Rapid adaptation in visual cortex and its impact on population coding

Abstract

History dependent adaptation of visual signals in the brain has been theorized to support novelty detection by creating a memory trace for recently experienced stimuli. It has been previously shown that neurons along the ventral visual stream, a system thought to be important for novelty detection, undergo increasing adaptation. Yet, it is not known whether this enhanced adaptation yields an increased discriminability of stimulus representations. To determine how adaptation impacts stimulus coding along the visual hierarchy, we used two-photon calcium imaging to measure activity in layer 2/3 neurons of mouse V1 and ventral higher visual areas LM and LI in response to novel and repeated stimulus orientations. We find that the increasing degree of adaptation along the ventral visual stream increasingly repels the tuning preference of individual neurons away from the adapter. This increasing tuning bias decorrelates the population representation of neighboring orientations gratings, particularly in LM and LI. In addition, we find that adaptation reduces the trial-to-trial variability of the magnitude of responses in all areas. However, these effects on single neuron tuning and population coding do not significantly impact the performance of population vector decoder or support vector machine discriminating oriented gratings on a novelty detection task following adaptation. Despite the increasing adaptation along the hierarchy, decoding performance is similar across areas. Together, these results demonstrate that while adaptation along the ventral visual stream progressively alters single neuron tuning and population coding, particularly through decorrelation of neighboring orientation representations and reduction of response variability, these neural changes do not necessarily enhance decoding performance in a novelty detection task.