Constellation: a tool for rapid, automated phenotype assignment of a highly polymorphic pharmacogene, CYP2D6, from whole-genome sequences

Abstract

An important component of precision medicine-the use of whole-genome sequencing (WGS) to guide lifelong healthcare-is electronic decision support to inform drug choice and dosing. To achieve this, automated identification of genetic variation in genes involved in drug absorption, distribution, metabolism, excretion and response (ADMER) is required. CYP2D6 is a major enzyme for drug bioactivation and elimination. CYP2D6 activity is predominantly governed by genetic variation; however, it is technically arduous to haplotype. Not only is the nucleotide sequence of CYP2D6 highly polymorphic, but the locus also features diverse structural variations, including gene deletion, duplication, multiplication events and rearrangements with the nonfunctional, neighbouring CYP2D7 and CYP2D8 genes. We developed Constellation, a probabilistic scoring system, enabling automated ascertainment of CYP2D6 activity scores from 2×100 paired-end WGS. The consensus reference method included TaqMan genotyping assays, quantitative copy-number variation determination and Sanger sequencing. When compared with the consensus reference Constellation had an analytic sensitivity of 97% (59 of 61 diplotypes) and analytic specificity of 95% (116 of 122 haplotypes). All extreme phenotypes, i.e., poor and ultrarapid metabolisers were accurately identified by Constellation. Constellation is anticipated to be extensible to functional variation in all ADMER genes, and to be performed at marginal incremental financial and computational costs in the setting of diagnostic WGS.

Figure 1

Graphical overview of the highly polymorphic CYP2D6/2D7/2D8 locus. (a) The reference Chr 22 locus comprising the CYP2D6*1 haplotype (white) and two non-functional paralogs, CYP2D7 (red) and CYP2D8 (grey). Note that the locus is on the minus strand and is shown in reverse. REP6 and REP7 are paralogous, Alu-containing, 600-bp repetitive segments found downstream of CYP2D6 and CYP2D7, respectively. The blue boxes indicate identical unique sequences downstream of CYP2D6 and CYP2D7 which are separated from REP7 by 1.6-kb in the latter. (b) Three CYP2D6 haplotypes, CYP2D6*2, CYP2D6*10 and CYP2D6*4. The CYP2D6 activity conveyed by these haplotypes is indicated by colour-coded boxes (red, non-functional variant; orange, decreased activity; green, fully functional reference activity; blue, increased activity). (c) The most common CYP2D6 copy-number variations. CYP2D6*5 is characterised by a deletion of the entire CYP2D6 gene and fusion of REP6 and REP7 (REP-del). Duplication haplotypes have two or more CYP2D6 copies, as exemplified by CYP2D6*2x2 (ultrarapid metaboliser) and CYP2D6*4x2 (non-functional). Less common are copy-number variants with three or more copies. (d) Hybrid genes composed of CYP2D7 and CYP2D6 fusion products that result from unequal recombination. A number of hybrid genes with a variety of switch regions have been described and are consolidated as the CYP2D6*13 haplotype. (e) Four tandem arrangements, featuring two or more, non-identical copies of CYP2D6.

Figure 2

In silico modelling of the uniqueness of alignments of simulated short-read sequences to the region of Chromosome 22 containing CYP2D6, CYP2D7 and CYP2D8 (hg19, chr22:42,518,000-42,555,000). Simulated singleton reads (a) and paired-end reads (b) from 50 to 5,000-nt in length were generated from this region. For paired-end reads, insert lengths varied from 300 to 800-nt. Exons, introns and genomic features to which reads mapped uniquely with GSNAP are shown as green‘1’; regions to which reads did not map uniquely are shown as red ‘0’.

Figure 3

Flow diagram of the assignment of CYP2D6 phenotype inferred by WGS and Constellation. Whole-genome sequence data are mapped to the reference human genome and CYP2D6 diplotypes called by Constellation. Predicted phenotype is determined by assigning an ‘activity score’ based on the individual diplotype.