Gut–liver-on-a-chip enables mechanistic study and risk assessment of drug-induced liver injury and drug–drug interactions

Abstract

Current preclinical models face challenges in recapitulating organ-level interactions affecting drug safety, and there has been little investigation into drug toxicity and related DILI. We present a pump-less gut–liver-on-chip, enabling integrated analysis of drug exposure–toxicity relationships and inter-organ pharmacological interactions. The platform incorporates a polarized intestinal barrier with a quadruple-cell co-cultured liver spheroid. Through simulation and comparative evaluation of oral versus systemic drug administration, we demonstrated the critical role of the intestinal barrier in modulating drug exposure, corresponding toxic responses and first-pass effects. Temporal profiling revealed progressive hepatic injury mechanisms involving mitochondrial dysfunction and activation of the apoptotic pathway. Pharmacological inhibition of cytochrome P450 attenuated victim-induced oxidative stress without affecting hepatic drug exposure, confirming enzyme-related bioactivation as the toxicity mechanism. Furthermore, transporter-mediated drug–drug interactions were functionally replicated, with perpetrator compounds altering substrate pharmacokinetics through competitive efflux inhibition and modified intestinal disposition. The ability of the platform to monitor drug exposure–toxicity relationships and drug–drug interactions was validated using combinations of perpetrator and victim drugs. This integrated approach advances applications of organ-on-chips by establishing causal relationships between drug exposure and toxicity, resolving the progression of temporal toxicity, and modeling drug–drug interactions, which are critical factors in predicting clinical hepatotoxicity and complex pharmacokinetic interactions.

---

1. Introduction

Drug-induced liver injury (DILI) is a main reason for drug failure in clinical trials and post-approval withdrawals, reflecting poor translation between preclinical animal models and human clinical outcomes. Because of this, preclinical strategies are incorporating more advanced in vitro models with multi-parametric endpoints.¹ From 2010 to 2019, new molecular entities constituted the majority of drug approvals, and orally administered new molecular entities have remained the main route over the decades.² According to the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) guidelines on the safety of pharmacology studies, researchers are expected to adopt a clinical administration route for research into drug efficacy and toxicity. Thus, a gut–liver system is suitable for mimicking the oral administration process, a common clinical administration routine, using small molecular compounds for the prediction of DILI.

Numerous cell-based methodologies have been fostered as alternatives to in vivo research and have contributed to an understanding of the mechanisms of toxicity.³ Organs-on-chips (OoCs), which recapitulate physiologically relevant cellular microenvironments, interconnect organs via flow controls, and enable the dynamic co-culture of different organ equivalents in vitro, are expected to be used to investigate relationships between drug exposure and potential hepatotoxic effects and, furthermore, to elucidate the absorption, distribution, metabolism, excretion, and toxicity (ADME-T) of the candidates.⁴ Gut–liver-on-chip, which recapitulates the uptake and metabolism of xenobiotics, simplifies complex pathophysiological events and provides physiological information about oral administration. The co-culture system recapitulates the two main barriers to orally consumed drugs and is capable of regulating drug transport, which can provide insights into the concentration available in the body as well as side effects under exposure; thus far, replicating drug exposure and relevant drug toxicity.⁵ There are relatively few research studies on this aspect,^6–8 illustrating that gut–liver-on-chips may serve as an advanced in vitro alternative model for deriving drug exposure and potential toxicity for further clinical research. With the development of organ-on-chip technology, it is expected to be used in more scenarios, including toxic studies: for example, drug–drug interaction (DDI).

DDI leading to adverse drug reactions represents a significant clinical concern that warrants serious attention.⁹ The main mechanisms underlying DDI involve two aspects: the inhibition/induction of cytochrome enzyme P450 families (CYP) and the regulation of drug transporters.¹⁰ Specifically, CYP3A-mediated DDI occurs when perpetrator drugs modulate the metabolic activity of CYP3A enzymes, thereby altering the metabolism of victim drugs and inducing hepatotoxicity;⁹ while transporter-mediated DDI arises from drug-induced changes in the expression of uptake and efflux transporters, which can have a significant impact on the absorption, distribution, metabolism, and excretion (ADME) profile of another drug, consequently affecting its pharmacokinetic properties and safety.¹¹ Thus, investigation of DDIs facilitates the elucidation of ADME characteristics of drugs and toxicity mechanisms, and provides critical insights into rational drug use in clinical practice.

