Computation by recurrent circuits of the cerebral cortex

National Institute Of Mental Health

Investigators

Linked publications, trials & patents

Abstract

Inhibition-stabilized network (ISN) models of the cerebral cortex predict that local (recurrent) synaptic connectivity plays a vital role in shaping cortical activity. Yet it has been experimentally unclear if cortical recurrent connectivity is strong enough to make ISNs good descriptions of cortical function. Here we test several ISN predictions, including the counterintuitive (paradoxical) suppression of inhibitory firing when the inhibitory network is stimulated. We pair, in awake subjects, optogenetic stimulation of all inhibitory subtypes with in vivo pharmacology to identify inhibitory cells. We observe responses consistent with an ISN in the upper layers of visual, somatosensory, and motor cortex. Stimulating parvalbumin (PV)-positive inhibitory neurons produces a population paradoxical effect only with transgenic, not viral, opsin expression. This effect is explained in a model where viral expression targets a subset of PV cells, showing inhibitory cell responses to stimulation can be highly dependent on the number of stimulated cells. We have also examined paradoxical neural responses that arise from input to not just inhibitory, but also excitatory neurons. We show that the recurrent connections in the cortex can create suppression when excitatory cells are excited, which intuitively should lead only to elevated firing rates in cortical cells. We first observe that intense visual inputs create a salt-and-pepper pattern of suppressed and excited neurons in mouse V1. To selectively examine the recurrent contributions to suppression, we stimulate excitatory neurons optogenetically, and measure responses using electrophysiological recordings and calcium imaging at the 2-photon cellular level and widefield population level. When cortical excitatory cells are excited directly, we find suppressed cells in a similar salt-and-pepper pattern as with strong visual input. Though we drive only excitatory neurons, both excitatory and inhibitory neural populations settle into similarly distributed steady states, with some cells elevated and some suppressed. Suppressed cells are distributed across the cortex, though widefield imaging reveals a surround showing suppression a few hundred microns away from the center of stimulation. A moderately or strongly coupled balanced-state cortical model with heterogeneous local connectivity can account for the results, displaying suppression due to recurrent connectivity even when only excitation is added to the network. The mechanism for paradoxical suppression with excitatory input is different than that for paradoxical responses to inhibitory input, though both depend on strong recurrent connectivity. For excitatory cell input, it is the variability in the number of connections each cell receives that accounts for suppression that is, some neurons receive few excitatory inputs and are suppressed, while others receive many excitatory inputs and their rates are elevated. This variability in local, recurrent connectivity is a novel mechanism to shape input-output transformations, and one that is essential to understand sensory cortical function, as this mechanism is needed to explain the suppression seen for strong visual input. Taken together, our results show that one function of recurrent cortical connectivity is to shape individual neurons steady-state responses to input. Such shaping of responses by recurrent connectivity is an essential feature of cortical computation, and thus of brain function. Our work helps explain why recurrent connections in cortical circuits are so pervasive, and in continuing work, we are further elaborating the computational role of cortical recurrent connections.