Combinations of common SNPs of the transporter gene ABCB1 influence apparent bioavailability, but not renal elimination of oral digoxin

Effects of different genotypes on the pharmacokinetics of probe substrates may support their use as phenotyping agents for the activity of the respective enzyme or transporter. Digoxin is recommended as a probe substrate to assess the activity of the transporter P-glycoprotein (P-gp) in humans. Current studies on the individual effects of three commonly investigated single nucleotide polymorphisms (SNPs) of the ABCB1 gene encoding P-gp (C1236T, G2677T/A, and C3435T) on digoxin pharmacokinetics are inconclusive. Since SNPs are in incomplete linkage disequilibrium, considering combinations of these SNPs might be necessary to assess the role of polymorphisms in digoxin pharmacokinetics accurately. In this study, the relationship between SNP combinations and digoxin pharmacokinetics was explored via a population pharmacokinetic approach in 40 volunteers who received oral doses of 0.5 mg digoxin. Concerning the SNPs 1236/2677/3435, the following combinations were evaluated: CGC, CGT, and TTT. Carriers of CGC/CGT and TTT/TTT had 35% higher apparent bioavailability compared to the reference group CGC/CGC, while no difference was seen in CGC/TTT carriers. No significant effect on renal clearance was observed. The population pharmacokinetic model supports the use of oral digoxin as a phenotyping substrate of intestinal P-gp, but not to assess renal P-gp activity.

were available for the population pharmacokinetic analysis. All subjects were assigned to four SNP combination groups based on the genotype of P-gp. The SNP combinations were CGC/CGC, CGC/ CGT, CGC/TTT, and TTT/TTT (as the respective combination of 1236/2677/3435 SNPs) with 9, 5, 15 and 9 subjects, respectively (see Table 1). Two subjects had different SNP combinations, CGT/TTT and CGT/CTT, and were excluded from the genotype covariate analysis due to the small group size. SNP frequencies of C1236T (0.425), G2677T (0.438) and C3435T (0.525) in the study population were similar to published data 26 , and did not show a significant deviation from Hardy-Weinberg equilibrium ( χ 2 of 1.32, 0.745 and < 0.001, respectively). For the SNP pairs C3435T/G2677T, C3435T/C1236T and G2677T/C1236T, a pronounced linkage disequilibrium was found (D′ of 0.78, 1.00, and 1.00; r 2 of 0.55, 0.67 and 0.95, respectively).
Digoxin empirical model. As a base model, a two compartment model with mixed first-and zero-order absorption and linear elimination described the data best. A summary of key model development steps is shown in Table 2. Visual predictive checks (VPCs) indicated a difference in apparent bioavailability and absorption shape between the two trials that was not captured well by the base model. Consequently, a significantly lower bioavailability was identified in trial I compared to trial II (-30.3%, drop in objective function value (OFV) by 72.6) and a significantly lower first-order absorption rate constant (Ka) (0.201 h −1 vs. 0.636 h −1 , drop in OFV by 46.0) was identified in the test period of trial II compared to trial I and the reference period of trial II. When introducing additional estimates for bioavailability in trial I and for Ka in the test period of trial II, VPCs did not show any further misspecification.
Covariate model for effects of ABCB1 SNP combinations. Apparent bioavailability was estimated separately for each of the defined SNP combinations in the population pharmacokinetic model, resulting in a significant drop in OFV by 26.1 points. In addition, renal clearances (CL R ), zero-order absorption durations (D2) and Ka were also computed separately for each SNP combination. However, the model did not improve significantly (drop in OFV by 1.27, 3.44 and 3.55 points, respectively). A non-parametric bootstrap with 1,000 samples was conducted for the model with separately estimated apparent bioavailabilities in each SNP combination group. Relative differences in apparent bioavailabilities and renal clearance between SNP combinations are summarized in Fig. 1. CGC/CGT and TTT/TTT carriers had an approximately 35% higher bioavailability com- Table 1. Observed distribution of genotypes for ABCB1. a 38 out of forty subjects in the two trials were included in our evaluation of the effect of SNP combinations. P-gp, P-glycoprotein. b Contains one noncompliant subject. Trial I Trial II   CGC/CGC a  3  6  1236 C/C + 2677 G/G + 3435 C/C   CGC/CGT a,b  2  3  1236 C/C + 2677 G/G + 3435 C/T   CGC/TTT a  8  7  1236 C/T + 2677 G/T + 3435 C/T   TTT/TTT a  3  6  1236 T/T + 2677 T/T + 3435 T/T   -1  1236 C/T + 2677 G/T + 3435 T/T   -1 1236 C/C + 2677 G/T + 3435 T/T  Table 3. VPC and Goodness of Fit (GOF) plots of plasma and urine data are shown in Figs. 2 and 3.

