CYP3A5*3 and CYP3A4*22 Cluster Polymorphism Effects on LCP-Tac Tacrolimus Exposure: Population Pharmacokinetic Approach

The aim of the study is to develop a population pharmacokinetic (PopPK) model and to investigate the influence of CYP3A5/CYP3A4 and ABCB1 single nucleotide polymorphisms (SNPs) on the Tacrolimus PK parameters after LCP-Tac formulation in stable adult renal transplant patients. The model was developed, using NONMEM v7.5, from full PK profiles from a clinical study (n = 30) and trough concentrations (C0) from patient follow-up (n = 68). The PK profile of the LCP-Tac formulation was best described by a two-compartment model with linear elimination, parameterized in elimination (CL/F) and distributional (CLD/F) clearances and central compartment (Vc/F) and peripheral compartment (Vp/F) distribution volumes. A time-lagged first-order absorption process was characterized using transit compartment models. According to the structural part of the base model, the LCP-Tac showed an absorption profile characterized by two transit compartments and a mean transit time of 3.02 h. Inter-individual variability was associated with CL/F, Vc/F, and Vp/F. Adding inter-occasion variability (IOV) on CL/F caused a statistically significant reduction in the model minimum objective function MOFV (p < 0.001). Genetic polymorphism of CYP3A5 and a cluster of CYP3A4/A5 SNPs statistically significantly influenced Tac CL/F. In conclusion, a PopPK model was successfully developed for LCP-Tac formulation in stable renal transplant patients. CYP3A4/A5 SNPs as a combined cluster including three different phenotypes (high, intermediate, and poor metabolizers) was the most powerful covariate to describe part of the inter-individual variability associated with apparent elimination clearance. Considering this covariate in the initial dose estimation and during the therapeutic drug monitoring (TDM) would probably optimize Tac exposure attainments.


Introduction
Tacrolimus (Tac) is a major immunosuppressant drug prescribed for over 30 years in kidney transplant patients to avoid graft rejection [1][2][3].High inter-and intra-individual variabilities in Tac exposure have been reported [4,5].Among the factors, there are hematocrit levels, time after transplantation, hepatic dysfunction, protein plasma concentrations, as well as patients' age, ethnicity, sex, and CYP3A4/A5 polymorphisms (SNPs), which have been described as being mainly responsible [4,[6][7][8][9][10][11].These factors and the Tac narrow therapeutic index are challenging for the establishment of an optimal dose.Indeed, Tac might be associated with several dose-dependent side effects such as neurotoxicity, nephrotoxicity, and post-transplant diabetes mellitus [12][13][14], which can be reduced by avoiding a Tac pre-dose concentration (C 0 ), too high in the long-term.Conversely, Tac under-dosing, resulting in too-low C 0 , should not be attempted because it increases the risk of acute rejection and immunologic sensitization [15].In this regard, the dosage of the Tac formulation can play an important role.Several formulations of Tac have been developed, demonstrating a very high efficacy in preventing an acute rejection episode.Thus, Tac was originally formulated as an immediate-release IR-Tac (Prograf ® , Astellas Pharma Europe Ltd., Staines, UK), administered twice daily.Then, the first once-daily prolonged-release formulation, (Advagraf ® , Astellas Pharma Europe Ltd., Staines, UK) was developed, and lastly, the once-daily extended-release formulation LCP-Tac was commercialized using a MeltDose ® delivery technology that provides a slower release/absorption rate, resulting in delayed Tmax and more decreased Cmax and fluctuation profile.Additionally, compared to the IR-Tac, reduction in Tac to the smallest possible particle size in LCP-Tac allows a better dissolution and absorption, increasing its bioavailability [16][17][18][19][20].Other factors, such as the varying distribution of CYP3A enzymes along the gut, may contribute to the increasing bioavailability of LCP-Tac compared to IR-Tac [21].Part of the Tac from the LCP-Tac formulation is absorbed in distal parts of the small intestine or even the colon, where the CYP3A enzyme exists to a lesser extent compared to the proximal parts, where most of the IR-Tac is absorbed.
The exposure and starting dose of Tac is significantly influenced by gene-encoding Tac-metabolizing enzyme cytochromes P450 3A4/5 (CYP3A4/5).CYP3A5 expresser (*1 allele carrier) requires higher Tac dose in comparison to non-expresser (*3/*3) to achieve similar exposure [22][23][24][25].It is suggested that a lower dose is required for patients with CYP3A4*22 SNP [26,27], as it leads to lower Tac oral clearance.In addition, ABCB1 is also thought to be a contributing factor to the low oral bioavailability of Tac.However, the impact ABCB1 SNPs on Tac pharmacokinetics has not been conclusively determined [28].
The influence of the predictive factors of Tac exposure widely studied after IR-Tac administration using modeling approaches [4,[29][30][31][32][33][34] has been widely studied.Although only a few studies describe the influence of these factors for LCP-Tac, of these, only two have been carried out in kidney transplant patients [35][36][37].This study aims at investigating the influence of CYP3A5, CYP3A4, and ABCB1 SNPs on the Tac exposure upon LCP-Tac administration, in stable adult renal transplant patients, using a populationmodeling approach.

