Combining functional annotation and multi-trait fine-mapping methods improves fine-mapping resolution at glycaemic trait loci

The Meta-Analysis of Glucose and Insulin-related traits Consortium (MAGIC) identified 242 loci associated with glycaemic traits fasting insulin (FI), fasting glucose (FG), 2 h-Glucose (2hGlu), and glycated haemoglobin (HbA1c). However, for the majority, the causal variant(s) remain(s) unknown. Modelling multiple traits and integrating functional annotations have each been shown to improve fine-mapping resolution. Here, we aimed to determine whether combining these techniques would further improve fine-mapping resolution. Using single-trait fine-mapping results from FINEMAP as input, we performed multi-trait fine-mapping with flashfm at 50 loci significantly associated with more than one glycaemic trait. We used fGWAS to build models of enriched annotations by considering 32 cell-type specific and 28 static annotations. We used these models to define prior probabilities to perform annotation informed fine-mapping with both FINEMAP (single-trait) and flashfm (multi-trait). Multi-trait fine-mapping of 106 locus-trait associations significantly (P = 1.23 × 10-17) reduced the median size of the credible sets accounting for 99% of the posterior probability of being causal (99CS) to 21.5 variants compared to the 60.5 variants in single-trait fine-mapping. Annotation informed single-trait fine-mapping of 211 locus-trait associations reduced (P = 4.24 × 10-12) the median 99CS size from 72 in agnostic single-trait fine-mapping to 52 variants. Annotation informed multi-trait fine-mapping of 110 locus-trait associations led to a further significant (P = 2.69 × 10-18) decrease in median 99CS size to 14.5 variants compared to 51.0 in annotation informed single-trait fine-mapping. In conclusion, by applying combined multi-trait and annotation informed fine-mapping to 50 loci, we refined the number of potential causal variants by 71.1% compared to single-trait agnostic fine-mapping.