CYP2C9 and CYP2C19: Deep Mutational Scanning and Functional Characterization of Genomic Missense Variants.

Single nucleotide variants in the open reading frames (ORFs) of pharmacogenes are important causes of interindividual variability in drug response. The functional characterization of variants of unknown significance within ORFs remains a major challenge for pharmacogenomics. Deep mutational scanning (DMS) is a high-throughput technique that makes it possible to analyze the functional effect of hundreds of variants in a parallel and scalable fashion. We adapted a "landing pad" DMS system to study the function of missense variants in the ORFs of cytochrome P450 family 2 subfamily C member 9 (CYP2C9) and cytochrome P450 family 2 subfamily C member 19 (CYP2C19). We studied 230 observed missense variants in the CYP2C9 and CYP2C19 ORFs and found that 19 of 109 CYP2C9 and 36 of 121 CYP2C19 variants displayed less than ~ 25% of the wild-type protein expression, a level that may have clinical relevance. Our results support DMS as an efficient method for the identification of damaging ORF variants that might have potential clinical pharmacogenomic application.

WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

☑ With the increasing application of next generation sequencing (NGS) to known “pharmacogenes,” large numbers of open reading frame (ORF) “variants of unknown significance” (VUS) are being identified. However, the functional implications of those VUS remains unclear.

WHAT QUESTION DID THIS STUDY ADDRESS?

☑ This study was designed to determine whether the application of DMS might make it possible to determine the functional effects of VUS that have been observed in the ORFs of cytochrome P450 family 2 subfamily C member 9 (CYP2C9) and cytochrome P4540 family 2 subfamily C member 19 (CYP2C19).

WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

☑ The results of the study demonstrate that deep mutational scanning (DMS) can be used to determine the functional implications of ORF VUS in CYP2C9 and CYP2C19.

HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

☑ The application of DMS to additional pharmacogenes would potentially expand the accuracy and clinical utility of the application of NGS to pharmacogenomics.

Genetic polymorphisms in or near pharmacogenes are a major cause of individual variation in drug response phenotypes.1 Cytochrome P450 family 2 subfamily C member 9 (CYP2C9) and cytochrome P450 family 2 subfamily C member 19 (CYP2C19) are genes that encode important cytochrome P450 enzymes that catalyze the phase I biotransformation of a variety of therapeutic drugs, including antiplatelet agents, selective serotonin reuptake inhibitors, and proton pump inhibitors.2, 3, 4, 5 Several years ago, the Mayo Clinic launched the RIGHT 1K study, in which next generation DNA sequencing (NGS) was performed with DNA from 1013 Mayo Biobank participants to identify variants in 84 pharmacogenes, including CYP2C9 and CYP2C19.1, 6 However, many of the polymorphisms observed in those patients were variants of unknown significance (VUS).1, 7, 8 In a recent publication, we functionally characterized six novel nonsynonymous open reading frame (ORF) variants in the CYP2C9 gene and seven nonsynonymous ORF variants in the CYP2C19 gene observed in DNA from participants in the Right 1K study.9 Conventional methods for the characterization of individual sequence variants “one‐at‐a‐time” are reliable, but they are also time‐consuming, labor‐intense, and not easily scalable. As a result, only a limited number of variants can practically be investigated in that fashion. To help address this challenge, predictive algorithms, such as Polyphen‐2, SIFT, and PROVEAN, among others, represent efforts to help identify deleterious variants, but their reliability is variable and inadequate for clinical application.10, 11, 12 Laboratory‐based assays are still required to reliably interpret the impact of ORF missense variants on protein function. As a result, a significant gap remains in the functional interpretation of the large number of missense VUS being discovered as DNA sequencing is applied ever more broadly. Deep mutational scanning (DMS) methods provide a platform with which a large number of missense variants can be interrogated in parallel.13, 14 For example, Fowler et al. developed a DMS assay for all possible variants in clinically important genes, such as BRCA1, TPMT, and PTEN, linking genotypes to functionally determined phenotypes.15, 16 Specifically, engineered “landing pad” HEK 293T cells were used as a platform to integrate pooled variant libraries, resulting in one variant per cell for further functional assessment.15 DMS includes the creation of variant libraries for a gene, selecting the library for function (e.g., resistance to drug or fluorescence markers of protein quantity) and—finally—high‐throughput DNA sequencing of variants to link them with the “activity” assayed in a functional test.

In our previous “one‐at‐a‐time” functional genomic study, we identified a series of missense variants in the ORFs of CYP2C9 and CYP2C19 that resulted in decreased protein expression as a result of proteasome‐mediated degradation, presumably due to an alteration in protein folding. In the present study, fluorescence‐activated cell sorting (FACS) was used as a selection mechanism for cells expressing variant sequences (i.e., the cells were sorted according to the abundance of reporter protein expression). We should point out that saturation mutagenesis can also be used to create a “saturation” library, including all possible mutations for an ORF sequence in a single reaction, and that approach has been applied in some previous DMS studies.15 However, many of the variants created are unlikely to occur clinically. Therefore, we applied a different mutagenesis approach, nicking mutagenesis, to generate libraries that contained human genomic variation known to be present in the general population. Generation of variants by nicking mutagenesis is not exhaustive, but it is much more efficient than the generation of mutants one at a time by site‐directed mutagenesis.17

In the present study, we have used DMS to analyze the functional implications of missense variants that have been observed in the ORFs of CYP2C9 and CYP2C19, ORFs that were fused to green fluorescent protein (GFP). Recombinase was used to integrate the variant libraries into landing pad cells, one per cell. Multiplexed functional selection performed with FACS was then used to separate the cells into different “bins” based on fluorescence readout at the single cell level. Amplicon sequencing of DNA in each bin, followed by computational analysis of the frequency of variants appearing in each bin, was used to determine levels of protein expression. To identify potentially severely damaging variants for these two important CYP pharmacogenes, we analyzed 230 observed missense variants in ORFs present in a publicly available database consisting of exome sequencing  data for 60,000 general population subjects. We included genetic variants with allele frequencies (> 0.00001) observed by the Exome Aggregation Consortium (ExAC; http://exac.broadinstitute.org/) as well as novel ORF VUSs from the Mayo Right 1K and Right 10K projects.6, 9, 18 The RIGHT 1K samples served as a control because we had already studied them individually.9 We found that 19 of 109 CYP2C9 and 36 of the 121 CYP2C19 missense variants that were studied displayed less than ~ 25% of the wild‐type (WT) protein expression, a level that might have clinical relevance.9 We also compared variant calling by the DMS method with the predictions of computational algorithms and, finally, we validated serverely damaging variants by the use of Western blot analysis. Our findings suggest that DMS can be an efficient method for the high‐throughput identification of low protein abundance ORF variants that might have potential clinical implications.