Discussion
In this evaluation based on a detailed characterization of digoxin pharmacokinetics in healthy volunteers, ABCB1 SNP combinations had a significant albeit small influence on the (apparent) bioavailability but not on the renal elimination of the drug. This result is in line with a previous non-compartmental analysis (NCA) of one of the  Table 3 for further information. In previous studies, no systematic evaluation of the effect of CGT on the pharmacokinetics of digoxin has been conducted. Several previous studies showed the effect of 3435CT on the pharmacokinetics of digoxin. However, www.nature.com/scientificreports/ we cannot identify the subjects as belonging to the CGT group since the SNPs at positions 1236 and 2677 were not considered. Therefore, there is currently not enough data to support a genotype effect of CGT. Kim et al. included CGC/CGT carriers in an ABCB1 genotype evaluation regarding the pharmacokinetics of fexofenadine. However, no statistical evaluation was conducted since there was only a single CGC/CGT carrier. We cannot explain why in our evaluation homozygous (but not heterozygous) carriers of TTT as well as the CGC/CGT genotype group had a different apparent bioavailability compared to homozygous wildtype carriers. Beyond the relatively small effects of the combined genetic variants and the pronounced inter-individual variability of apparent bioavailability, other possible explanations would include (1) the mechanisms by which each single SNP and/or the SNP combination may modify overall P-gp expression/activity; indeed, an effect of C3435T could be attenuated by the presence of the two further SNPs studied here; (2) whether a dominant model, a co-dominant model or a recessive model would be most appropriate to describe the relationship between genotype and P-gp expression/activity; for the presence of all three SNPs (TTT group), a recessive model would be in agreement with our findings; (3) effects of other covariates not assessed in the present analysis, such as further SNPs of ABCB1, SNPs in xenobiotic receptors involved in P-gp regulation etc.; (4) a chance finding based on the relatively low number of subjects in the CGC/CGT group (n = 5); this needs further investigation, and to this end, it would be interesting to assess individuals with two respective mutated alleles (CGT/CGT) but these were not present in our population. Although the employed model is of an empirical nature, net apparent bioavailability reflects the sum of intestinal absorption and intestinal as well as biliary secretion, which presumably results in enterohepatic circulation. The 35% higher bioavailability in TTT/TTT and in CGC/CGT carriers thus cannot be translated directly into a 35% lower activity of P-gp in these individuals. The results also do not allow to draw further conclusions on the underlying molecular mechanisms, such as reduction in ATP binding affinity, loss of ATP hydrolysis, modification of protein folding, or reduction in P-gp expression 46 . Published data on the effect of the TTT SNP combination on P-gp expression in the duodenum or the intestine is equivocal 53 . Although there are contradictory data on whether ABCB1 genotype would influence P-gp activity in vitro or in vivo, the present study showed an effect of combined SNPs on digoxin apparent bioavailability along with the absence of a respective effect on renal clearance of digoxin. This casts further doubts on the role of P-gp as a rate-limiting transporter for digoxin elimination in the kidney. However, we cannot rule out that our limited sample size was insufficient to detect minor differences in renal clearance between groups. Also, in our previous NCA analysis, no significant difference in renal clearance was found between CGC/CGC and other SNP combination groups 52 . When digoxin was given intravenously or orally with a strong P-gp inducer, renal elimination of digoxin was not relevantly affected, but the AUC 0-144 h for both administration routes and AUC 0-3 h , C max , and bioavailability for oral administration were decreased. In addition, non-renal clearance (which might reflect intestinal and/or biliary clearance) for intravenous administration and t max for oral administration were increased 19 . These findings also indicate that P-gp activity is not rate-limiting for the elimination of digoxin in the kidney. However, the impact of SLCO4C1 polymorphisms on the pharmacokinetics of substrates is unknown 54 . Whether SLCO4C1 (cytogenetic location: 5q21.1) is in linkage disequilibrium with ABCB1 (cytogenetic location: 7q21.12) has not been investigated yet. Therefore, we assume that polymorphisms of SLCO4C1 are randomly distributed in our study subjects, while we cannot exclude whether such polymorphisms may have an additional impact on the pharmacokinetics of digoxin.
