DECIPHERING THE ASSOCIATION BETWEEN POLYGENIC RISK FOR SCHIZOPHRENIA AND HIPPOCAMPAL FUNCTION

Background In recent years, Polygenic Risk Profile Scores (PRS's) are being increasingly used to indirectly measure the aggregate genetic effect of many weakly associated markers on the risk for a neuropsychiatric disorder. While the use of PRS's has supported the polygenic architecture of schizophrenia, a link between PRS's and the pathophysiology of this disorder is still missing. In this study, we analyze the association between PRS's for schizophrenia and hippocampal function during a simple declarative memory task in healthy volunteers. We then used regularized regression, a machine learning method, to further investigate the genetic structure that drives the association between schizophrenia PRS's and altered hippocampal engagement. Methods Two hundred and five healthy volunteers underwent BOLD fMRI (3 T) during a Simple Declarative Memory Task (SDMT), which included incidental encoding and retrieval of visual scenes. For both the encoding and retrieval sessions, the scenes were presented in a blocked fashion, with 4 blocks of neutral scenes and 4 blocks of visual scenes alternating with 9 blocks of resting state (fixation cross hair). In the current study, we focused on the encoding phase for neutral scenes only (Rasetti et al. 2014). PRS's were calculated for each individual as the weighted sum of the number of reference alleles on preselected markers based on PGC GWAS study (PGC 2014), with p-value threshold at 0.05 (24,670 SNPs). Imaging space association study was conducted using multiple regressions in SPM8. Region of interest (ROI) was defined as averaged signal within a 6 mm sphere around the peak of association between PRS and right hippocampus. Regularized regression was run using ROI in R using glmnet package (Zou et al. 2005). Results The PRS constructed with 24,670 SNPs (with association with schizophrenia at scoring threshold of p=0.05) is significantly associated with hippocampal activation (z=4.48, p=0.004, FWE corrected in bilateral hippocampal ROI), so that higher PRS corresponds to blunted hippocampal activation. Using the regularized regression with shrinkage factor lambda=0.001, we identified 255 SNPs driving this association, so that the association between the PRS based on these 255 SNPs (PRS255) and hippocampal activation increases dramatically (z=7.14, p Discussion Our results show that PRS's for schizophrenia are significantly associated with altered hippocampal function during a simple declarative memory task even in healthy volunteers. Using regularized regression, we are able to find a small subset of SNPs that drive the association and account for high percentage of variance in hippocampal activation. The use of machine learning techniques for dimension reduction allows disentangling the cumulative effect of risk genes on brain function.