Methods

Generation of landing pad HEK 293T cells

The landing pad construct and the promoter‐less cassette for CYP ORFs was created by Gibson assembly (Supplemental Text). TALENs were used to create double stranded breaks at the AAVS1 site and homology‐directed repair was used to introduce landing pad constructs into HEK 293T cells.15 The landing pad construct expressed doxycycline inducible blue fluorescent protein (BFP), which was used as a selection marker for landing pad insertion. Thirty percent of HEK 293T cells display hypotriploid karyotypes.19 The assay requires that a single variant be integrated per cell. Single cells were sorted into each well of a 96‐well plate for cloning. To identify which cells contained a single landing pad, we used Bxb1 recombinase mediated integration of a 1:1 ratio mixture of GFP and mCherry promotor‐less plasmids and analyzed fluorescence by flow cytometry. Clones having the lowest percentage of both GFP and mCherry were most likely to be candidate for further assay. The attB‐mCherry and attB‐GFP plasmids were transfected 24 hours after Bxb1 recombinase transfection, and BFP induction by doxycycline. After 5 days, candidate clones were trypsinized, washed with phosphate‐buffered saline, and were fixed in 4% formaldehyde at 4°C for 10 minutes. The cells were analyzed by flow cytometer FACS CantoX (BD Biosciences, San Jose, CA) and by the use of FACSDiva version 8.0 software and Flowjo software version 10 (BD Biosciences). The FACS CantoX instrument utilizes colinear 405, 488, and 561 nm lasers plus forward and side angle light scatter. Fluorescence images of variants were visualized using fluorescence microscopy (EVOS FLoid Cell Imaging Station; Life Technologies, Waltham, MA).

Nicking mutagenesis for variant library preparation

Nicking mutagenesis methods were modified from Whitehead et al. to construct variant libraries for ORFs containing CYP2C9 and CYP2C19 missense variants.17 Nicking mutagenesis uses Nt.BbvCI and Nb.BbvCI sites and exonuclease cleavage to degrade WT template DNA. Nt.BbvCI enzyme with nicking endonuclease and exonuclease digestion of WT template DNA was used to form single stranded DNA. At the annealing step, five oligos carrying the variant of interest were annealed separately to single strand templates and then pooled with five reactions in one pot, using high fidelity DNA polymerase and Taq ligase to close the double strand. The second template strand was degraded by Nb.BbvCI endonuclease cleavage and exonuclease digestion, and a new second strand was synthesized with a common primer on a cassette plasmid backbone. Phosphorylated oligos for CYP2C9 and CYP2C19 variants were purchased from IDT (Coralville, IA). Sanger sequencing was used to validate the sequences of variant clones. Detailed protocols are provided in the Supplemental Text.

Fluorescence‐activated cell sorting

Library cells were washed, trypsinized, and resuspended in phosphate‐buffered saline containing 5% fetal bovine serum. Cells were then sorted on an FACSAria with 407, 488, and 532 nm lasers (BD Biosciences) into four bins, and the cells were collected in culture medium. BFP−/mCherry+ cells containing CYP2C9 or CYP2C19 variants were flow sorted and grown for 5 days. BFP‐/mCherry+ cells were sorted again to determine the protein expression of CYP2C9/CYP2C19 variants based on their GFP/mCherry ratios. Gates were set based on GFP/mCherry ratios for cells integrating known CYP2C9/CYP2C19 variants and WT proteins as gating references. Four gates were set to dissect the pooled libraries into four different bins based on GFP/mCherry ratios. Data were analyzed by FACSDiva version 8.0.1 software.

Variant calling

Variant frequencies were calculated by high‐throughput sequencing of the DNA collected in each sorted bin. Genomic DNA was extracted from sorted cells using DNA extraction kits (Qiagen) and amplicons were sequenced as decribed in the Supplemental Text. Fastq files were aligned with respective CYP2C9 and CYP2C19 reference sequences using BWA mem aligner version 0.7.15. Samtools mpileup version 1.5 was used with a custom python script for single nucleotide variation calling. VarScan pileup2indel version 2.3.9 was used for calling INDELS. A base quality score cutoff of 20 and a mapping quality score cutoff of 20 were applied for both single nucleotide variation and INDEL calling. Custom scripts were used to summarize the data and add allele frequencies at all positions in the reference sequence.Variant counts in each bin were tabulated and each variant’s frequency in each bin was calculated. The effects of variants were obtained based on the frequency of variants in each bin.

Results

Generation of landing pad cell lines

The landing pad construct was generated and a promotor‐less cassette was constructed with ORFs for CYP2C9 and CYP2C19. The ORFs were fused to GFP as an indicator of protein expression and mCherry, which was independently expressed as a transfection control. The DMS assay requires that only a single transgenetic cassette be integrated per cell, which could be guaranted by having only a single landing pad per cell (Figure 1 a). To generate a cell line that integrated only a single copy of the landing pad, the landing pad construct was inserted into HEK 293T cells and cells were selected by BFP. Single cells positive for BFP were sorted into each well of a 96‐well plate for cloning. We observed different levels of BFP expression among different landing pad clones, probably because HEK 293T cells have been reported to have hypotriploid karyotypes.19 To identify a candidate clone with only a single landing pad, we selected > 30 clones with low BFP expression levels. We then tested those clones by using bxb1 recombinase to integrate a mixture of 1:1 ratio GFP only and mCherry only promotor‐less plasmids (plasmid sequences are shown in the Supplemental Text) into landing pad clones, followed by flow cytometry analysis. If a candidate clone contained only a single landing pad, it would integrate either GFP or mCherry, but not both. For clone 20, quadrant Q2 for the flow cytometry showed a negligible percentage of cells with both GFP and mCherry expression (Figure 1 b). In addition, no overlay of green and red was observed in the fluroscence image for clone 20, once again indicating that clone 20 had only a single landing pad (Figure 1 c). As a result, clone 20 was selected for use in the following experiments. This system allowed us to monitor variant protein expression in a high‐throughput fashion.

Figure 1

Generation of HEK 293T landing pad cells. (a) Plasmid maps of the landing pad construct and the promoter‐less cassette for CYP open reading frames that were fused to green fluorescent protein (GFP) and engineered for the simultaneous expression of IRES‐mCherry. (b) Flow cytometry results for landing pad HEK 293T clone 20 that had a low percentage of both GFP and mCherry, compatible with the conclusion that clone 20 contains a single landing pad. (c) Merged fluorescence photos showing that mCherry and GFP did not superimpose for clone 20. The landing pad clone 20 integrated either GFP or mCherry, which indicated that clone 20 contained a single landing pad. BFP, blue fluorescent protein; CMV, human cytomegalovirus promotor; IRES, internal ribosome entry site.

