Fig. 6: Neurons reverse selectivity congruent with choice or reward-value.
From: Dynamic perceptual feature selectivity in primary somatosensory cortex upon reversal learning
a Average calcium signals in hits, misses, CRs, and FAs for two neurons that reversed selectivity during reversal learning. Neuron 1, responses remained high for Hit and FA trials throughout all phases. Neuron 2, responses decreased for hits and misses and increased for FA and CR trials during the naïve reversal phase, and then increased again for hits and misses in the reversal expert phase. b Example of the ROC analysis for calculating the choice index (CI) of a neuron. The calculation is similar to the DI except that calcium signals were sorted by the animal’s licking behavior (lick vs. no-lick trials). Top, probability distributions of the calcium signals for the lick and no-lick trials. Bottom, ROC curve from which the CI is calculated. The CI expresses the likelihood that neuronal calcium signals predicted the upcoming lick. Statistical significance was determined using a permutation test (Methods section). c Sub-classification of neurons that reversed selectivity during reversal learning based on changes in choice index (CI). Choice neurons (purple, N = 24), CI was significant and remained of the same sign throughout all phases (*). Value neurons (orange, N = 35 neurons), CI was either not significant or of the opposite sign in post-reversal naïve sessions (#). d Average calcium signals upon texture presentation for each neuronal class when separated by hit trials with low (light) and high (dark) whisking or licking rates. Calcium signals were neither influenced by whisking rates (Paired t-tests; choice: N = 24, P = 0.87, value: N = 35, P = 0.81) nor licking rates (choice, P = 0.67, value, P = 0.77). Error bars represent SEM. e Table showing the percentages of each category within the different functional classes. Notably, none of the choice and value neurons was categorized as predictive for licking (number of neurons per class for which corresponding categories were defined prior training: non-selective 271, Value 11, Choice 14, Texture 17, Gained 90, and Lost 75). 0% is represented in blue in the look up table.
