NMDA Modulates Working Memory Attractor Stability: Opposite Regimes Can Produce Schizophrenia-like Instability and OCD-like Overstability

Abstract

Working memory (WM) depends on the sustained, selective firing of prefrontal neurons during a delay period — a property implemented in computational models as a stable attractor state. NMDA receptors are central to the excitatory recurrent currents that maintain these attractors. Using a biologically plausible network with AMPA, NMDA, and GABA_A synaptic currents, we show that NMDA conductance may act as a bidirectional gain control on attractor depth. To quantify this, we introduce \lambda_{\mathrm{delay}}, an exponential decay constant of the selective population firing rate during the delay period, where low values reflect stable memory maintenance and high values reflect attractor collapse. A 5% reduction in NMDA (schizophrenia-like) destabilizes the delay-period attractor, causing memory collapse and high \lambda_{\mathrm{delay}} values across the full recurrent weight range. Conversely, a 10% increase in NMDA (OCD-like) over-stabilizes the attractor, locking activity into persistence even at low synaptic weights. These opposite failure modes, instability and overstability, emerge from the same parameter and suggest a possible mechanistic axis that could link two clinically distinct psychiatric conditions. This micropublication was created as part of the Neuromatch Impact Scholars Program 2025. Project Website: https://impact-scholars.github.io/tahiri-2026-working-memory-model Repository: https://github.com/impact-scholars/tahiri-2026-working-memory-model