Shared endo-phenotypes of default mode dsfunction in attention deficit/hyperactivity disorder and autism spectrum disorder

Abstract

Categorical diagnoses from the Diagnostic and Statistical Manual of Mental Disorders (DSM) or International Classification of Diseases (ICD) manuals are increasingly found to be incongruent with emerging neuroscientific evidence that points towards shared neurobiological dysfunction underlying attention deficit/hyperactivity disorder and autism spectrum disorder. Using resting-state functional magnetic resonance imaging data, functional connectivity of the default mode network, the dorsal attention and salience network was studied in 1305 typically developing and diagnosed participants. A transdiagnostic hierarchical Bayesian modeling framework combining Indian Buffet Processes and Latent Dirichlet Allocation was proposed to address the urgent need for objective brain-derived measures that can acknowledge shared brain network dysfunction in both disorders. We identified three main variation factors characterized by distinct coupling patterns of the temporoparietal cortices in the default mode network with the dorsal attention and salience network. The brain-derived factors were demonstrated to effectively capture the underlying neural dysfunction shared in both disorders more accurately, and to enable more reliable diagnoses of neurobiological dysfunction. The brain-derived phenotypes alone allowed for a classification accuracy reflecting an underlying neuropathology of 67.33% (+/-3.07) in new individuals, which significantly outperformed the 46.73% (+/-3.97) accuracy of categorical diagnoses. Our results provide initial evidence that shared neural dysfunction in ADHD and ASD can be derived from conventional brain recordings in a data-led fashion. Our work is encouraging to pursue a translational endeavor to find and further study brain-derived phenotypes, which could potentially be used to improve clinical decision-making and optimize treatment in the future.

Fig. 1. Target network definitions.

The regions of interest (ROIs) used for all present analyses are rendered on the MNI standard brain with frontal, diagonal, and top views. a The four main default mode network (DMN) nodes are subdivided into 12 ROIs reflecting distinct subregions (dmPFC1–4, PMC1–4, left and right TPJ1–2)^–. b The DMN subregions are supplemented by nine ROIs for the dorsal attention network (DAN) and salience network (SN), drawn from previously published quantitative meta-analyses. The DAN was composed of the dorsolateral prefrontal cortex (dlPFC) and intraparietal sulcus (IPS) bilaterally. The SN included the anterior insula (AI), midcingulate cortex (MCC), and amygdala (AM) bilaterally. NeuroVault permanent link to all ROI definitions used in the present study: http://neurovault.org/collections/2216/

Fig. 2. Workflow.

a DMN, DAN, an SN network coupling was studied in a composite sample of 1,305 TD, ADHD, and ASD individuals taken from two multisite open-data repositories (ADHD-200 and ABIDE). b In a data-driven fashion, Indian Buffet Processes (IBP) automatically derived the number of hidden properties in the connectional fingerprints across participants without recourse to their clinical status. Automatic detection and weighing of shared and distinct unknown biological causes prompts its use in the identification of endo-phenotypes. c Latent Dirichlet Allocation (LDA) then inferred three overarching factors of underlying brain variation. Importantly, LDA allowed to derive hidden variability factors with mixed membership. Therefore, each participant’s connectional fingerprint was modeled to be simultaneously caused by multiple implicit neurobiological factors. d Each individual composition of the three neurobiological factors (representing distinct network-coupling profiles, lower section) was related to their respective clinical diagnoses (TD, ADHD, and ASD). In a preliminary analysis based on t-distributed stochastic neighbor embedding (t-SNE; ref.), biological subtypes can be identified from network connectivity patterns that are partly shared across TD, ADHD, and ASD

Fig. 3. Hidden properties in connectivity profiles.

Healthy (middle section in the columns), ADHD (upper section in the columns), and ASD (lower section in the columns) participants are compared with regard to the relative occurrence of each distinct hidden component (columns). Each hidden property resulted directly from the Indian Buffet Process and is depicted here with its occurrence (present versus not present) added up across all participants. These were automatically discovered in the whole-brain connectivity profiles without knowing to which of the three groups each participant belonged. Visibly, the identified connectivity features are dispersed across the participant groups. No single connectivity feature was exclusively associated with only one group

Fig. 4. Three neurobiological factors of variation with distinct connectivity patterns.

Bayesian inference allowed extracting a hierarchy of brain-defined subgroups, without access to the clinical diagnoses. Each of the three biological factors reflected a coherent pattern of resting-state connectivity between the default mode network (dmPFC-1/2/3/4, PMC-1/2/3/4, and bilateral TPJ-1/2), dorsal attention network (bilateral dlPFC and IPS), and salience network (bilateral AI, MCC, and AM). In each TD, ADHD, or ASD individual, the resting-state measurements of overall network-coupling patterns were driven by flexible recombinations of these three factors of connectivity variation. L/R left/right hemisphere

Fig. 5. Evaluation of predictability, robustness, and expressiveness of the transdiagnostic brain phenotypes for clinical validation.

Evaluating intra-subject predictions, the clinical usefulness of the measured network connectivity strengths (blue) was systematically evaluated against the discovered neurobiological endo-phenotypes (green). Violin plots are similar to box plots in showing the median (white point), quartiles (thick black lines), and outliers (below/above thin black whiskers), but also expose the probability densities of the data points (sideways shapes). a Classification performance (1.0 = all subjects correct, 0.33 = chance as red line) of predicting the original diagnosis groups (TD, ADHD, and ASD) versus the neurobiologically derived groups (indicated by the most important factor in each participant) based on the overall brain connectivity. The data-derived disease factors could be much better predicted in connectivity profiles from new, previously unseen participants (p < 0.0001). b Classification performance of predicting the original diagnosis groups based on connectivity profiles versus connectivity profiles and additional factor weights. Knowledge of the brain-derived disease factors much decreased the variance (concentration around medium), thus decreasing the uncertainty of each prediction for a given participant. c Group prediction performance from full connectivity profile versus exclusive knowledge of the brain-derived factor weights. Without direct access to the original brain connectivity measurements, three factor weights summarizing each subject were sufficient for non-inferior prediction (p = 0.47). The brain-imaging-derived phenotypes hence improved predictability, robustness, and expressiveness