Fig. 4: Pupil dilation and brief facial movements are indicative of sequence statistics.

A During widefield imaging, the right facial area was recorded under infrared illumination, including the eye and the whisker pad. Facial motion was mostly driven by the whisker pad on the mouse’s snout, here given as average motion per pixel. B The pupil diameter (ø) was tracked using SLEAP (top¹¹⁴, while facial motion components were analyzed by evaluating the different components underlying frame-to-frame changes in the image (bottom⁴⁷. For the latter, we selected the singular value decomposition component with the highest variance when aligned to the stimulus onsets. This component showed fast positive transients time-locked to the occurrence of sounds. Pupil diameter (ø) changed more slowly overall with some intermittent transients, which we attributed largely to noise, e.g., by whiskers moving in front of the eyes or pixel jitter in the pupil estimate. C Pupil diameter increased significantly around the time of omission (p = 0.002, Kruskal–Wallis test, downsampled from 50 Hz to 6.25 Hz, baselined to a linear extrapolation of the pre-omission diameter. D Facial motion energy was closely time-locked to each sound’s onset at a latency of ~65 ms and lasted only for ~100 ms. FME is aligned to sound onset at t = 0 preceded by at least 5 sounds (to avoid influence of preceding sound omissions), averaged over all stimulus presentations from all animals (N = 320 per animal). Error hull indicates 1 SEM across animals in all panels. E FME dropped during an omission, but subsequent sounds elicited substantially greater movement, compared to the pre-omission sounds (p = 0.0047, Wilcoxon rank sum test). F FME in response to tones after an omission showed a significant decrease for positions later in the sound sequence (p < 0.001, Kruskal–Wallis test). G FME increased significantly with the number of stimuli preceding the omission (p < 0.001, Kruskal–Wallis test, difference between the time ranges indicated by gray bars in the left panel), i.e., reduced more during the omission and increased more after it for longer pre-omission stimulus sequences (left). Colors correspond to different positions after omission (see right panel). H Analogous to the adaptation of facial motion (F), neural tone responses in PR (blue) and ORR (red) decreased significantly with the number of standards (p < 0.001, Kruskal–Wallis test). The ORR showed a stronger adaptation than the PR (−57% and −23%, respectively). Source data are provided with this paper.