Fine-mapping a genome-wide meta-analysis of 98,374 migraine cases identifies 181 sets of candidate causal variants

Abstract

Migraine is a highly prevalent neurovascular disorder for which genome-wide association studies (GWAS) have identified over one hundred risk loci, yet the causal variants and genes remain mostly unknown. Here, we meta-analyzed three migraine GWAS including 98,374 cases and 869,160 controls and identified 122 independent risk loci of which 35 were new. Fine-mapping of a meta-analysis is challenging because some variants may be missing from some participating studies and accurate linkage disequilibrium (LD) information of the variants is often not available. Here, using the exact in-sample LD, we first investigated which statistics could reliably capture the quality of fine-mapping when only reference LD was available. We observed that the posterior expected number of causal variants best distinguished between the high- and low-quality results. Next, we performed fine-mapping for 102 autosomal risk regions using FINEMAP. We produced high-quality fine-mapping for 93 regions and defined 181 distinct credible sets. Among the high-quality credible sets were 7 variants with very high posterior inclusion probability (PIP > 0.9) and 2 missense variants with PIP > 0.5 (rs6330 in NGF and rs1133400 in INPP5A). For 35 association signals, we managed to narrow down the set of potential risk variants to at most 5 variants.

Figure 1.

A Manhattan plot of the inverse-variance weighted fixed effects migraine meta-analysis including 98,374 cases and 869,160 controls. X-axis presents the chromosomal location and y-axis the −log₁₀(P-value). Known loci are highlighted in purple and new loci in green. Variants with posterior inclusion probability (PIP) > 0.9 and missense variants with PIP > 0.5 in high-quality fine-mapping regions are annotated.

Figure 2.

Fine-mapping a region near TSPAN2 at chromosome 1 using three different LD sources. a) Plot of the GWAS results with the chromosomal location on x-axis and the strength of the association as −log10 P-values from the inverse-variance weighted fixed-effect meta-analysis with 98,374 migraine cases and 869,160 controls on y-axis. Variants are colored based on the squared correlation (r²) to the two variants in the top configuration suggested by FINEMAP with the in-sample LD. The suggested top configurations based on three LD panels are marked by lines with the in-sample LD and the UKB-FG reference LD giving the same top configuration and the UKB reference LD including three additional variants (highlighted in green). Posterior inclusion probabilities (PIPs) for the variants based on b) in-sample LD, c) UKB-FG reference LD and d) UKB reference LD.

Figure 3.

a) Scatter plot comparing the maximum PIP differences (maxΔ) between the in-sample and reference LD for 26 fine-map regions. X-axis shows the UKB-FG reference LD and y-axis the UKB reference LD. b) Strip chart shows the posterior expected number of causal variants (PENC) from fine-mapping for the two LD reference panels for the 102 fine-map regions. Red dots indicate large differences from the in-sample LD (maxΔ > 0.1), and grey color indicates regions for which only reference LD is available and therefore maxΔ is not known. Horizontal line shows PENC = 3 that we use as a threshold to define reliable results with the UKB-FG panel.

Figure 4.

Summary of the fine-mapping results across the 102 migraine risk regions.