Parvalbumin Neurons and Cortical Coding of Dynamic Stimuli: A Network Model

Abstract

Cortical circuits feature both excitatory and inhibitory cells that underlie the encoding of dynamic sensory stimuli, e.g., speech, music, odors, and natural scenes. While previous studies have shown that inhibition plays an important role in shaping the neural code, little is known about how excitatory and inhibitory cells coordinate to enhance encoding of temporally dynamic stimuli. Recent experimental recordings in mouse auditory cortex (ACx) have shown that optogenetic suppression of parvalbumin (PV) neurons results in a decrease of neural discriminability between dynamic stimuli with speech envelope modulations. Here, we present a multilayer model of a cortical circuit that mechanistically explains these results. The model incorporates characteristic short-term synaptic plasticity (STP) profiles of excitatory and PV cells. Crucially, the model performance is based on populations of PV cells that separately respond to stimulus onsets and offsets. We reveal that by tuning the relative strengths of inhibition from onset- and offset-responding PV cells, the cortical network model captures the broad range of neural discriminability profiles in cortical single-unit data, with varying contributions from rapid firing rate modulations and spike timing. The model also replicates and explains the experimentally observed reduction in neural discrimination performance during optogenetic suppression of PV neurons. These results suggest that distinct populations of PV neurons enhance cortical discriminability of dynamic stimuli by encoding distinct temporal features, enhancing temporal coding, and reducing cortical noise.

Overexpression of the schizophrenia risk gene C4 in PV cells drives sex-dependent behavioral deficits and circuit dysfunction

“…This specific feed-forward arrangement was selected such that all four basic connectivity sub-types (E->E, E->I, I->E, and I->I) sampled experimentally via mEPSCs and mIPSCs recorded in both PYR and PV cells could be included and thus be leveraged to better understand the effects of increased levels of mC4 in PV cells on the firing properties of model neurons. This feed-forward network architecture is an adaptation of that used by Nocon et al (76), but was further modified using arrangement principles similar to those used by Seay et al (77) and Moore et al (78). Notably, our model lacks feed-back circuitry and classical inhibitory action of other interneuron types, and favors single synapses over a population construction.…”

“…PSC rise (𝜏1) and decay (𝜏2) kinetics were derived from experimental values (Table S6). Values used for fp and 𝜏p were the same as those in Nocon et al (76). In total, PSCs were modeled as:…”

“…In line with previous models (67,76,77), short-term depression was employed in all synapses. PSCs were modeled via the difference of two decaying exponential functions (76) featuring time constants where 𝜏1>𝜏2:…”

“…When a presynaptic spike is detected by a postsynaptic cell, x (spike inputs to that postsynaptic cell) is increased by P -representing synaptic strength -which is then reduced by fp, a fraction of its value (67,76).…”

“…These weights were adjusted by the same percent change as that observed in the mEPSC or mIPSC frequency or amplitude of the experimental data. In line with previous models (67,76,77), short-term depression was employed in all synapses. PSCs were modeled via the difference of two decaying exponential functions (76) featuring time constants where 𝜏 1 >𝜏 2 : where B is a normalization constant, s = 1 at maximum, and u(t) is the unit step.…”