(Two additional neurons showed significant firing after the NS but not after the DS, and these were not analyzed.) The difference in DS- and NS-evoked firing was not due to differences in ongoing locomotor behavior during cue excitation because firing also differed in trials in which the locomotor onset

latency was >500 ms; average post-DS firing was 16.1 ± 1.7 spikes/s and post-NS firing was 8.3 ± 1.2 spikes/s (p < 0.001, Wilcoxon test). Consistent with this observation, the onset and peak of the DS-evoked excitation preceded locomotor onset in the vast majority of trials (Figures 2D and S2). To determine whether post-DS firing GSK J4 was time locked to cue onset or to the onset of locomotion, we focused on a subset of correct DS trials with >200 ms separation between cue onset, locomotion, and lever press (median of 21 trials selected per neuron; see Supplemental Experimental Procedures). Aligned ERK activity to cue onset, the greatest change in average firing rate was immediately after the cue (Figure 2D). In contrast,

these same data show little change in firing rate at the time of locomotion onset (Figure 2E) or in relation to lever press or receptacle entry (Figure S2). Consistent with previous reports (Nicola et al., 2004), DS-evoked firing was greater on trials in which an operant response was made (16.8 ± 1.8 spikes/s) compared to when it was absent (12.5 ± 1.7 spikes/s; p < 0.001, Wilcoxon test; n = 54 neurons recorded in sessions with at least one missed DS trial). Thus, because cue-evoked firing consistently preceded locomotor onset and was greater when a reward-seeking response was subsequently made, cue-evoked excitation could influence the initiation or maintenance of cued reward-seeking behavior. We next determined the relationship between cue-evoked firing and the subsequent

reward-seeking movement using a generalized linear model (GLM). We analyzed only the DS trials in which a lever press response was made so that the cue value and the Phosphatidylinositol diacylglycerol-lyase ultimate outcome were identical in every trial. First, we determined which aspects of locomotor behavior to test for a relationship with neural activity. Because the locomotor responses in this task can begin at any point in the behavioral chamber, these movements can be described by many different variables. To select an appropriate set of locomotor features, we first calculated a large and redundant set of locomotor variables for each trial ( Table S1). We then used principal components analysis and factor analysis (PCA/FA) to identify a small number of underlying factors that accounted for the majority (74.2%) of cross-trial variability among all of the locomotor variables ( Table S2; Supplemental Experimental Procedures).