Resolving discordant CYP2D6 genotyping results in Thai subjects: platform limitations and novel haplotypes

Abstract

Aim:

Several CYP2D6 Luminex xTAG genotype calls were identified as inconsistent or suspicious among Thai subjects and further characterized to identify the root causes.

Material & methods:

Forty-eight subjects were followed-up with long-range-PCR, quantitative copy number assays and/or Sanger sequencing.

Results:

Most of the Luminex-duplication calls were either negative or had hybrid structures involving CYP2D6*36 in various configurations. Ten samples were inaccurately called as CYP2D6*2, *29 or *35 alleles. Sequencing revealed three novel haplotypes, CYP2D6*142, *143 and *144 of which two are nonfunctional.

Conclusion:

The Luminex platform produced a relatively high number of false genotype calls for Thai subjects. Our findings underscore the need for the systematic characterization of the CYP2D6 locus in diverse populations and rigorous platform validation.

CYP2D6 metabolizes over 20% of clinically prescribed drugs [1,2], including many antidepressants, antipsychotics, antiarrhythmics, antiemetics, the estrogen receptor antagonist tamoxifen, and opioids such as codeine, and tramadol.

Genetic variability in CYP2D6 is affected by SNPs, insertions or deletions of one or multiple nucleotides, and copy number variations (CNVs), which encompass gene deletions, duplications, multiplications and rearrangements with its CYP2D7 pseudogene [3,4]. These are often referred to as hybrid genes and can have CYP2D6–2D7 and CYP2D7–2D6 structures [5–8]. The highly polymorphic nature of the gene locus (as of 10 May 2021 the Pharmacogene Variation Consortium [PharmVar] [9,10] lists 137 star (*) alleles (*1-*145) not counting suballeles and structural variants) together with complex structural variation makes CYP2D6 genotyping and allele characterization a difficult task as commented on by Nofziger and Paulmichl [11].

Pharmacogenetic testing is increasingly implemented into routine clinical practice and facilitates individualized drug therapy [12–15]. It is therefore essential to accurately determine a patient’s genotype. One of the platforms utilized by clinical testing laboratories is the Luminex xTAG CYP2D6 v3 assay (referred from here on as ‘Luminex’ for simplicity), which employs multiplex PCR and allele-specific primer extension. The Luminex assay covers the following clinically relevant CYP2D6 alleles: *2-*11, *12, *14A and *14B, *15, *17, *29, *35, *39 and *41, as well as the CYP2D6*5 gene deletion and duplications.

In a previous study performed on a large Thai population sample, differences in allele frequencies, but also genotype calls, were observed between Luminex and three other platforms, in other words, the AmpliChip CYP450 test, Agena MassArray and TaqMan-based genotyping for a limited set of alleles [16]. Some of the differences could be explained by rare sequence variants that were not consistently tested by all platforms. In other cases, the Luminex allele calling algorithm did not recognize certain haplotypes, which caused inaccurate genotype calls or resulted in ‘no calls.’ However, inconsistent Luminex calls were not further investigated to determine the cause of those inconsistencies.

Due to ease of use, low cost and accommodating low-to-medium sample throughput, the Luminex platform is widely used for research as well as in the clinical setting [17–22]. Among our clinical testing and research samples, we continued to observe CYP2D6 ‘no calls’ and calls that were deemed suspicious upon manual comparison of the SNP results and automated genotype calls. Furthermore, the number of duplication calls was higher than expected.

The primary aim of this investigation was to confirm Luminex CNV calls, resolve Luminex genotype calls that were inconsistent based on SNP results and PharmVar allele definitions, and explain ‘no calls.’ Our hypothesis was that the Luminex assay does not accurately determine CYP2D6 genotype in some Thai patients due to inadequate allele coverage and/or presence of sequence variations including CNVs and/or allele calling software that is not reflecting the current state of allele designations as cataloged by PharmVar.

Materials & methods

Subjects

Forty-eight samples were selected from subjects who underwent clinical genotype testing (n = 34) and or were part of a research study (n = 14) investigating risperidone (RIS) metabolism in autistic children at the Laboratory for Pharmacogenomics, Somdech Phra Debaratana Medical Center, Ramathibodi Hospital. Sample selection was based on DNA availability, a ‘no-call’ CYP2D6 genotype result and/or inconsistent findings with orthogonal CNV spot-testing using the TaqMan gene copy number assay (Hs04083572_cn) interrogating intron 2. All subjects gave written informed consent for CYP2D6 genotype testing. The study was approved by the Ethics Committee of the Faculty of Medicine Ramathibodi Hospital, Bangkok, Thailand (MURA2017/556).

