A model-based workflow to benchmark the clinical cholestasis risk of drugs
Authors:
Vanessa Baier1, Olivia Clayton3, Ramona Nudischer3, Henrik Cordes1, Annika R. P. Schneider1, Christoph Thiel1, Timo Wittenberger2, Wolfgang Moritz5, Lars M. Blank1, Ulf P. Neumann7, Christian Trautwein8, Jens Kelm5, Yannick Schrooders4, Florian Caiment4,, Hans Gmuender2, Adrian Roth5, José V. Castell6, Jos Kleinjans4, Lars Kuepfer1,9,*
Affiliations:
(1) Institute of Applied Microbiology, RWTH Aachen, Germany
(2) Genedata AG, Basel, Switzerland
(3) Roche Pharmaceutical Research and Early Development, Roche Innovation Center Basel, Basel, Switzerland
(4) Department of Toxicogenomics, Maastricht University, Maastricht, Netherlands
(5) InSphero AG, Schlieren, Switzerland
(6) Instituto de Investigación Sanitaria. Hospital Universitario La Fe, Valencia, Spain
(7) Department of Surgery, University Hospital Aachen, Aachen, Germany
(8) Department of Medicine III, University Hospital RWTH Aachen, Aachen, Germany
(9) Institute of Systems Medicine, University Hospital RWTH Aachen, Aachen, Germany
Abstract
We present a generic workflow combining physiology-based computational modeling and in vitro data to assess the clinical cholestatic risk of different drugs systematically. Changes in expression levels of genes involved in the enterohepatic circulation of bile acids were obtained from an in vitro assay mimicking 14 days of repeated drug administration for ten marketed drugs. These changes in gene expression over time were contextualized in a physiology-based bile acid model of glycochenodeoxycholic acid. The simulated drug-induced response in bile acid concentrations was then scaled with the applied drug doses to calculate the cholestatic potential for each compound. A ranking of the cholestatic potential correlated very well with the clinical cholestasis risk obtained from medical literature. The proposed workflow allows benchmarking the cholestatic risk of novel drug candidates. We expect the application of our workflow to significantly contribute to the stratification of the cholestatic potential of new drugs and to support animal-free testing in future drug development.
Introduction
Drug-induced cholestasis is a severe incident in clinical practice. About 18 – 32% of patients in clinical studies investigating cases of drug-induced liver injury (DILI) can be classified as cholestatic (1). The associated patient mortality of drug-induced cholestasis is estimated to be up to 10% (2). Cholestasis describes an impaired bile flow from the liver to the gastrointestinal tract. In consequence, toxic bile acids (BAs) accumulate inside the liver and other tissues during intrahepatic cholestasis, ultimately leading to symptoms such as dilated bile canaliculi or bile stasis (3). However, a mechanistic understanding of the diverse molecular events underlying drug-induced cholestasis is incomplete to date. This is owed to the systemic nature of cholestasis involving the enterohepatic circulation (EHC) of BAs with biochemical transformation steps in different tissues along the gut-liver axis. Hence, a pure in vitro assessment is hampered by the fact that assays usually represent only one single isolated tissue.
We present a model-based in vitro workflow integrating in vitro data into a previously developed physiology-based bile acid (PBBA) model of glycochenodeoxycholic acid to predict the cholestatic drug potential (please see Figure S11 and Table S1) (4). The PBBA model was used to contextualize in vitro expression data of key genes in the BA metabolism obtained from a specifically designed in vitro assay using human liver spheroids (5). The different incubation times allow tracking the adaptation of hepatic tissue in response to repeated drug administration. The results are a quantitative estimate of changes in the BA metabolism induced by in vivo drug exposure and allow an evaluation of the cholestatic potential for each drug reflected by the specific change of BA levels in the clinical situation. In addition, the presented workflow can be used to assess the cholestatic potential of novel drugs by benchmarking this functional change against those compounds considered as reference hepatotoxicants in this study.
