Architectural reorganization of functional connectivity after propofol anesthesia exceeds within-state measurement variability and is only marginally reversed during early recovery

Abstract

Propofol anesthesia disrupts functional connectivity, but the magnitude of whole-brain architectural reorganization has not been calibrated against the variability of repeated measurement within a single state, and it is therefore unclear how much apparent reorganization reflects state change rather than estimation noise. We applied Weighted Jaccard (WJ) similarity to fMRI from healthy adults scanned during wakefulness, propofol-induced unresponsiveness, and early recovery (OpenNeuro ds006623), computing full pairwise Spearman correlation matrices among 456 regions. To provide an interpretable reference we computed a within-condition floor: WJ between two non-overlapping, equal-length segments of the same run. Every matrix in the primary analysis was estimated from an identical number of timepoints, because run durations differ across conditions. Between-state similarity fell significantly below the within-condition floor for all comparisons (floor WJ = 0.544; wakefulness versus recovery 0.486, wakefulness versus unresponsiveness 0.468, unresponsiveness versus recovery 0.458; paired Cohen's d 2.23 to 2.79, all p = 2.4e-07). Expressed against this reference, the state-attributable share of architectural divergence is approximately 11 to 14 percent, substantially smaller than the raw between-state dissimilarity implies. The conclusion was unchanged under four sign treatments: absolute, positive edges only, negative edges only, and a signed variant (all twelve tests p = 2.4e-07). Direct stratified sign-flip rates exceeded the within-condition floor at every magnitude stratum, by 5-fold for the task-matched contrast and 17-fold for unresponsiveness at min|r| ≥ 0.3, indicating genuine sign reorganization among strong edges rather than magnitude change alone. Reorganization was strongly redistributive (net-versus-gross uniformity R = 0.775 to 0.871), meaning edge strengthenings and weakenings largely cancel, so scalar summaries of total connectivity are insensitive to it by construction. Consistent with this, the change in mean absolute correlation did not distinguish wakefulness from recovery (p = 0.096) while the full-matrix comparison did. Subject-level recovery was small but reliably non-zero (3.2 percent, 95 percent CI 1.8 to 5.7). Group-averaged matrices yielded a recovery fraction an order of magnitude larger; group-level WJ exceeded that of every individual subject in all three comparisons, and this inflation persisted when subject identity was randomly permuted, establishing it as a property of matrix averaging rather than of cross-subject correspondence. These results support a modest, redistributive, and largely unreversed architectural reorganization after propofol, and indicate that group-averaged connectivity comparisons systematically overstate recovery.