Luminex xTAG CYP2D6 v3 genotype analysis

Genomic DNA was extracted from EDTA blood with the MagNa Pure (Roche Diagnostics, IN, USA) automated extraction system according to the manufacturer’s instructions. The Luminex xTAG CYP2D6 Kit v3 (Luminex Corporation, TX, USA) is a bead array method, utilizing allele-specific primer extension and hybridization to oligonucleotide bound microspheres. Assays were performed as recommended by the manufacturer. The assay interrogates 21 sequence variants of which 19 are within the CYP2D6 gene (-1584C>G, 31G>A, 100C>T, 124G>A, 137_138insT, 882G>C, 1022C>T, 1660G>A, 1662G>C, 1708delT, 1759G>T, 1847G>A, 2550delA, 2616delAAG, 2851C>T, 2936A>C, 2989G>A, 3184G>A and 4181G>C), the CYP2D6*5 gene deletion and the presence of a duplication. The variant positions and nucleotide changes used herein are based on the NG_008376.4 reference sequence with the ATG start codon being +1. CYP2D6 genotype calls were generated with the Luminex xTAG CYP2D6 Kit software (TDAS CYP2D6).

Qualitative CNV analysis by long-range-PCR

Long-range (XL)-PCR was performed as previously described [5–7,23]. Briefly, the duplicated gene copy was amplified using a CYP2D6-specific forward primer and a reverse primer that is specific for the sequence between duplicated gene copies. The size of the amplification fragment (referred to as Fragment D here and previous publications) is 8.6 kb long in the presence of gene duplications with a CYP2D6-like downstream region (e.g., CYP2D6*2×2 and *4×2) or 10.2 kb when the duplicated gene contains a CYP2D7-like downstream region (e.g., CYP2D6*36+*10 or *64+*4). The CYP2D6*5 gene was also detected by XL-PCR [5]. A summary of all XL-PCR reactions used in this study can be found in Supplementary Table 1. All XL-PCR reactions were performed using 2x KAPA LongRange HotStart ReadyMix with dye (Roche Diagnostics). Amplicons were visualized on a 0.7% agarose gel with 1x final concentration of SYBR™ Safe DNA Gel Stain (Thermo Fisher Scientific, MA, USA).

Quantitative CNV analyses

For follow-up studies, gene copy number was determined using a previously described multiplex PCR Assay (MPA) [7,8] targeting for gene regions, exon 1, intron 5, intron 6 and exon 9. Briefly, each 8 μl reaction contained 10–20 ng genomic DNA (less for samples with limited DNA availability), KAPA2G Fast HotStart DNA Polymerase (Roche Diagnostics), supplied buffer at 1x (intron 5/exon 9 combo assay) or 1.5x (exon 1 and intron 6 assays) final concentration, DMSO (5%), dNTPs (10 μM) and primers (each at 10 μM). Reactions were cycled with an Eppendorf MasterCycler ep Gradient S instrument (Eppendorf, Hamburg, Germany). PCR products were separated on an ABI 3730xl DNA Analyzer (Thermo Fisher Scientific) and analyzed with the GeneMapper software v5.0. Samples with 0, 1, 2, 3 and 4 gene copies served as controls in each run. Copy number calls were within 95% confidence levels for each copy number cluster [8].

Inconclusive MPA results were resolved using droplet digital PCR (ddPCR) as previously published [7] with the following modifications: 30–100 ng of DNA were digested with 1.2 U of EcoRI-HF (New England BioLabs, MA, USA), and TERT (Applied Biosystems, CA, USA) was used as the reference assay. Three CYP2D6 regions were targeted (5′UTR, intron 6 and exon 9). All TaqMan copy number assays were obtained from Thermo Fisher Scientific; assay IDs are provided in Supplementary Table 2. Briefly, digested genomic DNA (10–25 ng) was combined with the TERT reference, the CYP2D6 TaqMan™ assay and ddPCR Supermix for Probes (Bio-Rad Laboratories, CA, USA) at 1x concentrations to a total volume of 22 μl. Droplets were generated using the Bio-Rad Automated Droplet Generator and cycled in the C1000 Touch™ Thermal Cycler at 95°C for 10 min initially, then 40x cycles of 94°C for 30 s and 60°C for 1 min, with a final step at 98°C for 10 min. Droplet analysis was performed with the QX200 Droplet Reader and the QuantaSoft™ Analysis Pro software (version 1.0.596). Clusters were manually inspected for separation quality and copy number was called within 0.15 to the nearest integer. For example, a copy number of 2 was called if the CNV data were between 1.85 and 2.15.

