The atorvastatin metabolic phenotype shift is influenced by interaction of drug-transporter polymorphisms in Mexican population: results of a randomized trial

Atorvastatin (ATV) is a blood cholesterol-lowering drug used to prevent cardiovascular events, the leading cause of death worldwide. As pharmacokinetics, metabolism and response vary among individuals, we wanted to determine the most reliable metabolic ATV phenotypes and identify novel and preponderant genetic markers that affect ATV plasma levels. A controlled, randomized, crossover, single-blind, three-treatment, three-period, and six-sequence clinical study of ATV (single 80-mg oral dose) was conducted among 60 healthy Mexican men. ATV plasma levels were measured using high-performance liquid chromatography mass spectrometry. Genotyping was performed by real-time PCR with TaqMan probes. Four ATV metabolizer phenotypes were found: slow, intermediate, normal and fast. Six gene polymorphisms, SLCO1B1-rs4149056, ABCB1-rs1045642, CYP2D6-rs1135840, CYP2B6-rs3745274, NAT2-rs1208, and COMT- rs4680, had a significant effect on ATV pharmacokinetics (P < 0.05). The polymorphisms in SLCO1B1 and ABCB1 seemed to have a greater effect and were especially important for the shift from an intermediate to a normal metabolizer. This is the first study that demonstrates how the interaction of genetic variants affect metabolic phenotyping and improves understanding of how SLCO1B1 and ABCB1 variants that affect statin metabolism may partially explain the variability in drug response. Notwithstanding, the influence of other genetic and non-genetic factors is not ruled out.


Results
Study population. All participants were healthy non-related male Mexicans who identified themselves as mestizos. Most (93.1%) were residents from northeastern Mexican states, namely Nuevo Leon (83.3%), Coahuila (1.6%), Tamaulipas (3.3%) and San Luis Potosi (4.9%). The volunteers had similar anthropometric data, and no significant differences in body composition. No adverse effects due to drug administration occurred 6 . Pharmacokinetic analysis. Under controlled conditions, the mean ± standard deviation (SD) of adjusted pharmacokinetic parameters used for phenotyping were: maximum plasma concentration (C max ) = 41.70 ± 21.51 ng/mL, the area under the plasma ATV concentration-time curve (AUC) from time 0 to the time of last measurement (AUC 0-t ) = 143.35 ± 84.64 ng/mL*h, AUC from time 0 extrapolated to infinity (AUC 0-∞ ) = 155.84 ± 85.79 ng/mL*h, and the total drug clearance (Cl) = 0.65 ± 0.31 L/h*kg ( Table 1). The coefficient of variation (CV) was greater than 50% for all parameters, except for Cl, which was 48%.

Effect of gene polymorphisms and haplotypes on atorvastatin pharmacokinetics. Five gene
polymorphisms had a significant effect on different pharmacokinetic parameters. C max values were influenced by ABCB1-rs1045642 polymorphism under a dominant model. C/C carriers had a significant lower C max (P = 0.038) than C/T and T/T carriers (Table 2). With respect to SLCO1B1-rs4149056, C carriers (homozygous or heterozygous) had a higher AUC 0-t , AUC 0-∞ , Cl and volume of distribution (V d ) than the homozygous T genotype (P ≤ 0.038; Table 2). Regarding the metabolizer enzyme variants, heterozygous carriers (C/G) of CYP2D6-rs1135840 had higher elimination rate constant in the terminal drug phase (Ke) values, but significantly lower half-life (T 1/2 ) values than those with homozygous genotypes (P ≤ 0.022; Table 2). Similarly, COMT-rs4680 G/G carriers had a higher AUC 0-∞ , but lower Cl than A/A and A/G carriers (P ≤ 0.048; Table 2). We also found a significant effect of the CCTT haplotype (of the combination of ABCB1 and SLCO1B1 drug transporters) on C max , AUC 0-t , AUC 0-∞ and Cl parameters (P ≤ 0.040). Subjects with a CCTT genetic profile in drug transporter genes had lower C max , AUC 0-t , and AUC 0-∞ , but a higher Cl than those with other combinations ( Table 2). The regression analysis confirmed that three gene polymorphisms (ABCB1-rs1045642, SLCO1B1-rs4149056 and COMT-rs4680) had an impact on ATV pharmacokinetics ( Table 3). The regression analysis also revealed that CYP2B6-rs3745274 and NAT2-rs1208 are involved in pharmacokinetic variability. The CYP2B6-rs3745274 polymorphisms influenced the variability of C max (in the over-dominant model; Table 3) and NAT2-rs1208 variants affect AUC 0-∞ (under the co-dominant and dominant models; Table 3). No other polymorphisms had a significant effect on ATV pharmacokinetics.