Methods
PBPK-assisted liver spheroid in vitro assay. The in vitro incubation experiments with 3D InSight™ Human Liver Microtissues were conducted with a specifically designed in vitro assay mimicking in vivo drug exposure (5). Assay concentrations of the hepatotoxicants were applied according to the simulations of drug-specific PBPK models predicting the in vivo liver exposure during repeated therapeutic dosing according to the drug label. 234 samples of spheroids were taken and analyzed with sampling time points at 2 h, 8 h, 24 h, 72 h, 168 h, 240 h, and 336 h. Samples were sequenced, and RNA fold changes of the genes coding for the liver proteins CYP7A1, BSEP, and NTCP were obtained at different time points during the two weeks of treatment (Figures S12-S14).
Transcriptome analysis. Raw RNA-seq data were processed using Genedata Profiler® software v.11.0. For each sample, sequenced reads were mapped to the human genome version hg38 with the splice junction mapper STAR (version 2.5.3a) (6) using as annotation the reference genome gencode version 26 (October 2016 freeze, GRCh38) – Ensembl 88 (see Supplementary Material). Gene expression was used as a surrogate for protein expression since protein abundance could not be measured for membrane-bound transporters.
PBBA model. A physiology-based bile acid (PBBA) model has been developed previously, which describes the enterohepatic circulation of the exemplary bile acid glycochenodeoxycholic acid (GCDCA) in the body (4). The PBBA model is based on concepts from physiologically-based pharmacokinetic modeling (PBPK) (7) and simulates the systemic distribution and EHC of BAs, including their synthesis, transport, distribution, and excretion (please see Figure S11 and Table S1 in the Supporting Information). The model was carefully validated using various experimental data sets, including blood plasma concentration-time profiles, post-prandial effects, synthesis rates as well as BA pool sizes (4). The PBBA model includes four active transporters (BSEP, NTCP, ASBT, and OSTα) and synthesis through CYP7A1 (see Figure S11 and Table S1 in the Supporting Information). Active transporters, such as the sinusoidal sodium/taurocholate co-transporting polypeptide (NTCP) and the canalicular bile salt export pump (BSEP), are explicitly considered in the PBBA model. Expression of CYP7A1 is also taken into account. Active intestinal uptake transport such as apical sodium–bile acid transporter (ASBT) and basolateral organic solute transporter α (OSTα) have also been implemented in the PBBA model.
Results
We present a workflow to assess a drug’s clinical cholestasis risk based on physiology-based computational modeling and specifically-designed in vitro experiments. The overall workflow consists of the following six steps (Figure 1): Step 1) drug-specific PBPK models simulating clinically relevant administration protocols were used to calculate in vivo-like liver concentration profiles, which were then translated into an in vitro experimental design; step 2) 3D human liver spheroids were incubated with drug concentrations that correspond to simulated in vivo liver PK profiles; step 3) omics data were generated as assay readout; step 4) expression fold changes were integrated into the PBBA model, and drug-provoked changes in BA levels were simulated; step 5) simulated BA levels were compared to the clinical cholestasis risk for each drug. For new test compounds, steps 1) to 5) can be performed, and the induced BA levels can then be benchmarked with the already categorized drug set to estimate the cholestatic potential of this new compound (Step 6).
In a preceding step, we selected ten compounds known to cause DILI events according to the literature, namely: 5-fluorouracil (5FU), acetaminophen (APAP), azathioprine (AZA), cyclosporine A (CSA), diclofenac (DIC), isoniazid (ISO), methotrexate (MTX), phenytoin (PHE), rifampicin (RIF), and valproic acid (VPA) (Figure 2). These compounds were chosen to include various types of hepatotoxicity. While some compounds are well-known examples for either drug-induced cholestasis (AZA, CSA) or drug-induced hepatocellular toxicity (APAP, 5FU), others induce mixed types. Still, the clinical DILI risk of a drug, particularly its clinical cholestasis risk, is hard to quantify due to the largely patient-specific rather than the dose-related character of clinical DILI and drug-induced cholestasis. We, therefore, used various data resources to assess the clinical cholestasis risk of a given drug (Figure 2). First, cholestasis case numbers were taken from the Spanish DILI repository (8). Similar information was obtained from cholestasis labels of LiverTox (9) and a comprehensive review of drug-induced cholestasis (10). Finally, own patient data were included from DILI patients at the Hepatotoxicity Clinical Unit of the Hospital HuLaFe in Valencia, Spain. In this data set, a total of 19 samples of patients displaying cholestasis after a DILI episode attributed to treatment with AZA or MTX were analyzed by MS-TOF to determine a set of BAs present in sera. Hence, four different sources were considered to quantify the clinical cholestasis risk of a drug (Figure 2).
