Learning Naturalistic Temporal Structure in the Posterior Medial Network

Abstract

The posterior medial network is at the apex of a temporal integration hierarchy in the brain, integrating information over many seconds of viewing intact, but not scrambled, movies. This has been interpreted as an effect of temporal structure. Such structure in movies depends on preexisting event schemas, but temporal structure can also arise de novo from learning. Here, we examined the relative role of schema-consistent temporal structure and arbitrary but consistent temporal structure on the human posterior medial network. We tested whether, with repeated viewing, the network becomes engaged by scrambled movies with temporal structure. Replicating prior studies, activity in posterior medial regions was immediately locked to stimulus structure upon exposure to intact, but not scrambled, movies. However, for temporally structured scrambled movies, functional coupling within the network increased across stimulus repetitions, rising to the level of intact movies. Thus, temporal structure is a key determinant of network dynamics and function in the posterior medial network.

Figure 1.. Stimuli and design.

Subjects viewed 90-s clips from The Grand Budapest Hotel. Each of three clips was assigned to one of three conditions. Each clip was viewed six times. Numbers below each screenshot refer to a segment within each clip, with segments being continuous chunks 2.4–5.5s long. The Intact clip was viewed in its original format. The segments for the Scrambled-Fixed clip were randomly re-ordered, but viewed in the same scrambled order for all repetitions. The segments for the Scrambled-Random clip were also randomly re-ordered, but the segments were viewed in a different order for each repetition. Movie stills are pixelated in this figure for copyright reasons.

Figure 2.. Schematic of fMRI analyses.

(A) Intra-subject correlation consists of calculating the correlation between a single brain region’s activity timecourses for different repetitions of a given stimulus, within an individual. (B) Intersubject correlation consists of calculating the correlation between a single brain region’s activity timecourse for different individuals watching the same stimulus, for the same repetition of that stimulus. (C) Intra-subject functional correlation consists of calculating the correlation between different brain regions’ activity timecourses, within a given stimulus repetition, within an individual. (D) Inter-subject functional correlation consists of calculating the correlation between a brain region’s activity timecourse in one individual and a different brain region’s activity timecourse in different individuals, for a given stimulus repetition. For simplicity, only two brains are depicted for the inter-subject analyses, but these analyses involve comparing a given individual to the mean of all other individuals. Inter-subject correlation (B) and inter-subject functional correlation (D) identify components of brain activity dynamics that are shared between individuals, and thus related to the common stimulus viewed by different individuals. Inter-subject correlation (B) identifies such commonalities within a brain region, while inter-subject functional correlation (D) identifies commonalities shared between different brain regions. Intra-subject correlation (A) and intra-subject functional correlation (C) identify components of brain activity dynamics that could either be related to the stimulus being viewed or idiosyncratic to each individual. Thus, comparing inter- and intra-subject analyses offers insights into which components of activity are shared between people (inter-subject analyses) and which may be idiosyncratic to each individual (intra-subject analyses). See Materials and Methods for more details.

Figure 3.. Temporal dynamics in the precuneus are more consistent for intact vs. scrambled movies.

(A) The timecourses of BOLD activity in the precuneus for the first and last repetitions of each clip were extracted and correlated within subject. The mean intra-subject correlation was higher for the Intact clip compared to the Scrambled-Fixed and Scrambled-Random clips, which did not themselves differ. (B) Reliability was also assessed by correlating, for the first repetition, each subject’s timecourse of BOLD activity with the mean of all other subjects. The mean inter-subject correlation was higher for the Intact clip compared to the Scrambled-Fixed and Scrambled-Random clips, which did not themselves differ. Dots indicate individual subjects. Error bars are ± 1 SEM. ** p < .01, *** p < .001.

Figure 4.. Temporal structure enhances intra-subject functional correlation within the posterior medial network.

(A) The timecourse of BOLD activity in the precuneus was correlated within-subject with that of the hippocampus for the first and last repetition of each clip. Functional correlations did not change over repetitions of the Intact or Scrambled-Random clips, but more than doubled over repetitions of the Scrambled-Fixed clip, rising to the level of the Intact clip. A similar pattern of results was obtained for (B) the precuneus and angular gyrus and (C) the precuneus and posterior cingulate. Dots indicate individual subjects. Error bars are ± 1 SEM of the within-subject difference between repetitions. * p < .05, ** p < .01, *** p < .001.

Figure 5.. Changes in posterior medial network functional correlations from first to last repetition.

Graph visualization of intra-subject functional correlation between all pairs of posterior medial ROIs, including the results from Figure 4, as well as pairs that did not include precuneus. The edges depict the t value comparing intra-subject functional correlation in repetitions 1 vs. 6, with thicker lines indicating greater enhancement over time. Solid lines are statistically significant changes, whereas dashed lines are not statistically significant. The most consistent enhancement in intra-subject functional correlation was for the Scrambled-Fixed clip (all six edges were statistically significant), compared to the Intact clip (two statistically significant edges) and the Scrambled-Random clip (three statistically significant edges). PCC = posterior cingulate cortex, HPC = hippocampus, PCN = precuneus, ANG = angular gyrus.

Figure 6.. Functional correlation between the precuneus and early visual cortex.

Intra-subject functional coupling between precuneus and visual cortex did not increase over repetitions in any condition. Dots are individual subjects. Error bars are ± 1 SEM of the within-subject difference between repetitions.

Figure 7.. Changes in whole-brain precuneus (intra-subject) functional correlations from first to last repetition.

Contrast depicts regions that increase in functional coupling with the precuneus (purple) from the first to last repetition of each clip (p < .05 corrected). No regions showed enhanced intra-subject functional correlation for the Intact (top) and Scrambled-Random (bottom) clips. For the Scrambled-Fixed clip (middle), greater precuneus intrasubject functional correlation (in green) was observed in the hippocampus and angular gyrus (and a medial parietal region that overlapped with the seed). The black mask (right) indicates to which voxels this analysis was applied (where > 90% of subjects had coverage).