Study S06 · generated 2026-10-13T21:15:11+00:00 · 19.0 s

Network reorganisation and drift

Similarity of neuron-level FC across all sessions: context-specific configurations vs slow drift.

Question

Does each CSDS context (baseline, defeat, sensory contact, SIT) evoke its own network configuration that recurs across days, and how quickly does the functional network drift over the 11 days?

Background & rationale

Comparing whole connectivity matrices across sessions, rather than summary statistics, captures which neurons are coupled. In longitudinal imaging, neural representations drift over days while preserving coarse structure (Ziv et al. 2013; Driscoll et al. 2017; Rule et al. 2019). Context-dependent reconfiguration of functional connectivity is well established in human network neuroscience (Cole et al. 2014; Bassett et al. 2011). For CSDS the question is whether defeat and sensory contact each impose a recurring network state, and whether that state bleeds into baseline sessions as stress accumulates.

Methodology

FC similarity. For every pair of sessions within an animal, similarity = Pearson correlation between the Fisher-z-transformed upper triangles of the two neuron × neuron FC matrices (inferred spikes). The same neurons are present in every session, so this compares the identity of coupled pairs, not just summary topology.

Drift. For pairs from the same condition (BL, SD or SS; days 1–9, day 7 excluded), similarity was modelled as a function of day lag, separately for each animal. Pairs share sessions and are therefore not independent, so inference uses one slope per animal (exact Wilcoxon signed-rank across animals, BH-FDR across the three contexts). A negative slope indicates progressive drift.

Same-day similarity. Different-context pairs recorded on the same day were compared with different-context pairs from different days (per-animal means, exact Wilcoxon).

Context specificity. All pairs from different days among BL/SD/SS (days 1–9 without day 7) were modelled as similarity ~ same_condition + |lag| + animal. Restricting to different days removes the same-day confound (shared focal plane and slow fluorescence state). The same_condition coefficient estimates how much more alike two sessions are when they share a context. Significance was assessed by permuting condition labels within day for each animal (1,000 permutations; statistic = same-context minus different-context similarity after removing each animal's mean), which preserves the day structure. The per-animal mean difference (same − different context) was also tested with an exact Wilcoxon test.

Day-11 sessions. The SIT (empty, CD-1) and post-SIT networks were compared with the mean BL, SD and SS networks of days 1–9, day 7 excluded, to ask whether the SIT evokes a defeat-like configuration (exact Wilcoxon, SD vs BL reference).

Embedding. Classical (Torgerson) MDS of 1 − similarity per animal, coloured by condition.

Parameters

Parameter	Value
`similarity`	Pearson r of Fisher-z FC upper triangles
`permutations`	1000

Highlights

  
  null
Context-specific networksΔsim = +0.005
Sessions sharing a context are more alike than different-context sessions on other days (within-day permutation p = 0.126; Wilcoxon p = 0.219; OLS β = +0.005).

  significant
Drift of defeat networks-0.0249 per day
Mean per-animal change in FC similarity per day of separation between SD sessions; 6/6 animals negative, exact Wilcoxon p = 0.031, q = 0.047 (FDR over BL, SD, SS).

  null
SIT network resembles…no preference
SIT (CD-1) FC similarity to defeat 0.588 vs baseline 0.592, p = 1.000.

Results

Overall structure. Figure 1 shows session-by-session FC similarity. Different-context sessions recorded on the same day had a mean similarity of 0.738, against 0.598 for different-context sessions on different days (6/6 animals higher on the same day; exact Wilcoxon p = 0.031).

Drift. Same-context similarity changed with lag at BL -0.0227/day (p = 0.031, q = 0.047), SD -0.0249/day (p = 0.031, q = 0.047), SS -0.0264/day (p = 0.156, q = 0.156) (mean per-animal slope, exact Wilcoxon across animals; Table 1, Figure 2). Negative slopes indicate progressive reorganisation of which neurons are coupled. Elapsed time and accumulated defeat are confounded in this design, so drift cannot be attributed to stress alone.

Context specificity. Among different-day pairs, sessions sharing a context were more similar by +0.005 on average (4/6 animals positive; within-day label permutation p = 0.126; Wilcoxon p = 0.219). Per-context tests are in Table 2: BL: within 0.628 vs other 0.605 (p = 0.047); SD: within 0.616 vs other 0.597 (p = 0.078); SS: within 0.567 vs other 0.593 (p = 0.844).

Figures

**Figure 1. Session × session FC similarity.** Pearson correlation between Fisher-z FC matrices of every pair of sessions, chronologically ordered; the coloured strip marks condition.

**Figure 2. Drift of same-context networks.** FC similarity between sessions of the same context as a function of days apart (mean ± SEM across animals).

**Figure 3. Context-specific network configurations.** Left: per-animal mean similarity of different-day session pairs that share vs do not share a context. Right: per-context within-context vs cross-context similarity.

