Research ArticleDRUG TESTING

Reproducing human and cross-species drug toxicities using a Liver-Chip

See allHide authors and affiliations

Science Translational Medicine  06 Nov 2019:
Vol. 11, Issue 517, eaax5516
DOI: 10.1126/scitranslmed.aax5516

Spotting species-specific toxicity

Candidate drug testing using standard preclinical models cannot accurately predict which compounds are likely to cause drug-induced liver injury in humans. To improve selection of promising drug candidates, Jang et al. developed a Liver-Chip consisting of rat, dog, or human hepatocytes, endothelial cells, Kupffer cells, and stellate cells. Using the microfluidic chips, the authors confirmed mechanism of action of several known hepatotoxic drugs and an experimental compound. A second experimental compound that induced fibrosis in a rat Liver-Chip did not alter hepatocyte function in human chips, whereas a third compound demonstrated increased toxicity in a dog Liver-Chip. Results support using multispecies chips to identify species-specific differences in drug metabolism and toxicity.


Nonclinical rodent and nonrodent toxicity models used to support clinical trials of candidate drugs may produce discordant results or fail to predict complications in humans, contributing to drug failures in the clinic. Here, we applied microengineered Organs-on-Chips technology to design a rat, dog, and human Liver-Chip containing species-specific primary hepatocytes interfaced with liver sinusoidal endothelial cells, with or without Kupffer cells and hepatic stellate cells, cultured under physiological fluid flow. The Liver-Chip detected diverse phenotypes of liver toxicity, including hepatocellular injury, steatosis, cholestasis, and fibrosis, and species-specific toxicities when treated with tool compounds. A multispecies Liver-Chip may provide a useful platform for prediction of liver toxicity and inform human relevance of liver toxicities detected in animal studies to better determine safety and human risk.


The U.S. Food and Drug Administration (FDA) and European Medicines Agency generally require the safety of new drug candidates to be evaluated in both rodent and nonrodent animal models, frequently rat and dog, before moving the new chemical entity into human clinical trials. An analysis of 150 drugs that caused adverse events in humans found that regulatory testing in rats and dogs correctly predicted just 71% of toxicities in humans (1). Moreover, whereas gastrointestinal, hematological, and cardiovascular toxicities were predicted with a relatively high concordance, the ability to predict drug-induced liver injury (DILI) was much lower. This was further confirmed by a more recent survey comparing target organ toxicities in animal and first-in-human studies that also found a low concordance of DILI between human and animals (2). DILI, a major cause for liver failure and drug attrition (36), can be categorized as intrinsic or idiosyncratic. Intrinsic DILI can manifest as hepatocellular necrosis, steatosis, or cholestasis and is usually dose dependent. In contrast, idiosyncratic DILI is a rare event that occurs in large patient trials or at postmarketing.

The poor prediction of DILI in humans is driven by poor nonclinical to clinical translation. Thus, one of the major challenges that the pharmaceutical and biotechnology industries face is selecting compounds with reduced risk for DILI. Given the scale of this challenge and its negative impact on health care costs and development of new therapeutics, there is a critical need for more predictive and human-relevant alternatives to animal models.

Organs-on-Chips technology is gaining increasing popularity as models that are more predictive of human outcome. These Organs-Chips recreate the three-dimensional (3D) organ microenvironments, such as the multicellular architecture, tissue-tissue interface, vascular perfusion, fluid flow, and other relevant physical microenvironment of human organs, through a clever combination of design, engineering, and biology (711). This has been shown to faithfully recapitulate the complex functions and pathophysiology of multiple human organs, including the lung, intestine, and kidney (7, 1013). Here, we explored whether human-microengineered Organs-on-Chips technology may be used to design a species-specific Liver-Chip that can be used to address these challenges in the drug development process.


Development and characterization of a rat, dog, and human Liver-Chip

Species-specific chips lined by living rat, dog, or human hepatic cells were constructed using microfluidic Organ-Chips. Primary rat, dog, or human hepatocytes were seeded in the upper parenchymal channel within an extracellular matrix (ECM) sandwich (14) on top of an ECM-coated, porous membrane that separates the two parallel microchannels. Relevant species-specific rat, dog, or human liver sinusoidal endothelial cells (LSECs), with or without liver Kupffer cells and/or stellate cells, were cultured on the opposite side of the same membrane in the lower vascular channel (Fig. 1A). We initiated these studies by analyzing dual-cell chips containing only hepatocytes and LSECs (fig. S1A), which revealed that all three species of primary hepatocytes formed characteristic branched bile canalicular networks lined by functional multidrug resistance–associated protein 2 (MRP2) efflux transporters and maintained their stereotypical in vivo–like liver epithelial morphologies and cytoarchitecture for at least 14 days in culture under continuous flow (fig. S1B). In contrast, the same human, dog, and rat hepatocytes failed to form well-developed bile canaliculi when maintained for 14 days without endothelium in the conventional static ECM sandwich culture plates (fig. S1B). It has been reported that sandwich culture can also form extensive canalicular networks, but these are not sustained over 14 days and generally persist up to 7 days in culture (15, 16). In the microengineered Liver-Chip, the underlying vascular channel containing LSECs also displayed the multifunctional scavenger receptor stabilin-1, which is expressed selectively on sinusoidal endothelial cells of the liver, spleen, and lymph nodes (fig. S1B) (17).

Fig. 1 Recapitulation of species-specific drug toxicities in a rat, dog, and human Liver-Chip.

(A) Schematic of the Liver-Chip that recapitulates complex liver cytoarchitecture. Primary hepatocytes are grown in the upper parenchymal channel in ECM sandwich format, and NPCs (LSECs, Kupffer, and stellate cells) are grown on the opposite side of the same membrane in the lower vascular channel. (B) Albumin secretion after daily administration of bosentan at 1, 3, 10, 30, or 100 μM for 3 days in dual-cell (hepatocyte and LSECs) human chips and plates (hepatocyte sandwich monoculture) and for 7 days in dual-cell dog and rat chips and plates (n = 3 independent chips and plate wells; hepatocyte donor lot HUM1591). (C) Representative images of CLF (green, BSEP substrate), BSEP [red, 4′,6-diamidino-2-phenylindole (DAPI) in blue], and CDFDA (green, MRP2 substrate) from the parenchymal channel after bosentan (Bos) treatment at 30 μM and vehicle control (Con) for 7 days in human chips. (D) Quantification of percent CLF-positive area in bile canaliculi (BC) from the parenchymal channel. Mann-Whitney U test (n = 3 independent chips with three randomly selected different areas per chip; detailed description on the analysis in the Supplementary Materials; hepatocyte donor lot HUM4166). (E) Quantification of BSEP-positive area and fold change of BSEP gene expression. Mann-Whitney U test (n = 1 chip for BSEP imaging and n = 4 independent chips from n = 2 independent experiments for gene expression; hepatocyte donor lots HUM1591 and HUM4166). (F) Quantification of percent CDFDA-positive area in bile canaliculi from the parenchymal channel. Mann-Whitney U test (n = 3 independent chips with 5+ randomly selected different areas per chip; detailed description on the analysis in the Supplementary Materials; hepatocyte donor lot HUM4166). Scale bar, 20 μm. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Error bars present means ± SEM. NPC, nonparenchymal cell; LSEC, liver sinusoidal endothelial cell; CLF, cholyl-lysyl-fluorescein; BSEP, bile salt export pump; CDFDA, 5(6)-carboxy-2′,7′-dichlorofluorescein diacetate; MRP2, multidrug resistance–associated protein 2.