CYP2D6 genotyping using TaqMan

Commercially available TaqMan genotyping assays (Thermo Fisher Scientific) were utilized for confirmatory genotyping. Assays IDs are provided in Supplementary Table 2. The TaqMan assays were performed with 1x KAPA Probe Fast qPCR Master Mix (Roche Diagnostics) and 10–15 ng of genomic DNA or diluted XL-PCR products (1:2000). Reactions were run on a QuantSudio 12 K Flex Real-Time PCR system (version 1.2.2) with an initial denaturing step of 3 min at 95°C, followed by 50 (genomic DNA) or 30 (XL-PCR products) cycles at 95°C for 3 s and 60°C for 30 s. Data were analyzed with the TaqMan Genotyper Software (version 1.4.0) as well as manually inspected. Each sample was performed in duplicate reactions. Samples obtained from the Coriell Institute for Medical Research (NJ, USA) served as controls. Individual TaqMan assays utilized are listed in Supplementary Table 2.

Sanger sequencing

To fully characterize haplotypes, alleles of interest were amplified by allele-specific XL-PCR (ASXL-PCR), utilizing heterozygous SNPs and subsequently sequenced as previously described [24]. If alleles could not be separated with this approach, both alleles were amplified with a CYP2D6-specific universal forward and reverse primers (this fragment is referred to here and in previous publications as Fragment A). Complete haplotype characterization included the entire gene, upstream and downstream regions as prescribed by PharmVar. Supplementary Table 1 summarizes the XL-PCR amplicons generated for this study. XL-PCR sequencing templates were purified using the QIAquick PCR Purification Kit (Qiagen, Sciences Inc., MD, USA) per manufacturer’s protocol except the final elution step: 20 μl of EB Buffer was warmed to 50°C, applied to the spin column, incubated for 1 min and then centrifuged. This step was performed twice for a total elution volume of 40 μl to maximize DNA recovery. Sequencing was performed with BigDye Terminator version 3.1 chemistry and a capillary 3730xl DNA Analyzer (Thermo Fisher Scientific). Sequence traces were aligned and analyzed using Sequencher software version 5.4 (Gene Codes Corp., MI, USA), and aligned to the NG_008376.4 reference sequence.

Allele nomenclature & designations of novel haplotypes

Allele nomenclature used herein is per the PharmVar [9,10]. Novel haplotypes were fully characterized as required by PharmVar and submitted for allele designation.

In vivo metabolic activity toward RIS

Trough plasma concentrations of RIS and its 9-OH metabolite (9-OH-RIS) were quantified in samples collected approximately 12 h after the bedtime dose, using a validated and previously published high-performance LC–MS/MS method [25]. Plasma ratios (RIS/9-OH-RIS) were available for 200 children genotyped with the Luminex xTAG CYP2D6 v3 kit. Genotypes were grouped by activity score (AS) as recommended by the Clinical Pharmacogenetic Implementation Consortium [26]. Fourteen of the 200 children were part of the current investigation. None of the remaining 186 subjects had ‘no calls’ or inconsistent Luminex genotype findings.

Results

Forty-eight samples were selected for follow-up testing to confirm Luminex CNV calls, resolve discordant Luminex test results and explain ‘no-calls’ and/or resolve genotype calls that may be inconsistent with the SNPs detected. Figure 1 provides an overview of the CNV results across platforms and Figure 2 summarizes the characterization of discordant Luminex results using orthogonal genotyping and/or gene resequencing. The following sections describe findings by call type, in other words, those involving Luminex CNV duplication calls (referred to as Luminex DUP), Luminex CNV deletion calls (Luminex DEL) and/or Luminex genotype calls (Luminex GENO). It is noted that some samples had results that were discordant for both CNVs and genotype calls.

Figure 1. . Overview of copy number variation results across platforms.

The figure provides a summary of the 48 samples selected for follow-up CNV analyses. Original Luminex calls were categorized as shown and are highlighted in pink. Revised CNV calls are subcategorized and are highlighted in blue. Dotted arrows indicate that a sample may have one or more CNV features, while the blue, double-pointed arrows indicate samples with shared features. Several samples called by Luminex as DUP positive revealed a hybrid structure as well as a gene duplication and thus were counted once for the category ‘duplication detected’ and again for the category ‘hybrid detected’ (e.g., samples with a CYP2D6*10/*36×2+*10 genotype).

CNV: Copy number variation.

Figure 2. . Overview of genotype calls across platforms: resolving discordant results.

The flow chart provides a summary of the samples that were regenotyped and/or Sanger sequenced. Samples were first categorized based on whether any inconsistencies were noted and subsequently grouped into types of inconsistency. One group included those with inaccurate diplotype calls due to the presence of a hybrid most of which were CYP2D6*36+*10 tandem alleles that were called as *10. The other group consisted of samples with Luminex CYP2D6*2, *29 or *35 calls in the absence of 2851C>T. These samples were further characterized by genotyping and Sanger sequencing (highlighted in gray boxes). There were ten subjects with a discordant allele in this group; since one subject was determined to have two novel alleles (i.e., CYP2D6*143/*144), the total number of novel alleles and suballeles is 11.