Interaction of gene polymorphism on atorvastatin pharmacokinetics. The interaction of
ABCB1-rs1045642 with CYP2B6-rs3745274, under the dominant and over-dominant model, respectively, predicted 11.1% of the C max variability (Table 3). Furthermore, under the co-dominant model, the interaction of www.nature.com/scientificreports www.nature.com/scientificreports/ SLCO1B1-rs4149056 with NAT2-rs1208 predicted 19.7% of the AUC 0-∞ variation. In another approach to test gene polymorphism interactions, after a genotype adjustment of intermediate metabolizers, i.e. three subjects with a C/C genotype for ABCB1-rs1045642 were removed from the group, the normal and intermediate metabolizers became significantly different (Fig. 1b). The new mean ± SD for C max of intermediate metabolizers became 59.23 ± 20.12 ng/mL (cut-off 46.45-72.02 ng/mL) and the adjusted R-square increased to 0.700 (P < 0.005). In addition, CCTT carriers were only ones that were statistically different from other subjects (Table 4).
Phenotype-genotype association. When the association between phenotypes and genotypes was assessed, we found that, under the recessive model, SLCO1B1-rs4149056 is associated with fast/normal metabolizers while the presence of allele C is related to slow/intermediate metabolizers (Table 5). This association Pharmacokinetic profiles of ABCB1-rs1045642 genotypes: carriers of C/C genotype (blue) and carriers of C/T or T/T genotypes (orange) (d) Pharmacokinetic profiles of SLCO1B1-rs4149056 genotypes: carriers of C/C or C/T genotype (blue) and carriers of T/T genotypes (orange). Mean peak plasma concentration-time curves after single 80-mg dose of ATV. Data shown are mean ± standard error (SE) concentrations.

Discussion
Despite similar health conditions, body composition, and controlled experimental conditions, ATV pharmacokinetic parameters varied greatly among Mexican male individuals. We applied a novel approach (cluster analysis based on more pharmacokinetic parameters than a previous method) that distinguished four metabolizer phenotypes with higher accuracy and reliability than a previously reported classification method 6 . Furthermore, it allowed the selection of a model with the highest prediction percentage and best cut-off limits for each metabolizer phenotype. In addition, the analysis allowed identifying inconsistencies. For example, after phenotyping, considerable variability of C max values was observed in the intermediate metabolizer group, suggesting that there are important differences between individuals in the absorption process.
Although the clustering analysis method has been successfully applied for classification purposes in other areas of biomedical research [14][15][16] , it has been little used in pharmacokinetics, pharmacogenetics and pharmacogenomics 6 . A reason might be the need to realize multiple analyzes to select the ideal number of clusters. In our experience, the cluster analysis method proved to be a) effective in identifying the different groups within a Genotypes n C max (ng/ml) AUC 0-t (ng/ ml*h) AUC 0-∞ (ng/ ml*h) K e T 1/2 (h) Cl (L/h*kg) V d (L/kg)  www.nature.com/scientificreports www.nature.com/scientificreports/ population, b) useful to determine which are the groups or subgroups that require more focus, and c) practical to evaluate the absorption, distribution, metabolism and excretion (ADME) process when no metabolic data are available, as only pharmacokinetic data are required.

