Neural state changes during movie watching relate to episodic memory in younger and older adults

Abstract

Event segmentation is a key feature underlying the ability to remember real-life occurrences. At the neural level, event boundaries have been shown to align with boundaries between neural states-stable patterns of brain activity maintained over time. These neural states provide a valuable window into the neural underpinnings of event perception. To investigate how neural state boundaries relate to memory across the lifespan, we used the data-driven Greedy State Boundary Search method to implicitly identify neural state changes in younger and older adults' electroencephalography data during movie watching. Memory for the movie was tested and related to (1) neural state correspondence across individuals and (2) the degree to which the pattern of activity changes at boundaries. Neural state boundaries significantly aligned across people, but did not differ with age nor relate to memory. The degree of change at neural state boundaries also did not differ with age, but was positively related to memory for the movie. These findings suggest that age differences in the perception of naturalistic events may be less pronounced than previously thought, at least when measured implicitly, and that greater distinction between successive neural states relates to better memory for one's experiences regardless of age.

Keywords: EEG; aging; episodic memory; event segmentation; neural states.

Fig. 1

Neural pattern similarity with GSBS outputs for one example participant. (A) Time by time pattern similarity overlaid with GSBS boundary output for an example participant (older adult). Comparing the full matrix to the inset matrix reveals the occurrence of temporally nested states. (B) t-Distance plot from GSBS output for the same participant showing the t-distance metric for each iteration of the GSBS algorithm. The optimal selected number of states we identified is marked with a dotted line.

Fig. 2

Overview of EEG analyses. (1) Neural state boundaries determined separately for each individual. (2) Average state boundary location density determined for the group. Vertical lines represent the neural state changes from a random single participant. (3) Correlation between each individual and the group-level boundary density (in a leave-one-out manner) computed and adjusted for the maximum possible correlation (based on number of boundaries) in each individual. More positive values reflect greater alignment with the group. (4) Degree of change around state boundaries and non-boundaries computed for each individual by finding the average correlation for all time points in the 2 s before compared to 2 s after boundary and non-boundary timepoints. (5) Relative change around boundaries computed by subtracting the median similarity around non-boundary timepoints from the change around state boundaries. More negative values reflect a larger shift around boundaries compared to other time points.

Fig. 3

Memory performance predicted by between-subject state alignment and pattern similarity around state boundaries. Proportion of internal details calculated as the number of internal details (reflecting some aspect of the movie) divided by the total number of details provided during free recall. (A) Between-subject boundary alignment calculated as similarity to all other individuals’ neural state boundary placement adjusted for the maximum possible alignment for each individual. (B) Pattern similarity in the 2 s surrounding neural state boundaries adjusted for individual’s pattern similarity in the center of each neural state. Shaded areas represent 95% CI.