To assess the liver-specific physiological function of the dual-cell chips, we measured the secretion of albumin and compared this to results obtained from the same human, rat, and dog hepatocytes cultured alone in the static sandwich culture plates. These studies revealed that all three species-specific chips maintained substantially higher (3- to 14-fold greater, P < 0.0001) amounts of albumin production than cells in conventional sandwich monoculture plates (fig. S1C). The quantitative range of albumin production we measured in the human Liver-Chip between days 7 and 14 (~20 to 70 μg day−1 per million cells) was very similar to that estimated for humans in vivo (50 μg day−1 per million cells) using in vitro–to–in vivo extrapolation (iViVE) techniques (18). In contrast, hepatocytes within conventional sandwich monoculture plates showed lower (2.8- to 3.9-fold lower) amounts of albumin production than cells in the human Liver-Chip over the same time period.

To further evaluate the physiological relevance of the dual-cell chips, the drug-metabolizing capacity of the hepatocytes was characterized over time in culture. We measured the activities of the three major cytochrome P450 (CYP) isoforms (CYP1A, CYP2B, and CYP3A) that represent key CYP families involved in drug metabolism. We used prototypical probe substrates in a cocktail approach using concentrations of their respective substrates (phenacetin, bupropion, and midazolam) or a single substrate (cyclophosphamide for CYP2B and testosterone for CYP3A in the human model) that mirror their Michaelis constant (Km) in humans (19). These three isoforms also represent the major CYPs regulated by the xenosensors aryl hydrocarbon receptor, constitutive androstane receptor, and pregnane X receptor (20). These studies revealed that CYP activities measured in the dual-cell human, rat, and dog chips over the 14-day culture period were comparable to, or in some cases greater than, those exhibited by freshly isolated hepatocytes (fig. S2), which are the gold-standard model currently used by pharmaceutical researchers. In contrast, there was a decline in CYP activities in all three species in sandwich monoculture plates over the same time period (fig. S2).

Recapitulation of species-specific drug toxicities in a rat, dog, and human Liver-Chip

To explore whether these dual-cell chips could be used to predict species-specific DILI responses, we used the three species models to evaluate hepatotoxic effects induced by bosentan, a dual endothelin receptor antagonist that causes cholestasis in humans but not in rats or dogs. Inhibition of the bile salt export pump (BSEP), resulting in hepatocellular accumulation of bile salts, is the putative mechanism for bosentan-mediated DILI (21, 22). Daily administration of bosentan at 1, 3, 10, 30, or 100 μM resulted in decrease albumin secretion in these species-specific chips, with different potencies in human, dog, and rat chips [half-maximal inhibitory concentration (IC50) of 10, 30, and >100 μM, respectively] (Fig. 1B). We observed a correlation between the effect in the human Liver-Chip and the clinical response. The concentration in the human Liver-Chip at which we observed toxicity approximated to the plasma concentration of bosentan (Cmax = 7.4 μM) that has been associated with DILI in humans (23). Furthermore, the observed effect on albumin secretion at high concentrations in dog, but not rat, correlates with in vivo findings in which transient transaminase elevations and variable increases in bile acids were observed at high doses and plasma concentrations (80 μM) in dog studies but not in rat studies (24).

In addition, the model was more sensitive in detecting bosentan toxicity compared to sandwich monoculture plates (Fig. 1B), which failed to demonstrate an in vivo–relevant toxic response. This could be attributed to improved hepatocyte functionality, including maintenance of drug-metabolizing enzymes activities, presence of sinusoidal endothelial cells, and culture longevity in the Liver-Chip that is exposed to dynamic fluid flow compared to the static plate sandwich monoculture plates (figs. S1B and S2). In other complex cell-based liver models including 3D human spheroid hepatic cultures, the IC50 was found to be more than 10-fold higher than in vivo (25). Cotreatment of bosentan (30 μM) with cholyl-lysyl-fluorescein (CLF), a BSEP substrate, inhibited efflux of CLF by 50% (Fig. 1C), resulting in its intracellular accumulation (Fig. 1D) in the Liver-Chip, consistent with the known mechanism for hepatotoxicity of bosentan in humans. Inhibition of BSEP activity was also accompanied by decreases in BSEP protein and mRNA quantities (Fig. 1E). Bosentan also inhibited efflux of 5(6)-carboxy-2′,7′-dichlorofluorescein diacetate (CDFDA) (Fig. 1, C and F), a substrate for MRP2, in the Liver-Chip, which is consistent with a known role of MRPs in hepatic disposition of this drug (26). Thus, in addition to recapitulating species-specific hepatotoxicities, these results illustrate that mechanisms of DILI, which involve hepatic transporters, can be studied in the Liver-Chip. This highlights the advantage of the Liver-Chip in integrating mechanisms of action (in this case BSEP transporter inhibition) to functional outcome (decrease in albumin synthesis) in the same model.

Detection of diverse phenotypes of hepatotoxicity using a quadruple-cell Liver-Chip

To add a higher order of biological complexity to the Liver-Chip necessary to study diverse phenotypes of liver toxicity, we integrated species-specific nonparenchymal cells (NPC), hepatic stellate and Kupffer cells into the vascular channel to develop the quadruple-cell Liver-Chip model (Fig. 1A). These species-specific quadruple-cell chips also exhibited high amounts of albumin secretion similar to those observed in the dual-cell chips (fig. S3A). The human and rat quadruple-cell chips also maintained high CYP enzyme activities that were similar to, or higher than, those observed in freshly isolated hepatocytes or in the dual-cell chips (fig. S3B) over long-term culture. We used tool compounds that are known to cause diverse phenotypes of DILI or transaminitis in humans to characterize the ability to use the Liver-Chip for drug safety and risk assessment in humans and to enable insights into mechanism of action driving the toxicity.

