Orthogonalization of spontaneous and stimulus-driven activity by hierarchical neocortical areal network in primates

Understanding how biological neural networks perform reliable information processing in the presence of intensive spontaneous activity (1-3) is an essential question in biological computation. Stimulus-evoked and spontaneous activities show orthogonal (dissimilar) patterns in the primary visual cortex (V1) of mice (4-6), which is likely to be beneficial for separating sensory signals from internally generated noise (5, 7-12); however, those in V1 of carnivores and primates show highly similar patterns (3, 13-17). Consequently, the mechanism of segregation of stimulus information and internally generated noise in carnivores and primates remain unclear. To address this question, we used primate-optimized functional imaging (18) to precisely compare spontaneous and stimulus-evoked activities in multiple areas of the marmoset visual cortical pathway (19). In marmoset V1, but not in mouse V1, spontaneous and stimulus-evoked activity shared similar activity patterns, suggesting a function of spontaneous activity specific to mammals with functional columns. However, in the higher-order visual areas of marmosets, spontaneous and stimulus-evoked activities exhibited dissimilar patterns. Analysis of neural activity geometry further revealed progressive orthogonalization of the two types of activities along the cortical hierarchy in marmosets, which reached a level comparable to that of mouse V1. Thus, orthogonalization of spontaneous and stimulus-evoked activity is a general principle of cortical computation, which, in primates, is implemented by the hierarchical areal network.