Abstract
Extrastriate visual neurons show no firing rate change during a working memory (WM) task in the absence of sensory input, but both αβ oscillations and spike phase locking are enhanced, as is the gain of sensory responses. This lack of change in firing rate is at odds with many models of WM, or attentional modulation of sensory networks. In this article we devised a computational model in which this constellation of results can be accounted for via selective activation of inhibitory subnetworks by a top-down working memory signal. We confirmed the model prediction of selective inhibitory activation by segmenting cells in the experimental neural data into putative excitatory and inhibitory cells. We further found that this inhibitory activation plays a dual role in influencing excitatory cells: it both modulates the inhibitory tone of the network, which underlies the enhanced sensory gain, and also produces strong spike-phase entrainment to emergent network oscillations. Using a phase oscillator model we were able to show that inhibitory tone is principally modulated through inhibitory network gain saturation, while the phase-dependent efficacy of inhibitory currents drives the phase locking modulation. The dual contributions of the inhibitory subnetwork to oscillatory and non-oscillatory modulations of neural activity provides two distinct ways for WM to recruit sensory areas, and has relevance to theories of cortical communication.
Highlights
Our findings suggest that inhibitory subnetworks that are selectively activated by working memory (WM) play a dual role in influencing excitatory cells, modulating both the sensory gain and phase locking to network oscillations
WM modulates the timing of spikes relative to the αβ oscillation (SPL), the Phase aspect
Despite the lack of oscillatory input to the network, there are quasi-regular oscillatory fluctuations in the local field potentials (LFPs)-proxy, indicative of the emergent rhythmic behavior of the network, similar to prior computational studies involving reciprocal i-cell connections (Van Vreeswijk et al, 1994; Brunel and Hakim, 1999; Whittington et al, 2000; Brunel and Wang, 2001). This simple network architecture is able to generate oscillatory activity, which we can compare to the experimental LFP data under various combinations of top-down and bottomup input
Summary
Top-down signals modulate responses to incoming sensory stimuli (Desimone and Duncan, 1995; Humphreys et al, 1998; Lee et al, 2005; Mitchell et al, 2007; Fries, 2009, 2015; Churchland et al, 2010; Bosman et al, 2012; Vinck et al, 2013; van Kerkoerle et al, 2014; Womelsdorf and Everling, 2015; Engel et al, 2016; Michalareas et al, 2016; Moore and Zirnsak, 2017), and have been explored in computational models (Brunel and Wang, 2001; Ardid et al, 2007; Lakatos et al, 2008; Kopell et al, 2011; Lee et al, 2013; Kanashiro et al, 2017). The goal of this study is to provide insight into the neural mechanisms giving rise to this non-linear interaction between sensory stimulus and top-down signal: to explain how extrastriate responses to sensory stimuli are modulated, without changes in baseline firing, to account for WM- and attention-induced changes in spiking behavior Both attention and WM induce oscillations in local field potentials (LFPs) (Fries et al, 2008; Siegel et al, 2008; Gregoriou et al, 2009; Liebe et al, 2012; Daume et al, 2017) and the timing of spikes relative to these oscillations (Lee et al, 2005; Vinck et al, 2013; Bahmani et al, 2018; Drebitz et al, 2018; Fiebelkorn and Kastner, 2020). To achieve our goal, we developed a minimal dynamical model mimicking oscillatory behavior in the network and investigated how top-down and bottom-up signals interact to enhance representations of sensory information
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