The generic analgesic acetaminophen (APAP) can produce DILI, resulting in whole organ failure and death when overdosed. APAP toxicity is caused by direct injury to hepatocytes mediated by the toxic and reactive metabolite N-acetyl-p-benzoquinone imine that depletes cellular glutathione (GSH), causing oxidative stress; it can also be detoxified by hepatocytes, resulting in formation of glucuronide and sulfate metabolites (27). To evaluate APAP toxicity in the human quadruple-cell Liver-Chip, we maintained a constant flow rate that was determined to best reproduce its metabolism rate and turnover (10 μl/hour of flow rate) on the basis of its known intrinsic clearance. Metabolism of APAP using the Liver-Chip was confirmed by detection of notable amounts of APAP glucuronide in both the parenchymal and vascular channels after daily administration of 3 mM APAP for 20 days (fig. S4A), which confirmed that all four cell types were exposed to the hepatocyte-derived metabolites as a result of diffusion through the porous membrane. Treatment with APAP resulted in dose-dependent depletion of total GSH and adenosine 5′-triphosphate (ATP) at all concentrations tested (0.5, 3, and 10 mM) in the hepatocytes within the parenchymal channel and even more potently in the NPCs in the vascular channel (Fig. 2A), highlighting that APAP toxicity is not limited to liver epithelial cells. The depletion of GSH is also suggestive of formation of reactive oxygen species (ROS), which we confirmed by using a fluorogenic probe, CellROX (Fig. 2B). APAP-induced depletion of GSH and ATP preceded a decline in hepatocyte morphology (fig. S4B) and function, as measured by decreased albumin synthesis and increased expression of oxidative stress–related injury markers such as α–glutathione S-transferase (α-GST) and microRNA 122 (miR122) (Fig. 2C). In addition, cotreatment of APAP (3 mM) with the glutathione-depleting agent buthionine sulfoximine (BSO; 200 μM) amplified sensitivity to APAP toxicity. This was seen by the increases in ROS (Fig. 2B), miR122, and α-GST (Fig. 2C) that were not detected at the same APAP concentration in the absence of BSO. Together, this confirms the reported role of ROS in APAP-induced hepatotoxicity (28).

Fig. 2 Detection of liver injury and release of various DILI biomarkers using a quadruple-cell human Liver-Chip.

(A) Total GSH and ATP content from the parenchymal and vascular channels after daily administration of APAP at 0.5, 3, or 10 mM for 7 days in human chips. (B) Representative images of ROS intensity (magenta, CellROX) after daily administration of APAP at 0.5, 3, or 10 mM and co-administration of 3 mM APAP and 200 μM BSO for 7 days in human chips. Right: Quantification of number of CellROX-positive events per field of view. Kruskal-Wallis tests (n = 3 independent chips with three to five randomly selected different areas per chip). Scale bar, 100 μm. (C) Albumin, α-GST, and miR122 secretions from the parenchymal channel after APAP treatment for 7 days in human chips. Dunnett’s multiple comparisons test (n = 10 to 18 independent chips for albumin and n = 3 to 9 independent chips for the rest). Hepatocyte donor lot Qhum15063 was used. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Error bars present means ± SEM.

To explore whether the quadruple-cell chips could be used to model DILI mechanisms that target Kupffer cells, we studied JNJ-1 (fig. S5A), a colony-stimulating factor-1 (CSF-1) receptor kinase inhibitor. This compound caused minimal elevations in transaminases in some subjects (fig. S6A) in a human phase 1 clinical trial that was attributed to mechanism-based Kupffer cell depletion, which can play a role in the clearance of transaminases (29). Very high circulating transaminases were reported in two individuals (>10-fold) and were considered idiopathic and unique to the compound (fig. S6B). Minimal dose-related elevations in transaminases were also observed in rat (<5-fold) and dog (<3-fold) studies (fig. S6C) but without any correlative microscopic changes in the liver. Kupffer cell depletion was detected by decreased number of CD68-positive cells in human Liver-Chip after administration of JNJ-1 at 3 μM (Fig. 3A), concentrations that approach human clinical Cmax (~2 μM). Kupffer cell depletion was associated with a decrease in interleukin-6 (IL-6) and monocyte chemoattractant protein-1 (MCP-1) (Fig. 3B) in the vascular channel but was without an effect on hepatocyte function. Decreased hepatocyte function as measured by lowering of albumin secretion was only observed at 30 μM JNJ-1, about 15-fold above the human Cmax (Fig. 3C), suggesting potential for intrinsic toxicity only at high concentrations. These results demonstrated the ability of the human quadruple-cell Liver-Chip to detect a clinically relevant mechanism of action that targets Kupffer cells independently of hepatocytes.

Fig. 3 Detection of Kupffer cell depletion, steatosis, and fibrosis in a human Liver-Chip.

(A) Representative images of Kupffer cells (CD68 in green, DAPI in blue) from the vascular channel after daily administration of JNJ-1 at 3 μM for 7 days in human chips. Right: Quantification of the number of CD68-positive cells per field of view. Mann-Whitney U test (n = 3 independent chips with three to five randomly selected different areas per chip). (B) IL-6 and MCP-1 release after daily administration of JNJ-1 at 3, 10, or 30 μM for 3 days from the vascular channel in human chips. (C) Albumin secretion after 3 days of JNJ-1 treatment from the parenchymal channel in human chips. Dunnett’s multiple comparisons test (n = 3 to 5 independent chips). (D) Representative images of lipid droplets (Nile red in yellow, and DAPI in blue) from the parenchymal channel and activated stellate cells (α-SMA, green) from the vascular channel after daily administration of MTX at 1, 10, or 30 μM for 7 days in human chips. (E) Quantification of Nile red–positive events per field of view and α-SMA–positive cells per field of view. Kruskal-Wallis tests (n = 3 independent chips with three to five randomly selected different areas per chip). (F) Albumin secretion from the parenchymal channel and IP-10 secretion from the vascular channel after MTX treatment for 7 days and 1 day, respectively, in human chips. Not significant (n = 3 to 6 independent chips). Hepatocyte donor lot HUM4166 was used. Scale bar, 100 μm. **P < 0.01, ***P < 0.001, and ****P < 0.0001. Error bars present mean ± SEM.