Based on this data collection, all ten hepatotoxicants were assigned to a low, medium, or high category of the clinical cholestasis risk (Figure 2). For this, we checked for each compound if it was reported as cholestasis-inducing in any of the four sources. We used the cumulated number of occurrences (one point per source, without relative weighting) for each drug to differentiate the clinical cholestasis risk in the categories low (0 reports: 5FU and APAP), medium (1 report: MTX, PHE, and VPA) and high (2+ reports: AZA, CSA, DCL, ISO, and RIF) (Figure 2). This review of cholestasis reports is an essential prerequisite for our further analyses since it allows for an objective assessment of the clinical cholestasis risk for each compound, even though the amount of data varies considerably between the sources.
Drug-induced cholestasis usually occurs after multiple administrations of a drug (10). Therefore, it is mandatory to use a long-term in vitro assays to reproduce the drug intake scenario and the underlying mechanisms. To account for the adaptation of hepatic gene expression following repeated drug dosing, a specifically designed in vitro assay was implemented (5). In order to mimic the actual in vivo pharmacokinetics, the in vitro treatment concentrations were simulated beforehand with drug-specific PBPK models (Figure 1, Step 1). PBPK models for all ten drugs have been developed and were validated (Figure 3, Figures S1-S10, see Supplementary information) (5, 11-14). The models were used to simulate concentration profiles in the interstitial space of the liver over two weeks of therapeutic dosing regimen according to the specific drug label (Figure 1, Step 2, Methods). To this end, the simulated PK profiles are discretized at multiple sampling times, and assay media, which correspond to the interstitial space of the liver, are replaced after 2 h, 8 h, and 24 h each day (5). The incubation concentrations for the 3D human liver spheroid thereby approximate the in vivo situation predicted by the various PBPK models. As a readout, the gene expression fold changes of the three genes coding for BSEP, NTCP, and CYP7A1 (see Methods) from the measured transcriptome data were extracted and used to approximate changes in protein concentration in the model (Figure 1, Step 3) since protein abundance could not be measured for membrane-bound transporters. These concentration changes over the treatment time were then integrated into the previously published PBBA model (4) (see Methods). The model simulates the systemic distribution and EHC of endogenous glycochenodeoxycholic acid (GCDCA) as a representative BA, including its synthesis, transport, distribution, and excretion. A scaling factor was used to calculate the total BA pool from GCDCA concentrations. The PBBA model was simulated with the updated transporter activities to investigate the effects on BA levels over two weeks of therapeutic drug treatment (Figure 1, Step 4).
The simulations were performed for all ten drugs in a virtual population of healthy individuals, and the resulting changes in BA levels for different tissues were analyzed. In a previous work, a virtual population of 1,000 individuals with variability in base anthropometry, physiology, and protein concentration was found to adequately describe the physiological variability of plasma BA levels (see Methods). Individuals with disadvantageous anthropometric and physiological parameters may display BA levels significantly affected by drug administration and, therefore, represent subgroups of patients who are highly susceptible to drug-induced cholestasis.
The population simulation illustrates the time-dependent development of BA levels in response to the adaptation of hepatic gene expression during repeated drug dosing (Figure 4, Figure S11). Changes in BA concentrations are exclusively caused by the measured alterations in gene expression, which are simultaneously contextualized within the PBBA model. BA levels were simulated for the therapeutic doses of all 10 model drugs. The results are summarised with boxplots in Figure 4. Each boxplot presents the simulation of the same 1,000 healthy individuals with the integrated in vitro changes after treatment with one of the ten drugs. The plots show a drug-specific development of BA levels over the treatment time.
