Introduction

Warfarin is an oral anticoagulant that is widely prescribed for the prevention of various thromboembolic events [1–5]. It is a drug with a narrow therapeutic index and, therefore, careful monitoring is needed together with accurate and frequent dose adjustment for the prevention of adverse drug reactions (ADRs) related to its use. The efficacy and safety of warfarin therapy are critically dependent on monitoring the prothrombin time (PT)—interpreted as the International Normalized Ratio (INR) [6]. Inter-individual variability to warfarin dose requirement is a major clinical problem and may occur due to the influence of many factors, such as age, sex, genetic variants, illness and drug interactions. Genetic variations in the genes encoding enzymes responsible for warfarin metabolism and its pharmacodynamic actions are major factors that influence the dose requirements.

The key enzymes involved in the pharmacodynamics and pharmacokinetics of warfarin include cytochrome P450 (CYP) 2C9 (CYP2C9), vitamin K epoxide reductase complex 1 (VKORC1), CYP4F2 and gamma glutamyl carboxylase (GGCX). Recent studies have demonstrated that pharmacogenetic models using certain genetic polymorphisms of CYP2C9 (CYP2C9*2 and CYP2C9*3), VKORC1 [−1639 G > A, (rs9923231) and other SNPs] and clinical factors facilitate more accurate predictions of warfarin dose in various populations [7–10]. However, to date few studies have included the CYP4F2 (rs2108622) variant in their models. Other studies have shown that genetic polymorphism of the gene encoding GGCX leads to diminished carboxylation and is associated with a lower warfarin dose [11–13]. However, these models were limited to the specified study populations.

South Indians constitutes about 20.97 % of the total population of India and reside in four states [Andhra Pradesh (6.99 %), Tamil Nadu (5.96 %), Karnataka (5.17 %), Kerala (2.75 %)] and a union territory [Pondicherry (0.10 %)] (http://www.censusindia.gov.in/2011-prov-results/paper2-vol2/data_files/India2/Table_1_PR_Districts_TRU.pdf). Although this population shares a common ancestry (Dravidians), South Indians of today differ in terms of language, culture and dietary habits with limited admixture. David Reich et al. [14] found that the ancestral South Indians were distinct from the ancestral North Indians and East Indians. In our previous studies, we demonstrated that South Indians show significant differences from Caucasians, African Americans, Asians and other ethnic population in terms of genetic polymorphisms of drug-metabolizing enzymes and other genes [15–18].

The pharmacogenomics of warfarin therapy has been well studied in many ethnic populations. However, little information on the association of warfarin therapy and genetic polymorphisms is available on the South Indian population. Hence, the aim of this study was to ascertain the influence of genetic variability of the CYP2C9, VKORC1, CYP4F2 and GGCX genes on warfarin maintenance dose requirement in the South Indian patients and to develop a pharmacogenetic algorithm to predict the required warfarin dose based on these genetic and clinical factors.

Materials and methods

Study subjects

Patients aged 18–65 years of either gender who were attending the outpatient clinics of the department of cardiothoracic and vascular surgery and department of cardiology at Jawaharlal Institute of Postgraduate Medical Education and Research (JIPMER) Hospital, Pondicherry were eligible for inclusion in the study. All patients were of South Indian ethnicity, which was assessed by a review of the family history of three generations living in four states (Tamil Nadu, Kerala, Karnataka and Andhra Pradesh) and a union territory Pondicherry. The study was approved by the Institute Ethics Committee and conducted according to declaration of Helsinki. The study was explained to all study participants and written informed consent was obtained.

All study participants were on stabilized anticoagulation with warfarin for prevention of thromboembolism due to various conditions (Table 1). The mean daily maintenance dose (mg/day) of warfarin was defined as “patients receiving the same dose of warfarin for a period of 3 months with three or more consecutive INR measurements in a 1-week interval within target range (2–3.5)”.

Table 1 Patient characteristics
Full size table

Dose modifications were made based on American College of Chest Physician (ACCP) guidelines (8th edition) for warfarin therapy. Data on each patient’s age, height, weight, body mass index, medication history, INR values and warfarin dose were obtained from the respective patient’s case records. Patients who were on concomitant therapy with drugs potentially interacting with warfarin, patients with liver or renal dysfunction, pregnant and lactating women, smokers and alcoholics were excluded from the study.