ABCB1-rs1045642
The classification for this Mexican population should not be extrapolated indiscriminately to other populations, because population-specific intrinsic genetic variation may shift pharmacokinetics. Thus, population-specific ATV pharmacokinetic stratification criteria should be determined. Hereto, we recommend to develop an analogous cluster analysis which can be applied retrospectively. The pharmacokinetic variability of ATV observed in our population differed from the one reported for a Bengali population 20 . In both studies, the volunteers were young adult males, but the dose (40 vs. 80 mg) and race (Bengali vs. Mexican genetic background) differed between the studies.    www.nature.com/scientificreports www.nature.com/scientificreports/ We assessed common genetic variants involved in drug metabolism to verify and confirm their influence on ATV pharmacokinetic variability. After genotyping validation, the genotype distribution of ABCB1-rs1045642 (0.23 for C/C; 0.53 for C/T and 0.23 for T/T) differed from a previously reported distribution that had been obtained with microarrays 6 . Importantly, the effect on C max values remains significant. The ABCB1 gene encodes a transporter protein that has affinity for multiple substrates, both endogenous and exogenous ones 21 . The polymorphism rs1045642 is located in exon 26 and produces a synonymous substitution, so the role of the ABCB1 variant in plasma drug concentrations is controversial 22 . Even so, rs1045642 variant may influence the folding time of the protein, alter its specificity, and therefore influence the concentration of a given drug 23 . In addition, synergistic or antagonistic interactions with other gene variants may create variability. So far, evidence on synergistic or antagonistic effect of ABCB1 on ATV plasma concentrations is scarce 13 .The significantly lower values in C/C carriers and the dominant effect of allele T suggest that the rs1045642 polymorphism significantly affects the absorption, bioavailability and blood concentrations of ATV. This hypothesis is supported by the intestinal expression of ABCB1, where this transporter actively participates in drug absorption 24 . These data on rs1045642 are consistent with previous findings, even for different drugs and populations. T/T carriers had a higher C max for ATV among Americans and for rosuvastatin among Chinese, while C/C carriers had a lower C max and AUC for edoxaban and also in Mexicans for amfepramone 17,18,25,26 . Contrasting data, however, were found in a Korean population, where T/T carriers had a lower C max and longer half-life than C allele carriers. The group size was small (n = 3), though 27 . As far as we know, there is no other report about the effect of ABCB1-rs1045642 on the pharmacokinetics of other statins 28,29 . Our findings suggest that a lower concentration of drug may yield a lower response, whereas an increased exposure to the medicine could cause adverse effects. Conversely, the lack of pharmacodynamic data did not allow us to evaluate the response to treatment and confirm our hypothesis. Despite this limitation, the association between ABCB1-rs1045642 and the pharmacological response to ATV has been well documented. For example, the C/C genotype among Australian patients treated with ATV associated with a lower treatment efficiency (i.e. less decrease in LDL values) as compared to other genotypes 30 . Similar data have been reported for Egyptian males 9 , Iranian 31 and Polish populations 32 , and in a meta-analysis where 395 patients were included and treated with statins 33 . Regarding adverse effects, a higher frequency of the T allele has been found in patients who presented with myalgia, but no association has been reported 30 .
In our analysis, the genotype frequency of SLCO1B1-rs4149056 was similar to the one we have previously reported 6 . The new clustering analysis method confirmed the effect of SLCO1B1-rs4149056 on ATV pharmacokinetics. The SLCO1B1 gene codes for a protein responsible for the transport of organic anions and other compounds, such as drugs. This gene is expressed exclusively in the liver, where it has an important role in metabolism 34 . The rs4149056 polymorphism generates a p.V174A substitution that causes a decrease in expression and transport activity 35 . SLCO1B1-rs4149056 had the greatest effect on ATV pharmacokinetics. The presence of the C allele in SLCO1B1-rs4149056 seemed to affect AUC 0-t , AUC 0-∞ , Cl, V d , and therefore impact the exposure to ATV, which was consistent with its occurrence in normal/faster phenotypes. The association between SLCO1B1-rs4149056 and aforementioned pharmacokinetic parameters indicates that the variant mainly affects the metabolism and excretion phases of ATV in the Mexican population as opposed to ABCB1-rs1045642, which apparently affects the absorption phase. This assumption is consistent with ATV pharmacokinetics in Chinese (n = 32), Japanese (n = 31), Caucasian (n = 30), and Korean (n = 28) populations where the presence of the C allele of polymorphism rs4149056 was associated with higher C max and/or AUC values 7,27,36 . In our study, rs4149056 has a preponderant role on ATV metabolism, since it was the only one that showed an association with metabolic phenotypes. Again, the higher AUC and slower clearance in C allele carriers could result in a better response or a higher susceptibility to adverse effects. The latter, may be explained by the statin response studies, where the C allele has been related to the risk of myopathy 37,38 . In all studies, the C/C genotype was the least frequent.
CYP2D6 codes for one of the main drug metabolizing enzymes, since it participates in the biotransformation of around 25% of all drugs 39 . The CYP2D6-rs1135840 polymorphism creates an alternative splicing site that eliminates exon 6, without a significant impact on gene expression 40 . Another CYP2D6 polymorphism, CYP2D6-rs3892097, has been associated with ATV-induced adverse effects on muscle 41 . However, there is little information about the role of CYP2D6 in ATV metabolism or pharmacokinetics. In a previous study, CYP2D6-rs1135840 seemed to have a significant effect on AUC values 13 , but this finding was not confirmed in this study. Here, we identified that CYP2D6-rs1135840 has a significant effect on Ke and T 1/2 parameters, but this was not confirmed with the regression analyses, so its effect on the metabolism and excretion phases is not clear.
