APPLYING MULTIVARIATE TECHNIQUES, INCLUDING PARALLEL ICA TO COMMON COMPLEX PSYCHIATRIC ENDOPHENOTYPES

Abstract

IntroductionEndophenotypes for common, complex psychiatric and medical disorders are often underpinned by multiple risk genes contributing individually to small amounts of disease risk via interlocking molecular/biological processes. However, current genotyping strategies for inherited phenotypes are mostly predicated on univariate analyses such as GWAS that require very large samples. Multivariate approaches such as parallel ICA (p-ICA) are statistically efficient, and enable analysis of samples in the N=100’s to low 1000’s (Pearlson et al. Front. Genetics Sept 2015, 6:276). This presentation will review application of the p-ICA approach to several such data sets, with both neuroimaging and behavioral endophenotypes derived from the Brain and Alcohol Research in College Students (BARCS) and Bipolar Schizophrenia Network on Intermediate Phenotypes (BSNIP-1) studies. MethodsFor each separate analysis, SNP’s, from Illumina 1M and/or 2.5M chips were QC checked for e.g. excess missing rate, MAF, H-W equilibrium and corrected for population stratification, and entered into a parallel-ICA fusion analysis along with quantitative measures of the relevant endophenotype, e.g. multiple impulsivity measures, resting state functional MRI or EEG data. Associations between the two types of data were made based on inferring a correlation across subjects (e.g. subjects with certain linear combinations of SNPs also tend to show particular linear combinations of voxels). SNP components emerging from this analysis were then entered into functional annotation tools e.g. IPA, DAVID, to determine associations with particular biological classifications to identify underlying biological themes. FindingsTo provide 1 example, in the BSNIP-1 psychosis resting state EEG data set we determined overlap in gene ontology process, process networks and pathway maps across 3 different electrophysiological phenotypes (resting EEG, auditory oddball P300 and auditory paired stimulus processing P50 ERPs, and resting state). The intersection of gene clusters obtained from three independent multivariate genetic associations of different EEG phenotypes derived from were entered into GeneGo’s enrichment analysis and significance (i.e. probability of over-representation of common genes in the data relative to GeneGo’s database by chance) estimated. Processes implicated in common among the 3 endophenotypes included neuron differentiation, neuron projection guidance, axon guidance, neuron projection morphogenesis, interneuron development, neuron differentiation, neurogenesis, synaptogenesis, and cell adhesion/synaptic contact, implicating CNS neurodevelopment. SummaryMany complex psychiatric diseases and complex quantitative traits are driven by multiple genes of individually small effect. Univariate strategies require very large sample sizes and rarely illuminate underlying molecular biological pathways. Multivariate approaches such as p-ICA render some of these problems tractable.