Fig. 3: Diverse responsiveness of neuronal populations to negative and positive RPEs.

a Schematic of the different response profiles of local neuronal populations (top), classified into four categories: Positive RPE (pRPE, red), negative RPE (nRPE, blue), unsigned RPE (uRPE, green), and signed RPE (sRPE, gold; see main text for details). For each category, the coding strategy could be regular (i.e., increasing/decreasing HFA power with increasing positive/negative RPE magnitude) or inverted (i.e., decreasing/increasing HFA power with increasing positive/negative RPE magnitude). uRPE populations are labeled by whether they increase or decrease activity regardless of valence. Note that classical RL (gold shade) and current asymmetric RL models (purple shade) only account for regular sRPE, pRPE, and nRPE responsiveness (i.e., partially asymmetric coding). In contrast, unsigned RPEs (i.e., RPE salience) and inverted coding strategies (gray shade) emerge from fully asymmetric coding profiles absent from current theories. Populations responding to the four categories (colored stars) can be projected on a positive vs. negative RPE plane (bottom). Note how pRPE and nRPE units spread along the x and y axes (positive nRPE coefficients indicate increases in HFA with larger nRPE magnitude), while uRPE and sRPE units spread along the diagonal and off-diagonal. b Positive and negative RPE peak coefficients for responsive channels in dMPFC (top) and INS (bottom) belonging to each of the four categories, projected on the two-dimensional RPE plane. c Proportion of channels per participant and ROI falling within each category. Box plots depict median and interquartile range across participants (n = 10), with whiskers covering most extreme values except outliers. d Anatomical ___location of responsive channels colored per category (Top: dMPFC, bottom: INS). All channels were mirrored to the right hemisphere for dMPFC and left hemisphere for INS. Source data are provided as a Source Data file.