Between November 2009 and August 2010 at the Asan Medical Center neurology outpatient clinic, we prospectively enrolled patients who were diagnosed as having ischemic stroke confirmed by brain imaging studies, aged ≥18 years, and administered warfarin regularly under strict instructions for secondary prevention of ischemic stroke. Although INR level 2-3 has been considered 'optimal' for secondary prevention of stroke,32 considering its variation secondary to diet and occasional higher dosages needed in certain patients, such as those with mechanical valve replacement,32 we considered the INR level 1.7-3.5 'relatively stable',33 and enrolled patients who displayed this INR level on more than three consecutive measurements over at least 3 months before enrollment. We excluded patients who were 1) considered incompliant or anticipated to alter anticoagulation therapy (i.e., due to scheduled surgery); 2) severely dementic or aphasic so that reliable communication was not possible; 3) taking long-term medication that could potentially influence anticoagulation intensity, such as anti-epileptics, antibiotics, anti-inflammatory drugs, and alcohol; 4) diagnosed with underlying diseases potentially affecting the level of coagulation, such as vitamin K deficiency, malabsorption syndrome, malignancy; and 5) those who refused to consent. Patients with non-detectable R and S warfarin levels on the sampling date were additionally excluded. The study was approved by Ethics Committee of the Asan Medical Center, and written informed consent was obtained from all the subjects.

We reviewed medical records and interviewed individual patients. Data obtained were reasons for prescribing anticoagulation (e.g., arrhythmia, valvular heart disease), recent myocardiac infarction, heart failure, recurrent strokes despite antiplatelet use or severe intracranial stenosis, daily warfarin dose prescribed, and the warfarin medication schedule. Demographic features and risk factors for stroke were also recorded including age, gender, hypertension (defined as receiving medication for hypertension or blood pressure >140/90 mmHg with repeated measurements), diabetes mellitus (defined as receiving medication for diabetes mellitus, fasting blood sugar ≥126 mg/dL or post-prandial 2 hours≥200 mg/dL with two separate measurements), hypercholesterolemia (defined as receiving cholesterol-reducing agents, overnight fasting low density lipoprotein (LDL) level >100 mg/dL or overnight fasting cholesterol level >200 mg/dL), and current cigarette smoking (defined as smoking within six months of enrollment). The body mass index (BMI) was measured in each patient.

Since 2000, we have provided patients receiving anticoagulation therapy official education with an instruction program. At the time of enrollment, a well-trained nurse confirmed that the patient had received the instructions. Blood samples (10 cc) were obtained 12 hours after warfarin intake. We measured INR using standard methods.34 To determine patient drug compliance, the plasma warfarin levels of patients (R- and S-warfarin) were measured using chiral-phase high-performance liquid chromatography-tandem mass spectrometry (LC-MS/MS; ThermoFinnigan, San Jose, CA, USA), as described elsewhere, with minor modifications.35,36 While all patients followed our diet education program, serum vitamin K levels (phylloquinone) were additionally measured to adjust for the possible confounding effects of diet on INR, using LC-MS/MS with atmospheric pressure chemical ionization, as described elsewhere.37

We assumed three variables as biological phenotypes influenced by genotype: INR, daily maintenance warfarin dose, and warfarin response index (WRI). WRI was defined as INR divided by the daily maintenance warfarin dose (mg). Since WRI seems to be a more reliable variable reflecting individual response than INR or warfarin dose (see below), the index was employed as the biological phenotype in subsequent analyses. Based on WRI data obtained on the date of blood sampling, patients were divided into three groups: high (upper, 33.3%), intermediate (33.4-66.5%), and low (lower 33.3%) using two cutoff values. Novel SNPs were examined in two (high vs. low) groups.