**Figure 4. Low-dimensional map of network states.** Classical MDS of FC dissimilarity (1 − similarity) per animal; colour = condition, marker size grows with CSDS day.

**Figure 5. Day-11 networks relative to CSDS contexts.** Similarity of SIT (empty, CD-1) and post-SIT networks to the baseline, defeat and sensory-contact networks of days 1–9, day 7 excluded (mean ± SEM across animals).

Tables

Table 1. Drift of same-context networks across days

Per-animal least-squares slope of similarity on day lag; exact two-sided Wilcoxon of slopes across animals; q = BH-FDR across contexts. Download CSV

condition	slope_per_day	n_negative	n_animals	p	similarity_lag_min	n_pairs	q_fdr
BL	-0.0227	6	6	0.0312	0.704	60	0.0469
SD	-0.0249	6	6	0.0312	0.689	60	0.0469
SS	-0.0264	4	6	0.156	0.633	56	0.156

Table 2. Within- vs between-context similarity (different days only)

Exact one-sided Wilcoxon across animals. Download CSV

condition	within_context	to_other_contexts	p
BL	0.628	0.605	0.0469
SD	0.616	0.597	0.0781
SS	0.567	0.593	0.844

Table 3. Context specificity per animal

Mean FC similarity for different-day pairs sharing vs not sharing a context. Download CSV

subject	different_context	same_context	specificity
4703	0.701	0.722	0.0212
6445	0.527	0.52	-0.00707
6597	0.622	0.633	0.0108
6669	0.346	0.341	-0.0047
6672	0.708	0.714	0.00619
6675	0.685	0.689	0.00398

Table 4. Day-11 sessions vs days 1–9, day 7 excluded contexts

Mean FC similarity to each context. Download CSV

subject	session	BL	SD	SS
4703	POST	0.461	0.492	0.486
4703	SIT_CD1	0.535	0.576	0.572
4703	SIT_NoCD1	0.475	0.528	0.535
6445	POST	0.527	0.494	0.436
6445	SIT_CD1	0.558	0.565	0.456
6445	SIT_NoCD1	0.629	0.647	0.525
6597	POST	0.565	0.527	0.567
6597	SIT_CD1	0.537	0.48	0.555
6597	SIT_NoCD1	0.508	0.389	0.52
6669	POST	0.0905	0.0977	0.134
6672	POST	0.691	0.68	0.718
6672	SIT_CD1	0.703	0.71	0.73
6672	SIT_NoCD1	0.549	0.607	0.553
6675	POST	0.626	0.605	0.573
6675	SIT_CD1	0.629	0.607	0.578
6675	SIT_NoCD1	0.53	0.511	0.497

Table 5. All pairwise session similarities

Showing 10 of 973 rows. Download CSV

subject	a	b	cond_a	cond_b	day_a	day_b	similarity	lag	same_condition
4703	4703_BL_D1	4703_SD_D1	BL	SD	1	1	0.712	0	0
4703	4703_BL_D1	4703_SS_D1	BL	SS	1	1	0.713	0	0
4703	4703_BL_D1	4703_BL_D2	BL	BL	1	2	0.792	1	1
4703	4703_BL_D1	4703_SD_D2	BL	SD	1	2	0.716	1	0
4703	4703_BL_D1	4703_SS_D2	BL	SS	1	2	0.731	1	0
4703	4703_BL_D1	4703_BL_D3	BL	BL	1	3	0.764	2	1
4703	4703_BL_D1	4703_SD_D3	BL	SD	1	3	0.662	2	0
4703	4703_BL_D1	4703_SS_D3	BL	SS	1	3	0.668	2	0
4703	4703_BL_D1	4703_BL_D5	BL	BL	1	5	0.653	4	1
4703	4703_BL_D1	4703_SD_D5	BL	SD	1	5	0.626	4	0

Limitations

FC similarity is sensitive to session length and noise. BL/SD/SS sessions have equal length (≈2,000 frames), but SIT phases are ~1,000 frames and are shown descriptively only.
Slow changes in imaging quality across days also produce apparent drift.

References

Ziv Y. et al. (2013) Long-term dynamics of CA1 hippocampal place codes. Nat Neurosci 16:264–266.
Driscoll L.N. et al. (2017) Dynamic reorganization of neuronal activity patterns in parietal cortex. Cell 170:986–999.
Rule M.E., O'Leary T. & Harvey C.D. (2019) Causes and consequences of representational drift. Curr Opin Neurobiol 58:141–147.
Cole M.W. et al. (2014) Intrinsic and task-evoked network architectures of the human brain. Neuron 83:238–251.
Bassett D.S. et al. (2011) Dynamic reconfiguration of human brain networks during learning. PNAS 108:7641–7646.