Permanent Synaptic Loss Renders Plasticity Enhancement Ineffective: Computational Validation of Prune-Without-Repair in Schizophrenia-Like Deficits

Background and Hypothesis: Cognitive deficits in schizophrenia are early, stable, and poorly responsive to existing treatments. Excessive adolescent synaptic pruning, driven by complement pathways, interacts with impaired glutamatergic plasticity, raising the possibility that permanent synaptic loss ("prune-without-repair") underlies treatment resistance. The degree of irreversibility remains unclear, prompting computational tests of this mechanism. Study Design: We developed a diagnosis-free recurrent neural network (GRU-based, with rule-learning task) to simulate excessive pruning (~95% sparsity), chronic plasticity impairment (gradient scaling factor 0.5), and varying irreversibility (0–100% of pruned connections permanently masked). A healthy calibrated network (75% sparsity, Cognitive Index [CI] ≈116) served as reference. "Treatment" involved gradient-guided regrowth of reversible connections (30% fraction) followed by consolidation. Irreversibility was swept across six levels, with 10 seeds per condition for robustness. Study Results: Pre-treatment CI fell to ~79–81 across conditions. Fully reversible pruning yielded near-complete recovery (+30.5 CI points on average), approaching healthy levels. Recovery remained substantial at 20–80% irreversibility (25–31 points) but collapsed to minimal gains (+3.5 points) when pruning was fully irreversible. A clear dose-response gradient emerged, with intermediate irreversibility producing partial but meaningful rescue. Conclusions: Irreversibility alone suffices to generate treatment-resistant cognitive deficits in this parsimonious model, directly supporting the prune-without-repair hypothesis. Partial permanence (consistent with 30–50% estimates from biological data) explains variable clinical responsiveness and underscores the need for early interventions preserving synaptic reversibility.