Identifying enhancer properties associated with genetic risk for complex traits using regulome-wide association studies

Abstract

Genetic risk for complex traits is strongly enriched in non-coding genomic regions involved in gene regulation, especially enhancers. However, we lack adequate tools to connect the characteristics of these disruptions to genetic risk. Here, we propose RWAS (Regulome Wide Association Study), a new application of the MAGMA software package to identify the characteristics of enhancers that contribute to genetic risk for disease. RWAS involves three steps: (i) assign genotyped SNPs to cell type- or tissue-specific regulatory features (e.g., enhancers); (ii) test associations of each regulatory feature with a trait of interest for which genome-wide association study (GWAS) summary statistics are available; (iii) perform enhancer-set enrichment analyses to identify quantitative or categorical features of regulatory elements that are associated with the trait. These steps are implemented as a novel application of MAGMA, a tool originally developed for gene-based GWAS analyses. Applying RWAS to interrogate genetic risk for schizophrenia, we discovered a class of risk-associated AT-rich enhancers that are active in the developing brain and harbor binding sites for multiple transcription factors with neurodevelopmental functions. RWAS utilizes open-source software, and we provide a comprehensive collection of annotations for tissue-specific enhancer locations and features, including their evolutionary conservation, AT content, and co-localization with binding sites for hundreds of TFs. RWAS will enable researchers to characterize properties of regulatory elements associated with any trait of interest for which GWAS summary statistics are available.

Fig 1. RWAS workflow overview.

In brief, the RWAS workflow involves annotating SNPs to enhancers and other regulatory regions (rather than genes). Enhancer-level summary statistics are computed for input into association testing. Then, we use the MAGMA linear modeling framework to compute genetic associations between supplied enhancer-level covariates and these enhancer-based GWAS summary statistics. This approach relies on high-quality enhancer annotations for the tissue of interest that capture genetic risk for the disorder. To ensure these conditions were met, we first thoroughly characterized a set of brain-specific enhancers and demonstrated that these enhancers capture genetic risk for schizophrenia.

Fig 2. Genome-level Jaccard similarity matrix demonstrates age- and experimental model- specific enhancer patterning.

Fetal brain and neurosphere samples cluster together when the tree is cut at the second level, while adult brain samples, ESC-derived clusters, and astrocytes form separate groups. Color denotes Jaccard similarity statistic. Groupings determined using hierarchical clustering.

Fig 3. Genetic risk for schizophrenia is enriched in ChromHMM-derived brain enhancers.

A) Partitioned heritability of enhancer annotations by tissue in schizophrenia. Brain enhancers are enriched for heritability in schizophrenia compared to other tissues. B) Partitioned heritability of individual brain samples. Adult brain enhancers had the most significant enrichment, followed by fetal brain enhancers.

Fig 4. Identification of brain-expressed enhancers associated with genetic risk for schizophrenia.

A) Enhancer-based GWAS allows the aggregation of non-coding SNPs into nearby enhancer regions and captures risk loci as enhancer-level risk associations. Individual points on the top panel denote SNPs, while the lines on the bottom panel represent enhancer regions. B) Brain enhancers capture risk loci missed by enhancers from other tissues. Enhancers from primary T cells captured fewer genome-wide significant signals when compared to enhancers from the inferior temporal lobe of the adult brain. The light gray shaded areas denote loci where a given enhancer annotation has more genome-wide significant enhancers when compared to the other annotation.

Fig 5. Features of brain-expressed enhancers associated with schizophrenia risk.

A) Risk-associated enhancers are in physical contact with gene sets previously implicated in risk for neuropsychiatric disorders. Shown here are gene sets with a median p-value across the 10 brain enhancer samples < 0.05. GWAS = genes at GWAS risk loci. Full description of gene lists available in Methods. B) Evolutionary conservation and AT-richness of enhancers were associated with schizophrenia risk. HGE/HAR status and distance to the nearest gene TSS were not associated with risk.

Fig 6. TFs with AT rich motifs are overrepresented in risk-associated enhancers and have neurodevelopmental functions.

A) Positive association between the AT-richness of a given TF binding site motif and the effect size in the RWAS model. B) Higher median motif AT percentage of a given TF is positively associated with the TF being annotated to the Gene Ontology term “cell morphogenesis during neuron differentiation”. C) TFs with higher median Z-score in the RWAS analysis are more likely to be annotated to “cell morphogenesis during neuron differentiation.” Grey dashed line is the median value of the background set of all TFs in our dataset. D) Cell type-specific expression in the prenatal human brain for TFs that recognize positively associated motifs in the schizophrenia RWAS analysis. The displayed TFs recognize a motif with an RWAS Z-score > 3 in a brain enhancer. Each cell is colored by expression Z-scores averaged across the specified cell type.