Gamma-glutamyl carboxylase and its influence on warfarin dose

Abstract

Background

Via generation of vitamin K-dependent proteins, gamma-glutamyl carboxylase (GGCX) plays a critical role in the vitamin K cycle. Single nucleotide polymorphisms (SNPs) in GGCX, therefore, may affect dosing of the vitamin K antagonist, warfarin.

Methods

In a multi-centered, cross-sectional study of 985 patients prescribed warfarin therapy, we genotyped for two GGCX SNPs (rs11676382 and rs12714145) and quantified their relationship to therapeutic dose.

Results

GGCX rs11676382 was a significant (p=0.03) predictor of residual dosing error and was associated with a 6.1% reduction in warfarin dose (95% CI: 0.6%-11.4%) per G allele. The prevalence was 14.1% in our predominantly (78%) Caucasian cohort, but the overall contribution to dosing accuracy was modest (partial R² = 0.2%). GGCX rs12714145 was not a significant predictor of therapeutic dose (p = 0.26).

Conclusions

GGCX rs11676382 is a statistically significant predictor of warfarin dose, but the clinical relevance is modest. Given the potentially low marginal cost of adding this SNP to existing genotyping platforms, we have modified our non-profit website (www.WarfarinDosing.org) to accommodate knowledge of this variant.

INTRODUCTION

Warfarin (Coumadin™ and others) is an anticoagulant with a narrow therapeutic window and wide inter-individual response. In 2007, and again in January 2010, the United States Food and Drug Administration approved label changes for warfarin to recommend lower initial doses for patients known to have single nucleotide polymorphisms (SNPs) in two genes: one affecting warfarin metabolism (CYP2C9), and the other affecting sensitivity (VKORC1) (1-3). Several pharmacogenetic algorithms incorporate SNPs in both genes (e.g. CYP2C9*2, *3 and VKORC1 -1639 G>A), but these SNPs alone explain only one-third of the variability in warfarin dose (4-8). The inclusion of non-genetic factors can explain an additional one-fifth of warfarin dosing variability (8, 9). Incorporating other SNPs may further improve the accuracy of warfarin dosing (10-13).

Because of the critical pharmacodynamic role gamma-glutamyl carboxylase (GGCX) plays in the generation of vitamin K-dependent proteins, SNPs in the GGCX gene may affect the dose of warfarin and other vitamin K antagonists. GGCX resides in the endoplasmic reticulum membrane, oxidizing reduced vitamin K to vitamin K-2,3-epoxide while adding a carboxyl residue to the gamma carbon on selected glutamic acids to produce functional clotting factors II, VII, IX, and X and other vitamin K dependent proteins (6, 14-16). Patients with defects in GGCX have elevated International Normalized Ratios (INRs), even in the absence of a vitamin K antagonist (17), and GGCX-knockout mice die at birth from massive hemorrhage (18).

Although previous investigators have examined GGCX polymorphisms in patients taking warfarin, initial reports of associations with warfarin dose have either been unsubstantiated or await validation. A SNP located in intron 2, rs12714145 G>A, was associated with a 3.3% increase in warfarin dose requirements per A allele in a Swedish study of 201 patients (6). This observation was not confirmed in two subsequent studies which genotyped only 186 European Americans and 318 Chinese patients, respectively (19, 20). However, if validated, the association would be important because this SNP explained 1.4-3.3% of the dose variability in the Swedish population (6).

More recently, GGCX rs11676382 C>G, located in intron 14, correlated with lower warfarin doses in a single-center study of 186 European Americans, explaining 2% of the total variance of the warfarin maintenance dose in a dominant genetic model (19). However, the putative association was not significant in an additive model or after adjusting for multiple comparisons. Furthermore, Lubitz et al. did not confirm the significance of this SNP in 145 multiethnic patients (21). The goal of the present study, therefore, was to further investigate the putative associations between warfarin therapeutic dose and GGCX rs12714145 and rs11676382 in a large, multiracial, multi-centered cohort.

MATERIALS AND METHODS

Study Design

This was a cross-sectional study of 985 patients who have been well characterized (8, 11, 22, 23). Patients provided written, informed consent for use of their DNA, and the participating institutions approved the protocol. We genotyped adult patients (Table 1) who had their warfarin therapy managed at one of three sites: the anticoagulation clinics affiliated with Barnes–Jewish Hospital at Washington University Medical Center in St. Louis, the Anticoagulation Clinics of North America (San Antonio), and subjects who participated in the PREVENT (PREvention of VENous Thromboembolism) study (23). A stable therapeutic warfarin dose was defined as a dose that was unchanged for at least six days and resulted in two or more consecutive therapeutic INR values at least 3 days apart.

