BRAIN project: Circuitry Underlying Response Summation: Theory and Experiment

National Institute Of Mental Health

Investigators

Linked publications, trials & patents

Abstract

We have obtained data showing that inhibitory stabilized network (ISN) models describe several areas of the cortex. This means that recurrent coupling is strong enough to make the excitatory neurons unstable in the absence of reactive inhibition. Together with the Brunel group, we used inference in several simple rate-based models to provide further evidence that inhibition-stabilized-network (ISN) models can describe the data. An ISN model captures responses to different stimulation strengths and the dynamics of responses simultaneously. The data and the model together provide strong evidence that cortical responses are strongly affected by recurrent excitatory-inhibitory interconnections; the data cannot be explained by intra-laminar feedforward or multi-inhibitory-population disinhibitory models. This work places important constraints on the possible set of network models that can describe cortical activity. We have also investigated, with the Brunel lab, how networks of spiking neurons respond to inputs, and characterized their transfer function (i.e., how the population firing rate depends on the mean inputs to the network). This was done in two classes of models, current-based and conductance-based. Current theories make distinct predictions about the shape of this transfer function. The balanced network model predicts a linear transfer function, while the stabilized supralinear network model (SSN) can generate various types of nonlinearities. We investigated numerically and analytically the response of networks of current-based spiking neurons as a function of coupling strength. We showed that, while a linear transfer function is obtained in the strong coupling (balanced network) limit, nonlinearities appear as the coupling is decreased. We found that these nonlinearities are present both at response onset and at saturation, and systematically characterized how they are shaped by single neuron properties and network connectivity. All SSN nonlinearities can be generated, but other regimes are also present. Work in future project periods will experimentally seek to determine the operating regime used in biological cortical networks. We have also examined the universality of the balanced state model for cortical function, a foundational model for the function of cortical networks (for review, see Ahmadian and Miller, 2019). First proposed to explain the irregular firing seen in cortical circuits, the balanced state theoretical work was performed in binary neurons (i.e., those that can fire only at rates zero or one), and then extended to current-based spiking networks. Current-based neurons are simplified versions of real neurons, and in current-based models, synaptic strengths are not dependent on the membrane potential of the neuron as they are in real neurons. Conductance-based network models take this feature of real neurons into account. We showed (Sanzeni et al., 2020) that conductance-based models fundamentally cannot be described by the balanced network theory. The synaptic scaling rule that defines the balanced state, which is that average synaptic strength varies as 1/sqrt(N), where N is the total number of inputs to a cell, is not possible in conductance-based networks. We show mathematically that this rule and any possible modification of the rule that preserves balance produces constant, regular firing rates, violating experimental observations in the cortex. Balanced states can still exist, but they exist in only well-defined regimes of synaptic strength and number of inputs. Continuing intramural work in the laboratory seeks to determine whether this tuned balanced regime is pervasive in many cortical areas, and if there are learning rules that maintain this state. Other collaborative work within this project has progressed in this reporting year. We have compared input-output functions (computations) in visual cortex of several species, showing that optogenetic perturbations in both species produce large heterogeneity in cortical responses. Continuing work aims to determine the possible circuit sources of the observed differences in responses when external input is applied to a subset of cells. The goal of this project has been to determine what kinds of models describe cortical networks in vivo. We have made several discoveries, including that the so-called strong coupling theoretical regime applies to many cortical areas, and that network phenomena shape the input-output functions of cortical neurons. The progress made here positions us to understand, using models that make testable predictions, how biological neurons embedded in their networks compute by transforming inputs to outputs. References: Sanzeni A, Histed MH, Brunel N. Emergence of irregular activity in networks of strongly coupled conductance-based neurons. bioRxiv, 2020 Sep. Available from: http://biorxiv.org/lookup/doi/10.1101/2020.09.24.312579 Ahmadian Y, Miller KD. What is the dynamical regime of cerebral cortex? arXiv:190810101 q-bio Internet. 2019 Aug 28. Available from: http://arxiv.org/abs/1908.10101