The catechol-O-methyltransferase, encoded by the COMT gene, is an enzyme that helps to eliminate endogenous or toxic metabolites, as well as exogenous polycyclic compounds 42,43 . Due to its regulatory function of catecholamines, more is known about its role in pharmacodynamics than its role in pharmacokinetics 44 . The A allele related to susceptibility to coronary artery disease 44 . We discovered a significant effect of the COMT-rs4680 polymorphism on the C max , AUC 0-∞ and Cl of ATV, although this effect was not supported by regression analysis. The rs4680 variant produces a non-synonymous amino acid change (p.V158M) causing impaired COMT activity 44,45 . As far as we know this is the first study that reports a possible effect of COMT-rs4680 on ATV pharmacokinetics.
CYP2B6 is a P450 family pharmacogene responsible for the metabolism of 4% of the main drugs 19,46 . The CYP2B6 gene is expressed primarily in the liver 19,47 and its expression can be induced by different substrates including ATV 48 . The rs3745274 is located in exon 4 and produces a p.Q172H substitution, which is related with a slight reduction in expression and activity 19 . The T allele variant of rs3745274 has been related to a lower propofol dose 49 and an increased exposure to efavirenz 50 . Although ATV induced CYP2B6 expression in cultured human hepatocytes 51 , the role of CYP2B6 on ATV pharmacokinetics in vivo is unknown. Our results show that, under an over-dominant model, rs3745274 affects the variation of C max , suggesting a slight role in ATV absorption of an unknown mechanism.
Scientific RepoRtS | (2020) 10:8900 | https://doi.org/10.1038/s41598-020-65843-y www.nature.com/scientificreports www.nature.com/scientificreports/ The N-acetyltransferase 2 gene (NAT2) encodes a phase II metabolic enzyme involved in the biotransformation of drugs and carcinogens. NAT2 has many variations and has been associated with different metabolic phenotypes 52 , especially for anti-tuberculosis drugs 53 . The NAT2-rs1208 produces a p.K268R substitution that associates with a rapid acetylation 52 . To date, the influence of NAT2-rs1208 on statin pharmacokinetics has not been reported. We found that NAT2-rs1208 affected AUC 0-∞ under co-dominant and dominant models.
Although the effect of ABCB1 and SLCO1B1 transporters on pharmacokinetics and pharmacodynamics, as well as on susceptibility to adverse effects has been well documented, there are few studies that assess the interaction of these genes and their relationship with the effectiveness of treatment. Pharmacogenetics data from 1844 subjects suggest that ABCB1 and SLCO1B1 variants may be useful for improving effectiveness and preventing the risks of adverse effects of statin treatment 10 . Our data on a Mexican population confirm that the transporters ABCB1 and SLCO1B1 have a significant impact on ATV metabolism (Fig. 1c,d). Although the SLCO1B1 transporter had the greatest impact on ATV metabolism, the influence of ABCB1 is underscored by the fact that the presence of the ABCB1-rs1045642 C/C variant is sufficient to shift from an intermediate to a normal metabolism phenotype. Thus, different combinations of these two variants may generate a broad spectrum of metabolism and therefore a variable response to treatment. In our knowledge, this is the first study that demonstrates how the interaction of genetic variants affect metabolic phenotyping and improves understanding of how SLCO1B1 and ABCB1 variants that affect statin metabolism may explain the variability in drug response.
In the present study, some limitations remain. First, as only men were included in this study, our findings should be confirmed in women. Second, the lack of data on secondary metabolites of ATV did not allow us to validate the metabolic classification by another method. Third, this study was limited to candidate polymorphisms. Thus, other polymorphisms or genes that could have an impact on the metabolism of ATV were not investigated.

Conclusions
Variants of the transporter-encoding genes ABCB1 and SLCO1B1 have an important impact on ATV pharmacokinetics in a Mexican male population. Hence, the metabolism also varies from population to population, even for the same drug. So, it is important that each study perform its metabolic classification. Our results improve the understanding of the mechanism by which variation in transporters may affect the therapeutic response to ATV in a Mexican population. ABCB1 and SLCO1B1 variants were not only congruent with, but could also explain, the metabolic phenotype classification at a genetic level.