We searched known SNPs related with warfarin metabolism from published papers,31,38 database (http://www.ncbi/nlm.nih.gov/). We also sequenced whole VKORC1 genes for confirmation of the presence of unreported SNP or linkage between the high WRI and low WRI groups. We selected 26 SNPs in CYP2C9 (three sites; 430C>T (rs1799853), 1075A>C (rs10 57910), 1076T>C), VKORC1 (14 sites; 381C>T, 861C>A (rs17880887), 2653G>C, 3673A>G, 5496G>T, 5808T>G (r17882154), 6009C>T, 6484T>C (rs9934438), 6853C>T (rs17886369), 7566T>C, 8767G>C, 8814T>C, 9041G>A (rs17880624), 9071G>T), gamma-glutamyl carboxylase (GGCX, nine sites; eight SNPs and one microsatellite; 268A>T, 412G>A, 6258A>G, 8015C>T, 8016G>A, 8240A>G, 8445C> T, 13875A>G, CAA repeat).

Sequencing method (resequencing analysis for the detection of genetic variations) was used to determine genotype. For sequencing analysis, genomic sequences were obtained from the GenBank (http://www.ncbi/nlm.nih.gov/) database and PCR primers were designed using the Primer3 software (http://www.genome.wi.mit.edu/cgi-bin/primer/primer3_www.cgi) to amplify coding regions and promoter regions (≥500 bp upstream from exon 1) of VKORC1 and CYP2C9 genes (PCR primer information used in this study is available on request). PCR amplification reactions were performed in 20-µL total volume containing 20 ng genomic DNA, 0.15 µM of each primer, 100 µM of each dNTP, 1 unit of AmpliTaq Gold™ (Applied Biosystems, Foster City, CA, USA) and reaction buffer with 1.5 mM MgCl₂. PCR was performed at 94℃ for 5 minutes, followed by 35 cycles at 94℃ for 45 seconds, 55℃ for 30 seconds, 72℃ for 1 minute and final extension at 72℃ for 10 minutes. The PCR products (1 µL) were directly used as template for sequencing using BigDye™ terminator v2.0 cycle sequencing kit (Applied Biosystems) on an ABI3730 DNA sequencer according to manufacturer's instructions. DNA polymorphisms were identified using the PolyPhred program (http://droog.gs.washington.edu/PolyPhred.html) after sequence chromatograms were basecalled with the Phred program and assembled with Phrap. In the VKORC1, we searched unknown SNPs related with clinical phenotype. We selected high and low WRI groups and analyzed whole VKORC1 sequence to try and identify unknown SNP between groups. Also, we confirmed genotype using the Taq-Man SNP genotyping assay (Applied Biosystems) according to the manufacturer's instructions. Two polymorphic sites (rs7294 for VKORC1 gene and rs1057910 for CYP2C9) were selected and genotyped for reconfirmation. Fluorescence intensities were read on a 7900HT Sequence Detection System (Applied Biosystems), and genotypes were determined using SDS 2.1 software (Applied Biosystems).

Genotyping data were tested for deviation from the Hardy-Weinberg equilibrium using the Chi-square test. We analyzed log-transformed data of variables for biological phenotypes, specifically, INR, warfarin dose, and WRI, which were skewed. Fisher's exact test was used to compare the allelic frequencies of VKORC1 and CYP2C9 variants among patients with different variables. We calculated the effective sample size as 155.35 using calculation fomula39 (power 90%, α=0.05, correlation coefficients=0.1, Z β=1.282, Z α/2=1.96, Fisher's transformed Z=0.5*ln (1.1/0.9)=0.1003, sample size=([1.96+1.282]/0.1003)² +3=155.35).

Relationships between items such as age, body weight, BMI, and WRI and warfarin, respectively, were analyzed using Pearson correlation while the relationship between WRI and warfarin dose with items such as gender, hypertension, diabetes, hypercholesterolemia, and cigarette smoking were tested with t-test. Multivariate linear regression analysis was performed to determine the formula for warfarin dosage. All statistical analyses were performed using SPSS 14.0 (SPSS, Chicago, IL), with the level of statistical significance being set at P<0.05.