Table 1.

Patient Demographics for GGCX analysis.

Variables
Demographic variables	N = 985
Age, mean (SD), y	60 (15)
Gender:
Women, N (%)	486 (49.3%)
Men, N (%)	499 (50.7%)
Race:
Caucasian, N (%)	772 (78.4%)
African-American, N (%)	180(18.3%)
Asian or East Indian, N (%)	3(0.3%)
Other or Unknown, N (%)	30 (3.0%)
Clinical variables
Therapeutic warfarin dose, geometric mean (SD), mg/d	4.8 (1.6)
Target INR, mean (SD), units	2.3 (0.4)
Body surface area, mean (SD), m2	2.0 (0.3)
Current smoker, N (%)	140 (14%)
Takes amiodarone, N (%)	33 (3%)

SD = standard deviation; INR = international normalized ratio

Genotyping

DNA was extracted from patients’ whole blood or buffy coat samples using Puregene Purification Kits (Gentra, now Qiagen, Hilden, Germany). Genotyping was performed for GGCX rs11676382 and rs12714145 using PCR and Pyrosequencing® (24, 25). PCR primers were designed using the UCSC Golden Path Human Genome Browser March 2006 Build (http://genome.ucsc.edu/cgi-bin/hgGateway) and GGCX Refseq NM_000821 (Appendix 1).

Statistical analysis

To test an additive effect of GGCX SNPs on warfarin dose requirements, we coded each GGCX SNP as 0, 1, or 2 for the number of variant alleles present. We then regressed the number of GGCX SNPs onto the residual of the prediction according to a validated pharmacogenetic algorithm (8, 21): residual=In(observed therapeutic dose)-In(predicted dose)= βo+β1*(number of alleles), where ln is the natural logarithm. By exponentiating the residual error, we obtain the ratio of the observed dose/predicted dose, which can be used to correct the validated algorithm for GGCX, when genotype is known (Appendix 2).

To reduce the effect of extreme therapeutic dose values on regression coefficients, we Winsorized the 6 doses that fell outside of the usual (1-18 mg/day) range. We tested the model fit using standard regression diagnostic methods. All analyses were performed using SAS version 9.1 (SAS Institute, Cary, NC).

RESULTS

Participants had an average age of 60 years and half were female (Table 1). One hundred seventeen (12%) inherited one or two copies of the GGCX rs11676382 G allele (Table 2). The geometric means (SD) of the therapeutic dose (mg/day), stratified by rs11676382, were: C/C 4.9 (1.6); G/C 4.1 (1.6); and G/G 3.8 (2.2). GGCX rs11676382 was a significant predictor of residual dosing error (p=0.03), and was associated with a 6.1% (95% CI: 0.6%-11.4%) reduction in warfarin dose, per G allele (Table 3). Adding rs1167382 to the validated pharmacogenetic algorithm (8) increased the R² by 0.2% in the total population and by 0.3% in Caucasians.

Table 2.

Genotype Frequencies.

GGCX Alleles	Caucasians	African-Americans	Asian or East Indians	Other or Unknown Race	Association (95% CI) with dose residual, per allele
rs11676382
C/C	662 (85.8%)	175 (97.2%)	3	28 (93.3%)
G/C	103 (13.3%)	5 (2.8%)	0	2 (6.7%)	6.1% (0.6%-11.4%)
G/G	7 (0.9%)	0 (0.0%)	0	0 (0%)
rs12714145
G/G	394 (51.0%)	56 (31.1%)	3	14(46.7%)
G/A	298 (38.6%)	91 (50.6%)	0	14 (46.7%)	NS
A/A	80 (10.4%)	33 (18.3%)	0	2 (6.7%)

NS = not significant; CI = confidence interval.

Table 3.

Adjustment Factors for Existing Pharmacogenetic Dosing Algorithms, Depending on GGCX Genotype.

GGCX rs11676382 genotype	Formula for Correction Factor*	Adjustment in African-Americans and Asians*	Adjustment in Caucasians*
CC	1 / (PC2 + 2PCPG × 0.939 + PG2 × 0.9392)	1.00	1.01
GC	0.939 / (PC2 + 2PCPG × 0.939 + PG2 × 0.9392)	0.94	0.95
GG	0.9392 / (PC2 + 2PCPG × 0.939 + PG2 × 0.9392)	0.88	0.89

^*multiply predicted dose by this correction factor.

^†P_C = allele frequency of the wildtype C allele.

^‡P_G= allele frequency of the G allele.