In our empirical pharmacokinetic model, different values for apparent bioavailability were estimated for the two trials due to differences in study designs (see Table 3), probably attributable to the use of different concomitantly administered drugs. Thus, correcting for trial-specific differences was necessary to allow identification of trial-independent effects of ABCB1 SNP combinations. Furthermore, we introduced a mixed zero-and first-order absorption model since some plasma concentration profiles exhibited a double peak phenomenon, with a second peak occurring after 4 h to 8 h post-dose. The double peak phenomenon could not be attributed to a food effect, but is in line with previous evaluations of digoxin exhibiting a similar phenomenon 55,56 .
Another limitation of this study is that plasma and urine samples were only available up to 24 h after drug administration, which covers less than one elimination half-life of digoxin. However, the assessment of drug elimination should be reliable once absorption is completed. Indeed, the estimated renal clearance of digoxin was comparable with previously published data 57 . Despite the limitations, the chosen population pharmacokinetic approach allowed to correct for differences between trials and to describe the effect particularly attributable to SNP combinations based on a large and detailed dataset.
In the previous studies assessing the effect of ABCB1 genotypes on digoxin pharmacokinetics, noncompartmental methods were used, including C max , AUC 0-t, time of maximum plasma concentration (t max ) or CL R [27][28][29][30][31][32][33][34][35] . In general, a population pharmacokinetic evaluation might be advantageous to identify effects on parameters more closely related to physiological processes. Although our model described the data well, the complex interplay of drug absorption, intestinal secretion and biliary elimination could not be captured in detail by the empirical model solely based on oral administration of digoxin. The information on genotype effects that could be obtained by the present evaluation turned out to be supportive but not superior to the information obtained by noncompartmental analysis. Using semiphysiological models applied to datasets including both oral and intravenous administration, as has been used for drugs undergoing first pass metabolism such as midazolam 58 , might be a promising approach to learn more about the mechanism of ABCB1 genotype effects on digoxin pharmacokinetics.
