Abstract
Recent studies on V1 report top-down modulation of input-driven responses of sensory neurons, implying that exogenous sensory drives and endogenous top-down drives jointly determine V1 responses. By measuring fMRI responses in conjunction with a classification task on ambiguous ring stimuli, we sought to understand how V1 carries out its encoding operation on afferent currents while being adaptively modulated by top-down currents associated with perceptual tasks. Population activity of V1, as in its raw eccentricity profiles, failed to resolve the threshold differences between the ring stimuli due to large moment-to-moment fluctuations. The analysis of variance indicated that stimulus-evoked responses explain only one-fifth of the total variance and fMRI responses were highly correlated among eccentricity-bins, implying that a substantial fraction of V1 responses fluctuate as a whole. This led us to decompose the raw fMRI responses into untuned and tuned components: average response across eccentricity-bins and residual responses from the average, respectively, the former varying only in time and the latter varying in both space and time. The tuned responses revealed the veridical encoding operation of V1 by readily distinguishing between the ring stimuli, which was impossible with the raw fMRI responses. In contrast, the untuned were correlated with two major aspects of choice behavior—inter-trial variability in response time and inter-subject variability in response bias. We propose that this cohabitation of stimulus and choice information in V1 indicates the presence of top-down exertion of gain modulation on the early processing stage by the high-tier stage that accumulates evidence for perceptual choices.