Effect of CYP2C9 and CYP2C19 variants on protein levels

In our previous study, CYP2C9 218C>T, CYP2C9 343A>C, CYP2C19*3 (636G>A), and CYP2C19 815A>G affected final protein quantity, resulting in varying levels of decreased protein expression.9 In the present study, the reporter consisted of a promoter‐less cassette containing the C‐terminus of the CYP2C9 or CYP2C19 ORFs fused to GFP. That construct was expressed once it was integrated into landing pad cells to ensure one variant per cell and the landing pad BFP was disrupted. Flow cytometry analysis of BFP‐/mCherry+ cells showed that known “damaging” variants expressed significantly reduced levels of GFP and lower GFP/mCherry ratios, indicating that those cells expressed less protein. Mean GFP/mCherry ratios of known damaging variants were compared with those of WT proteins. The value for CYP2C9 218C>T was 28.4%; CYP2C9 343A>C was 48.7%; CYP2C19*3 (636G>A) was 17.6%; and CYP2C19 815A>G was 61.2% of the mean WT GFP/mCherry ratios, percentages that were roughly identical with Western blot results that we had previously reported (Figure 2 a–c).9 Next, we created constructs for nonsynonymous variants with allele frequencies > 0.00001 as reported by the Exome Aggregation Consortium and by the Mayo Clinic Right 1K study and created variant libraries for CYP2C9 and CYP2C19. Pooled variant libraries for both genes were integrated into landing pad cells. Using already known damaging variants for CYP2C9 and CYP2C19 as well as WT constructs as references for FACS gating, the cells were sorted into different “bins” based on GFP/mCherry ratios (Figure 2 d). Variants of CYP2C9/CYP2C19 with the lowest GFP/mCherry ratios (< 25% protein expression) were sorted into bin1 and were classified as severely damaging. WT‐like variants were sorted into bin4. The gating for four‐way sorting of CYP2C9/CYP2C19 pooled variants libraries is shown in Figure 3 a,b. Pools of cells in each bin were collected and were then used as input material for DNA sequencing. We used NGS to monitor the frequencies of variants in each bin. Variant frequencies (F v) appearing in each bin were then used to determine protein abundance scores. Abundance scores for each CYP2C9 and CYP2C19 variant were calculated by use of the following equation:

Figure 2

Flow cytometry of cytochrome P450 family 2 subfamily C member 9 (CYP2C9) and cytochrome P450 family 2 subfamily C member 19 (CYP2C19) constructs with known variants. (a,b) Flow cytometry analysis of blue fluorescent protein (BFP)‐/mCherrry+ cells that integrated wild‐type or known damaging variants, CYP2C9 218C>T and CYP2C9 343A>C, CYP2C19*3 and CYP2C19 815A>G. Note that for both wild‐type (WT) allozymes, most of the cells eluted toward higher green fluorescent protein (GFP)/mCherry ratios, while allozymes containing damaging variants eluted at significantly lower GFP/mCherry ratios than did the cells containing the WT. Mean GFP/mCherry ratios for those variants were consistent with Western blot results obtained during our previous study.9 (c) Cells transfected with constructs expressing known severely damaging variants (CYP2C9 218C>T and CYP2C19*3) eluted at low GFP/mCherry ratios, whereas other variants eluted between the two extremes (CYP2C9 343A>C and CYP2C19 815A>G), compatible with their being “damaging” variants. Four “bins” were established based on WT CYP2C9 and CYP2C19 and known damaging CYP2C9 and CYP2C19 variants, as shown diagramatically in (d).

Figure 3

Fluorescence‐activated cell sorting‐sorting of pooled cytochrome P450 family 2 subfamily C member 9 (CYP2C9) and cytochrome P450 family 2 subfamily C member 19 (CYP2C19) variant libraries. Blue fluorescent protein−/mCherrry+ cells integrating CYP2C9 and CYP2C19 pooled variant libraries were sorted into four bins based on their green fluorescent protein (GFP)/mCherry ratios. Gates were set based on wild‐type CYP2C9 and CYP2C19 and known damaging CYP2C9 and CYP2C19 variants. Pools of sorted cells in each bin were collect and were used as input material for subsequent amplicon DNA sequencing.

For each experiment, an “abundance score” for each variant studied was obtained by multiplying the variant frequency by weighted values (0.25–1) with bin1 assigned 0.25 and bin4 with 1.0.16 The final abundance score for each variant was calculated by averaging mean abundance scores across replicate assays. Variants were scored in at least three experiments, as shown graphically in Figure 4 and Figure . Variants were classified as “severely damaging,” “damaging,” or “tolerated,” with thresholds chosen on the basis of abuandance scores for known CYP2C9/CYP2C19 variants.9 Using Western blot results and corresponding enzyme activities from our previous publication as a reference,9 CYP2C9 variants with abundance scores below 0.578 (CYP2C9 1076T>C), were classified as “severely damaging,” displaying ~ 25% protein abundance as compared with WT, whereas those with abundance scores above that threshold but lower than 0.670 (CYP2C9 343A>C) were classified as “damaging” displaying ~ 50% of the protein abundance of the WT allozyme. Variants with scores above this threshold were considered “tolerated.” CYP2C19 variants with abundance scores below 0.597 (CYP2C19 1349C>A), were classified as “severely damaging,” whereas those with abundance scores above this threshold, but lower than 0.635 (CYP2C19 557G>A), were classified as “damaging.” Variants with scores above this threshold were considered to be “tolerated.” We found that 19 of 109 CYP2C9 and 36 of 121 CYP2C19 missense variants displayed less than ~ 25% of WT protein expression.

Figure 4

Protein abundance scores for 109 cytochrome P450 family 2 subfamily C member 9 (CYP2C9) and 121 cytochrome P450 family 2 subfamily C member 19 (CYP2C19) variants. (a) Abundance score values for CYP2C9 and CYP2C19 variants. Variants having abundance scores lower than that for CYP2C9 (1076T>C) were classified as severely damaging, whereas variants having abundance scores lower than that for CYP2C9 (343A>C) were classified as damaging variants. The results shown are averages for four replicates. (b) Mean abundance scores for CYP2C19 variants are shown in the histogram. Variants having abundance scores lower than that for CYP2C19 (1349C>A) were classified as severely damaging, whereas those with abundance scores lower than that for CYP2C19 (557G>A) were classified as damaging variants. The results shown are averages abudance scores for four replicates. SD values are listed in Tables and .

The variant calling results obtained by the use of DMS for variants from the ExAC study that had allele frequencies > 0.00001 and variants from the Mayo RIGHT 10K study are listed in Tables 1 and 2. The DMS results were also compared with predictions obtained using SIFT, Provean, and Polyphen2, and those results are listed in Tables , and , for CYP2C9 and CYP2C19, respectively. For variants that resulted in dramatically reduced protein expression levels, CYP2C9*11, CYP2C9*21, and another 12 CYP2C9 variants, as well as CYP2C19*8, CYP2C19*22, and another 26 CYP2C19 variants, the results were in good agreement among all three of these predictive algorithms. However, five CYP2C9 variants and seven CYP2C19 variants displayed < 25% of WT protein expression, whereas SIFT predicted that they were “tolerated.” In summary, we found that the three commonly applied algorithms, which we tested on our 230 variants, disagreed among themselves 30.4% of the time, and they disagreed with our DMS assays, for at least one algorithm, 54.7% of the time. We also searched PharmVar, a repository for pharmacogene variation that records functionally validated variants. For CYP2C9*21 and CYP2C19*2B, *8 and *22, our results were consistent with those reported in PharmVar.20 However, PharmVar currently lists several allozymes for both CYP2C9 and CYP2C19, which we found to be severely damaging as having only limited or moderate evidence. As a result, our results provide additional information with regard to the functional implications of these variants. Finally, 15 CYP2C9 and 30 CYP2C19 variants that we found to be severely damaging had not previously been reported or were reported to have uncertain function in PharmVar.20

