Inferring Psychological Traits Using Functional MRI Connectivity in Genetic Studies

Early magnetic resonance imaging (MRI) and functional magnetic resonance imaging (fMRI) studies in behavioral genetics sought associations between brain-based phenotypes and psychological traits of interest. While the... [ view full abstract ]

Early magnetic resonance imaging (MRI) and functional magnetic resonance imaging (fMRI) studies in behavioral genetics sought associations between brain-based phenotypes and psychological traits of interest. While the association patterns themselves were a step forward in the literature, integrating more complex fMRI procedures could improve our capability and utility to conduct gene-finding studies of cognitive/behavioral outcomes, even expanding beyond measured phenotypes. “Connectivity-Based Predictive Modeling” (CBPM) is an approach for creating brain-based measures of a psychological variable. In this approach, individual associations across the human functional connectome are used to construct a single score that predicts a behavioral trait. Once a score is estimated, other fMRI samples that did not measure the behavioral trait can use the brain as a proxy measure for that behavior. While these approaches are becoming popular in the MRI literature, they have not been used for genetic studies to date. Further, aspects of their functionality may be improved by using genetically informative samples. This study is the first to conduct a GWAS of a CBPM variable, validating the procedure for future use. In addition, we use twin data to improve the utility of this approach, possibly increasing the power to detect genetic effects. To this end, we use general intelligence as a practical example phenotype and a training sample 232 individual twin pairs from the Colorado Longitudinal Twin Study (LTS), a matched test sample of 199 twins from the LTS to test the properties of the out of sample predictions of the trained model, and a test sample ~8500 individuals from the UKBiobank to test the effectivieness of these phenotype techniques for GWAS. We found that the standard CBPM of intelligence was significantly predictive within the train sample and in the out of sample test set.  The CBPM estimated with only phenotypic information was not significantly heritable in the test set. Heritability of the CBPM was significantly improved by incorporating information from twin models into the training procedure. Finally, using a LD score, we found that our CBPM estimated in the UKBiobank was significantly genetically correlated with intelligence.  All effect sizes fell within the moderate range.  Further steps should be taken to improve model prediction, matching between samples and integeration of multimodal brain imaging data. This approach holds the possibility for integrating information from twin and fMRI studies into a broader statistical genetics literature.