Whether the observed relationship between the SNP combinations/haplotypes of ABCB1 and digoxin can be transferred to other P-gp substrates is currently not clear. For instance, in a previous study conducted in a Chinese population, 1236CC carriers had a lower C max (− 53%; p = 0.013), AUC 0-∞ (− 40%; p = 0.04), and cumulative amount excreted in urine over 6 h (− 52%; p = 0.027) and a higher apparent oral clearance (+ 35%; p = 0.013) of cloxacillin, another P-gp substrate, compared to carriers of 1236CC and 1236CT. Moreover, the homozygous CGC carriers also had a lower C max (p = 0.017), AUC 0-∞ (p = 0.032), and cumulative amount excreted over 6 h in urine (p = 0.026) and a higher apparent oral clearance of cloxacillin (p = 0.002) compared to homozygous carriers Scientific RepoRtS | (2020) 10:12457 | https://doi.org/10.1038/s41598-020-69326-y www.nature.com/scientificreports/ of TTT. Renal clearance of cloxacillin was not altered by the SNP combination TTT/TTT 59 . This result is very similar to the one obtained for digoxin. In contrast, in a study in Japanese subjects, homozygous carriers of TTT (defined as homozygous ABCB1*2 in the corresponding publication) compared to subjects who did not carry TTT had a lower renal clearance (p < 0.05) und urinary recovery (p < 0.01) of the P-gp substrate irinotecan and its metabolites, while there was no significant difference in the ratio AUC 0-∞ /dose 60 . Furthermore, Kim et al. showed that homozygosity for TTT (defined as homozygous ABCB1*2 in the corresponding publication) was related to a difference in AUC 0-16 h of orally administrated fexofenadine, another P-gp probe substrate, in the opposite direction: The AUC was 40% lower compared to homozygous CGC (defined as homozygous ABCB1*1/*1) 47 , suggesting increased intestinal secretion for this SNP combination. Moreover, the effect of different ABCB1 SNP combinations on the pharmacokinetics of cyclosporine was also inconsistent in previous reports, which may be attributable to the high variability in the pharmacokinetics in the heart and renal transplantation patients [61][62][63][64] . As a potential explanation for discrepant findings, digoxin was suggested to bind to different sites of P-gp unlike other typical P-gp substrates 65 . Whether the SNP combination of P-gp alters the structure of the binding site for digoxin, but not for other substrates, is also unknown. Thus, the whole protein structure of different P-gp SNP combinations and related drug-protein binding needs further to be studied in the future. Additionally, the DNA methylation level in the ABCB1 promoter may also influence the ABCB1 activity. Wu et al. evaluated both the effect of the SNP combination of ABCB1 and DNA methylation level in the ABCB1 promoter on digoxin pharmacokinetics. mRNA expression in intestinal epithelial cells showed no difference between homozygous CGC and homozygous TTT carriers (p = 0.087). However, mRNA expression of homozygous TTT-HM carriers, who had a higher degree of DNA methylation, was significantly decreased compared to homozygous TTT-LM carriers (lower degree of DNA methylation), homozygous CGC-LM and homozygous CGC-HM carriers by 31.1, 27.9 and 43.6% (p = 0.02, 0.013 and 0.008), respectively. Subjects who carried homozygous TTT-HM had a significant higher AUC 0-72 h , AUC 0-∞, C max and lower apparent oral clearance compared to homozygous CGC-LM (p < 0.05) 40 . These results suggest that DNA methylation should be considered as a covariate when assessing the effect of different ABCB1 SNP combinations on the pharmacokinetics of digoxin.
A further method to analyze the activity of membrane transporters is to evaluate the concentration of endogenous substrates (e.g., coproporphyrin I and III for OATP1B1/1B3 transporter) 66,67 . Recently, testosterone, aldosterone, and cortisol have been demonstrated to be endogenous substrates of P-glycoprotein in vitro 7,68 . In vivo, the genotype of P-gp affected the elimination of aldosterone in urine, which might be a potential biomarker for evaluating renal P-gp activity 68 . Thus, endogenous markers might help to further characterize the functional role of ABCB1 SNP combinations.

Conclusion
The empirical population pharmacokinetic evaluation showed that homozygous carriers of the TTT and CGC/ CGT have a 35% higher apparent bioavailability of oral digoxin, while no effects of CGC/TTT on apparent bioavailability and of any ABCB1 variants on renal elimination were observed. These results support the use of digoxin as a phenotyping substrate of intestinal but not of renal P-gp activity. Our study suggests considering the effect of a combination of SNPs on the pharmacokinetics of digoxin, rather than focusing on single SNPs. This might play an important role in the design and refinement of transporter phenotyping studies, including the development of appropriate sampling schedules, and the mode of administration. Furthermore, protein expression, protein structure and/or P-gp affinity to digoxin or other substrates in different genotypes need to be further investigated.

