Conference on Cognitive Computational Neuroscience

The limbic system plays a vital role in the evaluation of emotional stimuli. Human single-cell recordings imply the existence of “affect detectors”, neurons in the limbic system that fire preferentially during the presentation of stimuli containing one kind of emotional information (e.g., positive or negative). Neurons of this kind would support reliable classification of novel stimuli by their emotional content. To test this, we trained multivariate classifiers on single-cell electrode recordings from the amygdala, hippocampus, anterior cingulate cortex, and ventromedial prefrontal cortex to discriminate positive- and negative-affect trials. Recordings were obtained from 14 patients while they viewed and judged the emotional expression of computer-generated faces. While prior univariate analyses of these data indicate neurons with affect-selective spike-profiles, the cross-validated accuracy of these models was at chance across subjects—even when models were trained only on neurons that were circularly-selected based on a univariate analysis. This inability to generalize prior neural activity is inconsistent with these neurons being affect detectors, suggesting a more complex representation of emotional information in the brain.