Table 1

Protein abuduance scores of CYP2C9 variants from ExAC browser and Mayo Right 10K study

Exact cDNA	Exact amino acid	RSID	Common allele name	Allele frequency	Right 10K (variant prevalence)			DMS			PharmVar
					WT	Heterzygous	Homozygous		Functional study	Abundance score
c.752A>G	p.His251Arg	rs2256871	CYP2C9*9	0.006715	10079	5	0	Damaging		0.653893822	No record
c.449G>A	p.Arg150His	rs7900194	CYP2C9*8	0.005242	10074	1	0	Damaging		0.617433538	Uncertain function
c.1003C>T	p.Arg335Trp	rs28371685	CYP2C9*11	0.003784	10033	51	0	Severely Damaging		0.549201141	possibly decreased
c.374G>A	p.Arg125His	rs72558189	CYP2C9*14	0.002953	10081	3	0	Damaging		0.627008929	No record
c.1465C>T	p.Pro489Ser	rs9332239	CYP2C9*12	0.001912	10007	77	0	Damaging		0.583353141	No record
c.1080C>G	p.Asp360Glu	rs28371686	CYP2C9*5	0.001163				Damaging		0.629538267	Moderate evidence
c.801C>T	p.Phe267hPhe	rs149158426		0.000855				Damaging		0.597268637	No record
c.835C>A	p.Pro279Thr	rs182132442	CYP2C9*29	0.0004702	10080	4	0	TOLERATED		0.672275959	possibly decreased
c.1324G>A	p.Gly442Ser	rs368545396		0.0004286	10072	12	0	Damaging		0.603761357	unknown function
c.14T>C	p.Val5Ala	rs138957855		0.0003048	10077	7	0	TOLERATED		0.688673248	No record
c.394C>T	p.Arg132Trp	rs199523631	CYP2C9*45	0.0002804	10078	6	0	Damaging		0.639522141	unknown function
c.1264A>G	p.Ser422Gly	rs776769484		0.0002477	10082	2	0	Damaging		0.627667871	No record
c.895A>G	p.Thr299Ala	rs72558192	CYP2C9*16	0.0002473				TOLERATED		0.673073074	No record
c.520C>G	p.Pro174Ala	rs199539783		0.0001649	10079	5	0	Damaging		0.651742423	No record
c.269T>C	p.Leu90Pro	rs72558187	CYP2C9 *13	0.0001401				Damaging		0.616189783	possibly decreased
c.431G>A	p.Arg144His	rs141489852		0.0001401				Damaging		0.651784299	No record
c.1084C> G	p.Leu362Val	.		0.000132				Damaging		0.586379329	No record
c.1088C>T	p.Pro363Leu	rs150663116		0.0001319	10082	2	0	Damaging		0.581270896	possibly decreased
c.449G>T	p.Arg150Leu			0.0001319	10074	9	0	TOLERATED		0.689018975	No record
c.290G>C	p.Arg97Thr	.		0.0001236				Damaging		0.608683972	No record
c.515G>C	p.Cys172Ser	rs147617899		0.0001155	10079	5	0	TOLERATED		0.748597212	No record
c.1034T>C	p.Met345Thr	.		0.0001154				Damaging		0.601859628	No record
c.1341A>C	p.Leu447Phe	rs59485260		0.0001072	10079	5	0	Damaging		0.609375	No record
c.343A>C	p.Ser115Arg	rs771237265		0.000101	10082	2	0	Damaging		0.670161008	No record
c.1095C>A	p.Ser365Arg	rs139532088		0.00009894				Damaging		0.578622159	Uncertain function
c.980T>C	p.Ile327Thr	rs57505750	CYP2C9*31	0.0000907				Damaging		0.611262013	No record
c.89C>T	p.Pro30Leu	rs142240658	CYP2C9*21	0.00009067	10082	2	0	Severely Damaging		0.483953225	Uncertain function
c.448C>T	p.Arg150Cys	rs17847037	CYP2C9*48	0.00009066				TOLERATED		0.710010892	No record
c.688C>G	p.His230Asp	.		0.0000837				Damaging		0.65625	unknown function
c.389C>T	p.Thr130Met	rs200965026	CYP2C9*44	0.00008248	10080	4	0	Damaging		0.639755882	No record
c.517G>A	p.Ala173Thr	rs373758696		0.00008247	10080	4	0	Severely Damaging		0.548579579	No record
c.395G>A	p.Arg132Gln	rs200183364	CYP2C9*33	0.00008246	10083	1	0	TOLERATED		0.711347628	No record
c.1362G>C	p.Gln454His	.	CYP2C9*19	0.00008243				Damaging		0.645897332	No record
c.680C>T	p.Pro227Leu	.		0.0000756				TOLERATED		0.705389146	possibly decreased
c.1439C>T	p.Pro480Leu	rs530950257		0.00007417	10082	2	0	Severely Damaging		0.479818446	possibly decreased
c.556C>T	p.Arg186Cys	rs150435881		0.00006602				Damaging		0.592060606	possibly decreased
c.1166T>C	p.Ile389Thr	.		0.00006597				Damaging		0.631849327	No record
c.1421A>G	p.Asn474Ser	rs141011391		0.00006593	10082	2	0	Damaging		0.647619048	No record
c.373C>T	p.Arg125Cys	.		0.00005775				TOLERATED		0.677726163	uncertain function
c.1187A>G	p.His396Arg	rs187793133		0.00005772				TOLERATED		0.700935174	No record
c.1370A>G	p.Asn457Ser	.	CYP2C9*61	0.0000577				Severely Damaging		0.543305795	uncertain function
c.263T>C	p.Ile88Thr	rs139656048		0.00005768				TOLERATED		0.676032437	No record
c.218C>T	p.Pro73Leu	rs762081829		0.0000569	10083	1	0	Severely Damaging		0.571077996	No record
c.595G>A	p.Glu199Lys	.		0.00004956				Damaging		0.651252305	No record
c.371G>A	p.Arg124Gln	rs12414460	CYP2C9*42	0.0000495				Damaging		0.656819836	uncertain function
c.458T>C	p.Val153Ala	rs373993395		0.00004945	10081	3	0	Damaging		0.58402074	uncertain function
c.719T>C	p.Met240Thr	.		0.00004163				Damaging		0.640173483	No record
c.619A>T	p.Ile207Phe	.		0.00004134				TOLERATED		0.674486125	No record
c.371G>T	p.Arg124Leu	rs12414460		0.00004125				Severely Damaging		0.566767249	uncertain function
c.539C>T	p.Ser180Phe	rs200149294		0.00004125	10083	1	0	Severely Damaging		0.568764708	uncertain function
c.164C>T	p.Thr55Ile	rs771905380		0.00004124	10082	2	0	TOLERATED		0.701629687	No record
c.1004G>A	p.Arg335Gln	.	CYP2C9*34	0.00004122				TOLERATED		0.695204529	No record
c.709G>A	p.Val237Ile	.		0.00003334				Damaging		0.634745896	No record
c.624G>C	p.Leu208Phe	.		0.00003308				TOLERATED		0.681854013	No record
c.538T>C	p.Ser180Pro	.		0.000033				Severely Damaging		0.529048923	No record
c.541A>G	p.Ile181Val	.		0.000033				TOLERATED		0.689873007	No record
c.1297C>T	p.Arg433Trp	.		0.00003299				Severely Damaging		0.564814815	No record
c.386T>A	p.Met129Lys	rs750097042		0.00003299	10083	1	0	Damaging		0.643411363	No record
c.122A>T	p.Asn41Ile	.		0.00003298				Severely Damaging		0.458665155	No record
c.1024A>G	p.Arg342Gly	.		0.00003298				Severely Damaging		0.575108971	normal function
c.1036C>T	p.Pro346Ser	.		0.00003298	10081	3	0	Damaging		0.654218921	No record
c.1429G>A	p.Ala477Thr	.	CYP2C9*30	0.00003297				Severely Damaging		0.558712121	uncertain function
c.632C>T	p.Pro211Leu	.		0.00002482				TOLERATED		0.692533067	No record
c.629G>A	p.Ser210Asn	.		0.00002481				Damaging		0.644966918	No record
c.600G>T	p.Lys200Asn	rs766903671		0.00002478	10082	2	0	Damaging		0.663003421	No record
c.370C>T	p.Arg124Trp	.	CYP2C9*43	0.00002475				TOLERATED		0.737650497	No record
c.1153A>T	p.Thr385Ser	.		0.00002474				Severely Damaging		0.509235183	No record
c.863A>G	p.Glu288Gly	.		0.00002474				Damaging		0.601528578	No record
c.1135T>C	p.Tyr379His	.		0.00002474				Damaging		0.66603031	No record
c.433G>T	p.Val145Phe	.		0.00002473				Severely Damaging		0.460289634	No record
c.1022A>G	p.Asp341Gly	.		0.00002473				Severely Damaging		0.470510412	No record
c.704A>G	p.Lys235Arg	.		0.00001668				TOLERATED		0.68009768	uncertain function
c.791T>G	p.Ile264Ser	rs761895497		0.00001656	10082	2	0	Damaging		0.599331078	uncertain function
c.773A>G	p.Asn258Ser	.		0.00001656				TOLERATED		0.684042216	No record
c.638T>C	p.Ile213Thr	.		0.00001655				Damaging		0.648500178	No record
c.637A>C	p.Ile213Leu	.		0.00001655				TOLERATED		0.676605037	No record
c.618G>T	p.Lys206Asn	.		0.00001654				TOLERATED		0.690691147	No record
c.356A>C	p.Lys119Thr	.	CYP2C9*41	0.00001651				Damaging		0.601033475	decreased function
c.1076T>C	p.Ile359Thr	rs56165452	CYP2C9*4	0.0000165				Severely Damaging		0.578003356	possibly decreased
c.389C>A	p.Thr130Lys	.		0.0000165				Damaging		0.668748634	unknown function
c.820G>A	p.Glu274Lys	.		0.0000165				TOLERATED		0.687495665	No record
c.1080C>A	p.Asp360Glu	rs28371686		0.00001649				Damaging		0.605882567	No record
c.137G>A	p.Gly46Asp	.		0.00001649				Damaging		0.611297817	No record
c.109C>T	p.Pro37Ser	.		0.00001649				Damaging		0.62487462	No record
c.1045G>A	p.Asp349Asn	.		0.00001649				Damaging		0.648954995	No record
c.908G>T	p.Ser303Ile	.		0.00001649				Damaging		0.653594136	No record
c.986G>A	p.Arg329His	.		0.00001649				Damaging		0.66037359	unknown function
c.1136A>G	p.Tyr379Cys	rs773704286		0.00001649	10083	1	0	Damaging		0.665754704	No record
c.1214A>T	p.Glu405Val	.		0.00001649				TOLERATED		0.675922464	No record
c.146A>G	p.Asp49Gly	.	CYP2C9*37	0.00001649				TOLERATED		0.678200758	uncertain function
c.422T>C	p.Ile141Thr	rs148615754		0.00001649				TOLERATED		0.681659253	No record
c.1159A>G	p.Ile387Val	rs764211126	CYP2C9*56	0.00001649	10083	1	0	TOLERATED		0.688530952	No record
c.968T>C	p.Val323Ala	.		0.00001649				TOLERATED		0.700837059	No record
c.989T>C	p.Val330Ala	.		0.00001649				TOLERATED		0.727107621	No record
c.257C>A	p.Ala86Asp	.		0.00001648				Severely Damaging		0.513086248	No record
c.38G>A	p.Cys13Tyr	.		0.00001648				Damaging		0.590756013	uncertain function
c.445G>A	p.Ala149Thr	.	CYP2C9*46	0.00001648				Damaging		0.604943716	No record
c.312A>T	p.Glu104Asp	.		0.00001648				Damaging		0.638095238	No record
c.271G>A	p.Gly91Arg	.		0.00001648				Damaging		0.651222774	No record
c.296T>A	p.Ile99Asn	.		0.00001648				Damaging		0.651222774	No record
c.1415T>C	p.Val472Ala	.		0.00001648				Damaging		0.668443691	No record
c.247G>A	p.Val83Met	.		0.00001648				Damaging		0.669561409	uncertain function
c.460G>A	p.Glu154Lys	.		0.00001648	10083	1	0	TOLERATED		0.688904038	decreased function
c.296T>C	p.Ile99Thr	.		0.00001648				TOLERATED		0.6911887	No record
c.8C>A	p.Ser3Tyr	.		0.00001648				TOLERATED		0.694461693	No record
c.791T>C	p.Ile264Thr	rs761895497		0.0000109	10082	2	0	TOLERATED		0.683929331	unknown function
c._del707	p.Asn236Thrfs*5	.		0.000004099				Severely Damaging		0.25	No record
c.709G>C	p.Val237Leu	.		Not recorded				Damaging		0.623703953	No record
c.229C>T	p.Leu77Met	.		Not recorded				TOLERATED		0.72516812	No record

