Permanent Scars and Excitotoxic Pruning: A Dynamic Computational Framework for Bipolar Progression and Sensitization

Background: Bipolar disorder often follows a progressive course, with early episodes linked to stressors and later ones becoming more autonomous—a pattern captured by the kindling hypothesis. Recent genetic evidence supports excessive, inhibition-biased synaptic pruning as a core mechanism distinguishing bipolar from unipolar depression. Building on static pruning simulations, we developed a dynamic model to test whether episode-induced permanent damage naturally gives rise to sensitization and cross-pole vulnerability. Methods: We extended a gated recurrent unit (GRU)-based network trained on sequential classification. Phenotypes were induced via magnitude pruning (75–95% sparsity; 1.0–2.0× inhibition bias). Sensitization was implemented through chained episodes: depressive events caused permanent scarring (60% of acute pruning irreversible), manic events triggered excitotoxic loss of high-magnitude excitatory weights, and thresholds decayed multiplicatively (85–90%) per confirmed episode, with bidirectional cross-pole effects. Three experiments examined progression in BD-Classic (detailed chain), cross-phenotype patterns, and fixed-trigger decay. Results: In BD phenotypes, few confirmed episodes produced substantial threshold decay (e.g., depressive trigger 0.300 → 0.217), stress sensitivity amplification (up to 1.24×), and functional decline, with relative E/I preservation enabling bidirectional vulnerability. MDD showed unidirectional collapse. Fixed moderate triggers yielded escalating responses over time, recapitulating kindling dynamics. No rapid cycling emerged in these runs, but alternation potential was evident in BD variants. Conclusion: Inhibition-biased pruning provides a mechanistic substrate for kindling-like progression, reconciling mixed clinical evidence by linking initial circuit profile to sensitization style. The model underscores early intervention's role in halting scarring and suggests pruning pathways as preventive targets.