Study Design
Thirty patients were selected from an open-label, prospective, non-randomized investigatordriven single-center clinical trial (clinicalTrials.govNCT02961608), whereas the rest of the patients were taken from routine check-ups at the hospital with Tac C 0 .All the patients were treated with an immunosuppressive drug regimen of oral twice-daily IR-Tac (Prograf; Astellas Pharma Europe Ltd., Staines, UK), for at least six months before conversion to a once-daily LCP-Tac oral regime (Envarsus; Chiesi Farmaceutici, Parma, Italy) with a dose conversion ratio of 0.7.All patients received triple immunosuppression therapy combining Tac, mycophenolate mofetil, and prednisone.
A total of 98 stable renal transplant patients who had undergone a kidney transplant at least six months prior to the study were included.Patients with current infections, hepatitis B or C, severe gastrointestinal disorders, and patients receiving concomitant drugs that could interact with the CYP3A enzyme were excluded from this study.The study was conducted in compliance with the Declaration of Helsinki and approved by the ethics committee of the Bellvitge University Hospital (Barcelona, Spain).

Tacrolimus Measurement and Data Recording
Tac was measured using an LC-MS/MS method, previously developed and validated [38].Chromatographic determination was performed using the Acquity ( ® ) UPLC ( ® ) with a C18 BEH ™ reversed-phase column (2.1 × 50 mm id, 1. 7 µm).The limit of quantitation was set at 1.0 ng•mL −1 .At the start of the treatment, Tac daily doses and demographic characteristics of the patients were retrieved from the medical files.For hematocrit (%) and serum creatinine concentrations (µmol•L −1 ), their concentrations on each occasion were also monitored and recorded.The clinical outcome variables that were assessed were renal function (eGFR), estimated using the Chronic Kidney Disease Epidemiology Collaboration formula, delayed graft function (DGF), and graft loss.

Statistical Analysis
Unpaired t-tests were used to evaluate differences in log-transformed values of exposure metrics (C 0 normalized by dose and AUC concentration-time curves) between two groups, and at the same time, one-way analysis of variance tests were used for comparisons among three the groups.Results with p values of less than 0.05 were considered statistically significant.Statistical analyses were performed with the R program version (4.0.3).

