Learning Dynamics in Biophysical Spiking Network Models Are Shaped by KCC2/NKCC1 Cotransporter Stoichiometry

Neuronal chloride homeostasis, governed by the opposing actions of the KCC2 exporter and NKCC1 importer, plays a critical role in regulating excitability and plasticity. While their individual functions are well characterized, the quantitative impact of their combined dynamics on learning processes remains poorly understood. In this computational study, we systematically varied KCC2 and NKCC1 activity levels within a biologically grounded realistic spiking neural network (RSNN) model to examine their influence on learning performance. Simulations were conducted with 441 combinations of parameters, under two different NKCC1 stoichiometries: the canonical 1 Na + :1 K + :2 Cl − ratio and a proposed alternative 1 Na + :4 K + :5 Cl − configuration. Our results demonstrate that successful learning emerges only within a constrained range of KCC2 and NKCC1 activity. Moreover, the shape and boundaries of this functional window are qualitatively altered by the NKCC1 stoichiometry. Under the 1:1:2 configuration, increased KCC2 activity consistently enhanced learning by stabilizing chloride dynamics. In contrast, under the 1:4:5 stoichiometry, the relationship was non‐monotonic, indicating complex, context‐dependent effects of transporter activity on network performance. These findings highlight the critical role of biophysical transporter properties—such as ion coupling ratios—in shaping the computational capabilities of neural circuits. They underscore the importance of incorporating non‐synaptic, ion homeostatic mechanisms into models of learning, particularly when investigating brain development and neurological disorders where chloride regulation is disrupted.