Methods
Clinical trials. Data from 40 healthy Caucasian subjects receiving single oral doses of 0.5 mg digoxin in two clinical trials were analyzed 52,58 . In trial 1, a single dose of 0.5 mg digoxin was given concomitantly with 2 mg oral midazolam, 1 mg intravenous midazolam, 125 mg tolbutamide, 150 mg caffeine, 20 mg omeprazole and 30 mg dextromethorphan in the reference period. In the test period, the drugs of the reference period were combined with ethanol 58 . In trial 2, a single dose of 0.5 mg digoxin was given alone in the reference period and in combination with 10 mg adefovir, 500 mg metformin, 2 mg pitavastatin, and 100 mg sitagliptin in the test period 52 . Blood and urine samples were collected up to 24 h after drug administration. The study design and the timing of blood and urine samples are summarized in Table 4.
Genotyping of ABCB1 was carried out using the DMET Plus Array (Affymetrix, Santa Clara, California, United States) 69 . To define SNP combination groups, the common ABCB1 SNPs C1236T, G2677T and C3435T were taken into account 49 . Deviation from Hardy-Weinberg equilibrium was assessed using a chi square test.
Blood and urine samples of digoxin were processed with solid-phase extraction (Strata-X 30 mg/3 mL, product number: 8B-S100-TBJ, Phenomenex, Aschaffenburg, Germany). Digoxin-d3 was spiked in blood and urine sample as internal standard (product code: TRC-D446577-2.5MG, Toronto Research Chemicals, Toronto, Canada). Concentrations were quantified with a validated high-performance liquid chromatography-tandem mass spectrometry method (Agilent 1,260 Infinity, Agilent Technologies, Waldbronn, Germany/API 5,000, AB Sciex Germany GmbH, Darmstadt, Germany) as described previously 52,58 . The calibration range of digoxin in plasma and urine was 0.128 nmol/L to 38.4 nmol/L and 1.28 nmol/L to 384 nmol/L, respectively. All assays fulfilled the bioanalytical method validation criteria according to the FDA and the EMA guidelines 70,71 . Intra-day and inter-day inaccuracy and imprecision of all quality control samples were < 15%. Data analysis. Digoxin 72 . Model diagnostics were carried out using XPOSE 4.5.0.17 73 .
The empirical model was developed by starting with a one-compartment model with linear elimination and increasing the model complexity step-wise. Different absorption models were tested, including zero-order, firstorder with/ without lag time and enterohepatic cycling (EHC), to describe observed double peak phenomena. Clearance and volume of distribution parameters were scaled allometrically with body weight. Inter-individual variability (IIV) and between occasion variability (BOV) were computed for all pharmacokinetic parameters. A combined proportional and additive error model was applied. To account for potential systematic differences between trials, study effects were tested on all fixed effects parameters.
For all modelling steps, changes in OFV, the Akaike information criterion (AIC) for non-nested models, GOF and VPC were considered. For statistical tests based on the OFV, a chi squared distribution with the appropriate number of degrees of freedom was assumed. Furthermore, a non-parametric bootstrap analysis with 1,000 samples was conducted.
Effect of different ABCB1 SNP combinations. After identification of a reasonable base model, relationships between SNP combinations and pharmacokinetic parameters were evaluated by introducing different Ka, D2, apparent bioavailabilities and CL R in each SNP combination group in the population pharmacokinetic model. The changes in OFV were considered to identify significant relationships.
Compliance with ethical standards. Both clinical trials were approved by the Ethics Committee of the Faculty of Medicine, University of Cologne, Germany, and conducted in accordance with applicable regulations and the ethical principles described in the Declaration of Helsinki and the International Conference on Harmonization guidelines for Good Clinical Practice. The clinical trial I and clinical trial II are registered at clinicaltrials.gov with the IDs NCT02515526 and NCT02743260, respectively. Informed consent was obtained from all participants.

Data availability
The datasets analyzed during the current study are available from the corresponding author on reasonable request.
Received: 1 March 2020; Accepted: 10 July 2020 Table 4. Summary of study designs used for pharmacokinetic analysis of digoxin. C max, maximal observed plasma concentration.