DMS, deep mutational scanning; ExAC, Exome Aggregation Consortium; RSID, Reference SNP ID number; WT, wild‐type.

Table 2

Protein abuduance scores of CYP2C19 variants from ExAC browser and Mayo Right 10K study

Exact cDNA	Exact Amino acid	RSID	Common allele name	Allele Frequency	Right 10K (variant prevalence)			DMS		PharmVar
					WT	Heterzygous	Homozygous	Functional study	Abundance score
c.991G>A	V331I	rs3758581		0.06242	8802	1249	33	Damaging	0.599924823	Nomal function
c.276G>C	E92D	rs17878459	CYP2C19*2B	0.0236	9480	598	6	TOLERATED	0.713318402	No record
c.518C>T	A173V	rs61311738		0.005818	10080	4	0	TOLERATED	0.643519769	No record
c.481G>C	A161P	rs181297724		0.004786				Severely damaging	0.475593771	No function
c.449G>A	R150H	rs58973490	CYP2C19*11	0.002652	9985	98	1	TOLERATED	0.69271491	normal function
c.55A>C	I19L	rs17882687	CYP2C19*15	0.002127	10079	5	0	Damaging	0.623042396	normal function
c.358T>C	W120R	rs41291556	CYP2C19*8	0.00154	10045	39	0	Severely damaging	0.47978285	No function
c.1228C>T	R410C	rs17879685	CYP2C19*13	0.001517	10082	2	0	TOLERATED	0.676655086	Normal function
c.431G>A	R144H	rs17884712	CYP2C19*9	0.001079	10080	4	0	Damaging	0.621114695	decreased function
c.365A>C	E122A	rs17885179		0.0009637	10082	2	0	TOLERATED	0.753787879	No record
c.784G>A	D262N	.		0.0008586				Damaging	0.614038591	No record
c.448C>T	R150C	rs142974781		0.0004859	10083	1	0	Damaging	0.62953438	No record
c.680C>T	P227L	rs6413438	CYP2C19*10	0.000448	10079	5	0	Damaging	0.624712679	decreased function
c.985C>T	R329C	rs59734894		0.0003872	10078	6	0	Severely damaging	0.554502254	No record
c.394C>T	R132W	rs149590953		0.0003871	10081	3	0	Damaging	0.635242943	No record
c.241G>A	E81K	rs149072229		0.0003871	10068	16	0	TOLERATED	0.667726879	No record
c.374G>A	R125H	rs141774245		0.0002965	10062	21	0	Damaging	0.615675954	No record
c.337G>A	V113I	rs145119820		0.0002718	10079	5	0	TOLERATED	0.683885811	No record
c.1295A>T	K432I	rs146991374		0.000264				Damaging	0.609742945	No record
c.217C>T	R73C	rs145328984	CYP2C19*30	0.0002553	10082	1	0	Damaging	0.623329263	uncertain function
c.1078G>A	D360N	rs144036596		0.0002471				TOLERATED	0.693150526	None
c.395G>A	R132Q	rs72552267	CYP2C19*6	0.0002389	10082	2	0	TOLERATED	0.718026114	No function
c.164C>G	T55S	rs572853437		0.000132				TOLERATED	0.718673467	No record
c.557G>C	R186P	rs140278421	CYP2C19*22	0.0001236	10083	1	0	Severely damaging	0.560050139	No function
c.1004G>A	R335Q	rs118203757	CYP2C19*24	0.0001236	10080	1	0	Damaging	0.601312932	no function
c.557G>A	R186H			0.000108				Damaging	0.635198497	No record
c.831C>A	N277K	.		9.888E‐05	10080	3	0	TOLERATED	0.681062109	No record
c.1048G>A	A350T	rs201509150		9.884E‐05				Severely damaging	0.536022841	No record
c.1127T>A	F376Y	.		9.884E‐05	10083	1	0	TOLERATED	0.711485306	No record
c.25C>G	L9V	.		0.0000908				TOLERATED	0.731784659	No record
c.1007G>T	S336I	rs143833145		9.061E‐05				TOLERATED	0.637387252	No record
c.1036C>T	P346S	.		0.0000906				Severely damaging	0.580346079	No record
c.556C>T	R186C	rs183701923		8.242E‐05				Severely damaging	0.564313994	No record
c.389C>T	T130M	rs150152656		8.237E‐05	10082	2	0	TOLERATED	0.684802678	No record
c.527A>G	N176S	rs57700608		7.417E‐05	10074	10	0	TOLERATED	0.671875	No record
c.1075A>G	I359V	.		7.414E‐05				Damaging	0.603645833	No record
c.440A>G	E147G	rs147453531		7.413E‐05	10081	3	0	TOLERATED	0.643455877	No record
c.781C>T	R261W	.		6.605E‐05				Severely damaging	0.501301905	No record
c.836A>C	Q279P	rs61526399		6.591E‐05				Severely damaging	0.570907901	No record
c.1001A>T	N334I	.		0.0000659				TOLERATED	0.648759438	No record
c.221T>C	M74T	.		6.589E‐05				Severely damaging	0.58131891	No record
c.1324C>T	R442C	rs192154563	CYP2C19*16	5.772E‐05				Damaging	0.62128858	decreased function
c.1021G>C	D341H	rs770829708		5.766E‐05	10081	3	0	TOLERATED	0.656054948	No record
c.85C>T	L29F	.		4.946E‐05				TOLERATED	0.692408379	No record
c.837G>T	Q279H	.		4.944E‐05				TOLERATED	0.678912244	No record
c.1003C>T	R335W	rs368758960		4.942E‐05	10083	1	0	Severely damaging	0.497372294	No record
c.1034T>A	M345K	rs201132803		4.942E‐05	10080	1	0	Severely damaging	0.497916398	No record
c.371G>A	R124Q	rs200346442		4.942E‐05				Damaging	0.602183853	No record
c.925G>A	A309T	.		4.942E‐05				TOLERATED	0.636880409	No record
c.218G>A	R73H	.		4.119E‐05				Severely damaging	0.476206097	No record
c.1072T>C	Y358H	.		4.119E‐05				TOLERATED	0.648167478	No record
c.370C>T	R124W	.		4.118E‐05				Severely damaging	0.505962416	No record
c.726T>A	S242R	.		3.311E‐05				TOLERATED	0.702093879	No record
c.801C>A	F267L	rs377674118		3.303E‐05	10083	1	0	Severely damaging	0.537597325	No record
c.1171C>G	L391V	.		3.298E‐05				Severely damaging	0.569318436	No record