For 5FU, median plasma BA levels are reduced over the whole treatment period (Figure 4). For APAP, VPA, AZA, and DIC BA levels are fluctuating, but they ultimately increase. The remaining five compounds (MTX, PHE, CSA, ISO, and RIF) rise continuously at later time points. Comparison with measured patient BA levels after an MTX-induced cholestasis event (red line) shows good agreement with our simulated BA levels. This finding strongly supports the general relevance of the computational prediction. For AZA, the clinically measured BA levels were even higher than in our simulations. In both cases, our results indicate the need for longer exposure times above seven days to identify clear trends in the drug-induced effect.
We next analyzed the predicted BA concentrations in liver cells where increased BA levels may ultimately induce apoptosis of hepatocytes (3). Of note, it has been shown before that BA levels in tissue may differ from those in systemic blood plasma (4, 15). We, therefore, compared the mean of the 10% maximal BA levels reached in blood plasma with those simulated in liver cells (Figure 5). It was found that liver concentration levels exceed plasma BA levels at 14 days for population outliers. This is in contrast to the average population, where such an effect is not observable (mean AUC as well as mean Cmax, results not shown). For population outliers, all drugs are dense together after three days of treatment, while they diverge on day 7 and day 14 and ultimately cover a wide range of BA concentrations. In agreement with another study (16), the regression line reveals that the liver exposure disproportionately exceeds the blood plasma levels the longer the treatment period is. This confirms that the systemic plasma levels are not sufficient for describing the accumulation of BA in the liver.
This observation is further confirmed by the percentage of individuals in the virtual population whose tissue concentration levels lie above a particular threshold value. From in vitro measurements, it has been estimated that BA levels above 15 µmol/l induce integrin signaling, BA levels above 50 µmol/l induce apoptosis, and BA levels above 200 µmol/l lead to necrosis (3). Liver concentrations of BA induced by all ten drugs reach values above 15 µmol/l and 50 µmol/l, respectively. Even the critical value of 200 µmol/l is exceeded by AZA, MTX, PHE, CSA, and DIC. Consequently, for these drugs, one would expect cholestasis and liver damage because of BA accumulation, at least in susceptible patients represented through increased tissue concentration levels (Table S1, Supplementary material).
The various analyses indicate significant differences in the plasma BA levels induced by different drugs at therapeutic doses (Figure S13). Since the observed drug response, however, is exposure-driven due to the underlying PBPK-based assay design, the simulated BA levels were again divided by the therapeutic drug dose applied (Figure 6). This normalization, hereafter referred to as “cholestatic potential”, is conceptually similar to the potency of a drug in pharmacology. Of particular note, this normalization ensures that drugs with a similar cholestasis risk show a similar BA response at therapeutic dose levels. In our analysis, the ten tested drugs were first ranked according to their cholestatic potential (Figure 6 and Figure 1, Step 5). In a second step, the bars were colored according to the clinical cholestasis risk of each drug (Figure 6, Table 1). This visualization shows a good correlation of the cholestatic potential, which is quantified by the size of each bar, and the clinical cholestasis risk indicated by the color code (Figure 2). On the one hand, CSA, ISO, DIC, and AZA show a large cholestatic potential in perfect agreement with their high clinical cholestasis risk. On the other hand, 5FU, VPA, and APAP were all found to have a small cholestatic potential, again corresponding to the low clinical cholestasis risk. Importantly, the choice of the PK parameter (AUC or Cmax) or the compartment (venous blood plasma or liver intracellular) has only a slight impact on the ranking with single drugs changing positions among each other in the low/medium-risk range (VPA, 5FU and APAP) or the high-risk range (ISO and CSA or AZA and DIC). The eminent outlier in the ranking of the cholestatic potential is MTX which is in the top position concerning the cholestatic potential but has only a low clinical cholestasis risk in clinical practice.
