Patterned perturbation of inhibition can reveal the dynamical structure of neural processing

Abstract

Perturbation of neuronal activity is key to understanding the brain’s functional properties, however, intervention studies typically perturb neurons in a nonspecific manner. Recent optogenetics techniques have enabled patterned perturbations, in which specific patterns of activity can be invoked in identified target neurons to reveal more specific cortical function. Here, we argue that patterned perturbation of neurons is in fact necessary to reveal the specific dynamics of inhibitory stabilization, emerging in cortical networks with strong excitatory and inhibitory functional subnetworks, as recently reported in mouse visual cortex. We propose a specific perturbative signature of these networks and investigate how this can be measured under different experimental conditions. Functionally, rapid spontaneous transitions between selective ensembles of neurons emerge in such networks, consistent with experimental results. Our study outlines the dynamical and functional properties of feature-specific inhibitory-stabilized networks, and suggests experimental protocols that can be used to detect them in the intact cortex.

How many neurons are sufficient for perception of cortical activity?

“…The matched suppression that we observed in the local L2/3 network is in accordance with the general net inhibitory effect of pyramidal neuron stimulation observed in vivo ( Chettih and Harvey, 2019 ; Kwan and Dan, 2012 ; Mateo et al, 2011 ; Russell et al, 2019 ) and in detailed network models of cortex ( Cai et al, 2020 ). This supports the idea that such networks operate in an inhibition-stabilised regime where one role of inhibition is to control strong recurrent excitation ( Denève and Machens, 2016 ; Murphy and Miller, 2009 ; Ozeki et al, 2009 ; Pehlevan and Sompolinsky, 2014 ; Sanzeni et al, 2020 ; Tsodyks et al, 1997 ; van Vreeswijk and Sompolinsky, 1996 ; Wolf et al, 2014 ), although since our perturbations are not targeted to inhibitory neurons with specific tuning we cannot assess how functionally specific this architecture might be ( Sadeh and Clopath, 2020 ). However, these results also seemingly challenge recent work demonstrating that activation of co-tuned ensembles in V1 predominantly activates other similarly tuned neurons in the surrounding network ( Carrillo-Reid et al, 2019 ; Marshel et al, 2019 ) and that ablation of some neurons within functional sub-groups reduces activity in the spared neurons ( Peron et al, 2020 ).…”

“…Indeed, the hypothesis that brains use sparse, distributed activity patterns is supported computationally ( Kanerva, 1993 ; Olshausen and Field, 1996 ), energetically ( Attwell and Laughlin, 2001 ; Lennie, 2003 ; Schölvinck et al, 2008 ), and experimentally ( Barth and Poulet, 2012 ; Olshausen and Field, 2004 ; Wolfe et al, 2010 ). A major factor thought to govern such sparse coding is neuronal inhibition ( Haider and McCormick, 2009 ; Isaacson and Scanziani, 2011 ) which serves to balance and control recurrent excitation ( Denève and Machens, 2016 ; Haider et al, 2013 ; Murphy and Miller, 2009 ; Packer and Yuste, 2011 ; Pehlevan and Sompolinsky, 2014 ; Sadeh and Clopath, 2020 ; Tsodyks et al, 1997 ; van Vreeswijk and Sompolinsky, 1996 ; Wehr and Zador, 2003 ; Wolf et al, 2014 ) and shape neuronal output ( Borg-Graham et al, 1998 ; Cardin et al, 2010 ; Isaacson and Scanziani, 2011 ; Lee et al, 2012 ; Wilson et al, 2012 ). Two key questions are therefore: (1) What is the lower bound of activity that can be behaviourally salient?…”

Training recurrent spiking neural networks in strong coupling regime

“…The initial network was set up to be an inhibition-stabilized network (ISN), where strong recurrent excitatory connections were stabilized by inhibitory feedback without which the network would exhibit run-away excitation (Tsodyks et al, 1997;Sanzeni et al, 2019;Sadeh and Clopath, 2020). An ISN exhibits a paradoxical phenomenon where an external stimulus that excites inhibitory neurons decreases their firing rates.…”

“…The reduction in recurrent excitation is larger than the external stimulus resulting in a net decrease in excitatory input to inhibitory neurons. Such large reduction in recurrent excitation occurs because strongly connected excitatory neurons respond to changes in input with large gain (Tsodyks et al, 1997;Sanzeni et al, 2019;Sadeh and Clopath, 2020).…”

Dynamics of a Large-Scale Spiking Neural Network with Quadratic Integrate-and-Fire Neurons

“…The decision-making process depended on the system tending towards one of the attractors for the two alternative choices separated by an unstable saddle in the decision space. Besides, different networks in visual processing [13][14][15][16], neural disease [17][18][19], and movement [20] have also been researched by mean-field dynamics.…”