Heritability of movie-evoked brain activity and connectivity
- Medical Scientist Training Program, Columbia University Irving Medical Center, United States
- New York State Psychiatric Institute, United States
- Department of Psychiatry, Columbia University Irving Medical Center, United States
Figures
Figure 1 with 2 supplements
Blood oxygen level-dependent (BOLD) time course similarity scales with genetic relatedness across the cortex.
(A) Group differences in average BOLD time course similarity (indexed by intersubject correlation, ISC) show that BOLD time course similarity is greater among dyads who are more genetically related (51 MZ dyads, 34 DZ dyads, 690 UR dyads). Here, top and bottom rows reflect data acquired on different days of data collection while subjects viewed largely non-overlapping sets of movie clips. (B) Group-average ISC values used to create the difference maps in A, plotted in order of average ISC across all subject pairs, show that group differences are most pronounced in parcels with medium to high ISC (shading = SEM).
Figure 1—figure supplement 1
Blood oxygen level-dependent (BOLD) time course similarity by group.
Cortical surfaces show the group-level intersubject correlation (ISC) maps used to create the group difference maps in Figure 1A.
Figure 1—figure supplement 2
Blood oxygen level-dependent (BOLD) time course similarity scales with genetic relatedness across the cortex.
Scatterplots show the same data as in Figure 1B for each group comparison (each dot is one of 400 Schaefer parcels), highlighting greater genetic similarity in parcels with medium-to-high intersubject correlation (ISC).
Figure 2 with 3 supplements
Blood oxygen level-dependent (BOLD) time courses are heritable across the cortex.
(A) Cortical surfaces show heritability of BOLD time courses parcellated using the Schaefer 400 atlas, controlling for age, gender, and head motion (mean h2 Day 1/Day 2=0.064±0.034/0.068±0.036). (B) Residuals after regressing parcel-level intersubject correlation (ISC) from parcel-level heritability show that BOLD time courses in auditory cortices are less heritable than would be expected based on ISC, whereas the opposite is true for lateral prefrontal and temporo-parieto-occipital junction parcels.
Figure 2—figure supplement 1
Blood oxygen level-dependent (BOLD) time courses are heritable across the cortex.
Dorsal and ventral views of the same surfaces shown in Figure 2.
Figure 2—figure supplement 2
Blood oxygen level-dependent (BOLD) time course heritability is largely unaffected by GSR.
Cortical surface show that global signal regression (GSR) mildly increased BOLD time course heritability (average Day 1 h2 with/without GSR = 0.064/0.060; Day 2: 0.068/0.061) and had almost no effect on its spatial pattern (With GSR/without GSR Spearman = 0.99, pBrainSMASH <0.001 on both Day 1 and Day 2).
Figure 2—figure supplement 3
Blood oxygen level-dependent (BOLD) time course heritability magnitudes and spatial patterns are consistent in smaller subsamples.
Scatter plots show that average BOLD time course heritability magnitudes (left) and spatial patterns in smaller subsamples of our data are consistent with those observed in the full sample (shading = standard deviation across 100 random subsamples at each percentage).
Figure 3 with 1 supplement
Blood oxygen level-dependent (BOLD) time course heritability is greater in slower frequency bands, especially for more associative parcels.
(A) Purple/yellow cortical surfaces (upper row) show unfiltered BOLD time course heritability (upper left is identical to Figure 2A) as well as the heritability of BOLD time courses filtered with five frequency bands, with greater heritability in slower bands for Day 1 data. Red/blue cortical surfaces show BOLD time course heritability residuals after regressing out parcel- and frequency-level differences in ISC (lower left is identical to Figure 2B), with greater residuals in slower frequencies and more associative parcels. (B) Scatter plot shows heritability averaged across the cortex for each frequency band (i.e. the averages of the upper row of surfaces in A; shading = jackknife SEM). (C) Scatter plot shows the difference in heritability between the slowest and fastest BOLD-sensitive frequency bands for each of the Schaefer 400 parcels plotted against parcel ranks from the Sydnor et al. sensorimotor-association hierarchy (higher = more associative). Least squares lines were added to highlight the positive relationships between average h2 and parcel ranks but note that these relationships were formally tested with Spearman correlations. (D-F) Same as A–C for Day 2 data.