Since the cholestatic potential of each of the ten hepatotoxicants has been quantified independently, the clinical cholestasis risk of a novel drug candidate can be analyzed with the same model-based in vitro workflow. The relative ranking between known hepatotoxicants, as done in this study, allows a direct assessment of the to-be-expected clinical cholestasis risk of a new pharmaceutical development candidate. The proposed steps for benchmarking of the cholestatic potential of novel drugs are as follows:
1. build a PBPK model for a drug candidate from in vitro measurements and simulated in vivo-like liver drug concentrations, which can then be applied in an in vitro cell assay;
2. contextualize the measured expression data with the PBBA model and predict alterations of BA levels for the drug candidate;
3. calculate the cholestatic potential of the drug candidate by dividing the BA levels by the applied therapeutic dose of the drug candidate;
4. rank the cholestatic potential of the drug candidate within a set of known benchmark compounds.
This benchmarking of the cholestatic potential finally allows the assignment of the clinical cholestasis risk to a novel drug candidate at an early phase of clinical development based on standard preclinical in vitro information without the need for animal sacrifices.
Discussion
Identifying and assessing the cholestasis risk of drugs is a challenge in drug development and clinical practice (17). We applied a model-based integrative workflow in which time-series expression data were contextualized in a physiology-based bile acid model to simulate changes in BA levels after repeated drug dosage (Figure 1). In a first step, gene expression data were obtained from a 3D human liver spheroid assay designed to reproduce in vitro the hepatic drug exposure occurring in vivo after multiple drug administration at physiologically relevant concentrations (5). The hepatic drug exposure was estimated for each substance by a drug-specific PBPK model, which requires basic physicochemical information as well as a functional ADME understanding of the respective compound. Such a level of information is currently available at late preclinical and early clinical phases and does not rely on animal data, which makes our approach complementary to current workflows in preclinical development. In a second step, gene expression data were contextualized in the PBBA model (4) to simultaneously integrate changes in expression of multiple hepatic transporters and enzymes over time. This is an extension of current approaches to identify cholestatic compounds from isolated targets since it allows the observation of time-resolved long-term effects and the complex interplay of multiple transporters and enzymes in BA metabolism within an organism. In that sense, our generic workflow is an example for analysis at the systems level in pharmacology and toxicology by integrating in silico, in vitro, and in vivo layers and accounting for tissue interplay at the organ level.
A computational model of bile acid homeostasis has been developed and validated in the commercial DILIsym® platform, which has been applied in several studies (17). However, since DILISym® is a closed source tool, customized modifications such as in the present work are at least difficult to implement. Our simulations allow analyzing the functional effect of repeated drug administration on BA metabolism, including changes of BA levels in blood plasma and organs such as the liver. The time-series data account for the adaptation of hepatic gene expression following multiple drug administration to cover a potentially delayed onset of cholestasis. The results thereby enhance a mechanistic understanding of physiological processes underlying chronic toxicity, for example, increased accumulation of BAs in hepatocytes due to a locally impaired excretion transport. Furthermore, the normalization of the simulated BA profiles by the therapeutic drug dose allows calculating the cholestatic potential of a drug which is then ranked among different hepatotoxic drugs. This ranking, which is only obtained from the model-based in vitro workflow, correlates very well with the clinical cholestasis risk of the drugs (Figure 6). The presented workflow thus provides a possibility for benchmarking the to-be-expected cholestasis risk for novel drug candidates to a set of marketed compounds with a known risk profile (Figure 1, step 6). Since the workflow is based on the whole-body PBBA model, it accounts for different interfering factors of BA metabolism at the systems level along the enterohepatic circulation of BAs and could not have been achieved with a standard in vitro assays alone (18). We also included the simulation of a virtual population to account for individuals with a physiological predisposition for cholestasis (4). Remarkably, our workflow does not rely on any animal data and, therefore, can significantly contribute to applying the 3Rs principles for animal welfare in pharmacology and toxicology (19).