Base Model Development
The popPK analysis was performed with the non-linear mixed-effects model approach using the NONMEM ® version 7.5 (ICON Development Solutions, Hanover, MD, USA).Perl-Speaks-NONMEM (PsN) version 5.2.6, R package version 4.0.3,Pirana Modeling Workbench version 3.0 (Certara L.P. (Pharsight), St. Louis, MO, USA), and Xpose 4.7.2 were used as support tools for model evaluation.The first-order conditional estimation (FOCEI) method with interaction was used throughout the modeling process.
One-and two-compartment open models with linear elimination and first-order absorption were tested.The standard lag model and transit compartment models, in which the optimal amount of transit compartments is estimated [39], were used to model the absorption delay.Inter-individual variability (IIV) and inter-occasion variability (IOV) [40] were tested in all PK parameters assuming log-normal distributions.Additive, proportional, and combined error models were tested to describe the residual error (RE) variability.To compare the different tested models statistically, the likelihood ratio test, based on the reduction in the minimum objective function value (MOFV) at a significance level of p < 0.005 (change in MOFV [∆MOFV] = -7.879for 1 degree of freedom) was considered.For non-hierarchical models, the most parsimonious model according to the Akaike information criterion (AIC) was chosen [41].The decrease in MOFV, parameter precision expressed as a percentage of relative standard error (RSE%), reductions in IIV associated with parameters, ηand ε-shrinkage values [42], model completion status, and visual inspection of goodnessof-fit plots were also considered for model selection.

Covariate Model
The effect of factors that could be physiologically and clinically meaningful on the PK parameters was investigated.Specifically, the influence of patient age, gender, total body weight, body mass index, and hematocrit were evaluated.The effects of CYP3A4/A5 SNPs, cluster phenotypes, and ABCB1 genotypes were also tested on Tac CL/F.
Firstly, correlations between continuous covariates were explored.Then, univariate analysis, forward inclusion, and backward elimination stepwise procedures were carried out to explore the covariates [43].Significance levels of 5% (∆MOFV = -3.841units, p < 0.05) and 0.1% (∆MOFV = 10.8 units, p < 0.001) were considered during the forward addition and backward elimination steps, respectively.Only covariates providing a reduction in IIV associated with parameters of at least 10% were considered clinically relevant and were retained in the model.

Model Evaluation and Internal Validation
Goodness-of-fit plots were analyzed throughout the modeling process.The predictive capability was evaluated using prediction-corrected visual predictive checks (pcVPC) based on 1000 simulations [44].The median and 2.5th and 97.5th percentiles of the simulated data and their respective 95% prediction intervals were calculated and visually compared with the same percentiles obtained from the original raw data.Also, Npde [45] (normalized prediction distribution errors) diagnostics were performed.Model adequacy was evaluated by checking the even distribution of predicted discrepancies and comparing the shape, location, and variance of distribution parameters to the theoretical normal distribution.

Patient Characteristics and Datasets
Blood Tac concentration-time data (n = 655) from 98 renal transplant recipients were simultaneously analyzed.From this data, 480 out of 655 were obtained from the rich AUC sampled group (n = 30), at least six months after transplantation, and 175 were obtained from C 0 values (n = 68).
The median total daily administered dose was 2 mg with a wide variation between patients ranging from 0.5 mg to 12 mg.Figure 1 shows the overlaid individual full concentration time and normalized-by-dose concentration time profiles.In both cases, a large variability was observed, especially in the absorption phase, with a wide range of variation in Tmax among patients and more than one peak in some cases.
Demographic, biochemical, and genetic characteristics of the patient population are summarized in Table 1.Patient characteristics were similar when both groups were compared (clinical and follow-up study).Similar values in the hematocrit, serum creatinine, and renal function values between groups were observed.Table 2 displays comparative C 0 and AUC values sorted by the CYP3A4/A5, ABCB1 SNPs groups, and cluster phenotypes (Table 2).Statistically significant (p < 0.001) differences were found when comparing normalized by dose AUC values between CYP3A5 *1 expressers vs. non-expressers.Similarly, statistically significant (p < 0.001) differences were found when comparing C 0 normalized by a dose of CYP3A5 *1 expressers vs. non-expressers.Demographic, biochemical, and genetic characteristics of the patient population are summarized in Table 1.Patient characteristics were similar when both groups were compared (clinical and follow-up study).Similar values in the hematocrit, serum creatinine, and renal function values between groups were observed.Table 2 displays comparative C0 and AUC values sorted by the CYP3A4/A5, ABCB1 SNPs groups, and cluster phenotypes (Table 2).Statistically significant (p < 0.001) differences were found when comparing normalized by dose AUC values between CYP3A5 *1 expressers vs. nonexpressers.Similarly, statistically significant (p < 0.001) differences were found when comparing C0 normalized by a dose of CYP3A5 *1 expressers vs. non-expressers.
When cluster combinations were considered, the C0 normalized by dose was significantly different among all phenotypes (p < 0.01) (Table 2).For AUC/D values, statistically significant differences were found among all phenotypes, except between IM and PM (Table 2).
When cluster combinations were considered, the C 0 normalized by dose was significantly different among all phenotypes (p < 0.01) (Table 2).For AUC/D values, statistically significant differences were found among all phenotypes, except between IM and PM (Table 2).
In contrast, no statistically significant differences were found when AUC values were compared; that is, the normalized-by-dose ABCB1 *C carriers (high pumper) vs. noncarriers (low pumper) (p > 0.05).