Figure 3—figure supplement 1
Blood oxygen level-dependent (BOLD) time course heritability is greater in slower frequency bands, especially for more associative parcels (for uncensored data).
Same as Figure 3 using full time courses for each day (including 20 s rest and clip onset blocks).
Figure 4 with 4 supplements
Hyperalignment reduces blood oxygen level-dependent (BOLD) time course heritability.
(A) Cartoon illustrates the difference between shared cortical topographies and shared (topography-independent) information content. (B) Diagrams illustrate the inputs to response and connectivity hyperalignment (RHA and CHA, respectively) using the Schaefer 100 atlas. RHA topographies were learned using BOLD time course data from the other day’s movie-watching scans, while CHA topographies were learned from vertex-level functional connectivity (FC) profiles (i.e. correlations between one vertex’s BOLD time course and the average time course from each of the 99 other parcels) calculated from the other day’s resting state scans. (C) Vertex-level BOLD time course heritability is highest for data aligned via MSM (multimodal surface matching) and lower for data hyperaligned within 100 Schaefer atlas parcels using both response hyperalignment (RHA) and connectivity hyperalignment (CHA). (D) Differences between the MSM-only and hyperaligned heritability maps shown in (C) are distributed across the cortex but are most apparent in visual areas. (E) BOLD time course heritability decreases as a function of hyperalignment parcel size according to a power law (purple and orange lines); each dot corresponds to average cortex-wide heritability for data hyperaligned using one of the 10 Schaefer atlas resolutions (shading = jackknife SEM).
Figure 4—figure supplement 1
Blood oxygen level-dependent (BOLD) time course heritability is resolution-dependent.
Scatter plots show that average BOLD time course heritability across all parcels is higher for coarser Schaefer parcellation resolutions (shading = SEM).
Figure 4—figure supplement 2
Hyperalignment increases intersubject correlation (ISC).
Scatter plots show that, as expected, hyperalignment increases ISC (quantified as the median ISC value across all subject pairs and vertices), and that this increase is greater for response (vs. connectivity) hyperalignment and for coarser (vs. finer-grained parcellation resolutions).
Figure 4—figure supplement 3
Hyperalignment reduces blood oxygen level-dependent (BOLD) time course heritability.
Dorsal and ventral views of the surfaces in Figure 4C–D.
Figure 4—figure supplement 4
Linear, quadratic, and logarithmic fits of average blood oxygen level-dependent (BOLD) time course heritability and hyperalignment resolution.
Same as Figure 4E for non-power law models of the relationship between hyperalignment resolution and heritability.
Figure 5 with 2 supplements
Controlling for neural timescale (NT) reduces heritability of blood oxygen level-dependent (BOLD) time courses.
(A) Bar plots show average pairwise differences in cortex-wide NT across monozygotic (MZ), dizygotic (DZ), and unrelated (UR) dyads on both days of data collection. (B) Cortical surfaces show decreases in BOLD time course heritability after NTs calculated from the other day of data collection were included as covariates in the multidimensional heritability analyses for multimodal surface matching (MSM)-aligned and response hyperalignment (RHA)-aligned (using the Schaefer 100 parcellation) data, most prominently in mid-level auditory and visual regions. These maps are thresholded at Δh2 = ±0.01 to aid comparisons of MSM- and RHA-aligned results. The maximum differences in h2 after controlling for NTs were −0.025 for MSM-aligned data and −0.007 for RHA-aligned data, respectively.
Figure 5—figure supplement 1
Subject pairs with longer neural timescales (NTs) have more correlated blood oxygen level-dependent (BOLD) time courses.