c.1205C>T	P402L	.		3.298E‐05				TOLERATED	0.642305732	No record
c.1216A>G	M406V	rs144056033		3.298E‐05	10082	2	0	TOLERATED	0.699731714	No record
c.1060G>A	E354K	.		3.295E‐05				Severely damaging	0.475659325	No record
c.896C>G	T299R	.		3.295E‐05				Severely damaging	0.531790263	No record
c.305T>C	L102P	.		3.295E‐05				Severely damaging	0.574479079	No record
c.905C>T	T302I	rs58259047		3.295E‐05				Damaging	0.602678118	No record
c.1120G>A	V374I	rs113934938	CYP2C19*28	3.295E‐05				Damaging	0.627074235	normal function
c.961G>T	A321S	.		3.295E‐05				TOLERATED	0.639949178	No record
c.928C>T	L310F	.		3.295E‐05	10082	2	0	TOLERATED	0.679301841	No record
c.907A>G	S303G	.		3.295E‐05				TOLERATED	0.717662336	No record
c.648C>G	C216W	.		0.0000274				TOLERATED	0.656662248	No record
c.671A>T	D224V	.		2.561E‐05				Severely damaging	0.53699972	No record
c.753C>G	H251Q	rs148247410		2.478E‐05				Severely damaging	0.563668153	No function
c.769A>T	I257F	.		2.477E‐05				TOLERATED	0.681306328	No record
c.631C>A	P211T	.		2.476E‐05				TOLERATED	0.659306032	No record
c.629C>A	T210N	.		2.476E‐05				TOLERATED	0.684317643	No record
c.1316G>T	G439V	.		2.474E‐05				TOLERATED	0.665474123	No record
c.1371C>A	N457K	.		2.474E‐05				TOLERATED	0.682756843	No record
c.1414G>A	V472I	.		2.473E‐05				TOLERATED	0.670673077	No record
c.60G>C	W20C	.		2.473E‐05				TOLERATED	0.672724033	No record
c.562G>A	D188N	rs370803989	CYP2C19*35	2.473E‐05	10080	4	0	TOLERATED	0.677997182	uncertain function
c.82A>G	K28E	.		2.473E‐05				TOLERATED	0.680943598	No record
c.185G>C	G62A	.		2.472E‐05				Severely damaging	0.508683191	No record
c.190G>A	V64M	rs150045105		2.472E‐05				TOLERATED	0.650769501	No record
c.848C>T	T283I	.		2.472E‐05				TOLERATED	0.65615054	No record
c.1037C>T	P346L	.		2.471E‐05				Severely damaging	0.502953907	No record
c.1080C>A	D360E	.		2.471E‐05				Severely damaging	0.549265027	No record
c.1078G>C	D360H	rs144036596		2.471E‐05	10081	3	0	Severely damaging	0.567361111	No record
c.373C>T	R125C	rs200150287		2.471E‐05	10083	1	0	TOLERATED	0.669159602	No record
c.850A>G	I284V	.		2.471E‐05				TOLERATED	0.685908298	No record
c.430C>T	R144C	.		2.471E‐05				TOLERATED	0.688887841	No record
c.721G>A	E241K	.		1.657E‐05				Damaging	0.619158302	No record
c.728A>G	D243G	.		1.655E‐05				TOLERATED	0.696277128	No record
c.1288G>C	A430P	.		1.652E‐05				Severely damaging	0.545128622	No record
c.813G>A	M271I	.		1.652E‐05				TOLERATED	0.663920987	No record
c.169C>G	L57V	.		1.651E‐05				Damaging	0.634196461	No record
c.1465C>T	P489S	.		0.0000165				Severely damaging	0.577527563	No record
c.1373T>G	L458R	rs761587034		1.649E‐05	10083	1	0	Severely damaging	0.51802352	No record
c.1349C>A	T450N	rs141690375		1.649E‐05				Severely damaging	0.597025262	No record
c.1439C>T	P480L	rs779501712		1.649E‐05	10083	1	0	Damaging	0.626171339	No record
c.1325G>A	R442H	rs138112316		1.649E‐05				TOLERATED	0.646762738	No record
c.65A>G	Q22R	rs144928727		1.649E‐05				TOLERATED	0.656465743	No record
c.1398C>A	D466E	.		1.649E‐05				TOLERATED	0.661470643	No record
c.593T>C	M198T	rs186489608		1.649E‐05				TOLERATED	0.667063402	No record
c.1330G>C	E444Q	.		1.649E‐05				TOLERATED	0.694466074	No record
c.1213G>C	E405Q	.		1.649E‐05				TOLERATED	0.711027827	No record
c.518C>A	A173D	rs61311738		1.648E‐05	10080	4	0	Severely damaging	0.499614976	No record
c.1076T>A	I359N	.		1.648E‐05				Severely damaging	0.522206232	No record
c.197C>G	T66S	.		1.648E‐05				Damaging	0.626403544	No record
c.862G>A	V288I	.		1.648E‐05				TOLERATED	0.673350956	No record
c.537C>G	C179W	.		1.648E‐05				TOLERATED	0.694502569	No record
c.865A>G	I289V	.		1.648E‐05				TOLERATED	0.714496776	No record
c.445G>A	A149T	.		1.647E‐05				Severely damaging	0.521050873	No record
c.905C>G	T302R	rs58259047		1.647E‐05				Severely damaging	0.521776385	No record
c.331G>A	G111R	.		1.647E‐05				Severely damaging	0.541454278	No record
c.1021G>A	D341N	.		1.647E‐05				Severely damaging	0.571470548	No function
c.409G>C	G137R	.		1.647E‐05				Severely damaging	0.571889977	No record
c.1013G>T	C338F	.		1.647E‐05				Damaging	0.609881164	No record
c.217C>A	R73S	rs145328984		1.647E‐05	10082	1	0	TOLERATED	0.651405999	No record
c.410G>A	G137E	.		1.647E‐05				TOLERATED	0.653990969	No record
c.326G>C	G109A	.		1.647E‐05				TOLERATED	0.658559311	No record
c.347A>G	N116S	.		1.647E‐05				TOLERATED	0.666187952	No record
c.271G>C	G91R	rs118203756	CYP2C19*23	1.647E‐05				TOLERATED	0.668239425	uncertain function
c.1002C>A	N334K	rs563052490		1.647E‐05	10081	3	0	TOLERATED	0.687674269	No record
c.1112C>G	T371S	rs568155950		1.647E‐05	10083	1	0	TOLERATED	0.715706169	No record
c.578A>G	Q193R			4.06E‐06				TOLERATED	0.658206476	No record