Modeling steatosis and markers of fibrosis using Liver-Chip

Methotrexate (MTX) causes liver injury in humans characterized by steatosis, stellate cell hypertrophy, and fibrosis at maximal plasma concentrations of ~1 μM in some patient populations (30). These findings were recapitulated in the human quadruple-cell Liver-Chip, in which daily administration of MTX at 1, 10, or 30 μM for 7 days resulted in microscopic evidence of lipid accumulation as detected by Nile red staining and stellate cell activation as indicated by increased expression of α–smooth muscle actin (α-SMA) (Fig. 3, D and E). These changes were also associated with increases in interferon-γ–induced protein 10 kDa (IP-10), a chemokine whose elevation is associated with liver inflammation and fibrosis (31). There were no abnormalities in albumin secretion (Fig. 3F), which is consistent with the lack of predictive or diagnostic biomarkers for monitoring these toxicities in humans. These studies suggest that inclusion of microscopic end points for steatosis and markers of fibrosis in the quadruple-cell Liver-Chip could be an approach to identify compounds with a potential risk for these toxicities.

To investigate whether cross-species Liver-Chip models could be used to predict human-specific steatosis, fialuridine (FIAU) was tested in rat and human quadruple-cell chips. Development of FIAU, an antiviral nucleoside analog, was discontinued in phase 2 clinical trials because of liver failure and deaths in 5 of 15 patients caused by microvesicular steatosis (32). A review of the animal toxicology data concluded that the studies could not have predicted severe liver injury in humans caused by FIAU (32). Daily administration of FIAU at 1, 10, or 30 μM for 10 days in the human Liver-Chip resulted in a dose-dependent increase in lipid accumulation (Fig. 4A). There was also a concomitant dose-dependent decline in albumin secretion at concentration of ≥1 μM and release of liver injury markers including miR122, α-GST, and keratin 18 (Fig. 4, B and C). In contrast, there were no effects on lipid accumulation or hepatocyte function after treatment in the rat Liver-Chip with FIAU at the same concentrations and treatment duration as the human Liver-Chip (Fig. 4, A and B), which is consistent with previous nonclinical data (32) and studies in humanized hepatocyte mouse models that demonstrate species differences in steatosis compared to wild-type mice (33).

Fig. 4 Comparison of species differences in steatosis after FIAU treatment using a rat and human Liver-Chip.

(A) Representative images of lipid droplets (Nile red in yellow, and DAPI in blue) from the parenchymal channel after daily administration of FIAU at 1, 10, or 30 μM for 10 days in rat and human chips. Right: Quantification of Nile red intensity. a.u., arbitrary units. (B) Normalized albumin secretion (% control) after FIAU treatment for 7 days in rat and human chips. Dunnett’s multiple comparisons test [n = 3 independent chips for rat and n = 3 to 12 human chips from n = 2 different experiments using hepatocyte donor lot HUM4166 (all conditions) and HUM8305 (0 and 30 μM)]. (C) miR122, α-GST, and keratin 18 secretions after FIAU treatment for 10 days in human chips. Dunnett’s multiple comparisons test (n = 3 independent chips, hepatocyte donor lot HUM4166). Scale bar, 100 μm. *P < 0.05, **P < 0.01, and ****P < 0.0001. Error bars present means ± SEM.

Use of a species-specific Liver-Chip to query human relevance of animal liver toxicities

It is not uncommon for compounds to be discontinued due to liver toxicity observed in rats or dogs before testing in humans because of uncertainties on the clinical translation of these findings. To evaluate whether species-specific Liver-Chip could be used to assess human relevance, a Janssen proprietary compound (JNJ-2) that was discontinued because of liver toxicity in rats was characterized in the cross-species Liver-Chip. Daily oral administration of JNJ-2 (150 mg/kg; Cmax = 11.6 μg/ml; 24 μM; fig. S5B) for 2 weeks resulted in liver fibrosis, supported by increased α-SMA staining within stellate cells, which was persistent 3 months after compound washout in rats (Fig. 5A). These findings were associated with chronic inflammation of portal areas and decreases in albumin with no changes in transaminases in rats, and as a result, JNJ-2 was discontinued before testing in nonrodent species. Daily treatment of JNJ-2 at 3, 10, or 30 μM in rat quadruple-cell Liver-Chip for 4 days resulted in a dose-dependent increase in expression of α-SMA, a marker for fibrosis, within stellate cells (Fig. 5, B and C); decreases in albumin were observed but did not reach significance (Fig. 5D). In contrast, treatment of human Liver-Chip at the same concentrations did not produce these abnormalities, even when extended for 14 days of treatment (Fig. 5, B to D); there was also no effect on hepatocyte function after 4 days of treatment. These results suggest a potential species differences in response to JNJ-2 between rats and humans.

Fig. 5 Comparison of species differences in fibrosis using a rat and human Liver-Chip and elevation of transaminases using a dog and human Liver-Chip.

(A) Hematoxylin and eosin (H&E) and α-SMA images of rat liver after daily oral administration of JNJ-2 (150 mg/kg) for 2 weeks. (B) Representative images of α-SMA (red; DAPI in blue) from the vascular channel after daily administration of JNJ-2 at 30 μM for 4 days in rat chips and 14 days in human chips. (C) Quantification of α-SMA intensity from the vascular channel after daily administration of JNJ-2 at 3, 10, or 30 μM for 4 days in rat chips and 14 days in human chips. D, day. (D) Normalized albumin secretion (% control) after daily administration of JNJ-2 for 4 days in rat chips and 4 and 14 days in human chips. (E) H&E and Masson’s trichrome staining images of dog liver after daily administration of JNJ-3 (65 mg/kg) for 14 days. Low magnification showing multifocal areas of necrosis, inflammation, and portal fibrosis in the hepatic parenchyma. High magnification showing necrotic hepatocytes with inflammatory cells (left) and moderate portal fibrosis (collagen stained as blue), inflammation, and bile duct hyperplasia. (F) AST and ALT measurements after JNJ-3 treatment (15 and 65 mg/kg per day) in dogs for 6 days. (G) Fold increases of albumin and GLDH secretion from the parenchymal channel after administration of JNJ-3 at 1, 3, or 10 μM for 1 day in dog and human chips. (H) Fold increases of AST and ALT secretion from the parenchymal channel after administration of JNJ-3 for 1 day in dog and human chips. Dunnett’s multiple comparisons test (n = 3 to 4 independent chips). Human hepatocyte donor lot HUM4166 was used. Scale bar, 100 μm. **P < 0.01, ***P < 0.001, and ****P < 0.0001. Error bars present means ± SEM.

