A biophysically-detailed model of inter-areal interactions in cortical sensory processing

Mechanisms of top-down modulation in sensory perception and their relation to underlying connectivity are not completely understood. We present here a biophysically-detailed computational model of two interconnected cortical areas, representing the first steps in a cortical processing hierarchy, as a tool for potential discovery. The model integrates a large body of data from rodent primary somatosensory cortex and reproduces biological features across multiple scales: from a handful of ion channels defining a diversity of electrical types in hundreds of thousands of morphologically detailed neurons, to local and long-range networks mediated by hundreds of millions of synapses. Notably, long-range connectivity in the model incorporates target lamination patterns associated with feed-forward and feedback pathways. We use the model to study the impact of inter-areal interactions on sensory processing. First, we exhibit a cortico-cortical loop between the two model areas (X and Y), wherein sensory input to area X produces a response with two components in time, the first driven by the stimulus and the second by feedback from area Y. We perform a structural and functional characterization of this loop, finding a differential impact of layer-specific pathways in the feed-forward and feedback directions. Second, we explore stimulus discrimination by presenting four different spatially-segregate stimulus patterns. We observe well-defined temporal sequences of functional cell assembly activation, with stimulus specificity in early but not late assemblies in area X, i.e., in the stimulus-driven component of the response but not in the feedback-driven component. We also find the earliest assembly in area Y to be specific to pairs of patterns, consistent with the topography of connections. Finally, we examine the integration of bottom-up and top-down signals. When presenting a second stimulus coincident with the feedback-driven component, we observe an approximate linear superposition of responses. We find the implied lack of interaction consistent with the naive connectivity in the model and the absence of plasticity mechanisms that would underlie the learning of top-down influences. This work represents a first step in the study of inter-areal interactions with biophysically-detailed simulations.