Population PK Analysis
The PK profile of the Tac whole-blood concentration versus time data was best described by a two-compartment model with linear elimination, parameterized in terms of apparent elimination and distributional clearances (CL/F and CL D /F, respectively) and apparent central compartment (V c /F) and peripheral compartment (V p /F) distribution volumes.A time-lagged first-order absorption process was characterized using transit compartment models and the number of absorption compartments were fixed to 2. According to the base model, the MeltDose ® Tac shows a mean absorption transit time of 2 h 55 min.Inter-individual variability could be associated with CL/F, V c /F, and V p /F. Adding IOV on CL/F caused a statistically significant reduction in the model MOFV (∆MOFV = −232 units, p < 0.005) and was therefore retained in the final model.Further, inclusion of the IOV on other parameters did not improve the model.A proportional error model best described the RE distribution.When covariates were entered univariately on distribution Vc/F and V p /F volumes or clearances, neither body weight nor body mass index or age provided a significant drop in the MOFV (p > 0.05).Allometric inclusion of body weight on these PK parameters worsened the model.No improvement of the model was observed when residual error associated with concentrations was standardized by hematocrit values of 45%.
Genetic polymorphism in CYP3A5 (*1 expressers vs. non-expressers) statistically significantly influenced Tac CL/F, resulting in an MOFV drop of 25 points (p < 0.001), and a reduction in unexplained IIV associated with CL/F of 19%.Two different clearance values were observed for each group, i.e., 20.4 L•h −1 for expressers vs. 10.1 L•h −1 for non-expressers.In contrast, when CYP3A4 genotypes were tested on CL/F (CYP3A4*22 carriers vs. non-carriers), no statistically significant difference was found and no reduction in the IIV associated with CL/F was observed.A cluster of CYP3A4 and CYP3A5 polymorphism was created, combining both SNPs.The different cluster phenotype groups resulted in significantly different means of C 0 normalized by dose (Table 2), 1.23, 2.94, and 4.04 ng•mL −1 for HM, IM, and PM groups (p < 0.001), respectively.The inclusion of cluster as a covariate significantly decreased the MOFV with respect to the base model, resulting in an MOFV drop by 37.5 points (p < 0.001).Three different CL/F values were identified, one per each cluster phenotype, i.e., CL/F values were 19.6 L•h −1 , 10.6 L•h −1 , and 7.37 L•h −1 , for high, intermediate, and poor metabolizers, respectively.After the inclusion of this covariate, a reduction of 22.5% was observed in IIV associated with CL/F.
On the other hand, the ABCB1 *C/*C SNP failed to influence the model significantly.The unexplained IIV associated with the CL/F was reduced from 49% to 38% from the base to the final covariate model.The final model parameters were estimated with good precision, the RSE% of all the parameters being lower than 35%.The eta shrinkage for CL/F in the final model was 13%.Residual unexplained variability (RUV) associated with the final model was 9.67% and the corresponding shrinkage was 23.5%.The final population pharmacokinetic parameter values are displayed in Table 3.The descriptive capability of the model was confirmed by the goodness of fit plots.Good correlations between the observed and population and individual predicted concentrations were found with a random distribution around the identity line.Individual weighted residuals (IWRES) did not show any trend when plotted against individual predicted concentrations, confirming that residual error was adequately modeled.Similarly, conditional weighted residuals (CWRES) indicated that the structural part of the model adequately described the data (Figure 2).The scatter plots of NPDE vs. time and individual predicted concentrations (Figure 3) showed a random distribution around the null line with most of the predicted NPDE values within the 95% confidence interval of the theoretical normal distribution, proving the descriptive capability of the model.The pcVPC (Figure 4) showed that the model properly describes the mean tendency of the entire data.In general, the median, 5 th , and 95 th percentiles of the observations were within the 95% confidence intervals for the median, the 5 th , and 95 th percentiles of the simulated profiles.The pcVPC (Figure 4) showed that the model properly describes the mean tendenc of the entire data.In general, the median, 5 th , and 95 th percentiles of the observations wer within the 95% confidence intervals for the median, the 5 th , and 95 th percentiles of th simulated profiles.The pcVPC (Figure 4) showed that the model properly describes the mean tendency of the entire data.In general, the median, 5th, and 95th percentiles of the observations were within the 95% confidence intervals for the median, the 5th, and 95th percentiles of the simulated profiles.