ExAC, Exome Aggregation Consortium; RSID, Reference SNP ID number; WT, wild‐type.

Validation of severely damaging variants of CYP2C9 and CYP2C19

In silico predictions were not always consistent with our DMS results, as described in the preceeding paragraph. That fact emphasizes the need for the validation of variant classification using DMS or other functional assays. Although the efficiency of calling damaging variants using DMS exceeded the throughput of classical mutagenesis methods, we still needed to confirm the accuracy of calling for the variants that we studied. Therefore, we validated our variant protein expression calls by the use of Western blot analyses. As shown in Figure 5 a,b, our newly identified severely damaging variants for CYP2C9 and CYP2C19 displayed significantly decreased protein expression (< 25% protein expression with the exception of CYP2C9 371G>T which had ~ 50% protein expression) compared with the WT protein—shown at the far right in each of the panels. The binning patterns for variant frequencies for selected allozyme are shown in Figures 5 c,d.

Figure 5

Western blot validation of cytochrome P450 family 2 subfamily C member 9 (CYP2C9) and cytochrome P450 family 2 subfamily C member 19 (CYP2C19) allozymes identified as containing severely damaging variants. (a,b) The protein expression of CYP2C9 and CYP2C19 in blue fluorescent protein−/mCherry+ cells integrating severely damaging variants for were validated by Western blot analysis. The mCherry and β‐actin were used as loading controls. Each image includes a control lane for wild‐type (WT) CYP2C9 or CYP2C19. (c,d) Variant frequencies for three representative “severely” damaging variants for CYP2C9 and CYP2C19 showing their distributions in each of the four bins.

