIT: ~1,000 frames with an empty enclosure, then ~1,000 frames with a CD-1) and a post-test. Mouse 6669 has no SIT imaging block. SIT was not analysed in the previous article.

Neural data and preprocessing

The lab preprocessed the raw video as follows: black frames replaced by neighbouring averages, 3× spatial downsampling (630 → 210 px), rubberband baseline correction of mean frame intensity, and motion correction. Epochs were merged within each day, and across days, to check the field of view. Minian could not reliably detect neurons during defeat, so 848 ROIs were drawn manually on the maximum projection and aligned by hand across days. ΔF/F was computed on the full concatenated recording (F₀ = 10th percentile in 12-s windows). CASCADE then produced a spike probability for every ΔF/F sample (the model and noise level were not recorded). Each TS_{id}.xlsx workbook holds one row per neuron (95–187 per animal) and one column per frame, for both ΔF/F (dff_mc) and spike probability (predictions_dff_mc). The part1/2/3 sheets exist only because of Excel's column limit, and the pipeline concatenates them. Recordings are 71–78 frames shorter than 2,000 × blocks, and population-activity steps drift progressively later than a fixed 2,000-frame grid. Epoch boundaries are therefore placed at k·δ, where δ is a per-animal mean epoch length (1,998–2,001 frames, as reported by the lab) estimated from reliable steps. Study S00 documents the boundaries and validates the block-to-session mapping.

Behavioral data

BORIS-style state annotations (start, stop, behavior) per session video. ANY-maze triggered the miniscope, so both streams start together. Files of about 305 s (about 155 s for each SIT half) are shifted back by 5 s before alignment. Labels: standing, exploring, grooming, digging, physical / non-physical interaction, approach, attack, submissive, freezing, escape, danger.

Analysis philosophy

• Animal is the unit of inference. Neurons and sessions are nested within animals. Condition contrasts use per-animal means over days on which every compared condition exists (Friedman, exact Wilcoxon). Day trends are tested on per-animal slopes. Session-level OLS with animal × day block fixed effects is reported as a descriptive within-day estimate, because its p-values treat sessions as independent.
• Order confound. BL, SD and SS are always recorded in that order within a day, so condition contrasts cannot be separated from time-in-recording effects.
• Nulls that respect autocorrelation. Every encoding, decoding and connectivity test is compared with circular-shift surrogates, which preserve each signal's temporal structure.
• Density-controlled graphs. Topology is compared at a fixed edge density (10%), so condition differences cannot arise from global changes in correlation strength alone.
• Multiple comparisons. Families of tests are Benjamini–Hochberg corrected. With n = 6 the minimum exact two-sided Wilcoxon p is 0.031, so effect sizes and consistency (k/6 animals) are always reported.
• Sensitivity. Results are repeated on ΔF/F, and network effects are repeated after global signal regression and with a firing-rate covariate.

Studies

0. Data provenance, quality control and session mapping — Recovering session boundaries from concatenated recordings and validating them against behavior.
1. Behavioral phenotyping across chronic social defeat — Time budgets, the evolution of defeat responses across days, and a social-interaction phenotype.
2. Neural activity across CSDS phases — Inferred firing, active-neuron fraction and behavior-locked population responses.
3. Behavior-tuned neurons and ensemble reactivation — Which neurons encode defeat behaviors, and are defeat-responsive neurons re-engaged in the SIT?
4. Functional network topology across CSDS phases — Neuron-level co-activity graphs, null-model thresholding and graph-theoretic contrasts.
5. Communities, hubs and rich-club organisation — Ensemble (module) structure, connector hubs, rich clubs and their stability across days.
6. Network reorganisation and drift — Similarity of neuron-level FC across all sessions: context-specific configurations vs slow drift.
7. Behavior-state-dependent networks — Functional networks computed separately within each annotated behavior, frame-count matched.
8. Population decoding of behavior and context — Cross-validated linear decoders with temporally structured null models.
9. Population dynamics: dimensionality and hidden states — Participation-ratio dimensionality and hidden Markov model population states across CSDS.
10. Susceptible vs resilient: neural correlates of the CSDS phenotype — Which per-animal network, activity and coding measures separate the lab-classified susceptible and resilient mice?