Unpredictable chronic stress redistributes novelty-evoked exploratory space use and yields method-dependent identification of resilient-like archetypes in zebrafish

Unpredictable chronic stress (UCS) produces heterogeneous behavioural outcomes, but identifying resilient individuals remains challenging because post-stress behaviour is multidimensional and classification depends on the analytical framework. We characterised novelty-evoked exploration in 178 adult zebrafish (89 unstressed reference, 89 UCS-exposed) using four Novel Tank Test variables: distance travelled, line crossings, upper-zone entries and upper-zone time, from a historical LAPCOM dataset from four independent cohorts. Behaviour was robustly normalised to control distributions within cohorts. In the unstressed population, principal component analysis revealed a dominant exploratory gradient (PC1 = 74.9%; PC2 = 18.7%), whereas weighted k-nearest-neighbour Louvain analysis yielded a stable two-profile partition. However, null-model inference and resampling did not support discrete behavioural populations, indicating a predominantly continuous reference architecture. UCS significantly redistributed individuals within this architecture (PERMANOVA: F(1,173) = 8.10, R2= 0.0423, p = 0.0016), reducing occupancy of the higher-exploration reference profile from 55.1% to 24.7%, while multivariate dispersion, Mahalanobis distance and global correlation structure showed no detectable differences. De novo Louvain analysis identified lower- and higher-exploration configurations, but correspondence with the control reference was asymmetric: all 44 lower-exploration fish mapped to the lower reference profile, whereas only 22 of 45 higher-exploration fish mapped to the higher reference profile. Thus, higher exploration after UCS was insufficient to define a resilient-like response. Across five analytical frameworks, resilient assignments ranged from 9 to 45 individuals and susceptible assignments from 23 to 68, with partial individual-level convergence. These findings support reference-integrated behavioural stratification for identifying resilient- and susceptible-like candidates for physiological, cellular and molecular validation.