Discussion
To the best of our knowledge, this is the first population approach study focused on the influence of genetic polymorphisms of CYP3A5, CYP3A4, and ABCB1 on Tac pharmacokinetics upon LCP-Tac administration in stable kidney transplant patients.Only in one previous study from Woillard et al. [37], a similar number of data was analyzed (637 from two full PK profiles vs. 655 in our study) from fewer patients (47 vs. 98 in our study), but no influence of genetics was explored.In another study, Henin et al. [35] included a larger set of data and also addressed the influence of CYP3A5 SNPs but not of CYP3A4 and ABCB1, which are also involved in Tac pharmacokinetics, in "de novo" patients.
Unlike previous studies in kidney transplant patients [35,37], the disposition profile of Tac after LCP-Tac administration was best described by a two-compartment model.The delayed first-order absorption was described by two transit compartments and a mean transit time of 2.91 h.The number of transit compartments had to be fixed in the final model to better estimate the central compartment distribution volume and inter-individual variability associated with this parameter.The description of the delayed absorption and disposition profile provided by our final model was consistent with that of Martial et al. [36] in stable adult hepatic transplant patients (two compartmental models, NN = 1.6 and MTT = 3.4 h).In contrast, other studies in renal transplantation [35,37] omitted one disposition compartment to better describe the complex Tac-delayed absorption from LCP-Tac, hereby characterizing two and three absorption phases with rapid/slow [37] or rapid/medium/slow [35] absorption kinetics.According to these models, the highest fraction of drug absorbed occurred in the slow (60%) [37] and the medium kinetic absorption phases (61%, MTT = 4.82 h) [35], which probably represent what occurs at the more distal parts of the small intestine as described before for LCP_Tac [46].Thus, a scintigraphy study carried out in healthy volunteers revealed that LCP-Tac reached the colon at around 3.8-4.8h post-administration [46].Interestingly, our MTT (2.91 h) was between the fast (1.06 h) and medium (4.82 h) mean times reported by Henin for the absorption process.On the other hand, in our study and previous [35], in general, a high inter-individual variability was observed for parameters characterizing the absorption process.Herein lies the reason LCP-Tac is not immediately released at the proximal part of the intestine, where due to the higher expression of CYP3A4/A5, a higher first-pass effect takes place.Regarding the effect of ABCB1 SNP (*T/*T, *C/*T, and *C/*C SNPs), no influence on CL/F was detected, nor differences between exposure metrics of ABCB1*C vs. non-carriers (Table 2).This result is in agreement with previous studies carried out with the Tac-IR formulation exposure [7,25].
Our CL/F values were lower in CYP3A5 expressers vs. non-expressers (20.4 L•h −1 and 10.1 L•h −1 , respectively) than those reported by Henin et al. [35] in "de novo" patients.Specifically in this population, considering a mean bodyweight of around 70 kg, CL/F of 33.2 and 20 L•h −1 were estimated for CYP3A5 expressers.These differences could be due to co-administered corticosteroids at high doses in the early period post-transplant but also to hematocrit concentrations between "de novo" and stable patient sub-populations.Patients of our study received reduced doses or discontinued corticosteroids, compared to the higher doses given to "de novo" patients, in accordance with standard practice.Higher doses of corticosteroids can lead to greater induction of the CYP3A4 metabolizing enzyme and thus lead to higher Tac clearance [48].This finding again supports the role of CYP3A4 on the metabolism of Tac.On the other hand, hematocrit concentrations of our study were higher (from 37.4 to 44.0% (IQR)) than those reported in "de novo" patients [35] (from 22.9 to 37%).As a restrictive clearance drug, Tac is expected to decrease its clearance as the free fraction decreases, as occurs in our stable patients where hematocrit concentrations reach its almost normal values compared with "de novo" patients.This was also the reason why probably this covariate was not predictive of CL/F variability in the final model.All these findings support the development of the current model for LCP-Tac in stable patients.
However, this study has some limitations.It was performed in stable kidney transplant recipients more than 6 months after transplantation.Therefore, it may not be appropriate for de novo transplant recipients receiving LCP-Tac.Another limitation is the low prevalence of PM patients in the Caucasian population.Further studies with a larger sample size are required to improve the model and confirm the current results regarding the influence of CYP3A4*1 and CYP3A5*3 genetic polymorphisms on Tac exposure in stable kidney transplant patients treated with LCP-Tac.Finally, external validation of this PopPK model would be needed before using the final model as a support tool during therapeutic drug monitoring.