Material and methods
Design. This pharmacogenetic study used pharmacokinetic data from a controlled, randomized, crossover, single-blind, three-treatment, three-period, and six-sequence clinical study after a single 80-mg oral dose of ATV (tablets; Pfizer Pharmaceuticals LLC, Caguas Site, Caguas, PR), conducted in 60 healthy Mexican men. The clinical protocol complied with national and international ethical regulations, guidelines and norms, as described previously 6  Study population. Between January and February 2011, sixty healthy, 18-to-45-year-old, non-smoking, Mexican males with a weight ≥ 50 kg and a body mass index between 20 and 26 kg/m 2 participated in the bioequivalence study. Their health status was assessed based on physical examination, medical history, and clinical and biochemical tests. Exclusion criteria included abnormal laboratory results, drug abuse, ingestion of alcohol 1 week prior to the study, the use of medication three weeks before enrollment, and participation in a clinical research study within the previous 3 months.
Dosing regimen and sampling. The dosing and sampling involved a three-treatment (R = reference drug, T1 = Test 1 and T2 = Test 2 drugs are first and second treatment, respectively), three-period, and six-sequence schedule with a 2-week washout period between treatments. Before starting any drug administration, the statistical department of Ipharma S.A. randomized drug allocation (R, T1 or T2, file code, and subject code) with a balanced design using the Mersenne Twister algorithm and R statistical software. The participants were blinded to treatment. About 18 h before the first dose administration, the subjects were admitted to the clinical site and served a standard dinner (<800 kcal). The next morning, after an overnight fast, ATV was administered orally as a single 80-mg dose tablet (Pfizer, New York, NY, USA) with 240 mL of water. A standardized lunch was served 4 h after dosing and standardized dinner at 12 h after dosing. Peripheral blood (4-mL samples) was collected in K 2 EDTA-coated Vacutainers (BD Diagnostics, Franklin Lakes, NJ, USA), a pre-dose sample was taken (time 0), while other samples were taken at 0.25, 0.5, 0.75, 1, 1.5, 2, 2.5, 3, 3.5, 4, 5, 6, 8, 12, 24, 36 and 48 h after drug administration. Cells and plasma were separated (10 min at 1600 × g at 4 °C). Plasma was stored at −65 ± 15 °C until use while DNA was extracted from cells.
Plasma ATV quantitation. Proteins were eliminated by acetonitrile precipitation. Briefly, 400 µL acetonitrile was added to a 100-µL plasma sample, vortexed (70 rpm, 4 min), and centrifuged (9600 × g, 10 min, 10 °C). The supernatant (300 µL) was injected into a high-performance liquid chromatography-tandem mass spectrometry equipped with an Agilent 1100 control system (Agilent Technologies, Inc., Santa Clara, CA, USA) using a method validated by Ipharma S.A 6 . that respects Mexican regulations 54 and guidelines for the validation of bioanalytical methods of the European Medicines Agency (EMA) 55 . An electrospray ionization tandem mass spectrometry system with a precursor ion (+) 559.3 m/z and a product ion (+) 440.3 m/z was used for detection. The linearity of the analytical method was assessed with calibration curves (0.5, 2.5, 5, 10, 25, 50 and 100 ng/mL), which yielded a correlation coefficient (r) = 0.99292 and a lower limit of quantification (LLOQ) = 0.5 ng/mL of ATV in plasma. Precision was assessed with control samples (1.7, 7.5, 35, and 75 ng/mL) and was considered within an acceptable range: intraday CV < 5% and interday CV < 8%. An accuracy error ≤7% was considered to be acceptable. The recovery was 87% and no effects were detected for the matrix, hemolyzed, and lipemic plasma (CV < 4%).
Pharmacokinetic analysis. The C max and time to reach C max (T max ) were obtained from the concentration-time data. The pharmacokinetics parameters were determined by non-compartmental methods. The AUC 0-t and AUC 0-∞ were calculated with the log-linear trapezoidal rule. The Ke was estimated via log-linear regression from the terminal portion of the log-transformed concentration-time plots. T 1/2 was estimated by dividing 0.693 by Ke. The Cl was calculated by dividing the dose by AUC 0-∞ and adjusting for weight. The V d was calculated as Cl divided by Ke. The AUC's and C max values were adjusted for dose and weight (AUC's/dW and C max /dW) 17,56 . The pharmacokinetic analysis was performed using WinNonlin software v5.3 (Pharsight Corp., Mountain View, CA, USA).

Determination of ATV metabolizer phenotypes.