Genotyping of CYP2C9, VKORC1, CYP4F2 and GGCX

Samples of venous blood (5 mL) were collected from the study participants for estimation of PT and calculation of INR. Residual blood samples (cellular portion) after estimation of the INR were used for genotyping. DNA was extracted by the phenol–chloroform extraction procedure, and genotyping for polymorphisms of CYP2C9 (*2 and *3 alleles), VKORC1 (rs9923231, rs7196161, rs2884737, rs9934438, rs8050894, rs2359612 and rs7294), CYP4F2 (rs2108622) and GGCX (rs11676382) were performed in the Real-Time Thermo Cycler (model 7300; Applied Biosystems, Life Technologies Corp, Carlsbad, CA) using TaqMan SNP genotyping assays [Electronic Supplementary Material (ESM) Table 1]. The real time (RT)-PCR assays were carried out in triplicate in a 25-μL final volume that contained 12.5 μL of TaqMan universal PCR Master Mix (2×), 1.25 μL of 20× working stock of TaqMan SNP genotyping assay, 5.0 μL of genomic DNA diluted in DNAase-free water and 6.25 μL of MilliQ water (Millipore Corp, Billerica, MA). The thermocycling conditions consisted of one cycle at 95 °C for 10 min to activate the AmpliTaq gold DNA polymerase followed by 45 cycles of denaturation at 92 °C for 15 s and annealing/extension at 60 °C for 1 min. The genotype and allele calls were analyzed using Applied Biosystems PRISM 7300 Sequence Detection System (SDS) ver. 1.4.0 software.

Statistical analysis

GraphPad Instat® ver. 3.06 (GraphPad Software, San Diego, USA) and IBM® SPSS® Statistics ver. 19.0 (SPSS, Chicago, IL) software were used for statistical analysis. The genotype frequencies were analyzed for Hardy–Weinberg equilibrium using the chi-square test. The mean daily maintenance doses were compared between the genotype groups using the Kruskal–Wallis test (Dunn’s post hoc test) and Mann–Whitney U test.

Pairwise linkage disequilibrium (LD) pattern and haplotype frequencies were estimated using HAPLOVIEW ver. 4.1 (Daly Lab, Broad Institute, Cambridge, MA) [19]. All of the SNPs with minor allele frequencies of 0.01 % were excluded, and minimum haplotype frequency was set as 1 %. Haplotype blocks were defined using the solid spine rule incorporated by analysis in HAPLOVIEW software. Haplotypes were estimated by accelerated expectation–maximization (EM) algorithm in HAPLOVIEW. The confidence interval range for LD was set between 0.5 and 0.99. D′ values from 0.8–1.0 indicate strong LD between pair of SNPs, while a D′ value of < 0.8 indicates moderate LD and a D′ value of <0.2 indicates no LD.

The association between the genotype and drug dose was evaluated using linear regression analysis. Stepwise multivariate regression analysis was used to ascertain the influence of the independent variables (clinical and genetic) on the dependent variable (logarithmic transferred daily maintenance dose). All SNPs were included in univariate and multivariate analysis, the homozygote wild type, heterozygote and the homozygote variant genotypes were coded as 0, 1 and 2, respectively. Age, body weight, height and body mass index (BMI) were included as continuous variables. Clinical conditions were grouped and coded: (1) patients with various other conditions were coded as 1; those with mechanical heart valve replacement (required INR 3.0–3.5) were coded as 2. Green vegetable intake was categorized and coded as 0, no intake; 1, one time/week; 2 more than 1 time/week. A p value of <0.05 was considered to be statistically significant.

Results

Demographic and clinical characteristics of the subjects

A total of 257 patients were recruited for the study. Of these 17 patients were excluded due to either the lack of complete data or the lack of DNA samples as a result of technical problems, leaving 240 eligible patients ( 36.7 % male, 63.3 %, female). In approximately 74.2 % of patients, the indication for warfarin therapy was either rheumatic heart disease with mitral stenosis, mitral valve regurgitation or post atrial valve replacement. The demographic and clinical characteristics of the patients are presented in Table 1. All of the patients were under maintenance warfarin therapy for at least 3 months; the INR of these patients ranged from 2.0 to 3.5, and furosemide and phenoxymethyl penicillin were found to be co-prescribed with warfarin.

