Reduced short-lag autoregressive carry-over in resting-state latent dynamics in major depressive disorder

Major depressive disorder (MDD) is increasingly characterized as a disorder of brain dynamics, but the temporal scale on which resting-state activity differs from healthy controls remains unclear. We analyzed resting-state fMRI data from 71 participants (51 MDD, 20 controls; OpenNeuro ds002748) using bilateral grid-extracted left-right difference signals projected into a control-defined latent space. Across 30 analysis configurations spanning five denoising pipelines, three latent dimensionalities, and two left-right correction settings, MDD showed reduced mean lag-1 autocorrelation in 24 of 30 configurations (integrated β = -0.0321, p = 0.036) and reduced AR(1) gain φ in 24 of 30 (β = -0.0316, p = 0.037), both robust to adjustment for pipeline, age, sex, and head motion. The broader autocorrelation area under lags 1-10 was not significantly reduced (β = -0.0099, p = 0.22), and lag-by-lag analysis localized the group difference specifically to lag 1. A model-comparison analysis identified AR-gain reduction as one of three competing candidate generative transformations rather than the unique winner. MDD is thus associated with a focal short-lag autoregressive carry-over reduction, a candidate latent-dynamical phenotype rather than a fully identified mechanism.