Fig. 3: Large mesolimbic dopamine manipulations drive value-like learning.
From: Mesolimbic dopamine adapts the rate of learning from action
a, Left: experimental design for VTA–DA stimulation (stim.) predicted by a 0.5-s light flash at the front of the behavioural chamber. Right: mean uncued NAc–DA reward responses versus VTA–DA stimulation responses (right, individuals in grey, mean in black) for mice that received either large, uncalibrated (Uncalib.) stimulation (top; 30 Hz, 12 mW for 500 ms) or stimulation calibrated to reward responses (bottom; 30 Hz, 1–3 mW for 150 ms). b, Left: jRCaMP1b NAc–DA cue responses across training for mice that received large stimulations (5× the size of reward responses; filled circles, n = 7) or calibrated (Calib.) stimulations (1× the size of reward responses; open circles, n = 10). Right: quantified and raw data for mean NAc–DA traces after 750 training trials (5 sessions) with uncalibrated (top, two-tailed signed rank versus baseline, *P = 0.02, n = 7) and calibrated (bottom, two-tailed signed rank versus baseline P = 0.8, n = 10) stimulation. c, Predicted dopamine cue responses for the experiment in a simulated with a TD value learning model (light pink) versus ACTR (light blue). d, Top: mean NAc–DA signal after seven sessions of uncalibrated stimulation training, on probe trials with omitted stimulation that were delivered 10% of the time, with quantification over training (bottom). e, Predicted dopamine responses by TD (top) and ACTR (bottom) models, for uncalibrated (bold line) and calibrated (dotted line) dopamine stimulation. All error bars denote ±s.e.m.