Luminex CNV duplication calls versus revised calls

Thirty-six of the 48 samples were selected due to positive Luminex DUP or DEL calls, or CNV ‘no-calls’ (Figure 1). Samples were subjected to additional testing, using qualitative XL-PCR, a quantitative MPA assay targeting four gene regions, and ddPCR, if the MPA calls were inconclusive. As summarized in Figure 1, a duplication was detected in only eight of the 26 samples with a Luminex DUP-positive result (30.7% of calls). Of these eight, only two samples had a conventional gene duplication, i.e., CYP2D6*10/*41×2, while the other six were identified to have a CYP2D6*36+*10/*36+*10 or CYP2D6*10/*36×2+*10 diplotype. Although CYP2D6*36+*10/*36+*10 is the more likely configuration our results support both possibilities. While most of the false-positive samples (15/26; 57.7%) could be explained by the presence of a hybrid structure (i.e., CYP2D6*1/*36+*10), there were also n = 10 duplication-positive samples (10/26; 38.5%) that were negative for any duplication or hybrid structures per XL-PCR, MPA and/or ddPCR testing. The diplotypes of these samples contained CYP2D6*1, *2, *4, *5, *10 and *41 alleles; there was no commonality as to why they had the false-positive Luminex DUP calls. Of note, six samples with a possible CYP2D6*36×2 allele and one sample with a tentative CYP2D6*1×2/*36+*10 genotype were counted as ‘duplication detected,’ as well as ‘hybrid structure detected’ explaining n = 26 Luminex DUP-positive calls versus the sum of all revised CNV calls (n = 33) in Figure 1.

Of the seven Luminex DUP ‘no call’ samples, one had a possible duplication (the genotype could either be CYP2D6*10/*36×2+*10 or CYP2D6*36+*10/*36+*10) and three had a hybrid detected (CYP2D6*10/*36×2+*10 or CYP2D6*36+*10/*36+*10, CYP2D6*4/*36+*10 or CYP2D6*10/*68+*4, and CYP2D6*5/*36+*10). Furthermore, we detected a CYP2D6*5 gene deletion in three of the four ‘no duplication or hybrid detected’ samples, revising their genotypes as CYP2D6*4/*5, *5/*10 and *5/*41, respectively. The one other ‘no call’ sample was ultimately determined to have a CYP2D6*2/*144 genotype.

A significant portion of alleles (18/30; 60.0%) identified by Luminex as CYP2D6*10 were revised to CYP2D6*36+*10. Luminex does not detect the presence of CYP2D7-derived exon 9 sequences (commonly known as exon 9 conversion), explaining why CYP2D6*36+*10 were ‘defaulted’ to *10. To the best of current knowledge, CYP2D6*10 and *36+*10 are functionally equivalent.

For a complete list of samples, their initial and revised genotypes, and respective result groupings, we refer to Supplementary Table 3. Here, we also provide the possible diplotypes for samples with complex structures, which we were unable to fully resolve (e.g., CYP2D6*36+*10/*36+*10 or CYP2D6*10/*36×2, CYP2D6*4/*36+*10 or CYP2D6*10/*68+*4).

Luminex gene deletion calls versus revised calls

Eleven samples were identified by Luminex as having a CYP2D6*5 gene deletion, of which one was not confirmed (Figure 1). Additional characterization of this sample by XL-PCR and Sanger sequencing revealed that the false-positive Luminex DEL result was likely caused by a novel structural variant, in other words, a CYP2D6*36+*10 tandem with a CYP2D7-like downstream region, which may have interfered with the assay designed to detect the CYP2D6*5 gene deletion. This is the first report of a CYP2D6*36+*10 tandem with both gene copies having a CYP2D7-like downstream region. Additional information about these samples can be found in Supplementary Table 3.

Luminex genotype calls versus revised calls

Several samples had Luminex GENO calls that could not be reconciled based on the SNPs listed on their report. In fact, as shown in Figure 2, ten samples were miscalled as having a CYP2D6*2, *29 or *35 allele, although the *2 core SNP (2851C>T, p.R296C) was absent. Of note, three of these samples were identified by Luminex to contain the CYP2D6*41 allele; however, upon further investigation, these allele calls were determined to be correct as the discrepancy was found to be caused by the other allele (e.g., CYP2D6*2/*41 was revised to CYP2D6*1.032/*41). The Luminex xTAG CYP2D6 Kit v3 software appeared to make such calls based on the presence of -1584C>G and 1662G>C, even in situations when one or both CYP2D6*2 core SNPs were absent. Sequence analysis revealed that two of these samples had a CYP2D6*39 allele, five had *1 suballeles, of which three were positive for -1584C>G, and three had a novel allele, which was designated as CYP2D6*144 by PharmVar.