Discussion

The functional characterization of ORF missense variants in clinically important pharmacogenes remains a major challenge for pharmacogenomics. In a recent study, we identified and functionally characterized six novel nonsynonymous ORF variants in CYP2C9 and seven in CYP2C19 based on Mayo Right 1K data.9 We found that the enzyme activities of those variants generally correlated well with protein expression levels.9 Missense variants in CYP2C9/CYP2C19 may alter protein folding, leading to decreased protein expression as a result of accelerated proteasome‐mediated degradation, a major factor responsible for decreased enzymatic activity in pharmacogenomics.9, 21, 22, 23 The loss of function of allozymes containing nonsynonymous CYP2C9/CYP2C19 ORF single nucleotide polymorphisms due to decreased protein expression made it possible to analyze their function by fluorescence reporter assays.9 Because of the very large number of missense variants in ORFs, it is practically difficult to link the genotypes of these variants to their functional phenotypes using “one‐at‐a‐time” expression systems. Fowler and colleagues developed DMS to analyze variants for all possible amino acid alterations in several genes. Saturation mutagenesis was also used for DMS in several previous studies.9, 24, 25 That approach has advantages for use in pre‐emptive pharmacogenomics and makes it possible to interpet variant function based on protein structrual mapping.16, 26 Degenerate codons were used to generate the saturation libraries but some variants of interest may be missed due to codon bias, with up to 30–50% of possible variants missing from the final data sets.16 CYP2C9 and CYP2C19 each have ORFs that are 1.6 kb in length, so it would be difficult to generate saturation mutation libraries. As a result, we chose to apply a modification of the nicking mutagenesis method developed by Whitehead et al.17 to create focused variant libraries for missense variants that had reported to occur in humans for use in our study. Specifically, we analyzed 230 nonsynonymous ORF variants for CYP2C9 and CYP2C19 from the ExAC study that had minor allele frequencies > 0.00001. All of those variants had been shown to occur in humans and were not so rare as to be “private.” FACS sorting was used to separate variants of differing protein expression levels, all of which were subsequently analyzed by NGS to make it possible to calculate the frequency of each of the variants studied in each of our four FACS bins (see Figures 2 and 3).

X‐ray crystal structures have been determined for CYP2C9 and CYP2C19.27, 28 Six substrate recognition sites (SRS) have been identified in CYP2C enzyme sequences: amino acids 96–117 (SRS1), 198–205 (SRS2), 233–240 (SRS3), 286–304 (SRS4), 359–369 (SRS5), and 470–477 (SRS6).29 Nineteen of the 75 CYP2C9 and 9 of the 58 CYP2C19 damaging variants listed in Tables 1 and 2, which displayed reduced protein expression of at least 50% protein, mapped to known SRS sites, so they may also influence substrate binding. Other damaging variants listed in Tables 1 and 2 fall outside of those sites but may influence activity due to the disruption of active sites, although they have no influence on protein abundance. For example, we have reported that CYP2C9 709G>C and CYP2C19 65A>G displayed significantly reduced enzyme activity, but their protein levels were similar to that of that the WT.9 In silico predictions have been widely applied to predict variation in protein structure and function. Our previous work and that of others supports the importance of the application of additional, functional methods to validate results obtained by using predictive algorithms.30, 31 Therefore, we compared variant calling by use of DMS with the predictions of computational algorithms, and differences were found between our results and those of prediction algorithms, as listed in Tables and . Those differences may be due to either the accuracy of the prediction algorithms or to underlying molecular mechanisms. For example, CYP2C9 709G>C and CYP2C19 65A>G and CYP2C19*13 have WT‐like protein expression but loss of enzyme activity.9, 32 We also determined whether variants identified by DMS were truly damaging by the use of Western blot analyses, as shown in Figure 5.

A limitation of DMS based on fluorescence is that some genes have been found to not be amenable to this assay, because damaging variants for those genes had similar fluorescent intensities as did WT proteins.16 We found that DMS seems to be most sensitive for screening for severely damaging variants, which displayed clear fluorescence separation from WT‐like variants. However, it required careful interpretation of intermediate‐fluorescing variants. The validation of functional studies for variants characterized in this fashion will be essential if we are to incorporate these results into clinical decision making and electronic health records. To validate the most severely damaging variants that we identified by DMS, we used Western blot assays as our standard functional assay for validation, even though those studies were laborious and time‐consuming—but still necessary at this time. Protein expression is obviously an important aspect of functional genomics—but only one aspect. Additional functional validation based on enzyme activities for different CYP substrates should be performed in the future to further extend the functional characterization of the variant allozymes that we have studied. The regulation of CYP activity is a complex process involving multiple mechanisms, which include transcripton regulation by nuclear receptors, such as the pregnane X receptor, the constitutive androstane receptor, the glucocorticoid receptor and by members of the TSPYL gene family.33, 34, 35, 36 DMS, as we have used it, has the limitation of not addressing upstream DNA sequence variants, such as CYP2C19*17 that results in an increase in transcriptional activity.37, 38, 39 High‐throughput methodology for studying variants outside of ORFs will obviously be required for the interpretation of CYP variants that map outside of protein coding sequences. As a result, DMS is not a “final answer” but rather represents a significant step forward in our efforts to link genomic variation to variation in drug response phenotypes.

In summary, we have identified and validated 15 CYP2C9 and 30 CYP2C19 severely damaging variants that had not previously been reported in PharmVar.20 Those variants are potentially clinically actionable. Functional studies of those variants showed decreased protein expression, which could result in decreased drug metabolism. Our results add information that may help to improve the accuracy of current prediction algorithms and they may also provide test data sets for machine‐learning methods that might “learn” to predict the effects of ORF missense variants. The Mayo Clinic recently expanded the RIGHT 1K study to include an even larger cohort, the RIGHT 10K study that includes an additional 10,085 DNA samples with sequencing data for 77 pharmacogenes.6 Single nucleotide polymorphisms from both RIGHT 1K and RIGHT 10K studies were included in our analyses. This same DMS methodology can now be implemented to study other important pharmacogenes and preemptive NGS Mayo RIGHT 10K data as ever larger numbers of ORF missense variants are identified.

Funding

This study was funded by National Institutes of Health (NIH) grants U19 GM61388 (The Pharmacogenomics Research Network), R01 GM28157, R01 GM125633, and T32 GM08685, and by the Mayo Clinic Center for Individualized Medicine.

Conflict of Interest

Both Drs.Weinshilboum and Wang are co‐founders of and stockholders in OneOme, LLC. All other authors declared no competing interests for this work.

Author Contributions

L.Z., V.S., J.Y., D.L., and R.W. wrote the manuscript. L.Z., V.S., J.Y., L.W., and R.W. designed the research. L.Z., I.M., S.D., and J.R. performed the research. L.Z., V.S., and K.K. analyzed the data. All authors have given final approval of the manuscript for submission.

Figure S1.

Click here for additional data file.

Table S1.

Click here for additional data file.

Table S2.

Click here for additional data file.

Table S3.

Click here for additional data file.

Table S4.

Click here for additional data file.

Table S5.

Click here for additional data file.

Table S6.

Click here for additional data file.

Supplemental Text.

Click here for additional data file.