Resource > Scientific publications >
Quantitative Assessment of Population Variability in Hepatic Drug Metabolism Using a Perfused Three-Dimensional Human Liver Microphysiological System
Filed under: ADME and Drug metabolism
Summary
The study used the LiverChip, CN Bio’s perfused three-dimensional (3D) liver microphysiological system (MPS) platform, to culture primary human hepatocytes from five individual donors and measure how each donor’s microtissues metabolized six compounds: phenacetin, diclofenac, lidocaine, ibuprofen, propranolol, and prednisolone. Intrinsic clearance differed substantially between donors, with a coefficient of variation (CV) between 24.1% for phenacetin and 66.8% for propranolol, and a population physiologically based pharmacokinetic (PBPK) model built on the liver MPS data predicted both the average and the spread of lidocaine plasma concentrations observed in clinical subjects. The work sets out a framework for using liver MPS data to estimate population variability in hepatic drug metabolism, which is relevant to first-in-human dose setting and to drug metabolism and pharmacokinetics (DMPK) studies.
Study facts at a glance
| Publication | Tsamandouras N, Kostrzewski T, Stokes CL, Griffith LG, Hughes DJ, Cirit M. Quantitative Assessment of Population Variability in Hepatic Drug Metabolism Using a Perfused Three-Dimensional Human Liver Microphysiological System. Journal of Pharmacology and Experimental Therapeutics. 2017 Jan;360(1):95-105. |
| DOI | 10.1124/jpet.116.237495 |
| CN Bio product used | PhysioMimix® LC12 and PhysioMimix Core System |
| How the platform was used | Cryopreserved primary human hepatocytes from five individual donors and a five-donor pool were seeded at 6 × 10⁵ cells per scaffold, formed perfused 3D microtissues over 3 days, and on day 4 received a 1 µM bolus of one of six compounds, with medium sampled over 24 to 72 hours to measure drug depletion (flow rate not specified). |
| Biological context | Healthy human hepatic drug metabolism; no disease model. Hepatocyte-only liver MPS microtissues, with no non-parenchymal cells, from five donors (lots Hu1601, Hu1604, Hu1624, Hu8150, and Hu8181; three male and two female, all Caucasian, aged 21 to 72 years) plus a pooled sample containing equal numbers of cells from each donor, maintained in serum-free medium. |
| Comparator | Individual-donor cultures compared with pooled hepatocytes; liver MPS microtissues (day 6) compared with freshly thawed hepatocytes for expression of 90 metabolism-related genes; predicted compared with observed human in vivo hepatic clearance; and population physiologically based pharmacokinetic (PBPK) predictions compared with published clinical lidocaine plasma profiles. No static or two-dimensional (2D) culture comparator was included. |
| Key readouts | Parent compound depletion for phenacetin, diclofenac, lidocaine, ibuprofen, propranolol, and prednisolone by liquid chromatography with tandem mass spectrometry (LC-MS/MS); unbound intrinsic clearance with interdonor, interwell, and residual variability from nonlinear mixed-effects modeling; formation of four metabolites; albumin, urea, and lactate dehydrogenase (LDH) before and after dosing; messenger RNA (mRNA) levels of 90 drug metabolism-related genes; unbound drug fraction in medium; nonspecific binding to the device; and microtissue imaging. |
| Main interpretation | Hepatocytes from different donors cultured in the liver MPS showed substantial, quantifiable differences in intrinsic clearance, liver function markers, and gene expression, and a population PBPK model built from these data reproduced the clinically observed variability in lidocaine plasma concentrations, although absolute hepatic clearance was underpredicted by an average of 4.2-fold or 4.5-fold, depending on the liver model used. |
Table of Contents
Which CN Bio product was used?
The study used the LiverChip, a commercially available CN Bio platform for perfused 3D liver culture and the forerunner of the PhysioMimix Core system. Each well holds a scaffold on which primary human liver cells form an array of 3D tissue structures of approximately 0.2 mm, and an on-board micropump, driven by pneumatics, perfuses culture medium through the scaffold to control oxygenation and shear stress on the tissue, supporting long-term culture.
Cryopreserved primary human hepatocytes from each of five donors, plus a pool containing equal numbers of cells from all five, were seeded at 6 × 10⁵ cells per scaffold in 1.6 mL of medium per compartment. Each donor and the pool occupied 21 wells: 18 for the drug metabolism study (six compounds in three replicate wells each) and three sacrificed on day 6 for RNA analysis. Microtissues formed over 3 days, with plating medium on day 1 and serum-free maintenance medium thereafter. On day 4, each compound was added as a 1 µM bolus during a full medium change to 1.8 mL, and medium was sampled over 24 hours (phenacetin), 48 hours (diclofenac, ibuprofen, lidocaine, and propranolol), or 72 hours (prednisolone). Hepatocyte-free LiverChip wells were dosed in the same way to test for nonspecific binding to the device. The paper does not specify the flow rate.
