How the layer-dependent ratio of excitatory to inhibitory cells shapes cortical coding in balanced networks

Abstract

The cerebral cortex exhibits a sophisticated neural architecture across its six layers. Recently, it was found that these layers exhibit different ratios of excitatory to inhibitory (EI) neurons, ranging from 4 to 9. This ratio is a key factor for achieving the often reported balance of excitation and inhibition, a hallmark of cortical computation. Although previous studies have explored various biologically realistic aspects of classical balanced networks, such as the effects of more complex neuronal properties including spike frequency adaptation using adaptive exponential integrate and fire (AdEx) neuron models (1), the temporal evolution of randomly connected balanced networks through second order mean field approximations (2), predictions of nonlinear dynamics across diverse neuron models ranging from integrate and fire to Hodgkin-Huxley formulations, thus bridging population dynamics to neuronal electrophysiology (3), the impact of spatial heterogeneity in balanced networks (4-6), and variations in synaptic timescales (7), the specific effects of differences in excitatory-inhibitory (EI) ratios on layer-specific dynamics and computational properties have, to the best of our knowledge, not yet been addressed. We investigate this question using a sparsely connected network model of excitatory and inhibitory neurons. To keep the network in a physiological range of firing rates, we varied the inhibitory firing threshold or the synaptic strength between excitatory and inhibitory neurons. We find that decreasing the EI ratio allows the network to explore a higher-dimensional space and enhance its capacity to represent complex input. By comparing the empirical EI ratios of layer 2/3 and layer 4 in the rodent barrel cortex, we predict that layer 2/3 has a higher dimensionality and coding capacity than layer 4. Furthermore, our analysis of primary visual cortex data from the Allen Brain Institute corroborates these modeling results, also demonstrating increased dimensionality and coding capabilities of layer 2/3.