Supplementary Figure 1: Behavioral measures of motor function, motivation, and arousal do not change appreciably with associative learning
From: Frequency-specific hippocampal-prefrontal interactions during associative learning
In all panels in this figure, behavioral metrics were calculated in identical sliding trial windows (width = 10%, step = 0.5% of each total session length) on the same set of sessions (except for e), are plotted in a similar relative scale (~10× average s.d. across trial windows), and are assayed for changes with learning using the same statistical test (2-sided permutation test on means of early vs. late learning stage [first vs. last third of trials]). (a) Learning performance. Across-session (all sessions meeting learning criteria; n = 61) mean ± s.d. of logit-transformed percent of correct trials, plotted as a function of the percentile of each session’s trials. Performance robustly increases across trials (P ≤ 10−4). This difference is not due to restricting analysis to sessions with successful learning, as it remains significant when all sessions (n = 87) are included (P ≤ 10−4; dashed curves). Note that this is the same data plotted in main text Fig. 1c, but with learning curves pooled (averaged) across all four associations in each session, to match the number of observations for other data in this figure. Also plotted is the across-session mean ± s.d. (“MTS” to right of main plot) performance for an identity match-to-sample control task (i.e., matching an object to itself, rather than to a learned associate), which was significantly better than for the associative learning task (P ≤ 10−4). (b) Reaction time. Across-session (n = 61) mean ± s.d. of log-transformed reaction times to response targets. This metric—which may reflect both motor preparatory and motivational factors—does not change with learning (P = 0.49). Note this null result is likely due in part to the enforced delay in our task between choice object onset and response (cf. Fig. 1b), though the fact that match-to-sample task reaction times are significantly faster (P = 0.008) indicates that reliable reaction time modulations are possible with this task structure. (c) Saccade traces from a typical session. Eye position is plotted for each trial in the early, middle, and late learning stages (top to bottom) to left (green) and right (red) targets, for -130–130 ms relative to saccade onset. Dashed circles indicate fixation and saccade windows. Scale bar indicates 1 degree of visual angle. (d) Saccadic endpoint variability. Across-session (n = 61) mean ± s.d. of the variability of saccade endpoints (across-trial standard deviation of position 30–130 ms post-saccade, when the eyes were typically stable on the target), for vertical (top) and horizontal (bottom) dimensions and saccades to left (green) and right (red) targets. Though saccades do become significantly more variable with learning (all P < 2×10−4), the magnitude of this change is rather small (compare maximum increase of ~0.1 deg. to 1 deg. scale bar in panel c; see also panel g). (e) Pupil size. Across-session (n = 61) mean ± s.d. of pupil diameter during delay period (100–850 ms after start of delay), when pupil size is least influenced by external factors. Within each session, pupil size is expressed as a z-score relative to the fixation period mean and s.d. This metric—which is strongly linked to global arousal—does not significantly change with learning (P = 0.07), but is significantly decreased for the match-to-sample task (P ≤ 10−4). (f) Lip EMG. Across-session mean ± s.d. of lip EMG during outcome feedback period (100–1350 ms after outcome feedback), normalized by its mean value for each session. EMG was obtained from two animals performing a working memory–guided saccade task (purple; 4 sessions) or a visuomotor associative learning task (yellow; 6 sessions). Lip EMG—a proxy for reward-related orofacial movements—also shows little change with learning for either the working memory (P = 0.1) or learning (P = 0.43) tasks. (g) Summary of behavioral results. To compare behavioral changes across all reported metrics, relatively independent of the number of observations, we calculated a d′ statistic between the early and late learning stages: |meanearly–meanlate|/SDpooled. These results reiterate that across-trial changes in motor behavior, motivation, and arousal are relatively minor compared with learning-related changes in performance.
