Divergent Mechanisms of Antidepressant Efficacy: A Unified Computational Comparison of Synaptogenesis, Stabilization, and Tonic Inhibition in a Model of Depression

TL;DR

This study compares how different antidepressants work by simulating their effects on brain plasticity, showing each has unique benefits and drawbacks in treating depression.

Contribution

The paper introduces a unified computational model to compare three distinct antidepressant mechanisms in a shared framework.

Findings

Ketamine-like synaptogenesis improved stress resilience and durability with minimal relapse.

SSRI-like refinement showed moderate stress resilience but significant relapse vulnerability.

Neurosteroid-like inhibition provided rapid recovery but was state-dependent and less effective under extreme stress.

Abstract

Background: Major depressive disorder (MDD) is increasingly viewed as a disorder of impaired neural plasticity, yet the mechanisms underlying diverse antidepressant classes - glutamatergic (e.g., ketamine), monoaminergic (e.g., selective serotonin reuptake inhibitors (SSRIs)), and GABAergic (e.g., neurosteroids) - remain incompletely integrated. The objective of this study was to extend a pruning-plasticity model of depression and directly compare, from an identical severely pruned baseline state, the efficacy, stress resilience, durability, and relapse vulnerability of three mechanistically distinct interventions: ketamine-like targeted synaptogenesis, SSRI-like gradual refinement of existing connectivity, and neurosteroid-like tonic inhibition. Computational models offer a controlled means to compare these pathways, but prior work has typically examined single mechanisms.

Methods: We extended a pruning-plasticity model of depression by applying 95% magnitude-based synaptic elimination to overparameterized feed-forward networks trained on a four-class Gaussian classification task. From identical pruned states, three interventions were tested: ketamine-like gradient-guided regrowth (50% reinstatement) with consolidation; SSRI-like prolonged low-learning-rate training with gradual internal noise reduction; and neurosteroid-like global tonic inhibition (30% damping plus tanh activations) with brief consolidation. Outcomes included baseline accuracy, resilience to graded internal activation noise (up to σ = 2.5) plus input perturbation, and relapse vulnerability after an additional 40% pruning.

Results: All treatments restored near-ceiling performance on unchallenged inputs. Ketamine-like synaptogenesis uniquely reduced sparsity (to ~47%) and conferred superior stress resilience (extreme noise accuracy 84.5%) with near-zero relapse drop (−0.2%). SSRI-like refinement improved combined stress accuracy to 83.5% but showed limited extreme noise tolerance (44.0%) and substantial relapse vulnerability (10.8% drop). Neurosteroid-like inhibition achieved rapid combined stress recovery (97.5%) while active, but was state-dependent (decline upon removal) with poor extreme noise buffering (42.5%) and moderate relapse drop (4.1%).

Conclusions: These simulations demonstrate that antidepressants operate through mechanistically distinct routes-structural rebuilding (ketamine), gradual optimization of existing connectivity (SSRIs), or reversible dynamic stabilization (neurosteroids)-yielding trade-offs in onset speed, durability, and stress resilience. The findings support a multifaceted plasticity framework for depression and provide computational rationale for mechanism-based treatment selection and combination strategies.

Figures