We also tested another Janssen proprietary compound, JNJ-3 (fig. S5C), that was discontinued from further development because of hepatocellular necrosis, portal fibrosis, and biliary hyperplasia after daily dosing at 15 and 65 mg/kg for 14 days in dogs, with maximal plasma concentrations of 12.7 and 19.4 μM, respectively (Fig. 5E). These findings were associated with elevations of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) after 14 days of treatment (Fig. 5F). JNJ-3 was administered to dog and human chips for 4 days, after which the study was stopped because of damage to hepatocytes at the highest concentrations tested; biochemical end points were measured in samples collected 24 hours after dose. Administration of JNJ-3 significantly decreased (P < 0.0001) albumin secretion at ≥1 μM in dog chips and at ≥10 μM in human chips (Fig. 5G). ALT, AST, and glutamate dehydrogenase (GLDH) were also elevated in dog and human chips, but only at the highest concentrations of 10 and 30 μM, respectively (Fig. 5, G and H), indicating that for this compound, albumin is a more sensitive marker of hepatocyte dysfunction in the Liver-Chip model. Transaminase elevations in dog Liver-Chip occurred at concentrations that bridged dog plasma concentrations. Thus, the dog Liver-Chip corroborates in vivo results. Although toxicity by JNJ-3 was 3- to 10-fold more potent in dog Liver-Chip than human chip, it is highly probable that liver toxicity could have occurred in patients if this compound had progressed into clinical trials.

Identifying risk for idiosyncratic DILI using the human Liver-Chip

One of the most difficult forms of hepatotoxicity to predict in the clinic relates to idiosyncratic DILI responses that are often missed during nonclinical and early clinical testing. To explore whether the human Liver-Chip might be useful to predict these types of response, we tested TAK-875, a G protein–coupled receptor 40 (GPR40) agonist that was discontinued in phase 3 trials because of low incidence (2.7%) treatment-related elevations in transaminases (>3-fold rise in upper limit of normal) combined with a few individual cases of serious DILI (34). In vitro and in vivo studies identified formation of reactive acyl glucuronide metabolites, suppression of mitochondrial respiration, and inhibition of hepatic transporters by TAK-875 as potential mediators of its hepatotoxic effects (35). Daily administration of TAK-875 at 10 μM (equivalent to human Cmax) in the human quadruple-cell Liver-Chip for about 2 weeks resulted in formation of the oxidative metabolite (M-1) (6.4% of parent) (Fig. 6A) at similar relative amounts as that reported in humans (10% of parent) (36) and production of high amounts of the acyl glucuronide metabolite (54% of parent) (TAK-875AG), again consistent with reports that glucuronidation of TAK-875 represented a major clearance pathway in humans. Glucuronide metabolites are substrates for canalicular and basolateral hepatic MRP transporters, but at high intracellular concentrations, they may inhibit their own efflux and accumulate in hepatocytes. Daily administration of TAK-875 at 3, 10, or 30 μM in the Liver-Chip resulted in a dose-dependent decrease in biliary efflux of the MRP2 substrate CDFDA, implying that MRP2 was inhibited by TAK-875AG formed in the human Liver-Chip (Fig. 6B). This allowed us to probe the consequences of prolonged exposures to TAK-875 and its de novo formed reactive metabolite TAK-875AG in the Liver-Chip in a 2-week study. We found an effect on mitochondrial membrane potential, confirmed by a dose-related and time-dependent redistribution of the mitochondrial potential sensitive dye tetramethylrhodamine methyl ester (TMRM) detected at 1 week of treatment (Fig. 6, B and C) and lipid droplet accumulation and formation of ROS (Fig. 6, B and C) at the end of the 2-week treatment; these end points may be consequential to perturbation of the mitochondria.

Fig. 6 Identifying risk for idiosyncratic DILI using human Liver-Chip.

(A) Formation of TAK-875 metabolites after day 1 and day 14 of TAK-875 treatments at 10 μM in human chips. (B) Representative images of MRP2 transporter activity (CDFDA in green), mitochondrial depolarization (TMRM in red), ROS (CellROX in cyan), and lipid droplets (AdipoRed in red) after daily administration of TAK-875 at 10 or 30 μM for 8 or 15 days in human chips. DAPI is shown in blue. Scale bars, 100 μm. (C) Quantifications of number of CDFDA-positive fractions in bile canaliculi area, number of redistributed TMRM fractions, and CellROX-positive events per field of view after daily administration of TAK-875 at 3, 10, or 30 μM for 15 days in human chips. Kruskal-Wallis tests (n = 3 independent chip with five randomly selected different areas per chip). (D) MCP-1 and IL-6 released from the vascular channel and albumin and keratin 18 secreted from the parenchymal channel after 14 days of TAK-875 treatment in human chips. Dunnett’s multiple comparisons test (n = 3 independent chips). Hepatocyte donor lot HUM4166 was used. ****P < 0.0001. Error bars present means ± SEM.

We then investigated whether an innate response could be detected in chips treated with TAK-875 on the basis of the prevailing hypothesis for immune-mediated DILI that an innate response caused by covalent protein binding, cell stress, and release of damage-associated molecular patterns can trigger an adaptive immune attack of hepatocytes in a few susceptible individuals (37). Treatment with TAK-875 caused a biphasic release of the inflammatory cytokines MCP-1 and IL-6 at 10 μM (human equivalent Cmax) but not at 30 μM (Fig. 6D). A correlative increase in stellate cell activation was also observed at 10 μM but not at 30 μM (fig. S7). On the basis of analysis of nuclei staining images (Fig. 6B), there was no detectable cell death or detachment in the hepatocyte cell layer at 30 μM, whereas loss of cells was detected in the nonparenchymal layer (fig. S7) that contains immune cells, which likely explains decreased cytokines measured at this concentration. In addition to Kupffer cells, activated stellate cells can also contribute to an inflammatory response in the liver, and so activation of stellate cells and possibly Kupffer cells may also have contributed to cytokine release at 10 μM.


These results suggest that a species-specific Liver-Chip can be used to predict rat, dog, and human hepatotoxicities and that they may be useful for safety and risk assessment to estimate human relevance of drug-induced liver toxicities seen in animals. They could also potentially be used to predict human idiosyncratic hepatotoxicities and elucidate mechanisms of action.

Ongoing research on DILI ranges from basic understanding of adverse pathways and mechanisms by which some compounds cause DILI to clinical research aimed at identifying predictive DILI biomarkers. This research is spurred by human safety concerns for liver failure and the associated high costs incurred when drugs are withdrawn from clinical trials or from the market based on incorrect predictions from nonclinical testing. Recently, the FDA released a predictive toxicology roadmap aimed at qualifying in silico and in vitro models to enable their use in regulatory decision-making; qualification of these models will encourage their uptake and use, which would also meet objectives of reduction, refinement, and replacement of animals (38). Although there have been major advances in the development of in vitro models that predict DILI, key gaps remain that need to be addressed. For example, lack of metabolic competence in plated primary hepatocytes has been addressed by more complex 2D coculture or 3D spheroid models, which can maintain metabolic competence for prolonged durations (15, 3941), but these models are static and closed; therefore, metabolites may accumulate to quantities that are not physiologically relevant or they may be underrepresented due to loss after medium replenishment that is required to maintain cell survival. In addition, dense 3D constructs often suffer from limited oxygen and nutrient transport, and even when spheroid cultures include endothelial cells and Kupffer cells, they lack physiologically relevant tissue-tissue interfaces and cytoarchitecture that has been demonstrated to be a key driver of cell function (10, 42). It is also impossible to apply physiologically relevant fluid flow and associated mechanical forces, in which drugs pass specifically through an endothelium-lined interface as they do in vivo. Mechanical forces have also been demonstrated to be key players in gene expression and in vivo cell function (10). In addition, in spheroids, it is difficult to visualize dynamic changes in cell position, morphology, or function within these large, disorganized, and multicellular structures.