One sample was reported as CYP2D6*1/*29 by Luminex, which was suspicious, given that the CYP2D6*29 allele is generally only observed in subjects of African ancestry. TaqMan genotyping did not detect the CYP2D6*29 core SNPs, while subsequent Sanger sequencing revealed two heterozygous sequence variants, i.e., -498C>A and a 16 bp deletion in intron 3 (1781–1795del, rs750569685). Due to limited amounts of DNA, we were unable to conduct further analysis to determine whether these SNPs are in cis or trans. Therefore, this sample may have a diplotype consisting of a CYP2D6*1.010 and a novel *1 suballele. Regardless of the phase, the 16 bp deletion may have interfered with the Luminex assay, detecting the 1660G>A and 1662G>C CYP2D6*29 core SNPs in exon 3.

One sample with a CYP2D6*10/*10 Luminex call triggered further investigation due to heterozygosity of 1662G>C. Since this SNP is part of all current CYP2D6*10 allele definitions, the sample would be expected to be homozygous for 1662G>C. Sequence analysis indeed revealed homozygosity, suggesting an erroneous Luminex SNP call.

Identification of novel alleles & suballeles

Sequencing other samples with inconsistent Luminex GENO calls revealed three novel CYP2D6 haplotypes and one novel suballele. PharmVar designated these alleles as CYP2D6*142, CYP2D6*143, CYP2D6*144 and CYP2D6*10.005.

CYP2D6*142: One of the two samples with this allele was originally called by Luminex as having a CYP2D6*5/*10 genotype, while the second sample was reported as CYP2D6*10/*10. The CYP2D6*142 allele contains both CYP2D6*10 core SNPs, 100C>T and 4181G>C, explaining the original Luminex calls, but also has the CYP2D6*22 core SNP 82C>T (Figure 3). The RIS/9-OH-RIS metabolic ratio of 0.187, measured in the plasma of the subject with the CYP2D6*10/*142 genotype, is compatible with those observed for subjects with an AS of 0.25, 0.5 or 0.75 (Table 1). The subject with the CYP2D6*5/*142 genotype was not part of the RIS study and thus, no RIS/9-OH-RIS was available. Since CYP2D6*5 is a no function allele, any observed activity reflects that of the novel allele.

Figure 3. . Novel allele and suballeles discovered.

This figure summarizes the SNPs present in the novel CYP2D6*142, *143 and *144 alleles, the novel *10.005 suballele and the fully characterized *39.001 allele discovered in this Thai cohort. The left most column indicates the variant position and nucleotide change using the NG_008376.4 as reference sequence counting from the sequence start. The second column provides respective coordinates counting from the translation start codon (ATG = +1). The black column represents the CYP2D6*1.001 reference sequence. SNPs present on a variant allele are highlighted by a blue background. SNPs in red/bold highlight the core SNP(s) of their respective allele definitions. 3785A>G (-1235A>G) is a SNP located between two A-tracks, which is difficult to sequence and could not be resolved for three alleles as indicated by semicolons. The seven SNPs comprising a CYP2D7-derived sequence are known as ‘intron 1 conversion SNPs’ and are annotated as such. SNPs causing an amino acid change or interfere with splicing are shown in bold; all other SNPs are intronic or in the upstream region and are not impacting function based on current knowledge.

Table 1. . Median (interquartile range) of risperidone/9-hydroxy-risperidone ratios per activity score groups among 197 children.

AS	n (%)	Ratio (RIS/9-OH-RIS)
AS = 0	1 (0.51)	1.50
AS = 0.25	17 (8.63)	0.35 (0.17–0.82)†
AS = 0.5	55 (27.92)	0.19 (0.06–0.35)†
AS = 0.75	15 (7.61)	0.26 (0.19–0.35)†
AS = 1.0	17 (8.63)	0.05 (0.02–0.11)
AS = 1.25	70 (35.53)	0.04 (0.02–0.08)
AS = 1.5	9 (4.57)	0.05 (0.04–0.07)
AS = 2.0	13 (6.60)	0.01 (0.00–0.02)
*1/*1	8 (4.06)	0.01 (0.00–0.02)

Of the 200 children with RIS/9-OH-RIS ratios, n = 3 had genotypes containing CYP2D6*142, *143 or *144 alleles and thus, were excluded; genotypes for n = 11 were validated in this study; genotypes of the remaining 186 samples correspond to those reported by Luminex.

