Fig. 2: Neural population analysis using principal component analysis.

a We used principal component analysis (PCA) to assess neural representations of various decision-making relevant signals by comparing pairs of trial types that differentiate key stimuli (e.g., CS+ vs. CS-) or behaviors (e.g., seeking vs. not seeking, drinking vs. not drinking). b, c Four example neurons modulated their trial averaged firing rates dependent on trial type (baseline subtracted, firing rates scaled to maximum (bar), seeking vs. not-seeking state, see Fig. 3 for more information on seeking state). Trial type 1 consisted of data recorded when the animal was seeking alcohol and trial type 2 consisted of data recorded when the animal was not seeking alcohol (details of behavioral identification of seeking discussed below). Neurons 1 and 2 primarily responded to the CS+, though neuron 2 more clearly differentiated between trial type 1 and trial type 2. Neuron 3 responded to the CS+ and at the beginning of access on trial type 1. Neuron 4 differentiated trial types prior to the CS+ onset. d PCA reduced the four neuron firing rates to population signals (PCs) that are linear combinations of individual neuron firing patterns. The PC trajectory for trial type 1 is larger, indicating a more robust population signal. e Neurons sorted by PC linear combination coefficient (neuron identity matched in right and left panels). Firing responses before CS+ onset, during the CS+, and during access are visible (event timings same as b, but time frame longer in e). f Seven PCs were found to be stable across subsampling trials, which were used to control for differences in neuron yield across experimental groups and will be explored in the subsequent analyses. These selected PCs accounted for 38% of the overall variance in neural firing. Note that these stable PCs consisted of the first six PCs and PC 10.