Conclusions
A new population pharmacokinetic model has been developed in stable renal adult transplant patients upon LCP-Tac administration.The influence of CYP3A4/A5 SNPs as a combined cluster including three different phenotypes has been identified as a powerful covariate to describe part of the inter-individual variability associated with the apparent elimination clearance.This suggests the inclusion of this covariate could improve dose optimization in routine clinical practice.Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Figure 1 .
Figure 1.(A) Overlaid individual steady-state PK profiles of Tac following LCP-Tac administration.(B) Overlaid normalized by dose individual steady-state PK profiles of Tac following LCP-Tac administration.

Figure 1 .
Figure 1.(A) Overlaid individual steady-state PK profiles of Tac following LCP-Tac administration.(B) Overlaid normalized by dose individual steady-state PK profiles of Tac following LCP-Tac administration.
around the null line with most of the predicted NPDE values within the 95% confidence interval of the theoretical normal distribution, proving the descriptive capability of the model.

Figure 2 .
Figure 2. Goodness-of-fit plots of the final model.(a) Observed concentrations vs. populationpredicted concentrations.(b) Observed concentrations vs. individual predicted concentrations.(c) Individual weighted residuals (IWRES) vs. individual predicted concentrations.(d) Conditional weighted residuals (CWRES) vs. time from the start of the study.

Figure 3 .
Figure 3. Normalized prediction distribution errors (NPDE) of the final PopPK model.(a) Scattered plot of NPDE vs. time from the start of the treatment.(b) Scattered plot of observed NPDE vs. individual predicted concentrations.The central dashed black line represents the null line, whereas the dotted and dashed black lines above and below the zero line represent the 95% confidence interval for the distribution of NPDEs.

