Excessive Synaptic Pruning Drives Perseverative Phenotypes in a Computational Model of OCD: Validation of a Neurodevelopmental Hypothesis and Implications for Relapse

Background: Polygenic studies now show that genes guiding synaptic pruning—especially those tied to complement activity and microglia—carry much of the heritable risk for obsessive-compulsive disorder (OCD), whereas classic glutamatergic theories gain less support. How pruning errors alter cortico-striato-thalamo-cortical (CSTC) circuits, and why many patients relapse after treatment, remain open questions. To address these gaps, we built a computational model that asks whether excessive pruning alone can create the cognitive rigidity seen in OCD and what conditions make recovery lasting. Methods: A gated recurrent unit (GRU) network stood in for CSTC circuitry and learned a rule-switching task similar to the Wisconsin Card Sorting Test. Developmental over-pruning was mimicked by removing mainly recurrent weights. Therapy was represented by gradient-driven regrowth of connections, followed by a consolidation phase of variable length. Relapse was triggered through a second round of pruning. Added input noise served as a stand-in for glutamatergic disruption. We varied pruning severity, consolidation time, single-shot versus repeated regrowth, and stress levels across three random seeds. Results: Pruning alone produced clear, dose-related perseveration: once sparsity reached roughly 85–93 %, perseverative errors more than doubled, even without stress or noise. Noise reduced overall accuracy but did not drive rigidity. Regrowing connections cut perseveration by 55–70 %, yet short consolidation led to easy relapse. Longer consolidation and moderate, repeated regrowth—three cycles worked best—gave the strongest protection. Networks that had undergone heavy initial pruning remained more relapse-prone than lightly pruned peers, indicating an intrinsic vulnerability. Conclusions: The simulations support pruning dysregulation as a primary cause of OCD-like rigidity and place glutamatergic noise in a secondary, modulatory role. Relapse appears to arise when freshly formed synapses are not consolidated long enough. Extended or phased treatment may therefore be essential, especially for individuals carrying higher genetic risk linked to pruning pathways, to secure durable remission.