Comparing low-pass sequencing and genotyping for trait mapping in pharmacogenetics

Abstract

Background: Low pass sequencing has been proposed as a cost-effective alternative to genotyping arrays to identify genetic variants that influence multifactorial traits in humans. For common diseases this typically has required both large sample sizes and comprehensive variant discovery. Genotyping arrays are also routinely used to perform pharmacogenetic (PGx) experiments where sample sizes are likely to be significantly smaller, but clinically relevant effect sizes likely to be larger.

Results: To assess how low pass sequencing would compare to array based genotyping for PGx we compared a low-pass assay (in which 1x coverage or less of a target genome is sequenced) along with software for genotype imputation to standard approaches. We sequenced 79 individuals to 1x genome coverage and genotyped the same samples on the Affymetrix Axiom Biobank Precision Medicine Research Array (PMRA). We then down-sampled the sequencing data to 0.8x, 0.6x, and 0.4x coverage, and performed imputation. Both the genotype data and the sequencing data were further used to impute human leukocyte antigen (HLA) genotypes for all samples. We compared the sequencing data and the genotyping array data in terms of four metrics: overall concordance, concordance at single nucleotide polymorphisms in pharmacogenetics-related genes, concordance in imputed HLA genotypes, and imputation r². Overall concordance between the two assays ranged from 98.2% (for 0.4x coverage sequencing) to 99.2% (for 1x coverage sequencing), with qualitatively similar numbers for the subsets of variants most important in pharmacogenetics. At common single nucleotide polymorphisms (SNPs), the mean imputation r² from the genotyping array was 0.90, which was comparable to the imputation r² from 0.4x coverage sequencing, while the mean imputation r² from 1x sequencing data was 0.96.

Conclusions: These results indicate that low-pass sequencing to a depth above 0.4x coverage attains higher power for association studies when compared to the PMRA and should be considered as a competitive alternative to genotyping arrays for trait mapping in pharmacogenetics.

Keywords: Genotype imputation; Low-pass sequencing; Pharmacogenetics; Trait mapping.

Fig. 1

Genotype concordance across platforms at specific variants relevant to pharmacogenomics. a. Concordance at SNPs in ADME genes. Variants were classified as “rare” if the minor allele was present in five or fewer copies in the sample (corresponding to an allele frequency of about 3%. Concordance rates are split according to the genotype calls on the PRMA, which was considered “truth”—reference concordance is at variants where the PRMA is homozygous reference, and non-reference concordance is for all other sites. b. Concordance in HLA genotypes across platforms. Shown are the concordance rates between sequencing and genotyping array data in imputed HLA genotypes. Concordance is shown for 0.4x and 1x sequencing

Fig. 2

Comparison of imputation quality across platforms. Alleles were binned according to their minor allele frequency (as measured on the genotyping array) and imputation r² averaged across all variants in the bin. For sequencing data, the array data was treated as ‘truth’ and imputation r² computed by correlating imputed dosages to array genotypes. For the array data, imputation r² for all genotyped variants was computed using a leave-one-out procedure implemented in minimac2