Most human conditions develop in genetically susceptible individuals from the interaction with environmental risk factors. These complex disorders result from the summation of effects from multiple genetic risk loci. Genome-wide association studies (GWASes) measure the association of single nucleotide polymorphisms (SNPs) with traits or conditions, and allow the creation of individualised polygenic risk scores. However, these explain only a small portion of a condition’s genetic heritability. Further, there is evidence that schizophrenia GWAS signals are enriched within genomic regulatory blocks, which are clusters of conserved non-coding elements that span key developmental loci and function as long-range enhancers activating transcription of target developmental genes. This suggests that enhancer-based annotations might be useful to refine polygenic signals for schizophrenia.

In this work, I aimed to increase the amount of variance explained by PRS for schizophrenia, and a comparison condition hypertrophic cardiomyopathy, using tissue-specific regulatory enhancer-promoter annotations. To do so, I developed neural- and cardiac-specific enhancer lists, which I tested for enrichment, respectively, in schizophrenia and hypertrophic cardiomyopathy (HCM) heritability. I found that neural-specific enhancers are highly enriched in schizophrenia heritability -- especially when overlapping genomic regulatory blocks. Then I created partitioned polygenic risk scores for enhancer-based and non-enhancer-based SNPs, where enhancer-based SNPs are prioritised. I further compared the amount of adjusted heritability for both conditions explained by original GWAS vs partitioned polygenic risk scores, and found up to a 6.5% increase in the Coefficient of Determination for schizophrenia, and similar amounts for HCM -- however, this was not statistically significant. The increasing trend was specific for brain-expressed enhancers in schizophrenia, while it was widespread for HCM. Finally, I considered whether neural-specific enhancer-based partitions might be better modelled in GWAS using nonadditive effects, however my results were inconclusive due to small sample sizes.