A novel, spatially-specific reward signal in the macaque visual cortex

Stimulus-reward relationships have been shown to modify activity within sensory cortex (Serences 2008, Franko et al. 2010). One problem interpreting these studies is the inherent difficulty disentangling reward and attentional effects (Maunsell 2004). A recent study, however, attempted to disambiguate these two explanations by temporally separating stimulus and reward (Pleger et al. 2008). Pleger et al. found reward-level contingent modulations upon reward delivery but not during stimulus discrimination. These results argue against an attentional explanation, as there is no modulation in the stimulus phase when attention would be beneficial. In light of these results, we employed contrast-enhanced event-related monkey fMRI to monitor whole-brain functional responses in a stimulus-reward paradigm. We used a passive fixation task and a 2 by 2 design with factors abstract visual cue and juice reward. Importantly this design included cue-reward trials (to establish and maintain a stimulus-reward relationship) and reward-only trials (to test for reward modulations in the absence of a stimulus). We found that fMRI activity in visual cortex during the reward-only trials was significantly reduced and this reduction co-localized the cue representation. To quantify this we calculated the spatial correlation of stimulus activity (as determined by an independent localizer experiment) and reward-only activity and found significant correlations in visual areas V3, V4, and TEO. To investigate this negative reward signal (NRS) further we manipulated the expected value of two distinct visual cues. Initially one cue was given a high expected value (EV) (66% reward trials) while a second cue had a relatively low EV (33% reward trials). After an initial training and scanning period this cue-value relationship was inverted and a second scanning period was started. Within a given cue’s representation we found a larger NRS when the cue had a higher EV. We then compared the time-course of the NRS within the two cue representations and found that the largest differences were time-locked to changes in the cue-value relationships indicating a potential role for the NRS in the learning of new cue-value relationships. These results demonstrate a novel negative reward signal that specifically modulates fMRI activity within the representation of reward predicting cues and which is sensitive to EV, especially EV changes. Interestingly, this cue-specific modulation in visual cortex is negative in sign, indicating it is of a different nature then previously documented attention/anticipation effects that can occur in the absence of a stimulus.status: publishe