State-dependent regulation of cortical processing speed via gain modulation

Abstract

To thrive in dynamic environments, animals can generate flexible behavior and rapidly adapt responses to a changing context and internal state. Examples of behavioral flexibility include faster stimulus responses when attentive and slower responses when distracted. Contextual modulations may occur early in the cortical hierarchy and may be implemented via afferent projections from top-down pathways or neuromodulation onto sensory cortex. However, the computational mechanisms mediating the effects of such projections are not known. Here, we investigate the effects of afferent projections on the information processing speed of cortical circuits. Using a biologically plausible model based on recurrent networks of excitatory and inhibitory neurons arranged in cluster, we classify the effects of cell-type specific perturbations on the circuit's stimulus-processing capability. We found that perturbations differentially controlled processing speed, leading to counterintuitive effects such as improved performance with increased input variance. Our theory explains the effects of all perturbations in terms of gain modulation, which controls the timescale of the circuit dynamics. We tested our model using large-scale electrophysiological recordings from the visual hierarchy in freely running mice, where a decrease in single-cell gain during locomotion explained the observed acceleration of visual processing speed. Our results establish a novel theory of cell-type specific perturbations linking connectivity, dynamics, and information processing via gain modulations.

Predicting perturbation effects from resting activity using functional causal flow

“…5A-B). These metastable dynamics were previously shown to capture physiological properties of resting state activity in cortical circuits (Litwin-Kumar and Doiron, 2012, Mazzucato et al, 2015, 2016, 2019, Rostami et al, 2020, Wyrick and Mazzucato, 2020).…”

“…We validated our theoretical framework for causal inference using two classes of models: a continuous rate network and a network model of resting state activity in a cortical circuit. The latter is a biologically plausible model based on a recurrent spiking network where excitatory and inhibitory neurons were arranged in functional assemblies (Litwin-Kumar and Doiron, 2012, Mazzucato et al, 2015, Rostami et al, 2020, Wyrick and Mazzucato, 2020). Experimental evidence including multielectrode recordings in behaving monkeys (Kiani et al, 2015) strongly supports the existence of functional assemblies in cortex (Lee et al, 2016, Perin et al, 2011, Song et al, 2005, Wong et al, 2016).…”

“…Clustered spiking networks capture complex physiological properties of cortical dynamics during resting and stimulus-evoked activity due to the metastable dynamics of cluster activations. Such physiological properties include context- and state-dependent changes in neural activity and variability (Deco and Hugues, 2012, Litwin-Kumar and Doiron, 2012, Mazzucato et al, 2015, 2016, Rostami et al, 2020, Schaub et al, 2015); as well as neural correlates of behavior and cognitive function such as expectation, arousal, locomotion, and attention (Mazzucato et al, 2019, Rostami et al, 2020, Wyrick and Mazzucato, 2020).…”

“…E and I neurons were arranged in p = 10 clusters of equal size Pairs of neurons belonging to the same cluster had potentiated synaptic weights by a ratio factor , for α, β = E, I . Network parameters were chosen to generate spontaneous metastable dynamics with a physiologically realistic cluster activation lifetime, consistent with previous studies (Jones et al, 2007, Mazzucato et al, 2015, 2019, Wyrick and Mazzucato, 2020). Parameter values are in Table 3.…”

“…When V hits threshold (for α = E, I ), the neuron emits a spike and V is held at reset V reset for a refractory period τ refr . Thresholds were chosen so that the homogeneous network (i.e.,where all ) was in a balanced state with rates ( r E , r I ) = (2, 5) spks/s (Amit and Brunel, 1997, Mazzucato et al, 2019, Wyrick and Mazzucato, 2020). Input currents contained a contribution I rec from the recurrent connections and an external current I ext = I 0 + I pert ( t ) (units of mV s − 1 ).…”