Recurrent Pruning Dynamics Differentiate Depressive Fragility from Manic Instability: A Computational Extension of the Pruning-Mediated Hypothesis to Bipolar Disorder

Background: Bipolar disorder (BD) shares substantial genetic overlap with schizophrenia and major depressive disorder (MDD), particularly in synaptic pruning pathways, yet manifests distinct episodic manic hyperactivity rather than chronic fragility. While prior pruning simulations capture MDD-like stress-induced collapse, few address BD's "pruned-but-potent" phenotype—efficient but unstable circuits amplified into mania by factors like cognitive reserve. Methods: A recurrent neural network (GRU-based, ∼2.76 million parameters) was trained on sequential classification tasks. Pruning varied by severity (95% for MDD analog; 80% for BD) and bias (inhibition-preferential for BD to skew E/I balance). Cognitive reserve was modeled via activation scaling (depleted to extreme). Evaluation contrasted depressive conditions (internal noise/input perturbations) with manic instability (sustained drive measuring variance, maxima, explosion). Gradient-guided regrowth (50%) tested plasticity rescue. Results: Severe unbiased pruning induced MDD-like collapse (accuracy ∼49–50% under stress) with subdued manic metrics except extreme reserve explosion. Moderate inhibition-biased pruning preserved partial stress tolerance (∼74%) but generated profound instability under reserve amplification (extreme variance >4.5×10¹⁷, consistent explosions), classified as manic phenotype. Reserve sweeps showed dose-dependent shifts from fragility to hyperactivity. Regrowth normalized accuracy in both but left lingering manic variance, yielding partial phenotypic rescue. Conclusions: Moderate pruning with E/I skew and reserve amplification converts shared structural liability into BD-specific manic proneness, distinguishing it from MDD's erosive collapse. These dynamics support pruning dominance—activation-skewed in BD—interacting with reserve to drive episodic instability, implying targeted neuroimmune interventions for manic prevention and nuanced plasticity approaches across mood disorders. The recurrent framework advances unified modeling of bipolar spectrum divergence.