The LiverChip supported all in vitro culture work. The mixed-effects analysis, in vitro to in vivo extrapolation (IVIVE) of clearance, and population PBPK simulations were computational steps applied to the data generated on the platform.
Find out more about CN Bio Liver-on-a-chip models here
Find out more about CN Bio DILI assays here
What this paper is about
Predicting how fast the liver clears a new compound is central to preclinical development, because it informs the first human dose and the dosing regimens needed to keep drug concentrations within the therapeutic window. Those predictions are most useful when they describe the range of people who will receive the compound, rather than a single “average individual”. Standard in vitro systems, such as human liver microsomes and cryopreserved hepatocyte suspensions, lose metabolic activity over time, which makes low-clearance compounds difficult to study, and they tend to underpredict in vivo clearance. They are also often run with material pooled from several donors, so the results describe an average and provide no measure of interindividual variability. CN Bio’s guide to choosing in vitro liver models for drug development tasks compares hepatocyte suspensions, co-cultures, spheroids, and perfused liver MPS.
This study, a collaboration between the Massachusetts Institute of Technology (MIT) and CN Bio, set out to measure population variability in hepatic drug metabolism directly in a liver MPS. The researchers cultured hepatocytes from five donors individually in the LiverChip, measured the depletion of six compounds alongside metabolite formation, liver function markers, and gene expression, and used nonlinear mixed-effects modeling (NONMEM software) to separate interdonor differences from well-to-well differences and measurement noise. They also tested whether pooled hepatocytes give an unbiased estimate of the donor average, a common practice that had not previously been evaluated in MPS technologies. CN Bio co-author Tomasz Kostrzewski, joint first author, conducted the experiments, and the work was funded by the microphysiological systems programs of the Defense Advanced Research Projects Agency (DARPA) and the National Institutes of Health (NIH).
Because in vitro metabolism work aims to predict human pharmacokinetics, the study carried the liver MPS results through IVIVE of hepatic clearance, using parallel tube and well-stirred liver models, and into a population PBPK model for lidocaine that predicts concentration-time profiles across a virtual population. The application area is in vitro drug metabolism and DMPK rather than disease modeling.
What the researchers found
The study reported the following, in order of importance.
Substantial interdonor variability in intrinsic clearance. Typical unbound intrinsic clearance ranged between 0.81 µL/min per million cells for prednisolone and 17.8 µL/min per million cells for diclofenac. Interdonor variability (CV) was 24.1% for phenacetin, 28.5% for lidocaine, 29.3% for prednisolone, 32.6% for ibuprofen, 36.2% for diclofenac, and 66.8% for propranolol. Interwell variability within a donor was lower than interdonor variability for every compound except phenacetin (26.1% versus 24.1%) and ranged between 6.0% for diclofenac and 32.9% for propranolol. Residual unexplained variability ranged from 8.4% (lidocaine) to 21.6% (propranolol). A one-compartment model of monoexponential decline described most compounds well; diclofenac showed a decline that a biexponential model might describe better, although further analysis indicated that the simpler assumption did not introduce substantial bias.
Metabolite formation tracked parent depletion. Metabolite levels (acetaminophen from phenacetin, 4-hydroxydiclofenac, 2-hydroxyibuprofen, and 6β-hydroxyprednisolone) correlated with intrinsic clearance in the same donor and well, with R² values of 0.91 for prednisolone, 0.82 for ibuprofen, and 0.77 for diclofenac, and a weaker correlation for phenacetin (R² = 0.4). Interdonor differences in parent depletion were therefore also reflected in metabolite formation.
Clearance predictions correlated with human data but were systematically low. Hepatic clearance predicted from the liver MPS correlated strongly with observed in vivo values (R² of 0.75 with the parallel tube liver model and 0.77 with the well-stirred model). In absolute terms, clearance was underpredicted by an average of 4.2-fold (parallel tube) or 4.5-fold (well-stirred), with the smallest gap for phenacetin (1.7-fold and 2.1-fold) and the largest for propranolol (8.2-fold and 8.5-fold). The authors note that this underprediction is similar to what has been observed with other traditionally used in vitro systems.
A population PBPK model reproduced clinical variability for lidocaine. The researchers built a population PBPK model for lidocaine that integrated in silico and experimental inputs, using liver MPS clearance and its interdonor variability, along with an empirical scaling factor derived across all six compounds to correct the systematic underprediction. Simulations of 1,000 virtual individuals receiving a 3 mg/kg intravenous infusion of lidocaine hydrochloride over 3 minutes closely matched arterial plasma profiles published for five clinical subjects (Tucker and Boas, 1971), capturing both the average trend and the extent of interindividual variability. None of the clinical concentration data were used to fit model parameters.