The original version of the PBBA model for healthy individuals was validated with BA concentrations in different tissues, transition times, or turn-over times (4). In the current extension, the model-based in vitro workflow presented was used to describe changes in BA plasma levels in patients receiving repeated drug treatment based on in vitro expression data. While usage of protein expression data would have been clearly preferable, transcriptome data can be assumed as a reasonable surrogate for protein abundance since the correlation between gene and protein expression is generally positive (20). The validation of the resulting BA profiles with clinical data (Figure 4) is not straightforward due to the limited availability of adequate patient data. In real life, patients are only hospitalized once the DILI event has occurred, and the damage is already apparent and diagnosable. Hence, the onset of the cholestatic event, which is what the presented workflow basically describes, is very difficult to obtain from clinical records. Besides, patients hospitalized after a DILI event have usually been treated simultaneously with several drugs. In this study, we had access to data from DILI patients after MTX or AZA monotherapy. Comparing the measured BA levels in these patients and our computational simulations showed overall a good agreement. In the future, more clinical records of DILI cases in individual patients would be a powerful data source for model refinement and further benchmarking.
While there is generally a good correlation between the clinical cholestasis risk and the estimated cholestatic potential (Figure 6), deviations can be observed for MTX. Among the investigated drugs, the highest rise in BA levels was consistently predicted for MTX, which seems to overestimate the clinical cholestasis risk where MTX is only ranked “medium” (Figure 2). Since MTX is usually known to induce miscellaneous damage, a possible explanation for this deviation is that the clinical cholestasis risk of this drug is masked by the predominant hepatocellular damage. Hence, the simulated elevated BA levels possibly induce cholestatic damage, but the hepatocellular damage triggers the clinically apparent symptoms. Our clinical data support this hypothesis, where we have an MTX-induced cholestasis patient with elevated BA levels matching those of our simulations.
For future applications, the inclusion of more compounds would be beneficial to enhance the predictive accuracy of the benchmarking. A current limitation of the PBBA model is that it describes the enterohepatic circulation of only a single BA and considers only 4 active processes to ensure identifiability of corresponding model parameters. Future versions of the PBBA model will include different BAs and account, in particular, for the biochemical conversion among them. Further extension of the computational model could include consideration of specific effect models (21-23). Likewise, direct drug effects may be described if inhibitory binding constants were available for all ten compounds. Alternative calculations of the cholestatic potential, for example, through a normalization by bioavailability, are conceivable.
Altogether, the contextualization of in vitro expression data in a physiology-based computational model allows describing the functional effect of drug administration on BA metabolism at the systems level. Our workflow enables a time-resolved investigation of cholestasis-inducing drugs by accounting for adaptation of hepatic gene expression in response to multiple dosages. Of note, changed BA levels in hepatic tissue can be quantified, which may differ considerably from BA plasma levels (15). We expect our workflow to significantly support the application of animal-free toxicity tests in drug development in the future.
References
(1) Petrov, P.D., Fernández-Murga, M.L., López-Riera, M., Goméz-Lechón, M.J., Castell, J.V. & Jover, R. Predicting drug-induced cholestasis: preclinical models. Expert opinion on drug metabolism & toxicology 14, 721–38 (2018).
(2) Sundaram, V. & Björnsson, E.S. Drug-induced cholestasis. Hepatology communications 1, 726–35 (2017).
(3) Jansen, P.L.M. et al. The ascending pathophysiology of cholestatic liver disease. Hepatology, 722– 38 (2017).
(4) Baier, V. et al. A Physiology-Based Model of Human Bile Acid Metabolism for Predicting Bile Acid Tissue Levels After Drug Administration in Healthy Subjects and BRIC Type 2 Patients. Frontiers in physiology 10, 1192 (2019).
(5) Kuepfer, L. et al. A model-based assay design to reproduce in vivo patterns of acute drug-induced toxicity. Arch Toxicol 92, 553-5 (2018).