Recently, there have been many attempts to develop liver-based cell systems to replicate organ-level functions using microengineered approaches such as microphysiological systems (MPS), and some have included the four cell types we included in the quadruple-cell chips in the present study (4345). Some of these studies demonstrated enhanced functionalities with flow (increased albumin secretion and CYP activities) but were limited to short-term (1 day) cultures (43). In contrast, in the present study, we showed that both the dual-cell and quadruple-cell Liver-Chip remain metabolically competent and maintain albumin production and activities of multiple key drug-metabolizing enzymes at in vivo–like amounts for at least 14 days in culture. Although DILI responses to various drugs were measured in other MPS liver models (41, 44, 46), the drug concentrations used were not clinically relevant. In contrast, we used drug concentrations that bridged those measured in the plasma of nonclinical models or humans in the present study, and as a result, we were able to generate results that closely recapitulated those previously reported in both nonclinical studies and human clinical trials.

End points assessed in most in vitro systems are limited to measures of cell viability as an initial assessment of potential hazards, but they often do not capture the mechanisms that underlie DILI or clinically relevant end points, and they are not effective for human risk assessment. We evaluated whether the Liver-Chip could detect more complex and mechanistically relevant DILI end points. Using a combination of microscopy, tissue staining, and measurement of DILI biomarkers, we detected diverse phenotypes of DILI, including hepatocellular injury, cholestasis, steatosis, Kupffer cell depletion, and stellate cell activation as a marker of fibrosis. Thus, the quadruple-cell Liver-Chip appears to be suitable for detecting toxicities that are attributable to direct effects on the four liver cell types included in our model; however, the chip is currently not capable of detecting toxicities of the bile duct because this cell type has not yet been included in the design. With the tool compounds tested so far, toxicities in the model were detected at concentrations that bridged human plasma concentrations associated with DILI, suggesting that the model has potential to be used for human risk assessment.

Although the enzymatic activities in the human Liver-Chip were robust, some activities were lower or higher in comparison to fresh human hepatocytes, which most likely reflect donor-to-donor variability. The CYP enzyme activities in rat chips were similar to those measured in fresh rat hepatocytes when a single substrate was used, but they were higher when a cocktail of CYP substrates were used, which suggests potential for interactions among these substrates in the cocktail approach.

One advantage of the microengineered Liver-Chip is the use of continuous flow in an open system, which ensures that all cells are exposed to sufficient amounts of the parent drug and its metabolites simply by adjusting the flow rate. The open system also allows for continuous collection or sampling of the effluents of both the vascular and parenchymal channels, which prevents accumulation of the parent and its metabolites, while enabling measurement of biomarkers and other biological end points over time. This can also contribute to the heightened sensitivity of the Liver-Chip system when compared to static in vitro liver models.

The ability to measure mechanistic end points and biomarkers in the model also makes it suitable for delineating pathways and mechanisms causing DILI. For instance, the observation that GSH and ATP depletions are early events in APAP-mediated toxicity followed by a decline in hepatocyte function and eventual oxidative stress and overt injury suggests that toxic metabolite–mediated mitochondrial dysfunction and ATP depletion are likely early events in the APAP toxicity cascade. Mitochondrial dysfunction has been identified as a hazard for APAP toxicity (47). Depletion of GSH and ATP in NPCs after treatment with APAP implies that the toxic metabolite can escape hepatocytes and mediate an effect on other cell types. This is supported by studies in mice in which APAP metabolites were detected bound to reticulocytes, which lack drug-metabolizing enzymes, indicating that these metabolites also escape hepatocytes in vivo (48). The increased sensitivity of NPCs to APAP toxicity compared to hepatocytes may be due to a reduced detoxification capacity relative to that in hepatocytes. The current studies in chips confirm that both hepatocytes and NPCs contribute to APAP hepatoxicity in vivo.

To make key decisions in the drug development process, specific contexts of use must be defined for predictive in vitro models such as the Liver-Chip before their qualification. These contexts of use can be developed in areas such as prediction of human liver toxicity, clinical translation of toxicity observed in animal studies, or identifying the DILI potential of compounds that form reactive metabolites. Results from our studies with bosentan, FIAU, MTX, and JNJ-1 show that the human Liver-Chip can be used to study diverse mechanisms of responses in the liver with compounds that target parenchymal and NPCs and that these responses occur at clinically relevant concentrations that bridge plasma concentrations where effects were observed in humans. Human-specific sensitivities to toxicity by bosentan and FIAU were also confirmed in the human Liver-Chip compared to the companion animal Liver-Chip, demonstrating species-specific in vivo–in vitro correlation. The putative mechanism for bosentan-induced liver toxicity—inhibition of bile acid efflux via BSEP resulting in intracellular accumulation of bile acids and inhibition of MRP2 that could be secondary to formation of bosentan glucuronide (26)—was also confirmed; an advantage of the Liver-Chip is that we could couple these mechanistic end points to a measurable decline in hepatocyte function. The species differences in bosentan toxicity may be related to species differences in the composition of toxic bile acids (22). The species differences for response to FIAU is explained by species differences in expression and activity of the equilibrative nucleoside transporter 1 (ENT1), which is absent in rats but present in humans. ENT1 facilitates entry of FIAU into the mitochondrial membrane, causing mitochondrial toxicity (49). The absence of mitochondrial hENT1 (PEΧN) in rodents limits the entry of the drug into mitochondria and thereby prevents the toxicity of this drug. The species-specific expression of nucleoside transporters in the mitochondria of humans, but not of other animals, can explain why the preclinical toxicology testing of FIAU in mice, rats, dogs, and primates failed to predict human-specific mitochondrial toxicity and hepatotoxicity of FIAU to humans (32, 49). Cumulatively, these findings further highlight the potential of the human Liver-Chip to serve as a safety testing tool for human-specific drug-induced hepatotoxicity.

