Abstract
Cortical responses to repeated presentations of a stimulus are variable and this variability is correlated between cells. Theoretical work has shown that correlated variability can strongly affect the ability of neuronal populations to encode sensory information, although the impact depends critically on both the structure of correlation and the way in which responses are decoded. Ultimately, however, the impact of correlated variability in a local network rests in how it affects downstream networks. To evaluate how correlated variability propagates through the visual system, we have recorded simultaneously from a population of neurons in macaque primary visual cortex (V1, using implanted arrays of 100 microelectrodes) and their downstream targets in the input layers of area V2. We measured responses to repeated presentations of drifting gratings and evaluated how well we could predict trial-to-trial fluctuations in V2 responsiveness by monitoring population activity in V1. We fit a generalized linear model (GLM) to a subset of the trials, and tested its ability to predict responses on novel data. We found we could predict a substantial portion of trial-to-trial fluctuations in V2 by monitoring V1 responses, with a performance level that was similar to our ability to predict a V1 neuron's response by monitoring its nearby neighbors; that is, it was as if V1 and V2 neurons were embedded in a single network with shared noise. The V1 neurons weighted most heavily in the GLM were those with spatial receptive fields most similar to that of the target V2 cell; relative orientation preference of the V1 and V2 cells played relatively little role. Our results suggest that a substantial portion of V2 variability can be explained by fluctuations in the response of V1. Cortical variability in V1 thus has a substantial impact on downstream networks.