(6) Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics (Oxford, England) 29, 15–21 (2013).
(7) Kuepfer, L. et al. Applied Concepts in PBPK Modeling: How to Build a PBPK/PD Model. CPT: Pharmacometrics & Systems Pharmacology 5, 516–31 (2016).
(8) Andrade, R.J. et al. Drug-Induced Liver Injury: An Analysis of 461 Incidences Submitted to the Spanish Registry Over a 10-Year Period. Gastroenterology 129, 512–21 (2005).
(9) Diseases., N.I.o.D.a.D.a.K. LiverTox: Clinical and Research Information on Drug-Induced Liver Injury (Bethesda (MD), 2012).
(10) Padda, M.S., Sanchez, M., Akhtar, A.J. & Boyer, J.L. Drug-induced cholestasis. Hepatology 53, 1377–87 (2011).
(11) Cordes, H., Thiel, C., Aschmann, H.E., Baier, V., Blank, L.M. & Kuepfer, L. A Physiologically Based Pharmacokinetic Model of Isoniazid and Its Application in Individualizing Tuberculosis Chemotherapy. Antimicrobial Agents and Chemotherapy 60, 6134–45 (2016).
(12) Thiel, C., Cordes, H., Baier, V., Blank, L.M. & Kuepfer, L. Multiscale modeling reveals inhibitory and stimulatory effects of caffeine on acetaminophen-induced toxicity in humans. CPT: pharmacometrics & systems pharmacology 6, 136–46 (2017).
(13) Thiel, C. et al. A Comparative Analysis of Drug-Induced Hepatotoxicity in Clinically Relevant Situations. PLoS computational biology 13, e1005280 (2017).
(14) Thiel, C. et al. Using quantitative systems pharmacology to evaluate the drug efficacy of COX-2 and 5-LOX inhibitors in therapeutic situations. npj Systems Biology and Applications 4, 28 (2018).
(15) Eggink, H.M. et al. Transhepatic bile acid kinetics in pigs and humans. Clinical nutrition (Edinburgh, Scotland) 37, 1406–14 (2018).
(16) Albrecht, W. et al. Prediction of human drug-induced liver injury (DILI) in relation to oral doses and blood concentrations. Archives of Toxicology 93, 1609–37 (2019).
(17) Woodhead, J.L. et al. Mechanistic Modeling Reveals the Critical Knowledge Gaps in Bile Acid- Mediated DILI. CPT Pharmacometrics Syst Pharmacol 3, e123 (2014).
(18) Funk, C. & Roth, A. Current limitations and future opportunities for prediction of DILI from in vitro. Archives of Toxicology, (2016).
(19) Russell, W.M.S., Burch, R.L. & Hume, C.W. The principles of humane experimental technique. Special ed. edn. (Universities Federation for Animal Welfare: Potters Bar, 1992).
(20) Kosti, I., Jain, N., Aran, D., Butte, A.J. & Sirota, M. Cross-tissue Analysis of Gene and Protein Expression in Normal and Cancer Tissues. Sci Rep 6, 24799 (2016).
(21) Cordes, H., Thiel, C., Baier, V., Blank, L.M. & Kuepfer, L. Integration of genome-scale metabolic networks into whole-body PBPK models shows phenotype-specific cases of drug-induced metabolic perturbation. npj Systems Biology and Applications 4, 10 (2018).
(22) Danhof, M. Systems pharmacology – Towards the modeling of network interactions. European journal of pharmaceutical sciences: official journal of the European Federation for Pharmaceutical Sciences 94, 4–14 (2016).
(23) Iyengar, R., Zhao, S., Chung, S.W., Mager, D.E. & Gallo, J.M. Merging systems biology with pharmacodynamics. Sci Transl Med 4, 126ps7 (2012).
(24) Heggie, J.R. et al. Validation of a high-performance liquid chromatographic assay method for pharmacokinetic evaluation of busulfan. J Chromatogr B Biomed Sci Appl 692, 437-44 (1997).
