The Functional Brain Organization of an Individual Allows Prediction of Measures of Social Abilities Transdiagnostically in Autism and Attention-Deficit/Hyperactivity Disorder

Abstract

Background: Autism spectrum disorder and attention-deficit/hyperactivity disorder (ADHD) are associated with complex changes as revealed by functional magnetic resonance imaging. To date, neuroimaging-based models are not able to characterize individuals with sufficient sensitivity and specificity. Further, although evidence shows that ADHD traits occur in individuals with autism spectrum disorder, and autism spectrum disorder traits in individuals with ADHD, the neurofunctional basis of the overlap is undefined.

Methods: Using individuals from the Autism Brain Imaging Data Exchange and ADHD-200, we apply a data-driven, subject-level approach, connectome-based predictive modeling, to resting-state functional magnetic resonance imaging data to identify brain-behavior associations that are predictive of symptom severity. We examine cross-diagnostic commonalities and differences.

Results: Using leave-one-subject-out and split-half analyses, we define networks that predict Social Responsiveness Scale, Autism Diagnostic Observation Schedule, and ADHD Rating Scale scores and confirm that these networks generalize to novel subjects. Networks share minimal overlap of edges (<2%) but some common regions of high hubness (Brodmann areas 10, 11, and 21, cerebellum, and thalamus). Further, predicted Social Responsiveness Scale scores for individuals with ADHD are linked to ADHD symptoms, supporting the hypothesis that brain organization relevant to autism spectrum disorder severity shares a component associated with attention in ADHD. Predictive connections and high-hubness regions are found within a wide range of brain areas and across conventional networks.

Conclusions: An individual's functional connectivity profile contains information that supports dimensional, nonbinary classification in autism spectrum disorder and ADHD. Furthermore, we can determine disorder-specific and shared neurofunctional pathology using our method.

Keywords: ADHD; Autism spectrum disorder; Functional MRI; Functional connectivity; Magnetic resonance imaging; Predictive modeling.

Figure 1.. LOO cross-validation CPM results for SRS and ADOS sub-scale score.

(A.) LOO cross-validation CPM results for SRS sub-scale scores. For each SRS sub-scale (i.-vi.), the sum of the predicted SRS score from +ve and −ve models are plotted against known scores. (B.) As in (A.) for ADOS sub-scale scores. The linear regression and 95% confidence interval are shown in black/grey.

Figure 2.. SRS and ADOS split-half cross-validation and permutation testing results, and application to all individuals (ABIDE-I/II).

(A.) Correlation (R-value) for split-half cross-validation. (i.) For each SRS sub-scale, the bar in the left-most column is reproduced for reference from the LOO cross-validation CPM results reported in Figure 1. N=352/260. The middle two columns are from split-half train/test cross-validation CPMs (n=200 iterations). The final column shows the null results from permutation testing where subjects and scores are scrambled prior to LOO cross-validation (n=1,000). For all SRS sub-scales, train/test results are greater than null results (P<2E-144). (ii.) Same as (i.) for ADOS sub-scales (P<0.03). See Supplementary Figure 3.A.&4.A. for results from SRS&ADOS +ve/−ve feature sets. (B.) From each iteration of the split-half cross-validation, the model was applied to all individuals from ABIDE-I/II less those in the training group (N=632 training) to predict clinical scores. Across iterations (n=200) mean predicted scores are compared between TD and ASD individuals. For all sub-scales, predicted SRS (P<1E-07) scores are greater for ASD than TD individuals (i.). Likewise, all but the severity ADOS sub-scale score was greater for ASD than TD individuals (P<0.02) (ii.). Between ASD and TD groups, motion (P>0.14), and age (P>0.96) were not different.

Figure 3.. Anatomy of SRS and ADOS sub-scale networks.

For SRS (A.) and ADOS (B.), edge overlap within (i./iii.) and between (ii./iv.) ten a priori atlas networks and our CPM networks are plotted for +ve (i./ii.) and −ve (iii./iv.) feature sets. Each layered plot shows the cumulative (sum) likelihood (1.0-Pvalue) estimated from the probability of edges being shared between a priori networks and each SRS/ADOS sub-scale network. Likelihoods greater than chance are indicated with an asterisk. Notice that in all plots, networks, and internetwork pairs, are ordered from greatest to least cumulative likelihood (i.e. the x-axis is ordered differently in each plot). Inlays show the edges of example SRS/ADOS +ve/−ve sub-scale networks as circle-plots as well as edges/nodes overlaid on glass brains.

Figure 4.. Results from ADHD/TD individuals from ADHD-200 data set.

As in Figure 1.A./B., (A.) LOO cross-validation CPM results for ADHD sub-scale scores. For each sub-scale (i.-iii.) the sum of the predicted ADHD score from the +ve/−ve models are plotted against known score. As in Figure 2.A., (B.i.) Correlation (R-value) of split-half cross-validation CPMs (n=200) for each ADHD sub-scale and null results from shuffled data (n=1,000). As in Supplementary Figure 1.A., (B.ii.) correlation matrix of ADHD behavior sub-scale scores. As with SRS and ADOS, ADHD sub-scale scores are highly correlated. As in Supplementary Figure 7.A.i./B.i., (C.) shows layer plots of the cumulative number of edges versus edge length for ADHD sub-scale networks. Networks with edge lengths which are not normally distributed are denoted by a cross (☨). For all not normally distributed networks, edges are skewed towards longer lengths. None of the networks are prone to outliers. There is a difference between +ve and −ve feature set edge lengths for all sub-scale networks (P<4E-03). As in Figure 3., (D.) ADHD edge overlap within (i./iii.) and between (ii./iv.) ten a priori atlas networks and ADHD networks are plotted for +ve (i./ii.) and −ve (iii./iv.) feature sets. Inlays show the edges of example sub-scale networks as circle-plots as well as edges/nodes overlaid on glass brains.

Figure 5.. Generalizability of composite networks and overlap of composite network edges: SRS and ADOS (+ve/−ve) & SRS and ADHD (+ve/−ve).

Plotted in (A.) and (B.) are correlations of predicted versus known behavior using composite networks applied across scales. All composite networks were thresholded at three. In (A.), the SRS (i.) and ADOS (ii.) composite networks were used to predict scores for individuals from ABIDE-I/II for whom only the other score was available (i.e. the SRS network was used to predict scores for individuals for whom ADOS scores (not SRS scores) were available). Composite networks were also applied across the ABIDE-I/II and ADHD-200 data sets. (B.) Predicted SRS scores correlate with known ADHD scores in individuals from the ADHD-200 data set. Layer plots showing the shared anatomy of SRS and ADOS (C.) and SRS and ADHD (D.) composite networks across thresholds. Composite network overlap of +ve and −ve feature sets was computed by taking the products: +ve/+ve (upper left, red/grey), +ve/−ve (upper right, purple/grey), −ve/+ve (lower left, purple/grey), and ve/−ve (lower right, blue/grey) of paired networks and computing the likelihood that each atlas network contribute the observed number of edges to each set of shared features.