Functional selectivity and specific connectivity of inhibitory neurons in primary visual cortex

Abstract

In the cerebral cortex, the selectivity of neurons for features of sensory stimuli arises through the interaction of excitatory and inhibitory synaptic inputs. Excitatory neurons receive inhibitory input that closely tracks excitation 1-4 , stabilizing network dynamics 5 while improving efficiency and robustness of the neural code 6-8 . However, how this balance of excitation and inhibition is achieved by cortical circuits is unclear, since inhibitory interneurons are thought to pool the inputs of nearby excitatory cells and provide them with non-specific inhibition proportional to the activity of the local network 9-13 . Here we show that although parvalbumin-expressing (PV) inhibitory cells make connections with the majority of nearby pyramidal cells, the strength of their synaptic connections is structured according to the similarity of the cells' responses. Individual PV cells strongly inhibit those pyramidal cells that provide them with strong excitation and share their visual selectivity. This fine-tuning of synaptic weights supports co-tuning of inhibitory and excitatory inputs onto individual pyramidal cells despite dense connectivity between inhibitory and excitatory neurons. Our results indicate that individual PV cells are preferentially integrated into subnetworks of inter-connected, co-tuned pyramidal cells, stabilising their recurrent dynamics. Conversely, weak but dense inhibitory connectivity between subnetworks is sufficient to support competition between them, de-correlating their output. We suggest that the history and structure of correlated firing adjusts the weights of both inhibitory and excitatory connections, supporting stable amplification and selective recruitment of cortical subnetworks.

Theory of Neuronal Perturbome: Linking Connectivity to Coding via Perturbations

“…We found that to obtain feature-specific suppression, strong and functionally-specific subnetworks of E and I were necessary. That is, both E and I neurons with similar receptive fields (RFs) should be connected together more strongly than their non-similar counterparts, which was consistent with recent results in visual cortex (Znamenskiy et al, 2018).Our modelling results shed light on the above mentioned controversy by showing that featurespecific amplification and suppression could both exist in the cortex, depending on the regime of functional similarity between the influencers and the influencees. Our model suggests specific predictions on how to observe this in the cortex.…”

A distinct class of bursting neurons with strong gamma synchronization and stimulus selectivity in monkey V1

“…Recent studies in rodents have began to genetically target distinct cell-types to study their precise contributions to sensory tuning and network oscillations. These studies support a role for inhibition in shaping orientation-and direction-selectivity of V1 neurons (Kerlin et al, 2010;Znamenskiy et al, 2018;Perrenoud et al, 2016). Furthermore, they have implied specific classes of inhibitory interneurons in the generation of V1 gamma oscillations in mice (Veit et al, 2017;Chen et al, 2017;Senzai et al, 2019;Perrenoud et al, 2016).…”

Excitatory and inhibitory subnetworks are equally selective during decision-making and emerge simultaneously during learning

“…Another recent study (Znamenskiy et al, 2018) suggested that targeted connectivity between 422 excitatory and inhibitory neurons allows for the existence of highly selective excitatory 423 subnetworks, while keeping the network stable. In circuits with non-selective inhibition, by 424 contrast, excitatory subnetworks had to be weakly selective in order to keep the network stable 425 (Znamenskiy et al, 2018). The permissiveness to highly selective excitatory subnetworks in 426 circuits with selective inhibition may be advantageous for situations that require precise 427 encoding of sensory stimuli for discrimination.…”