Pooled hepatocytes gave a relatively unbiased average. Intrinsic clearance in pooled hepatocytes ranged between 0.91 µL/min per million cells (prednisolone) and 18.6 µL/min per million cells (diclofenac). The ratio of individual-donor to pooled clearance averaged 0.97 across compounds (0.61 for propranolol up to 1.42 for ibuprofen), and its 95% confidence interval included 1 for every compound except propranolol, the only compound with marginal evidence of bias. At the gene level, only five of 90 genes in the liver MPS differed significantly between pooled hepatocytes and the donor average. The authors conclude that pooled hepatocytes are a reliable option for studying drug metabolism in the liver MPS when interindividual variability is not of interest.
Liver function markers predicted clearance better than cytochrome P450 gene expression. Before dosing, albumin and urea production correlated positively, and LDH release negatively, with the intrinsic clearance later measured in the same donor and well. A least absolute shrinkage and selection operator (Lasso) regression model identified albumin, urea, LDH, and cytochrome P450 (CYP) mRNA levels as significant predictors together, explaining a substantial part of the variability in intrinsic clearance (R² = 0.52). CYP mRNA accounted for only a very small portion, and the authors conclude that screening mRNA expression of metabolic enzymes across donors should not be used as a surrogate for interdonor variability in metabolic activity.
Gene expression was largely retained and less variable in the liver MPS. On day 6, only 10 of 90 drug metabolism-related genes differed significantly from freshly thawed hepatocytes (four downregulated and six upregulated). The average interdonor CV in mRNA levels was 34% in the liver MPS compared with 72% in freshly thawed hepatocytes, which the authors hypothesize reflects adaptation to the controlled and consistent culture environment (medium composition, flow, and oxygen gradient). Some genes still differed between donors, including CYP3A5, which was strongly upregulated in donor Hu8181 and downregulated in the other four.
Hepatocytes remained functional throughout the study. Albumin, urea, and LDH differed significantly between donors before dosing, and pooled cultures matched the merged donor data for albumin but showed higher urea and LDH. After dosing (days 5 to 7), albumin production had increased and LDH release had decreased significantly compared with day 4 for every compound, while urea tended to rise for all compounds except prednisolone. The authors attribute these changes to culture time rather than compound effects. Microtissue formation was consistent across donors and maintained throughout the culture period.
No detectable drug binding to the device. Hepatocyte-free LiverChip wells showed no evidence of nonspecific binding of any of the six compounds after 48 hours. Binding to medium components varied widely, with unbound fractions of 0.13 for diclofenac, 0.31 for ibuprofen, 0.88 for lidocaine, 0.94 for prednisolone, and 0.98 for both propranolol and phenacetin, and these values were built into the clearance calculations.
Why the paper matters
For drug developers, the practical task is to predict clearance for the population that will receive a compound, including the people at either end of the metabolic range. This study shows that a liver MPS can quantify interdonor variability in intrinsic clearance, and that well-to-well variation within a donor is generally smaller than the donor-to-donor differences being measured, which the authors see as further confidence in MPS for future population variability studies. A separate cross-site reproducibility study of CN Bio’s liver MPS addresses reproducibility across laboratory sites and cell batches. The size of the differences between donors also means that clearance predictions for new compounds based on hepatocytes from a single donor should be evaluated carefully.
The paper also sets out a complete analysis workflow. The liver MPS generates concentration-time data; nonlinear mixed-effects modeling derives unbound intrinsic clearance and its variability while accounting for medium volume and drug binding to medium components; and those parameters then feed IVIVE and a population PBPK model that adds physiological variability, such as hepatic blood flow and organ volumes. The authors describe the work as the first study of population variability in drug metabolism in an MPS and, to their knowledge, the first to combine experimental liver MPS data with a computational systems pharmacology framework to make in vivo pharmacokinetic predictions. Model-based analysis of liver MPS data is also central to Roche’s quantitative drug metabolism study on the PhysioMimix liver-on-a-chip, which develops cell-number normalization and medium evaporation corrections.
Two findings directly guide study design. Pooled hepatocytes provided a relatively unbiased estimate of average clearance, so pooling remains a reasonable choice when the question concerns the typical individual. Functional readouts (albumin, urea, and LDH) carried more information about metabolic capacity than CYP mRNA levels, supporting direct functional measurement when characterizing donors.