Genotype analysis

In our patient cohort, the genotype frequencies of CYP2C9*1/*1, CYP2C9*1/*2, CYP2C9*1/*3 and CYP2C9*2/*3 were 78.7, 6.7, 13.8 and 0.8 %, respectively, and homozygous CYP2C9*2/*2 and CYP2C9*3/*3 were not identified. The allele frequencies of CYP2C9*1, CYP2C9*2, and CYP2C9*3 were 88.9, 3.8 and 7.3 %, respectively. The genotype and allele frequencies of VKORC1, CYP4F2 and GGCX are given in Table 2. The genotype frequencies of all variants studied were found to be in Hardy–Weinberg equilibrium.

Table 2 Effects of the VKORC1, CYP4F2 and GGCX genotypes on mean daily warfarin dose in South Indian patients (n = 240)
Full size table

Haplotypes of VKORC1

All seven VKORC1 SNPs were in strong LD with each other. As a consequence, 26 haplotypes were inferred, with eight of these having a frequency of >1 %. The analysis revealed a very strong LD pattern (D′ > 0.8) between SNPs rs7294, rs2359612, rs8050894, rs9923231 and rs7196161 and a moderate LD pattern (0.5 < D′ < 0.8) between SNPs rs9934438 and rs2884737 (ESM Fig. 1). Among the eight major haplotypes, haplotype H1 (includes variant allele of rs7294) (ACGCTGT) was the most frequent (74.0 %) (ESM Table 2). The mean dose (5.46 ± 1.96 mg/day) in the H1 group was significantly different from that in the other haplotype groups, except for the H2 haplotypes. The reference haplotype H2 includes all of the wild-type alleles among the seven SNPs (GCGCTGT); the frequency of this haplotype was found to be 9.0 % and it was associated with a mean warfarin dose (5.83 ± 1.44 mg/day) that was significantly higher than the other combinations. The other haplotype groups consisted of variant alleles and were associated with significantly lower daily doses of warfarin.

The difference in daily maintenance dose of warfarin in the different genotype groups

The mean daily warfarin maintenance dose was found to be 4.7 ± 2.1 mg/day. Patients with the CYP2C9*1/*2, CYP2C9*1/*3 or CYP2C9*2/*3 variant genotype required a 51.0, 60.9 or 62.2 %, respectively, lower daily maintenance dose of warfarin (2.8 ± 1.5, 2.3 ± 1.3 or 2.2 ± 0.4 mg, respectively) than the normal CYP2C9*1*1 genotype group (5.2 ± 2.1 mg). The influence of CYP2C9 genotypes on mean daily warfarin dose is shown in Fig. 1. The relative influence of the VKORC1, CYP4F2 and GGCX genotypes on warfarin maintenance dose is given in Table 2. The effect of additive and dominant models on warfarin dose was studied only in the VKORC1, CYP4F2 and GGCX genotype groups. There were no homozygous variants observed in the VKORC1 rs2884737 and GGCX rs11676382 groups. Hence, the additive and dominant models were not compared in these genotypes. In other genotype groups we observed significant differences in warfarin doses in terms of additive and dominant models (Table 2).

Fig. 1
figure 1

Effect of cytochrome P450 2C9 (CYP2C9) variants on daily warfarin dose in a South Indian population (n = 240). Box and whisker plot showing the distribution of warfarin daily maintenance dose for CYP2C9 genotypes. Box Median and interquartile range (2.5–97.5 percentile), vertical line above and below boxes Minimum and maximum values, respectively, number above the whisker mean value ± standard deviation. p < 0.001 for *1/*1 vs. *1/*2 and *1/*1 vs. *1/*3; p < 0.05 for *1*/1 vs. *2*/3. No significant difference was found between the other genotypes

