Subtype Classification of Autism Spectrum Disorder Via Resting-State fMRI Reveals Distinct Brain Network Endophenotypes

Thursday, May 12, 2016: 11:30 AM-1:30 PM
Hall A (Baltimore Convention Center)
J. A. Richey1, K. M. Gates2, S. Lane2, A. D. Valdespino3, R. A. Müller4 and A. Di Martino5, (1)Virginia Tech, Blackbsurg, VA, (2)Psychology, University of North Carolina - Chapel Hill, Chapel Hill, NC, (3)Psychology, Virginia Tech, Blacksburg, VA, (4)San Diego State University, San Diego, CA, (5)NYU Child Study Center, New York, NY
Background:  Phenotypic heterogeneity has presented a significant obstacle to MRI/fMRI-based diagnostic classification because most optimization algorithms assume a single, mutually exclusive distinction between ASD and typically developing groups. However, extant literature very clearly indicates that autism is not a single clinical entity, but a manifestation of dozens or more likely hundreds of genetic and genomic disorders (Betancur, 2011). Lack of information about subgroups of ASD is a problem in both diagnostic and treatment domains, because effects of interest may only be observed in only a subset of cases, thereby reducing statistical power and obscuring mechanisms of change in currently available treatments. 

Objectives:  The primary objective of this project was to apply a novel group search algorithm (Group Iterative Multiple Model Estimation [GIMME]; Gates & Molenaar, 2013) to resting-state fMRI data to determine whether brain-based heterogeneity within ASD can actually be useful information, that facilitates the identification of subgroups whose brain network properties are similar. A second objective is to then identify whether ASD symptoms within subgroups also cluster together.

Methods: We evaluated resting-state fMRI from individuals with ASD (N=70) from the NYU site in the Autism Brain Imaging Data Exchange (ABIDE; DiMartino et al., 2014). Data were preprocessed in AFNI (motion corrected, censored/scrubbed[0.05 FD], lowpass filtered). We applied GIMME to timeseries data extracted from seven brain regions in the default mode network (DMN). GIMME identifies contemporaneous and temporally lagged paths that are (1) common to all members of the sample and (2) unique to subsets of cases.

Results:  Three distinctive subgroups of ASD emerged from our analysis of the DMN. Consistent with previous work on brain connectivity of DMN related to social-cognitive deficits (Lynch et al., 2013; Uddin et al., 2013), group 1 demonstrated hyperconnectivity of posterior cingulate, precuneus and left angular gyrus, perhaps relating to deficits in theory of mind and difficulty in episodic memory and self-related processing. Group 2 demonstrated increased connectivity between midline regions (dorsomedial PFC, rostral anterior cingulate), which has been previously linked to repetitive behaviors (Weng et al., 2010). Finally, group 3 demonstrated similar patterns to groups 1 and 2, but also enhanced connectivity of rostral anterior cingulate and posterior cingulate/precuneus. Preliminary analysis of phenotyping data indicates that groups differ on Vineland subscales (Deficits in Daily Living Skills domain: Grp 1; Deficits in Socialization domain: Grp 2).

Conclusions:  Effective connectivity maps at the individual level can be aggregated into verifiably homogeneous subgroups/communities, which share similar brain network properties. Analyses relating phenotyping data available through ABIDE are ongoing, and indicate that brain-derived subgroups meaningfully map onto symptoms.

Funding: This work was supported by R03 MH102651 “Data Mining for Autism Endophenotypes in a Large-Scale Resting State fMRI Repository.” PI: Richey.