Inhibition of CSF-1 receptor kinase by JNJ-1 caused elevation of transaminases in animals and humans that is considered secondary to Kupffer cell depletion. Studies in genetically altered mice previously found the CSF-1/CSFR pathway to be essential for survival of Kupffer cells (50). This was confirmed in our Liver-Chip in which treatment with JNJ-1 caused a depletion of Kupffer cells accompanied by decreases in cytokines. Transaminase elevations have been observed with other agents that inhibit CSF-1/CSF-1R (51). Kupffer cells depend on CSF-1 for viability, and evidence has been presented that transaminase elevations are downstream of Kupffer cell depletion and possibly a consequence of a role for Kupffer cells in transaminase clearance (29).

A gap exists in assigning human relevance of liver toxicities observed in animal studies, especially when these are observed in only one species. JNJ-2 and JNJ-3 are examples of compounds that caused liver toxicity in animal studies and were discontinued before clinical development; JNJ-2 caused fibrosis in rats, a finding that is not monitorable in humans, whereas JNJ-3 caused severe hepatocellular injury and biliary hyperplasia in dogs as early as 1 week after daily treatment. The rat Liver-Chip was very sensitive to treatment with JNJ-2, whereas no toxicity was observed in the human Liver-Chip at the same concentrations up to 14 days of daily treatment. Activation of stellate cells noted in rat but not in human Liver-Chip confirmed that we could reproduce the pathophysiology observed in rats. Moreover, lack of a similar response in the human Liver-Chip suggests potential for a species-specific difference in drug response. Unfortunately, in vivo data are not available for JNJ-2 in dogs or JNJ-3 in rats because these studies were never executed, and so a direct comparison of results across all three Liver-Chip platforms for both compounds is not possible. Although the results we obtained are interesting and could have influenced an internal decision to test the compound in nonrodents to address whether fibrosis was rat specific, the model would need robust qualification with a specific context of use to convince regulatory agencies to make a decision with regard to the lack of human relevance of the rat findings. As we observed the release of cytokines/chemokines including IP-10 in chips treated with MTX, FIAU, and JNJ-2, which has been identified as a potential marker for fibrosis (52), this model may also be amenable to biomarker discovery, especially for challenging disease areas such as steatosis and fibrosis for which suitable biomarkers for monitoring disease progression and response to treatment in humans are lacking.

Reactive metabolite formation has been identified as an important hazard associated with compounds that cause rare or idiosyncratic DILI. The microengineered Liver-Chip provides an opportunity to put formation of reactive metabolites in a cell, tissue, and organ context for a functional readout of their contributions to DILI. For instance, putative adverse pathways for TAK-875–mediated DILI were detected in discrete in vitro models (34); however, the physiological consequence of perturbing these mechanisms could not be studied. Treatment of Liver-Chip with TAK-875 showed that continuous and prolonged exposure to parent and reactive metabolites caused mitochondrial dysfunction, oxidative stress, formation of lipid droplets, and an innate immune response (cytokine release), all of which are harbingers of DILI for patients with increased susceptibility. In a 24-week phase 3 clinical trial with TAK-875, liver biopsies of five of seven patients with higher than normal transaminases also presented with steatosis (53). It was challenging to assign causality because of disease background in patients with type 2 diabetes, but a treatment-related effect could not be ruled out. The Liver-Chip studies suggest that steatohepatitis secondary to mitochondrial dysfunction and ROS formation could be a potential phenotype of DILI after treatment with TAK-875. This exemplifies the advantage of the Liver-Chip for assessing the pathophysiological consequence of reactive metabolite formation, which has been strongly associated with idiosyncratic DILI (37, 54, 55).

In conclusion, we have shown that a species-specific Liver-Chip have potential future applications for safety testing, disease modeling, mechanism of action determination, biomarker identification, and predicting human hepatotoxicities, including idiosyncratic responses. This approach could also be used to query human relevance of toxicities observed in nonclinical animal studies or for mechanistic investigations of DILI detected in nonclinical and clinical studies. It should also be possible to create next-generation Liver-Chip that capture donor-to-donor variability in hepatocytes, which is known to influence differences in drug-induced hepatotoxicity by incorporating induced pluripotent stem cells–derived hepatocytes and other liver cell types from multiple donors.


Study design

This study sought to explore whether a species-specific Liver-Chip can be used as a tool for hepatotoxicity safety testing, mechanism of action determination, biomarker identification, and predicting human hepatotoxicities. For each experiment, chips were assigned to experimental groups randomly and with an equal probability of receiving vehicle or treatment. At least three (n = 3) independent chips were used for each experimental group. Chips were excluded by a predetermined exclusion criterion, which evaluated potential flow issues based on the effluent volume collected at each time point. The chips that had any flow issues during the experiment were excluded from all analysis. Furthermore, to avoid potential bias in assessing the results during sample and data processing and during morphology scoring, laboratory personnel was blinded to the identity of the samples. All measurements were performed at least in duplicate, and some experiments were repeated at least two times. The precise number of experimental replicates and repeats is given in the figure legends.

All animals that were administered JNJ-1, JNJ-2, or JNJ-3 were maintained at Janssen Pharmaceuticals facilities in accordance with the guidelines established by the Association for Assessment and Accreditation of Laboratory Animal Care, and all procedures were approved by the Janssen Animal Care and Use Committee. Details of the protocols are in the Supplementary Materials.

The clinical study protocol and amendment(s) for JNJ-1 were reviewed by an Institutional Review Board. This study was conducted in accordance with the ethical principles that have their origin in the Declaration of Helsinki that are consistent with good clinical practices and applicable regulatory requirements. Subjects or their legally acceptable representatives provided their written consent to participate in the study after having been informed about the nature and purpose of the study, participation/termination conditions, and risks and benefits of treatment. Informed consent was obtained at screening before any study specific procedure was conducted. Personal data from subjects enrolled in this study were limited to those data necessary to investigate the efficacy, safety, quality, and utility of the investigational study drug(s) used in this study and were collected and processed with adequate precautions to ensure confidentiality and compliance with applicable data privacy protection laws and regulations. Details of the protocols are in the Supplementary Materials.

