Personalized Intrinsic Network Topography Mapping and Functional Connectivity Deficits in Autism Spectrum Disorder

Abstract

Background: Recent advances in techniques using functional magnetic resonance imaging data demonstrate individually specific variation in brain architecture in healthy individuals. To our knowledge, the effects of individually specific variation in complex brain disorders have not been previously reported.

Methods: We developed a novel approach (Personalized Intrinsic Network Topography, PINT) for localizing individually specific resting-state networks using conventional resting-state functional magnetic resonance imaging scans. Using cross-sectional data from participants with autism spectrum disorder (ASD; n = 393) and typically developing (TD) control participants (n = 496) across 15 sites, we tested: 1) effect of diagnosis and age on the variability of intrinsic network locations and 2) whether prior findings of functional connectivity differences in persons with ASD compared with TD persons remain after PINT application.

Results: We found greater variability in the spatial locations of resting-state networks within individuals with ASD compared with those in TD individuals. For TD persons, variability decreased from childhood into adulthood and increased in late life, following a U-shaped pattern that was not present in those with ASD. Comparison of intrinsic connectivity between groups revealed that the application of PINT decreased the number of hypoconnected regions in ASD.

Conclusions: Our results provide a new framework for measuring altered brain functioning in neurodevelopmental disorders that may have implications for tracking developmental course, phenotypic heterogeneity, and ultimately treatment response. We underscore the importance of accounting for individual variation in the study of complex brain disorders.

Keywords: Autism spectrum disorder; Child development; Connectome; Functional magnetic resonance imaging; Individuality; Resting state.

Figure 1

Schematic of Personalized Intrinsic Network Topography (PINT) Method. A) PINT starts with a template set of regions of interest (ROIs) selected from the Yeo et al. (11) atlas. B) Template (input) and C) Personalized (output) correlation matrices from a representative subject. D) PINT starts by calculating average mean timeseries from circular ROIs (of 6mm radius) around 80 central “template” vertices. In this depiction, we zoom in on an ROI from the FP network. The correlation of a vertex timeseries (dashed black line) is calculated with the FP network timeseries (orange line) after the timeseries of the five other networks are regressed from both. Then, the center of the ROI is moved to the vertex of maximal partial correlation (the direction of movement is show using the black arrow). E) Once all 80 ROIs have been moved, the network timeseries are updated and the algorithm repeats around the new vertex locations F) The algorithm iterates until all 80 ROIs are centered about a vertex of highest partial correlation (represented by the black circle). Abbreviations: DA = Dorsal Attention Network, DM = Default Mode Network, FP =Fronto-Parietal Network, VA = Ventral Attention Network, SM = Sensory Motor Network, VI = Visual Network. (Note: the Limbic Network was not included because it contains many areas of high fMRI signal susceptibility.)

Figure 2

Test-Retest and Longitudinal Stability of PINT (ABIDE Longitudinal and CoRR datasets). A) Measurement of within-subject and cross-subject distance for a personalized Fronto-Parietal Network (FP) ROI is depicted for two representative longitudinal participants. The orange outline represents the Fronto-Parietal Network as defined by the Yeo et al. atlas (11). “Template” vertex locations are shown in black for reference. B) The locations of PINT “personalized” vertices show consistency over time. Grey violin plots show the distribution of cross-subject distances (averaging across 80 ROIs) for the UCLA site (left, n=14) and UPSM site (right, n=17). Box plot and points, for the cross-subject measure, show the mean distance of each subject’s baseline scan to all other subjects’ follow-up scan. This measure is compared with the distance of each subject’s baseline scan to their own follow-up scan (within subject distance). C) This effect replicated in Test-Retest and Longitudinal comparisons in three samples of TD individuals obtained from CoRR (left to right) NYU_2 test-retest, n=135 scanned same session; NYU_2 longitudinal, n=53 scanned ~6 months apart); Utah_1 test retest; n= 23, scanned same session; Utah_2 longitudinal, n= 16, scanned ~2 years apart; UPSM longitudinal, n=44, scanned ~2 years apart.

Figure 3

A) Medial right hemisphere view of diagnosis effect on personalized ROIs locations in the Ventral Attention (VA) Network. The probability map of the personalized ROI location (i.e. one where the value at each vertex represents the proportion of participants in the sample, pooling ASD and TD, whose 6mm ROI emcompases that location) is plotted in grayscale. The template vertex locations are plotted as black dots. The overlaid colours show areas where a significant effect of diagnosis was observed (warm colors TD > ASD; cool colours ASD > TD). B) Lateral left hemisphere view of age effect on personalized ROIs locations in the Default Mode (DM) Network . The probability map of the personalized ROI location (all participants 30 and under) is plotted in grayscale. The template vertex locations are plotted as the black dot. The overlaid colours show areas where a significant effect of age group was observed (warm colors: children > adults; cool colours: adults > children). Children were defined as those under the age of 12 (n=183); adults were defined as between the ages of 18–30 (inclusive, n=251).

Figure 4

A) Locations (left) and chord diagram of network composition (right) for the 214 edges showing significant hypo-connectivity in ASD calculated from the Template ROIs (before PINT). B) After PINT, the number of significant hypoconnected edges (blue) is reduced and four hyper-connectivity connected (orange) edges are seen. For the chord diagrams, the width of the chord represents the number of significant edges and color of the chord represents the mean effect size for those significant edges.