Figure 2 .
Figure 2. Goodness-of-fit plots of the final model.(a) Observed concentrations vs. populationpredicted concentrations.(b) Observed concentrations vs. individual predicted concentrations.(c) Individual weighted residuals (IWRES) vs. individual predicted concentrations.(d) Conditional weighted residuals (CWRES) vs. time from the start of the study.

Figure 2 .
Figure 2. Goodness-of-fit plots of the final model.(a) Observed concentrations vs. population predicted concentrations.(b) Observed concentrations vs. individual predicted concentrations.( Individual weighted residuals (IWRES) vs. individual predicted concentrations.(d) Condition weighted residuals (CWRES) vs. time from the start of the study.

Figure 3 .
Figure 3. Normalized prediction distribution errors (NPDE) of the final PopPK model.(a) Scattere plot of NPDE vs. time from the start of the treatment.(b) Scattered plot of observed NPDE v individual predicted concentrations.The central dashed black line represents the null line, wherea the dotted and dashed black lines above and below the zero line represent the 95% confidenc interval for the distribution of NPDEs.

Figure 3 .
Figure 3. Normalized prediction distribution errors (NPDE) of the final PopPK model.(a) Scattered plot of NPDE vs. time from the start of the treatment.(b) Scattered plot of observed NPDE vs. individual predicted concentrations.The central dashed black line represents the null line, whereas the dotted and dashed black lines above and below the zero line represent the 95% confidence interval for the distribution of NPDEs.

Figure 4 .
Figure 4. Prediction-corrected visual predictive check (pcVPC) for the final model.Tac concentrations given in ng•mL −1 , time after dose given in hours.The solid line represents the median observed whole blood concentrations (ng•mL −1 ); prediction-corrected plasma concentration in the pcVPC to the right, and the red band represents a simulation-based 95% confidence interval for the median.The observed 5% and 95% percentiles are presented with dashed red lines, and the 95% confidence intervals for the corresponding model predicted percentiles are shown as blue bands.The observed whole blood concentrations (prediction corrected in the pcVPC) are represented by blue circles.

Figure 4 .
Figure 4. Prediction-corrected visual predictive check (pcVPC) for the final model.Tac concentrations given in ng•mL −1 , time after dose given in hours.The solid line represents the median observed whole blood concentrations (ng•mL −1 ); prediction-corrected plasma concentration in the pcVPC to the right, and the red band represents a simulation-based 95% confidence interval for the median.The observed 5% and 95% percentiles are presented with dashed red lines, and the 95% confidence intervals for the corresponding model predicted percentiles are shown as blue bands.The observed whole blood concentrations (prediction corrected in the pcVPC) are represented by blue circles.

Table 1 .
Demographic, biochemical, and clinical characteristics of the patients included in the study.
Values are given as arithmetic mean (interquartile range) for continuous variables, and as count (percentage) for categorical variables; BMI, body mass index; HTC, hematocrit; GF, glomerular filtration rate estimated by the CKD-EPI formula; Cr, serum creatinine; HM, high metabolizer; IM, intermediate metabolizer; PM, poor metabolizer.

Table 2 .
Comparative normalized-by-dose C 0 and AUC values sorted by the CYP3A5, CYP3A4, and ABCB1 SNP groups and by the three different cluster phenotypes.Values are reported as geometric means (95% CI).Abbreviations: AUC, area under the whole-blood concentrationtime curve; Ctrough: trough whole blood concentrations, Ctrough/D and AUC/D normalized by dose Ctrough and AUC values.N = number of occasions data analyzed is the same for Ctrough and Ctrough/D.Similarly, N is the same for AUC and AUC/D.Dose expressed as median (range).* p-values for Ctrough/D mean values statistical comparisons, z p-values for AUC/D mean values statistical comparisons, # differences between HM and IM, & differences between HM and PM, $ differences between IM and PM.

Table 3 .
Pharmacokinetic parameter estimates of the final model.Relative standard errors are given in parenthesis.

Table 3 .
Pharmacokinetic parameter estimates of the final model.Relative standard errors are given in parenthesis.