Half of the subjects (N = 492) carried either one or both copies of the GGCX rs12714145 A allele (Table 2), but it was not a significant predictor of dose residual (p=0.39). The geometric mean (SD) therapeutic dose was 4.7 (1.6) for subjects with either the GGCX rs12714145 A/A or A/G genotype, and was 5.1 (1.6) for those with G/G genotype.

DISCUSSION

Although GGCX rs12714145 was not confirmed as a predictor of therapeutic warfarin dose requirements, GGCX rs11676382 C>G was significantly associated with decreased warfarin requirements: each G allele was associated with a 6.1% reduction (e.g. 93.9% of prediction) in dose. The effect of GGCX rs11676382 C>G on R² was modest, however, and could explain why Lubitz et. al were unable to confirm this association in their smaller population of 145 diverse patients (21). Given the allele frequency of rs11676382 C>G, we estimate about 14% of Caucasians would carry this SNP, and without genotyping for it, their initial pharmacogenetic-based therapeutic warfarin dose estimates would be 6%–12% too high. Because of warfarin's narrow therapeutic index, even this modest over-estimation of warfarin dose could contribute to supratherapeutic INR values or bleeding.

With CYP2C9 (*2 and *3) and VKORC1 −1639 G>A SNPs explaining about one-third of the variability in therapeutic warfarin dose (4-6, 8, 26), and emerging data from recent high throughput genotyping studies of Swedish and Chinese populations (27-29), few other genes are likely to have major effects on warfarin dose. However, by genotyping for GGCX rs11676382 and other SNPs that raise R² modestly (10-13), one could increase the accuracy of existing pharmacogenetic algorithms significantly.

In their sample of 201 predominantly Swedish patients, Wadelius et al. 2005, found a 3.3% increase in dose per GGCX rs12714145 A allele (6, 30). While their univariate analysis was significant for this SNP (p=0.036), their multivariate analysis was not (6, 30). Reider et al., on the other hand, did not find any association with warfarin dose and rs10691423, a SNP that is in high linkage disequilibrium with rs12714145 (19). Similarly, our analysis of GGCX rs12714145 A SNP in a larger (n= 985), more diverse population found no association between this SNP and warfarin dose.

The scope of the present study was limited to validating two GGCX SNPs in a large and diverse cohort of therapeutic patients. We did not evaluate GGCX rs699664 (8016G>A; Q325R), which was associated with increased warfarin dosage in Japanese patients (31), even though prior studies in Caucasian populations did not confirm this putative association (19, 32, 33). A related limitation is that few Asian patients were included in the present study. However rs11676382 C>G has yet to be observed in Asian populations according to public databases such as HapMap Phase 3. Lastly, as with any warfarin related study, definitions of stable warfarin dose and selection of covariates vary among studies (34-38), and obfuscate attempts at repeatability.

In August, 2009 the United States Center for Medicare and Medicaid Services (CMS) announced it would reimburse for warfarin pharmacogenetic testing only in the setting of approved randomized clinical trials. We hope that CMS will remunerate laboratories involved in warfarin pharmacogenetic research not only for VKORC1 and CYP2C9 testing, but also for GGCX and other SNPs (10-13) that modestly improve accuracy. Although delaying warfarin therapy while waiting for a $400 genotype of 2 genes may not be cost effective now,(39) in the future, genotyping of multiple SNPs will become faster and less expensive, so the cost effectiveness of pharmacogenetic dosing will improve. Using all available pharmacogenetic information during warfarin initiation would allow for optimal dosing accuracy. With this vision in mind, we have updated our non-profit website, www.WarfarinDosing.org, to accommodate, but not require, GGCX rs11676382. We will use this website in our upcoming trial (ClinicalTrials.gov Identifier NCT01006733).

Table 4.

Warfarin Pharmacogenetics Snapshot: past, present and future.

1. What is known about this topic?

• Warfarin is an anticoagulant with a narrow therapeutic window and wide intra-individual response.

• In 2007 and again in 2010, the US FDA approved labeling changes for warfarin to recommend lower initial doses for patients known to have SNPs in genes affecting warfarin metabolism (CYP2C9) and sensitivity (VKORC1).

• SNPs in CYP2C9 and VKORC1 explain one-third of the variability in warfarin dose.

2. What does this paper add?

• GGCX rs11676382 was associated with a 6.1% reduction in warfarin dose (95% CI: 0.6%-11.4%) per G allele.

• Knowledge of GGCX rs11676382 genotype has been incorporated in www.WarfarinDosing.org for pharmacogenetic dose initiation.

• GGCX rs12714145 was not a significant predictor of warfarin dose