To estimate the ideal number of phenotypes, we applied a modified four-step multivariate analysis of the combined dose-and-weight-adjusted pharmacokinetic parameters C max , AUC 0-t , AUC 0-∞ and Cl 6,14,16,57 . First, hierarchical cluster analysis (HCA) was performed with the Ward linkage method and the distance matrix was calculated with the Manhattan measure. These analyses were applied to z-score transformed pharmacokinetic values to circumvent the comparability problems caused by the different scales of the non-transformed pharmacokinetic values. Aforementioned analysis were carried out with Minitab 16 software (Minitab Inc., State College, PA, USA). Second, we identified the subjects in each cluster, calculated the means of the adjusted pharmacokinetic parameters, and assigned the clusters to metabolizer phenotypes based on the means of the pharmacokinetic parameters. Third, we validated the phenotyping model by automatic linear modeling with the forward stepwise method, the Akaike Information Criterion (AICC) and the Overfit Prevention Criterion (ASE); followed by linear and logistic regression analysis. Finally, the cut-off limits were assessed by one-way analysis of variance and the Kruskal-Wallis test, considering P < 0.05 to be statistically significant different, with SPSS for Windows V.25 (IBM Corp., NY, USA).
Genotyping tests. Genomic DNA that had been isolated with the alkaline lysis method 58 was quantified by fluorometry using the Qubit dsDNA BR assay kit and a Qubit 3.0 fluorometer (Invitrogen; Thermo Fisher Scientific, Waltham, MA, USA). Compliance with DNA purity (OD 260 /OD 280 between 1.8 and 2) was assessed with a Nanodrop 1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). DNA was stored at 10 ng/µL at −20 °C until analysis.
The polymorphisms ABCB1-rs1045642, SLCO1B1-rs4149056, CYP2B6-rs3745274, CYP2D6-rs16947, CYP2D6-rs1135840, NAT2-rs1208 and COMT-rs4680 were genotyped using a QuantStudio 1 real-time PCR system and TaqMan genotyping assays (Applied Biosystems; Thermo Fisher Scientific, Wilmington, MA, USA) according to the manufacturer's protocol. Briefly, the PCR was prepared with 1X TaqMan Universal PCR Master Mix (Thermo Fisher Scientific, Inc.), 1X TaqMan genotyping assay mix, 10 ng DNA, and nuclease-free water to a total volume of 10 µl. Thermal cycling conditions were as follows: 95 °C for 10 min, 40 cycles of 95 °C for 15 sec and 60 °C for 1 min. Quality controls included a genotyping control using previously genotyped samples across different platforms, genotype call rate equal to 1.0, a Hardy-Weinberg equilibrium test with P > 0.05, and a minor allele frequency> 0.01. To detect the combination of the most important variants in the study population, a haplotype analysis was performed using the Haplotype Analysis Software V.1.05 59 , under a modified data entry method for diploid genomes.
Statistical analysis. The sample size calculation considered an intrasubject coefficient of variation (CV) of 45% for C max and AUC, a confidence interval (CI) of 90%, a significance level of 5%, a minimum power of 80%, and a Ω of 0.25. Thus, a sample size of 58 would suffice. The Hardy-Weinberg equilibrium was determined by comparing the genotype frequencies with the expected values using the maximum likelihood method 60 . To assess the effects of polymorphisms on the ATV pharmacokinetic parameters, comparisons between two and three groups were made. The Student's t-test and one-way analysis of variance were used for parametric distributions, while Mann-Whitney U and Kruskal-Wallis H tests were used for nonparametric distributions. Post hoc tests (Bonferroni's and Tamhane's T2) were used for pairwise comparisons. To confirm the contribution of genetic factors to the variability of pharmacokinetic parameters, automatic linear modeling with the forward stepwise method, AICC and ASE was performed, as well as linear regression analysis with various modes (default mode, stepwise, remove, backward and forward). Possible associations of genotypes or combinations of genotypes with phenotypes were evaluated using X 2 and Fisher's exact tests and validated by logistic regression analysis. The evaluation effect of polymorphisms and associations were assessed under four different models (co-dominant, dominant, over-dominant and recessive). The odds ratio (OR) was estimated with a 95% CI. All P values were two-tailed. Corrected P values (Pc) were obtained using the Bonferroni correction for exclusion of spurious associations. P < 0.05 was interpreted as statistically significant. The statistical analyses were performed with SPSS for Windows, V.25 (IBM Corp., NY, USA).