The Dynamical Regime of Sensory Cortex: Stable Dynamics around a Single Stimulus-Tuned Attractor Account for Patterns of Noise Variability

Abstract

SummaryCorrelated variability in cortical activity is ubiquitously quenched following stimulus onset, in a stimulus-dependent manner. These modulations have been attributed to circuit dynamics involving either multiple stable states (“attractors”) or chaotic activity. Here we show that a qualitatively different dynamical regime, involving fluctuations about a single, stimulus-driven attractor in a loosely balanced excitatory-inhibitory network (the stochastic “stabilized supralinear network”), best explains these modulations. Given the supralinear input/output functions of cortical neurons, increased stimulus drive strengthens effective network connectivity. This shifts the balance from interactions that amplify variability to suppressive inhibitory feedback, quenching correlated variability around more strongly driven steady states. Comparing to previously published and original data analyses, we show that this mechanism, unlike previous proposals, uniquely accounts for the spatial patterns and fast temporal dynamics of variability suppression. Specifying the cortical operating regime is key to understanding the computations underlying perception.

Task-evoked activity quenches neural correlations and variability across cortical areas

“…We first show empirically that task-evoked activity suppresses neural correlations and variability across large cortical areas in two highly distinct neural data sets: NHP spiking and human fMRI data (Figure 1). This confirms previous findings showing quenched neural variability during task states in both NHPs and humans (Churchland et al, 2010;He, 2011He, , 2013Hennequin et al, 2018) , while going beyond those previous studies to report globally quenched inter-area task-state neural correlations. In particular, we focused on neural variability and correlation changes across large cortical areas in our electrophysiology data set (rather than between pairs of neurons) given our focus on large-scale neural interactions, and to facilitate a comparison between different correlation approaches (FC in fMRI data and spike count correlation in electrophysiology data).…”

“…We found that variability was highest when there was no stimulation, while variability decreased for any type of evoked stimulation (e.g., negative or positive input amplitudes). Despite the model's simplicity, these findings are consistent with our (and others') empirical and model results demonstrating that task states quench time series variability in both human and animal data (Churchland et al, 2010;He, 2011;Hennequin et al, 2018) . Our minimal modeling approach directly links descriptive statistics (e.g., time series variability) with rigorous dynamical systems analysis (e.g., attractor dynamics).…”

“…In contrast, during task-evoked states, the output dynamics of the sigmoid transfer function are reduced, which correspond to evoked states (e.g., cortical "Up" states) that exhibit quenched variability (Harris and Thiele, 2011;Renart et al, 2010) . The quenching of output variability can be explained by different biological mechanisms, such as clustered excitatory connectivity in local circuits, tightening of E-I balance due to inhibitory activation, neural adaptation, and/or irregular synaptic vesicle release (Doiron et al, 2016;Hennequin et al, 2018;Litwin-Kumar and Doiron, 2012;Rosenbaum et al, 2012) . The manifestation of these biological mechanisms can be summarized at the mean-field by the reduction of the response variability due to the decreased slope in the sigmoid transfer function during highly active or inactive states.…”

“…Neural variability is quenched during task-evoked states in a neural mass model Here we rigorously ground the intuition that task-evoked activity reduces output variability using neural mass modeling and dynamical systems theory. A recent study provided evidence that an evoked stimulus drives neural populations in sensory cortex around a stable fixed point attractor (Hennequin et al, 2018) . We first extended these findings using a simplified neural mass model, which allows for a comprehensive dynamical systems analysis that is mathematically difficult in higher dimensions.…”

“…Measures of neural correlations and variability are widely used in neuroscience to characterize neural processes. During task states, neural variability has consistently been shown to be reduced during tasks across human functional magnetic resonance imaging (fMRI) (He, 2011(He, , 2013 , local neural populations (Churchland et al, 2010;Hennequin et al, 2018;Jacobs et al, 2018) , and both spiking (Hennequin et al, 2018;Litwin-Kumar and Doiron, 2012) and mean-field rate models (Deco and Hugues, 2012;Ponce-alvarez et al, 2015) . Despite this convergence in the neural variability literature, there are disparities in the use and interpretation of neural correlations.…”