Liver-Chip culture

Before cell seeding, chips (S-1 Chips, Emulate Inc.) were functionalized using Emulate’s proprietary protocols and reagents (Liver-Chip Protocols and ER, Emulate Inc.). After surface functionalization, both channels of the Liver-Chip were coated with species-specific combination of ECM. For the human Liver-Chip, we used a mixture of rat tail collagen type I (Corning) and bovine fibronectin (Gibco). For the rat Liver-Chip, we used a mixture of human placenta collagen type IV (Sigma-Aldrich) and fibronectin (Gibco). For the dog Liver-Chip, we used a mixture of collagen I (Corning), collagen type IV (Sigma-Aldrich), and fibronectin (Gibco). Primary hepatocytes were seeded in the upper channel of the Liver-Chip at a concentration of 3.5 million cells/ml and later overlaid with Matrigel (Corning) and then incubated at 37°C with 5% CO2. For the dual-cell culture model (hepatocytes and LSECs), the LSECs were seeded at a concentration of 2 to 4 million cells/ml in the lower vascular channel. For the quadruple-cell model (hepatocytes, endothelial cells, Kupffer, and stellate cells), a mixture of LSEC, Kupffer, and stellate cells were seeded in the channel of the Liver-Chip at the following concentrations: 3 million cells/ml for LSEC, 0.5 million cells/ml for Kupffer cells, and 0.1 million cells/ml for stellate cells. After cell seeding, the upper channel of the Liver-Chip was maintained in William’s E medium containing GlutaMAX (Gibco), ITS+ Premix [human recombinant insulin, human transferrin, selenous acid, BSA (bovine serum albumin), and linoleic acid] (Corning), dexamethasone (Sigma-Aldrich), ascorbic acid (Sigma-Aldrich), fetal bovine serum (Sigma-Aldrich), and penicillin/streptomycin (Sigma-Aldrich). The vascular channel of the Liver-Chip was maintained with species-specific endothelial media (Emulate Inc.). Two days after seeding, the chips were connected to the Human Emulation System (Emulate Inc.), and both of the chip channels were perfused at 30 μl/hour to provide a continuous supply of fresh media for the duration of the experiments.

Statistical analysis

As indicated in the figure legends, one-way analysis of variance (ANOVA), Sidak’s, and Dunnett’s multiple comparisons tests were used for parametric data, and the Mann-Whitney U test or Kruskal-Wallis tests was used for nonparametric data. All statistical analyses were performed using Prism 7 (GraphPad).


Materials and Methods

Fig. S1. Morphology and functionality of a species-specific dual-cell Liver-Chip.

Fig. S2. Cytochrome P450 enzyme activity in a species-specific dual-cell Liver-Chip.

Fig. S3. Comparison of hepatic functionalities between a dual- and quadruple-cell Liver-Chip.

Fig. S4. Detection of glucuronide metabolites of APAP and hepatocellular injury using a quadruple-cell human Liver-Chip.

Fig. S5. JNJ compound structures.

Fig. S6. Transaminase data from JNJ-1 treatment in human phase 1 clinical trial and rat and dog in vivo studies.

Fig. S7. Stellate cell activation after TAK-875 treatment.


Acknowledgments: We thank S. Dallas, J. Silva, P. Guzzie-Peck, and G. Bignan for the useful scientific discussions; L. Shanno for the gene expression analyses; J. Chen for the bioanalysis of APAP metabolites; M. Kelley for the critical review of the manuscript; V. Antontsev for the iViVE modeling; U. Doshi at In Vitro ADMET Laboratories (IVAL) for performing the liquid chromatography–mass (LC-MS) spectrometry analysis; and J. Resnikoff and P. Patel for their contribution in initial model validation. Funding: Primary funding support for this project was obtained from Janssen Biotech Inc., a Janssen Pharmaceutical company of Johnson & Johnson, with additional funding from AstraZeneca, the Wyss Institute for Biologically Inspired Engineering at Harvard University, and the Defense Advanced Research Projects Agency (DARPA) under Cooperative Agreement Number W911NF-12-2-0036. Author contributions: G.A.H., M.A.O., K.-J.J., L.E., K.K., and D.E.I. initiated the project, designed the model features for dual-cell model, defined the model development strategy, and provided guidance to solve initial challenges, successful model development, and characterization. G.A.H., M.A.O., K.-J.J., and D.E.I. initiated the project, designed the model features for quadruple-cell model, and defined model development and characterization strategy. M.A.O., K.-J.J., H.-K.L, D.S., and G.A.H. designed quadruple-cell model characterization study experiments. L.E., K.-J.J., and G.A.H. designed dual-cell model characterization experiments, which were included in the Supplementary Figures. M.A.O., D.S., H.-K.L., and K.-J.J. designed all compounds studies. K.-J.J., J.R., K.R.K, D.B.P., G.K., J.E.R., and D.C performed all Liver-Chip experiments. J.N. performed the CLF efflux imaging and analysis in bosentan study. R.B. performed image analysis of JNJ-2 and TAK-875 studies. H.P. performed dog Kupffer and stellate cell isolation. J.S. performed iViVE analysis. H.-K.L. and W.L. performed metabolite identification analysis of APAP, TAK-875, and compound analysis for polydimethylsiloxane absorption of JNJ compounds. M. Singer validated and performed liver biomarker analyses. M.D., M. Sonee, A.J.S., C.M., E.B., and B.S. designed, conducted, or analyzed in vivo experiment for JNJ compounds. B.J. designed a cocktail substrate for CYP enzyme activity test. A.S. and L.C.A. performed LC-MS analysis for CYP450 enzyme activity measurement. A.H. and S.H. were involved initial model development. J.R., K.R.K., J.E.R., M.S., E.B., B.S., K.-J.J., H.K.L., R.B., and M.S. wrote the methods. K.-J.J. prepared all figures. K-J.J. and M.A.O. wrote the rest of manuscript with editing and input from D.E.I., G.A.H., L.E., D.W., E.B., B.S., C.M., H.-K.L., J.R., A.H., D.B.P., J.E.R., R.B., and J.S. Competing interests: D.E.I. holds equity interest in Emulate Inc. and chairs its scientific advisory board. D.E.I. is an inventor on a patent application (PCT/US2009/050830) submitted by Harvard University that covers “Organ mimic device with microchannels and methods of use and manufacturing thereof.” K.-J.J, J.R., K.R.K., D.B.P., G.K., J.E.R., J.N., D.C., R.B., J.S., K.K., and G.A.H. are current or former employees of and hold equity interests or options to obtain equity interests in Emulate Inc. D.E.I., G.A.H., K.-J.J., S.H., and A.H. are inventors on a patent application (PCT/US16/64661) submitted by Harvard University that covers “Devices for simulating a function of a liver tissue and methods of use and manufacturing thereof” and G.A.H., K.-J.J., S.H., J.R., K.R.K., H.P., J.S., D.B.P., and M.A.O. are inventors on a patent application (PCT/US2016/064795) submitted by Emulate Inc. that covers “Devices and methods for simulating a function of a liver tissue.” Data and materials availability: All data and materials are available in the main text or the Supplementary Materials.

Stay Connected to Science Translational Medicine

Navigate This Article