Full size image

Among the VKORC1 variants, the −1639 G > A (rs9923231) SNP and the intronic SNPs (rs2884737, rs9934438, rs80850894 and rs235912) were found to be associated with a significant reduction in daily dose. Patients having one defective allele in VKORC1 rs9923231, rs2884737, rs9934438, rs80850894 and rs235912 required a significantly lower daily maintenance dose of warfarin (47.97, 32.94, 48.41, 36.73 and 33.33 % mg/day, respectively) than patients carrying the wild-type allele. Patients having two defective alleles in rs9923231, rs9934438, rs80850894 and rs235912 required an 85.5, 88.9, 78.4 and 80.6 % lower daily maintenance dose of warfarin, respectively, than carriers of the wild-type allele. Among the study population no homozygous variant was identified in rs2884737. The VKORC1 rs7196161 and the untranslated region (UTR) 3′ variant rs7294 carriers required significantly higher doses of warfarin than the carriers of the wild-type genotype. Carriers of the variant allele in CYP4F2 (GA and AA) required significantly higher doses of warfarin (19.7 and 49.0 %, respectively) than carriers of the CYP4F2 GG genotype. In contrast, carriers of the GGCX mutant allele required significantly lower doses of warfarin (45.86 %) than carriers of the GGCX (CC) genotype.

Regression analyses

Univariate analysis was performed to determine the influence of 19 individual factors on daily warfarin dose requirement (Table 3). The results revealed that 17 of these variables were significantly associated with warfarin dose requirements. Vegetable intake and serum albumin levels did not significantly influence the daily maintenance dose of warfarin in the study patients, but only 19 (7.9 %) of the patients reported consuming green vegetables. Our analysis revealed that nongenetic factors (including age, sex, clinical condition, height, weight, BMI) and genetic [in CYP2C9 (*2 and *3), VKORC1 (rs9923231, rs2884737, rs9934438, rs80850894, rs235912, rs7196161 and rs7294), CYP4F2 (rs2108622) and GGCX (rs11676382) genes] factors significantly influenced the daily warfarin maintenance dose. The genetic variation of VKORC1 rs9923231 alone contributed up to 27.5 % of the variation in the required dose.

Table 3 Univariate regression analysis of clinical and genetic variables on mean daily warfarin dose requirement
Full size table

A multivariate regression analysis was performed by considering the significant variables identified in the univariate analysis. Only age, weight, clinical conditions and genetic variations in VKORC1 (rs9923231, rs2884737, rs9934438, rs80850894, rs235912, rs7196161 and rs7294), CYP2C9 (*2 and *3), CYP4F2 and GGCX were retained in the final model (Table 4). The multivariate regression analysis revealed that clinical and genetic factors together contribute 62.1 % (adjusted r 2 = 0.602, p < 0.0001) of the variation in daily maintenance dose of warfarin, with the genetic factors being the major predictors (46.6 %) of warfarin dose in our South Indian population. The stability of the multivariate model was confirmed by 1,000 bootstrap replications from the original data. The bootstrap validation was performed to determine the percentage variance in warfarin dose requirement, and the bootstrap estimate was found to be 100 %. There was a significant correlation between the actual dose and the predicted dose calculated by using the multivariate model (ESM Fig. 2). Our model predicted 63.5 % of the warfarin dose based on the linear regression curve estimation analysis.

Table 4 Factors jointly associated with warfarin dose requirement in South Indian patients by multiple linear regression analysis
Full size table

To determine the clinical relevance, the performance of clinical and pharmacogenetic algorithms were compared in the low-dose (≤17.5 mg/week), intermediate-dose (≥17.5 to ≤49.0 mg/week) and high-dose (≥49.0 mg/week) groups (Table 5). For patients who required ≤17.5 mg/week (31.7 % of the total study patients), the pharmacogenetic algorithm yielded a better prediction than the clinical algorithm, with the pharmacogenetic algorithm showing only a 38.2 % overestimation in the low-dose group and the clinical algorithm showing a 86.8 % overestimation. The pharmacogenetic algorithm was found to predict the low and higher doses better than the clinical algorithm. However, there was no significant difference observed between the two algorithms in the intermediate-dose group. The performance of the pharmacogenetic algorithm of our study was compared then with the pharmacogenetic algorithm developed by the International Warfarin Pharmacogenetics Consortium (IWPC). The pharmacogenetic algorithm of our study was found to perform better than the IWPC algorithm in all dose groups, with underestimation being significantly higher in the IWPC algorithm compared with our new algorithm.