^†Statistically significant compared with AS = 1.0, 1.25, 1.5 and 2.0 (p < 0.05) from values expressed as median (interquartile range).

9-OH-RIS: 9-Hydroxy-risperidone; AS: Activity score; RIS: Risperidone.

Both samples were also Luminex DUP-positive. XL-PCR and Sanger sequencing revealed that the CYP2D6*142 allele contains a CYP2D7 gene with a CYP2D6-like downstream region, which likely caused the false-positive Luminex DUP call. This is substantiated by the amplification of an XL-PCR fragment (referred as fragment B here and in previous publications), which amplifies in the presence of CYP2D6 gene duplications such as CYP2D6*1xN, *2xN, etc. It has been shown, however, that this fragment also amplifies alleles containing a CYP2D7 with a CYP2D6-like downstream region [6], causing false-positive duplication calls.

CYP2D6*143: This novel allele was discovered in a sample that was initially miscalled by Luminex as CYP2D6*1/*35 (Figure 2). This allele shares a core SNP (31G>A) with CYP2D6*35, but also harbors 1626T>G that causes a premature stop codon (p.Y124X) and is thus predicted to be a no function allele. Notably, other core SNPs of CYP2D6*35 including 2851C>T are absent (Figure 2, note that there are 11 alleles in ten samples with discordant calls for CYP2D6*2, *29 or *35 in the absence of 2851C>T; this is due to one sample having both, *143 and *144).

CYP2D6*144: This allele was found in three subjects, two of which were reported by Luminex as CYP2D6*1/*2 (and DUP ‘no call’). The third individual was described above for CYP2D6*143 and revealed to also have the novel *144 allele (Figure 2 & Supplementary Table 3). In addition to seven SNPs (matching those found on the *1.032 suballele), *144 harbors 2441G>A, which is located at the intron 4/exon 5 splice junction. This SNP (rs377725912) is annotated by the dbSNP database as a ‘splice acceptor variant.’ The subject with the CYP2D6*143/*144 genotype presented with a RIS/9-OH-RIS ratio of 2.104, which is higher compared with that of a subject with two known no function alleles (Table 1). The second subject with a CYP2D6*1/*144 genotype had a RIS/9-OH-RIS ratio of 0.053, which is consistent with that observed for subjects with an AS of 1 (Table 1). The third subject, genotyped as CYP2D6*2/*144, presented with a RIS/9-OH-RIS ratio of 0.130, which was higher than expected for subjects with an AS of 1. The CYP2D6*2 allele was, however, not sequenced (due to limited amounts of DNA) and may contain additional SNPs. Taken together, the presented evidence suggests that CYP2D6*144 is a no function allele.

CYP2D6*10.005: This novel suballele was found in a sample for which the Luminex returned a CYP2D6*2/*10 genotype call despite the absence of the CYP2D6*2 core SNP (2851C>T) (Figure 2). Sequencing revealed a novel CYP2D6*10 suballele that was designated CYP2D6*10.005 by PharmVar (Figure 3). The subject’s other allele matched the existing definition of CYP2D6*39.001 but had additional SNPs. The existing definition of CYP2D6*39 was based on exon sequencing only and lacked information for upstream, downstream and intronic regions and thus was flagged by PharmVar as being a haplotype with limited evidence [3]. This allele was submitted to PharmVar, and based on the SNPs detected in this sample, its definition revised, and the evidence level raised to definitive.

Impact of revised genotypes on AS & phenotype prediction

To assess the impact of miscalls, ‘no-calls’ and other inconsistencies on phenotype prediction, we assigned AS to Luminex-based genotypes and the revised genotypes. As shown in Supplementary Table 3, the AS changed for 29 of the 48 samples (60.4%) based on their revised genotypes, which will or may also cause a change in phenotype classification for 14 of the 48 subjects (29.1%).

Discussion

CYP2D6 genotype analysis is complex due to the high number of known allelic variation, including gene CNV and structural rearrangements between the CYP2D6 and CYP2D7 genes [3,11,27,28]. Furthermore, there may still be limited information regarding the nature and degree of variation in the population of interest. Although there are CYP2D6 pharmacogenetic studies in Thai people [16,18,29–32], unresolved discordances among genotyping platforms are a concern, especially when used clinically [16]. Also of concerns is that commercially available platforms vary considerably in regard of sequence variations (or star alleles) tested and in the extent of determining CNVs [33–35]. Most of these platforms, regardless of whether they are approved by the U.S. Federal Drug Administration or not, only interrogate the commonly observed variants across populations. This, however, can be problematic, as rare population-specific variants may not be adequately identified. As a result, a patient’s predicted CYP2D6 metabolizer status, which is typically used for drug and/or dosing recommendations, may not be accurate.