(25) Odlind, B., Hartvig, P., Lindstrom, B., Lonnerholm, G., Tufveson, G. & Grefberg, N. Serum azathioprine and 6-mercaptopurine levels and immunosuppressive activity after azathioprine in uremic patients. Int J Immunopharmacol 8, 1-11 (1986).
(26) Zins, B.J. et al. A dose-ranging study of azathioprine pharmacokinetics after single-dose administration of a delayed-release oral formulation. J Clin Pharmacol 37, 38-46 (1997).
(27) Critchley, J.A., Critchley, L.A., Anderson, P.J. & Tomlinson, B. Differences in the single-oral-dose pharmacokinetics and urinary excretion of paracetamol and its conjugates between Hong Kong Chinese and Caucasian subjects. J Clin Pharm Ther 30, 179-84 (2005).
(28) Rawlins, M.D., Henderson, D.B. & Hijab, A.R. Pharmacokinetics of paracetamol (acetaminophen) after intravenous and oral administration. Eur J Clin Pharmacol 11, 283-6 (1977).
(29) Shinoda, S., Aoyama, T., Aoyama, Y., Tomioka, S., Matsumoto, Y. & Ohe, Y. Pharmacokinetics/pharmacodynamics of acetaminophen analgesia in Japanese patients with chronic pain. Biol Pharm Bull 30, 157-61 (2007).
(30) Gupta, S.K., Manfro, R.C., Tomlanovich, S.J., Gambertoglio, J.G., Garovoy, M.R. & Benet, L.Z. Effect of food on the pharmacokinetics of cyclosporine in healthy subjects following oral and intravenous administration. J Clin Pharmacol 30, 643-53 (1990).
(31) Crook, P.R., Willis, J.V., Kendall, M.J., Jack, D.B. & Fowler, P.D. The pharmacokinetics of diclofenac sodium in patients with active rheumatoid disease. Eur J Clin Pharmacol 21, 331-4 (1982).
(32) Willis, J.V., Kendall, M.J., Flinn, R.M., Thornhill, D.P. & Welling, P.G. The pharmacokinetics of diclofenac sodium following intravenous and oral administration. Eur J Clin Pharmacol 16, 405-10 (1979).
(33) Zhang, Y. et al. Diclofenac and Its Acyl Glucuronide: Determination of In Vivo Exposure in Human Subjects and Characterization as Human Drug Transporter Substrates In Vitro. Drug Metab Dispos 44, 320-8 (2016).
(34) Lauterburg, B.H., Smith, C.V., Todd, E.L. & Mitchell, J.R. Pharmacokinetics of the toxic hydrazino metabolites formed from isoniazid in humans. J Pharmacol Exp Ther 235, 566-70 (1985).
(35) Chladek, J. et al. Pharmacokinetics and pharmacodynamics of low-dose methotrexate in the treatment of psoriasis. Br J Clin Pharmacol 54, 147-56 (2002).
(36) Lund, L., Alvan, G., Berlin, A. & Alexanderson, B. Pharmacokinetics Glycochenodeoxycholic acid of single and multiple doses of phenytoin in man. Eur J Clin Pharmacol 7, 81-6 (1974).
(37) Acocella, G. Clinical pharmacokinetics of rifampicin. Clin Pharmacokinet 3, 108-27 (1978).
(38) Houin, G., Beucler, A., Richelet, S., Brioude, R., Lafaix, C. & Tillement, J.P. Pharmacokinetics of rifampicin and desacetylrifampicin in tuberculous patients after different rates of infusion. Ther Drug Monit 5, 67-72 (1983).
(39) Nagai, G., Ono, S., Yasui-Furukori, N., Nakamura, A., Mihara, K. & Kondo, T. Formulations of valproate alter valproate metabolism: a single oral dose kinetic study. Ther Drug Monit 31, 592-6 (2009).
(40) Perucca, E. Pharmacological and therapeutic properties of valproate: a summary after 35 years of clinical experience. CNS Drugs 16, 695-714 (2002).