Table 5 Percentage of patients with ideal, underestimated or overestimated dose of warfarin, as estimated by pharmacogenetic, clinical and IWPC algorithms (n = 240)a
Full size table

Discussion

The aim of our study was to develop a new dosing algorithm specifically designed for the South Indian population that considers clinical data and SNPs associated with warfarin sensitivity and resistance that have been largely unaccounted for in most of the previous studies. Our results in terms of genotype and allele frequencies of CYP2C9 and VKORC1 are in agreement with those of our previous studies. In the South Indian patients population of the present study, the warfarin dose requirement decreased with increasing age, and weight was a significant demographic factor that accounted for 9.2 % (p < 0.05) of the variability in warfarin dose. Among the genetic determinants of warfarin dose, CYP2C9, VKORC1, CYP4F2 and GGCX genetic variants accounted for 12.4, 32.4, 1.9 and 0.6 % of the variability, respectively. The GGCX genetic variant was found to be rare in our population, however its effect on dose variation was significant (univariate r 2 = 0.019, p < 0.05).

Many studies have reported that polymorphisms of VKORC1 are a predictive factor of warfarin dose requirement [8, 20–38]. Our study demonstrated that multiple SNPs of VKORC1 can be used to predict the warfarin maintenance dose. Of the seven VKORC1 SNPS, rs7294 G > A was a more frequent variant and associated with higher warfarin dose (haplotype H1). The wild-type allele (haplotype H2) was also associated with a higher warfarin dose requirement. The haplotypes observed in our study were similar to those observed in our previous study in the Tamilian population [39]. Lal et al. [40] reported that haplotype TCGTCA (H7) is more common in Indians. In agreement with this result, in our study population we found that the TCGTCAG haplotype (H1 but categorized as H7 in previous studies) was associated with a higher warfarin dose requirement and that our study subjects required an intermediate warfarin dose. Several studies have demonstrated that SNPs in the 5′ UTR VKORC1 influences warfarin sensitivity. Rieder et al. demonstrated that nine haplotypes constructed from ten polymorphisms in the VKORC1 gene were significantly associated with warfarin dose requirement [41]. In the same study, the haplotype containing the 1173 C allele was found to be associated with the requirement for a higher warfarin dose than the 1173 T allele. Our results are in agreement with these findings. Previous studies have also shown that CYP2C9 and VKORC1 genetic polymorphisms are associated with variations in warfarin dose of ⋍5–20 % and ⋍13–34 %, respectively [42, 43]. In our study we have observed that CYP2C9 and VKORC1 contributed up to 12.4 and 32.4 % of the variability in warfarin dose, respectively.

In our study we observed that the effect of genetic polymorphisms of CYP2C9 was lower than that of VKORC1 in terms of influencing warfarin maintenance dose. Many studies have been conducted in various ethnic populations, but the predicting factors in the respective pharmacogenetic models differ [20–23, 25–28, 30–32, 35–38]. Although most of the clinical variables in these algorithms are similar, there are large inter-ethnic variations in the genes evaluated for the pharmacogenetic predictive models. The common loci in these algorithms are CYP2C9*2, CYP2C9*3 and VKORC1 -1639 G > A. Only a few studies have included CYP4F2, GGCX and EPHX1 [43–45].

Large inter-ethnic variations in terms of distribution of the variant alleles have been found in each population studied. Asians completely lack carriers of the CYP2C9*2 allele, with the latter being more prevalent in Caucasians. The prevalence of the CYP2C9*2 allele has been reported to be 4 % in South Indians [17]. The prevalence of CYP2C9*3 has been reported to be 5.8 % in Caucasians, 5.3 % in Chinese, 2.3 % in Japanese and 8 % in South Indian populations. Further, the allele frequency of VKORC1-1639A has been reported to be 10.8 % in African Americans, 67.1 % in Asians, 38.2 % in Caucasians and 12 % in the Tamilian population [42]. The distribution of the variant alleles of CYP4F2 and GGCX has also been reported to vary among the world’s populations [13]. Further, evidence that ethnicity is a limiting factor of warfarin maintenance dose has been steadily accumulating. A study conducted in Malays, Chinese and Indians revealed that Indian patients required significant higher doses of warfarin than Malays and Chinese [24]. In this same study, Indians were found to have a larger number of CYP2C9 variant genotypes than Chinese and Malays, and Chinese and Malays had a larger number of VKORC1 variant alleles than Indians. The authors also demonstrated that there was a significant difference in warfarin dose among these different ethnic populations, with Asian patients requiring a lower warfarin dose than Caucasians and Indians [24]. In our study we observed that our South Indian patients required an intermediate warfarin dose.