Owing to previous [16] and continuing observations of concerning Luminex calls, we characterized 48 samples to identify underlying causes that may explain ‘no calls’ or inconsistent findings (Figures 1&2 & Supplementary Table 3). Only eight of the 26 samples with a positive Luminex duplication call were confirmed to have a duplication, ten were CNV-negative, and rather interestingly, 15 had structures involving a hybrid gene such as the CYP2D6*36+*10 tandem. Since samples were tested using two or three different methods for CNV evaluation, we are confident that the Luminex test results were indeed false-positive or false-negative (Supplementary Table 3). Since we did not systematically verify Luminex deletion-negative samples, it remains unknown, how often false-negative occur. At the time the study was designed, there was no indication for false-negative calls, our findings, however, suggest that false-negative calls can also occur.

We can only speculate as to why this platform produced such a high number of erroneous CNV calls. We hypothesize that the different structural variants containing a CYP2D6*36 gene, which are commonly found in people of East Asian ancestry, including the Thai, interfere with the Luminex testing methodology. The Luminex assays generate PCR products by targeting regions deemed to be specific to gene ‘duplications’ and ‘deletions.’ The presence of these amplification products is detected by the platform, and respective CNVs are called by the software. We speculate that certain alleles (including the CYP2D6*36+*10 tandem allele and variations thereof, such as CYP2D6*36×2, *36×2+*10, etc.) may inadvertently promote the amplification of the Luminex duplication-specific products, causing the false-positive calls. Our data strongly suggest that either the Luminex regions targeted for ‘duplication’ and ‘deletion’ may vary among our samples in terms of their downstream structure (i.e., regions that are CYP2D6-like rather than CYP2D7-like) or may contain unknown sequence variants. Downstream gene regions have been described for several structural variants (see the PharmVar Structural Variation document on PharmVar [28]) but are often only described as CYP2D6-like or CYP2D7-like, based on the absence or presence of a 1.7 kb long CYP2D7-derived ‘spacer’ sequence. There is little known about how polymorphic the downstream regions are, since these are not systematically interrogated. The CYP2D6*13+*2 tandem, for example, may be miscalled as a CYP2D6*2×2 duplication because the downstream region of the *13 CYP2D7–2D6 hybrid gene supports amplification of ‘duplication’-specific PCR products [5]. On the other hand, duplications may go undetected, if sequences deviate from their perceived structure. For example, the duplicated copy of a CYP2D6*17×2 may, in rare cases, have a CYP2D7-derived downstream region with the spacer. Here, the detection of the duplication is eluded altogether, when using XL-PCR targeting the CYP2D6-like region [8].

Another limitation of the Luminex Platform is its allele calling algorithm. Among the 48 validation samples, ten alleles were incorrectly assigned as CYP2D6*2, *29 or *35 (Figure 2 & Supplementary Table 3). Since the Luminex genotype caller is proprietary, it remains speculative of why these alleles were miscalled. For example, two samples lacking 2851C>T, but having 1662G>C and 4181G>C were called CYP2D6*2 and not *39. In addition, some calls may be incorrect due to how the algorithm interprets -1584C>G. While this SNP is indeed present on most CYP2D6*2 alleles in Europeans, many *2 suballeles found in predominantly non-European populations do not have -1584C>G. Moreover, other star alleles also contain -1584C>G. Outdated or incomplete allele definitions used by the Luminex algorithm may explain, why six samples were called CYP2D6*2 but ultimately shown to have a *1.032 suballele or a novel allele containing the -1584C>G SNP. We believe that these highlighted limitations of the Luminex algorithm affects East Asian populations including the Thai to a greater extent than those of European ancestry.

Sanger sequencing also discovered three novel alleles, CYP2D6*142, *143 and *144, as well as a novel suballele, *10.005, among the 48 study samples. In addition, the full-length sequence data obtained for the CYP2D6*39.001 allele allowed PharmVar to upgrade the allele’s evidence level from limited to definitive. CYP2D6*143 does not encode functional protein owing to a premature stop codon (1626T>G, p.Y124X). CYP2D6*144 has a splice acceptor SNP (2441G>A) that is predicted to interfere with splicing. In vivo RIS/9-OH-RIS ratios corroborate that CYP2D6*144 is indeed also not translated into functional protein. Owing to CYP2D6*142 being paired with a decreased function *10 allele, RIS/9-OH-RIS ratio reflects activity of both alleles together making it impossible to unequivocally determine the activity of the novel allele. The CYP2D6*142 allele has three SNPs: 100C>T (p.P34S, underlying SNP causing severely decreased function of *10), 82C>T (p.R28C, not impacting [36] or increasing function [37] of *22), and 4181G>C (present in many alleles, likely not impacting function). Based on the RIS/9-OH-RIS ratio of 0.187 (Table 1), CYP2D6*142 could be a decreased or no function allele.