The study focused on interdonor variability and so included a small compound set (six compounds). Its scope was also limited to five Caucasian adult donors, hepatocyte-only cultures, and a culture period of up to 7 days. The systematic underprediction of clearance was handled with an empirical scaling factor, and the authors call for studies with a wider and more diverse set of compounds to evaluate liver MPS clearance prediction and derive robust correction relationships, alongside mechanistic modeling methods designed around the features of liver MPS. The authors also note that liver MPS can be integrated with MPS of other organ systems to build platforms in which several organ modules interact. Because the liver plays a central role in such platforms, this work further supports their use in pharmacokinetic and pharmacodynamic investigations. Gut/liver multi-organ models are one example of this approach.
Key study takeaways
- The study used CN Bio’s LiverChip platform to culture perfused 3D hepatocyte microtissues from five individual donors and a five-donor pool, and measured depletion of six compounds in each.
- The liver MPS showed substantial interdonor variability in unbound intrinsic clearance, with a CV between 24.1% (phenacetin) and 66.8% (propranolol), and interwell variability was lower than interdonor variability for five of the six compounds.
- Compared with individual-donor data, pooled hepatocytes gave a relatively unbiased estimate of average clearance (mean individual-to-pooled ratio of 0.97, with marginal evidence of bias for propranolol only) but provide no information on variability.
- The workflow combined drug depletion, metabolite formation, albumin, urea, LDH, and a 90-gene expression panel; albumin, urea, and LDH predicted clearance, while CYP mRNA contributed little.
- A population PBPK model built from liver MPS data predicted both the average and the spread of clinical lidocaine plasma concentrations, supporting the use of modeling and simulation to translate liver MPS results to the in vivo context.
- Predicted hepatic clearance correlated with observed human values (R² of 0.75 and 0.77) but was underpredicted by about 4-fold on average; the lidocaine PBPK model corrected this with an empirical scaling factor, and the authors call for larger, more diverse compound sets to evaluate liver MPS clearance prediction.
Why this paper is worth reading
This paper is useful because it shows DMPK and pharmacometrics scientists how to extract population variability, rather than a single average, from liver MPS experiments, and how to carry that information through to clinical-scale predictions. It gives a worked example of the full chain (study design across donors and replicate wells, mixed-effects analysis, IVIVE with two liver models, and stochastic PBPK simulation), with the model equations and the separate sources of variability laid out. It also answers two practical questions that arise when designing hepatocyte-based clearance studies: when pooled hepatocytes are an acceptable choice, and whether gene expression screening can stand in for functional measurements. The reported underprediction of absolute clearance, and the correction used to handle it, help set realistic expectations for how liver MPS clearance data feed into PBPK work. Teams planning donor-resolved clearance or IVIVE studies in a perfused liver model can run them in their own laboratory or through CN Bio’s absorption, distribution, metabolism, and excretion (ADME) study services.
FAQ
The study used the PhysioMimix LC12 plate and the PhysioMimix Core System, both from CN Bio.
The LiverChip was used to culture cryopreserved primary human hepatocytes from five donors, and a pool of all five, as perfused 3D microtissues seeded at 6 × 10⁵ cells per scaffold. On day 4, the microtissues received a 1 µM bolus of phenacetin, diclofenac, lidocaine, ibuprofen, propranolol, or prednisolone, and researchers sampled the medium over 24 to 72 hours to measure drug depletion and metabolite formation.
The study modeled healthy human hepatic drug metabolism using hepatocyte-only liver MPS microtissues from five individual donors (three male and two female, aged 21 to 72 years) and a pooled sample of those donors. No disease model was used.
The liver MPS revealed substantial interdonor variability in intrinsic clearance, with a coefficient of variation between 24.1% and 66.8% across six compounds, while interwell variability was generally lower. A population physiologically based pharmacokinetic (PBPK) model built from the liver MPS data predicted the clinically observed variability in lidocaine plasma concentrations.
The study compared individual-donor hepatocytes with pooled hepatocytes; liver MPS microtissues with freshly thawed hepatocytes for expression of 90 metabolism-related genes; predicted human hepatic clearance with observed values; and population PBPK predictions with published clinical lidocaine plasma profiles. The paper does not include a static or 2D culture comparator.
The readouts were parent drug depletion by LC-MS/MS; unbound intrinsic clearance and its interdonor and interwell variability from mixed-effects modeling; metabolite formation; albumin, urea, and LDH production or release; expression of 90 drug metabolism-related genes; unbound drug fraction in medium; and nonspecific binding to the LiverChip.
The paper gives DMPK and pharmacometrics teams a worked framework for estimating population variability in hepatic drug metabolism from liver MPS data and translating it to in vivo predictions with PBPK modeling. It also shows that pooled hepatocytes reliably estimate average clearance, and that CYP mRNA screening should not be used as a surrogate for interdonor differences in metabolic activity.
Full citation
Related products and services
Contract research services
Discover how to utilize our cross-species models to inform next-step decision making via our DILI in vitro Contract Research Services here.
Add PhysioMimix Core in your lab
To develop your own cross-species Liver MPS models, you will need:
