Evaluation of transethnic fine mapping with population-specific and cosmopolitan imputation reference panels in diverse Asian populations

Abstract

There has been limited success in identifying causal variants underlying association signals observed in genome-wide association studies (GWAS). The use of 1000 Genomes Project (1KGP) allows the imputation to estimate the genetic information at untyped variants. However, long stretches of high linkage disequilibrium within the genome prevent us from differentiating between causal variants and perfect surrogates, thus limiting our ability to identify causal variants. Transethnic strategies have been proposed as a possible solution to mitigate this. However, these studies generally rely on imputing genotypes from multiple ancestries from 1KGP but not against population-specific reference panels. Here, we perform the first transethnic fine-mapping study across three Asian cohorts from diverse ancestries at the loci implicated with eye and blood lipid traits, using population-specific reference panels that have been generated by whole-genome sequencing samples from the same ancestry groups. Our study outlines several challenges faced in a fine-mapping exercise where one simply aims to meta-analyse existing GWAS that have been imputed against reference haplotypes from the 1KGP.

Figure 1

Histograms (gray vertical bars) and cumulative frequencies (red lines) on the ranks of the simulated causal variants out of 2000 rounds of simulations. The horizontal axes indicate the rank of the association evidence at the rediscovered causal variant in comparison with all neighboring variants. Three of the display panels indicate the causal variant ranking when the analysis is performed with the data from a single ancestry panel, whereas the fourth display panel indicates the rank when data from three ancestry groups are combined in a transethnic fine-mapping.

Figure 2

Regional plots of SNPs at the LDL-C locus, CELSR2, for the Chinese (SCES), Malays (SiMES), Indians (SINDI) and the transethnic fine mapping of all three cohorts. The vertical axes measure the statistical evidence of association with the log₁₀BF, and each SNP is indicated by a colored circle, diamond or triangle. In each panel, the index SNP rs611917 is indicated by the purple diamond, whereas all remaining SNPs are assigned colors according to the extent of LD, with the lead SNP in six categories: (i) r²≥0.8 (red); (ii) 0.6≤r²<0.8 (gold); (iii) 0.4≤r²<0.6 (green); (iv) 0.2≤r²<0.4 (cyan); (v) (ii) r²<0.2 (blue); and (vi) unknown r² (gray). Diamonds represent variants that are found within the ‘99% credible set' and function-altering variants are represented with triangles. Recombination rates estimated from the International HapMAp Project are superimposed with blue lines, and all gene annotations are obtained from the University of California Santa Cruz genome browser.

Figure 3

Regional plots of SNPs at the LDL-C locus TOMM40-APOE from two transethnic meta-analyses using either the population-specific reference panels or the cosmopolitan reference panel from the 1000 Genomes Project. In each panel, the index SNP is represented in purple with either a circle or a diamond, with the latter used if the variant is annotated to alter function by the University of California Santa Cruz genome browser. All remaining SNPs are assigned colors according to the extent of LD with the lead SNP in six categories: (i) r²≥0.8 (red); (ii) 0.6≤r²<0.8 (gold); (iii) 0.4≤r²<0.6 (green); (iv) 0.2≤r²<0.4 (cyan); (v) (ii) r²<0.2 (blue); and (vi) unknown r² (gray).