Of the 200 subjects in the autism study, we were able to retest all subjects with a miscall, ‘no call’ or inconsistent finding (n = 14 or 7%), suggesting that a significant portion of subjects may not be accurately genotyped by Luminex, thus leading to phenotype misclassification. For example, a patient may be misclassified as UM (instead of as NM) and be prescribed a higher dose, which could cause adverse events; the patient may also unnecessarily be switched to another medication based on a false-positive UM assignment. We were, however, unable to estimate the proportion of such samples among those genotyped for clinical purposes. Since we relied, at least in part, on opportunistic sampling for this investigation, we can only estimate the fraction of Thai subjects that may be affected by inaccurate Luminex calls. Further studies are warranted to more precisely determine that fraction. Another limitation is that not all of the investigated subjects have been sequenced; therefore, we cannot rule out with certainty that these do not have a novel allele or suballele. Finally, Luminex CNV-negative samples were not part of this validation study, so it remains to be seen whether false-negative duplication calls are of concern as well. We also acknowledge that some subjects with RIS ratios (i.e., those not further investigated here) may not have been accurately genotyped by Luminex, which may impact their AS grouping.

Conclusion

This investigation highlights that not all CYP2D6 genotyping platforms are equal. While we only investigated inconsistent cases based on Luminex results, other platforms using similar approaches for CNV testing and out of date algorithms for genotype calling, may face similar challenges. Therefore, it is important to recognize the limits of a platform used, as well as implement appropriate protocols for reflex testing, which would ideally involve CNV testing by quantitative methods, targeting the 5′UTR or exon 1 and the exon 9 regions, to confirm the presence or absence of duplications, deletions and/or hybrid genes, and resequencing to unequivocally determine the allelic variant(s) present. The Luminex (and other platforms) may have more robust results, when used for subjects of European descent. Our findings demonstrate that this platform may produce false calls for some Thai subjects and possibly other East Asian populations. This may lead to phenotype misclassification and suboptimal therapeutic recommendations. This validation study underscores the need for the systematic characterization of the CYP2D6 locus in racially and geographically diverse populations. Especially for clinical settings, it is necessary to develop platforms, or complement existing platforms with additional testing, to capture variants present in the population of interest to accurately predict a patient’s phenotype for individualized drug therapy.

Summary points.

• Full allele characterizations of CYP2D6 can be difficult due to its highly complex gene locus with numerous variants and structural rearrangements with its pseudogene, CYP2D7.

• Forty-eight samples tested with the Luminex xTAG CYP2D6 v3 Kit were chosen for validation, based on inconsistent and/or suspicious Luminex copy number variation and genotyping calls.

• Follow-up copy number variation testing was performed with long-range-PCR and quantitative PCR methods and confirmatory genotyping with TaqMan™ assays. Allele characterization was accomplished with Sanger sequencing of allele-specific long-range-PCR fragments.

• Fifteen (57.7%) of the 26 Luminex-duplication false-positive calls contained a CYP2D6*36 (CYP2D6–2D7 hybrid structure) in different configurations. Luminex-duplication ‘no calls’ can be attributed to the presence of either a hybrid (3/7; 42.9%) or a duplication (1/7; 14.3%). Of the 15 Luminex-duplication negative samples 13 (86.7%) were confirmed to not have a duplication or CYP2D6*36 structure. Ten (90.9%) of the 11 Luminex-deletion positives were confirmed to have a CYP2D6*5.

• Twenty-one (67.7%) of the 31 inconsistent Luminex genotype calls were due to the presence of a CYP2D6*36 in different configurations.

• Ten miscalls appeared to be due to out-of-date Luminex allele definitions causing some alleles such as some CYP2D6*1 suballeles, *39 and the novel alleles identified in this study to be ‘defaulted’ to CYP2D6*2, *29 or *35 assignments.

• Sanger sequencing revealed three novel alleles (CYP2D6*142, *143 and *144) and a novel suballele (CYP2D6*10.005). Based on in vivo risperidone/9-OH-risperidone data, the function of the CYP2D6*142 allele may be similar to that of CYP2D6*10. CYP2D6*143 is predicted to be nonfunctional due to a SNP causing a premature stop codon (p.Y124X). CYP2D6*144 contains a splice acceptor variant that is also predicted to be nonfunctional.

• We identified some root causes of false-positive and inconsistent Luminex xTAG CYP2D6 genotype calls in Thai subjects.

• These findings highlight the need to not only understand population specific CYP2D6 variation but also to extensively validate the platforms chosen for testing.