The IWPC has developed a pharmacogenetic algorithm based on a retrospective study in a relatively large sample size [derivation group n = 4,043; validation group n = 1,009). This algorithm predictes the required weekly warfarin dose more accurately than the clinical algorithm and fixed dose approach [25]. Large randomized clinical trials (RCTs) are currently underway to quantify how the pharmacogenomics-based dose initiation affects laboratory and clinical outcomes. Also, two multi-center RCTs, namely Clarification of Optimal Anticoagulation through Genetics (COAG; USA trial) and Pharmacogenomics Approach to Coumarin Therapy (EU-PACT; European trial) (ClinicalTrials.gov Identifiers NCT00839657 and NCT01119300, respectively) were conducted to compare the clinical algorithms with pharmacogenetic algorithms [46, 47]. The studies were conducted on Caucasians, African Americans and Asians (Chinese and Japanese), but they did not focus on ethnic diversity. However, in our study the population-specific algorithms were more useful than the multi-ethnic algorithms. The United States Food and Drug Administration (US FDA) updated the label of warfarin twice: in 2007, advising physicians to consider the use of “genetic tests to improve their initial estimate” of the initial dosage, and in 2010, adding a new table with the range of expected therapeutic warfarin doses based on CYP2C9 and VKORC1 genotypes (http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/2007/ucm108967.htm).

A study in a South Indian population by Pavani et al. [29] derived a pharmacogenetic algorithm which predicted up to 44.9 % of warfarin dose requirement. These authors reported that VKORC1 -1639 G > A, VKORC1*3, VKORC1*4, CYP2C9*3, CYP2C9*2, male gender, vitamin K intake, age and body BMI were the major predictive factors and that CYP2C9*3, VKORC1-1639 G > A and VKORC1*3 polymorphisms were associated with warfarin sensitivity. However, in our study we found that age, body weight and clinical conditions, such as post mechanical heart valve replacement, were the important clinical factors to predict warfarin dose. The important genetic factors to predict the warfarin dose in South Indian patients were the genetic polymorphisms of CYP2C9 (CYP2C9*2 and CYP2C9*3), VKORC1 (rs9923231, rs7294, rs9934438 and rs2359612), CYP4F2 (rs2108622) and GGCX (rs11676382). The clinical and genetic factors together contributed to 62.1 % of the variability in the required warfarin dose in our South Indian patients. Furthermore, many studies have explained the variation in warfarin dose requirement based on genetic polymorphisms, such as those in CYP2C18, MDR1, EPHX1, Factor II, Factor X, PROC, PROS and other genes in the warfarin metabolic pathway [43, 44, 48].

One limitation of our study is that it was conducted under strict exclusion criteria. Hence, the study result does not consider the impact of environmental factors, such as smoking and the concomitant consumption of alcohol and drugs. However, the patients attending the anticoagulant clinic of our hospital were advised by clinical pharmacologists on oral anticoagulation therapy and the complications of smoking and alcohol with oral anticoagulation. However, co-morbid conditions, such as liver and renal dysfunction and drug interactions, were not included in our analysis. Further prospective studies with the addition of other genetic factors, such as SNPs in the CYP2C18, MDR1 and EPHX1 genes, a larger sample size and consideration of concomitant drugs and illnesses may significantly contribute to the improvement of our dosing model.

In conclusion, the results of our study explain the effects of CYP2C9, VKORC1, CYP2C9 and GGCX genetic polymorphisms on warfarin daily maintenance dose in a South Indian population receiving warfarin maintenance therapy. The pharmacogenetic algorithm established will be useful for predicting the required starting dose of warfarin in South Indian patients.