Saccade Reward Signals in Posterior Cingulate Cortex

Human subjects orient their eyes to regions of interest in the visual world with high contrast and information content. What remains poorly understood, however, is how we accomplish this remarkable feat and choose, among the plethora of stimuli in the visual environment, where to look (Figure A). To examine the neural basis of the decision about where to look, monkeys were trained to make eye movements to visual targets for varying amounts of fruit juice reward, while the activity of neurons in posterior cingulate cortex (CGp) was monitored. CGp is an area of brain that is anatomically connected with both oculomotor and reward-related brain regions and is therefore well situated to relate information about the intrinsic value of visual stimuli to the selection and generation of gaze shifts. Models of reinforcement learning theory posit that when animals (or people) choose between two alternate movements, the choice they make depends on the likelihood of reward and the predicted value of reward to be expected from each of these movements. Expected value is reflected in the weights (w1 and w2) of the neural circuits underlying the production of the alternate movements (Figure B). When the outcome of a particular movement differs from the expected value of that movement, a prediction error is generated which modifies the synaptic weights associated with the behavioral alternatives. We therefore reasoned that if CGp neurons were involved in the selection of rewarding eye movements in a manner suggested by reinforcement learning theory, we would expect to see differences in reward value reflected in the firing patterns of these neurons for the prediction and the outcome of the movement. Indeed, we have found that the activity of CGp neurons correlates with the size of reward to be expected from eye movements made into the response field of a neuron under study (Figure C). This modulation of CGp neuronal activity by reward size occurs following the eye movement (Figure C, left) and following the receipt of the juice reward (Figure C, right). Across the population of neurons under study, Figure D plots the proportion of CGp neurons with activity significantly modulated by the independent effects of reward size (black line) and the peak velocity, amplitude, and latency of the eye movement. When the effect of reward size is considered independently of parameters of the eye movement, CGp neurons are modulated by reward size most notably following movement and reward delivery, suggesting these neurons carry information about the expected reward value as well as the actual reward outcome of eye movements. A separate experiment in which predicted rewards were randomly omitted showed that these neurons carry signals correlated with errors in reward prediction (McCoy et al. 2003).