Surface plots show that pairwise intersubject correlation (ISC) values from one day of data collection scale with summed NTs from the other day’s data, especially in auditory and visual cortices (n=15,753 unique dyads; 54% of cortical vertices significant at false discovery rate (FDR)-corrected pperm <0.05 on both days).
Figure 5—figure supplement 2
Blood oxygen level-dependent (BOLD) time course heritability with (bottom) and without (top) controlling for neural timescale (NT).
Surface plots show the heritability maps used to generate the difference maps in the top row of Figure 5B.
Figure 6 with 1 supplement
Functional connectivity (FC) profile similarity scales with genetic relatedness across the cortex.
(A) Group differences in average FC profile similarity show that FC profiles are more similar for dyads who are more genetically related (51 MZ dyads, 34 DZ dyads, 690 UR dyads). (B) Group-average FC profile similarity values used to create the difference maps in A, plotted in order of average FC profile similarity across all subject pairs (shading = SEM).
Figure 6—figure supplement 1
Movie-watching functional connectivity (FC) profile similarity by group.
Same as S1 for FC profile similarity during movie-watching.
Figure 7 with 1 supplement
Functional connectivity (FC) profiles are heritable across network combinations.
Heatmaps show heritability of FC profiles for all unique within- and between-network combinations of the 17 Kong networks after controlling for age, gender, and head motion. FC profiles during movie-watching (left column) were more heritable than resting state FC profiles (middle column) for more sensory-oriented networks (red rows in the right column).
Figure 7—figure supplement 1
Functional connectivity (FC) strengths are similarly heritable during movie-watching and resting states.
Same as Figure 3 but for FC strength (vs. profile) heritability. FC strengths were largely heritable during movie-watching (Day 1 mean h2-SOLAR=0.42±0.13, Day 2 mean h2-SOLAR=0.41±0.11; 84% of network combinations significant on both days at false discovery rate (FDR)-corrected pperm <0.05), but the cross-day reliability of the FC heritability patterns across network combinations was about half that of the FC profile analysis (Spearman ρ=0.39, pperm <0.001), and no movie FC strength heritability values were significantly greater than rest FC values on both days for any network combination (right column).
Figure 8 with 3 supplements
Hyperalignment reduces functional connectivity (FC) profile heritability.
(A) Heatmaps show decreased FC profile heritability for most combinations of 17 Kong networks following response hyperalignment (RHA) (left) and connectivity hyperalignment (CHA) (right) compared to the multimodal surface matching (MSM)-only baseline. (B) Scatter plots show that hyperalignment, especially with RHA, decreases FC profile heritability according to a power law function; each dot corresponds to average cortex-wide heritability for data hyperaligned using one of the 10 Schaefer atlas resolutions (or MSM-only alignment, shading = jackknife SEM).
Figure 8—figure supplement 1
Response (but not connectivity) hyperalignment decreases functional connectivity (FC) strength heritability.
(A–B) Same as Figure 5 but for FC strength (vs. profile) heritability. Response hyperalignment (RHA) using the Schaefer 100 atlas decreased average FC strength heritability across all network combinations by 24% (95% CI: 12–35%) on Day 1 and by 28% (15–42%) on Day 2. Although connectivity hyperalignment (CHA) at the 100-parcel resolution lowered FC strength heritability on both days, these decreases were not statistically significant (Day 1: 15% [-10–39%], Day 2: 10% [-13–23%]).
Figure 8—figure supplement 2
Linear, quadratic, and logarithmic fits of average functional connectivity (FC) profile heritability and hyperalignment resolution.
Same as Figure 4B for non-power law models of the relationship between hyperalignment resolution and FC profile heritability.
Figure 8—figure supplement 3
Linear, quadratic, and logarithmic fits of average functional connectivity (FC) strength heritability and hyperalignment resolution.
Same as Figure 4B for non-power law models of the relationship between hyperalignment resolution and FC strength heritability.
Author response image 1
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Heritability of movie-evoked brain activity and connectivity
eLife 14:RP106081.
https://doi.org/10.7554/eLife.106081.3
