Genomic and Behavioral Analysis of Autism Spectrum Disorders Based on Different Brain Imaging Modalities

Background:

Autism spectrum disorder (ASD) is a complex condition that can be defined as a group of diverse neurodevelopmental disorders. ASD is identified by early conditions which change information processing of the nervous system and impact the communication and social development of the patients. In addition to environmental factors, many genes of various pathways can be represented as varied risk factors for developing ASD.

Objectives: To merge genomic and behavioral data with multi modal MRI using Big Data Techniques

Methods: 200 individuals (80 with ASD and 120 controls) from the NDAR database were selected as having both T1-weighted MRI and single nucleotide polymorphism (SNP) genotyping. Of these, 74 also had available resting-state functional MRI (rs-fMRI) time series. Structural MRI scans were processed from which two vector-valued shape descriptors were derived. A “global” descriptor was obtained by approximating the surface with an 81st-order SPHARM model and calculating the power spectrum from the spherical harmonic coefficients. A vector of “local” descriptors was built by aligning each surface with a map of select Brodmann areas, 25 per hemisphere, and taking the average Gaussian curvature of the surface within each region. The pattern of activity from Rs-fMRI scans within each independent component was scored as consistent with ASD using a fuzzy, neural network-based classifier, and these scores form a vector of functional connectivity descriptors. All SNP genotyping had been performed using the Illumina HumanOmni 2.5-8 whole-genome kit. The extracted descriptors from both sMRI and fMRI were used as phenotypic information in the genomic linkage analysis performed using PLINK software to identify genomic variants (SNP Cclls from GenomeStudio software) significantly associated with each of the three descriptor vectors. Significant SNPs were identified by rs# in dbSNP, and any associated genes were cross-referenced with Uniprot, GeneCards, and the SFARI database

Results:

Using a false discovery rate (FDR) of 0.1, twelve SNPs were found to be significantly associated with regional brain curvature on sMRI, including four intron variants, two coding variants, and six within intergenic DNA. Of the SNPs most strongly associated with global brain shape, with FDR = 0.1, five were intronic and two intergenic. The top 5 SNPs linked with ASD-related functional connectivity, with uncorrected p < 10⁻⁵, were intronic or intergenic. However, none of these was significant at FDR = 0.1. The sample size being much lower than for sMRI analyses. Implicated genes include several associated with vesicles/vesicle transport (AP1G2, STON2, SYTL1), including synaptic vesicles, with metal ion transport including calcium ions (HCN1, SLC12A8, STIM1), and with embryonic development (STOX2) and neural migration in particular (ASTN2). Other affected genes may have a membrane or chromatin regulatory function (MDGA2, NXPE2/NXPE4, UHRF1) or are non-protein coding (LINC01800, MIR99AHG). Only two of the genes identified, ASTN2 and HCN1, are currently listed in the SFARI database.

Conclusions: In summary, distinct novel genes or traditional autism risk genes may be identified differently by Rs-fMRI, or local vs global descriptors of sMRI that map, along with behaviors, to RDoC defined neural circuits relevant to ASD.