Despite some research on drug permeability using organ-on-chips, there has been little investigation into drug toxicity and related DILI. Here, we report DILI assessment in a pump-less two-chamber chip (Fig. 1 and S11) by integrating gut equivalents and three-dimensional (3D) liver spheroids. As our aim is to mimic hepatotoxicity and related DDIs after oral administration, the design of the chip simulates the absorption process in vivo. By selecting acetaminophen (APAP) as a test compound, proof-of-concept is confirmed by simultaneously detecting toxic effects and drug exposure through biochemical analysis, fluorescence detection and liquid chromatography (LC). Then, to mimic DDIs in vitro, we first studied CYP-dependent toxic effects by pretreatment with ABT for 1 h and then administering APAP before examining hepatotoxicity. Then, we simulated transporter-related absorption, distribution, and metabolism changes in emodin with interference by cyclosporine, before ultra-performance liquid chromatography–tandem mass spectrometry (UPLC-MS) and detection of toxicity. The outcomes of the study demonstrate the proof-of-concept of the two-organ platform as an in vitro alternative model for DILI and DDI assessment.

Fig. 1 The IBAC M1 chip model. (A) Schematic of the chip. The chip has two chambers and allows for the co-culture of two different organs through a porous membrane. (B) To realize dynamic cultivation, a shear force was produced on an interval IBAC rocker with parameters of 1 cycle per min and less than 30° inclination (20° inclination was chosen in the experiments). (C) Schematic of gut-on-chips. (D) Schematic of gut–liver-on-chips. (E) After drug administration, the media in the chambers can be collected for the determination of concentration profile and biomarkers, as organic equivalents can be used for toxic determination with various fluorescence probes and antibodies in situ.

2. Methods and materials

2.1 Cell culture

The resources from cultivation of human colon adenocarcinoma cell line Caco-2, HT29-MTX-E12, human hepatocellular carcinoma cell line HepG2, the human umbilical vein endothelial cell line (HUVEC-T1), the human monocyte leukemia cell line THP-1 and human hepatic stellate cells (HHSC) are shown in SI 1, with reference to previous reports.¹²

2.2 Pump-less two-organs-on-chip

We used a pump-less two-organs-on-chip to realize the co-culture of intestine and liver equivalents. As shown in Fig. 1A and S8, the IBAC M1 chip (ME102411), consisting of a static chamber above a 0.07 cm² polyester membrane and a dynamic chamber comprising a fluid passageway with two reservoirs, was designed, assembled and kindly provided by Beijing Daxiang Biotech Co., Ltd. To provide a dynamic shear force, the chip was fixed on an interval IBAC rocker (MR100110, Beijing Daxiang Biotech Co., Ltd.) before the rocker was installed, and a shear force of 0.45 dyn cm⁻² was provided in the fluid passageway (Fig. 1B). According to the chip instructions, intestinal cells can be cultured above or beneath the membrane by turning the chip over (Fig. 1C) and forming a confluent intestinal barrier. Before we chose to cultivate the intestinal model above the membrane, we tested the constructed intestinal function in both ways. As for the gut–liver-on-chip, we briefly tested the function of the gut–2D liver-on-chip, but mainly the gut–3D liver-on-chip (Fig. 1D), and proof-of-concept studies for toxicity and exposure supervision, time-related toxicity procedure, and DDI-influenced toxic effect were carried out in gut–3D liver-on-chips. Drug concentration was determined using liquid chromatography, and drug toxicity was determined with a microplate reader, microscope, and high-content cell imaging analyzer after treatment (Fig. 1E). The working state of the chip and system is shown in Video S1 for illustrative purposes only.

2.3 Establishment of gut-on-chip

For the gut-on-chip model cultured on an IBAC M1 chip above a transparent membrane, the chambers were pretreated with D-PBS for at least 24 h. Tissue culture was treated with 300 μg mL⁻¹ Matrigengel (ABWbionova, 082706) on a rocker, switching between +20° and −20° inclinations for 1 circle per min (Fig. 1B) at 37 °C for at least 2 h. Basic electrical resistance was measured in D-Hanks. Then 10 μL of cell suspension (1.17 × 10⁶ cell per mL) was added to the top side of the membrane and kept in 37 °C for 1–4 h in after 200 μL of complete medium had been added to the opposite chamber. After the cells had attached to the surface of the membrane in the static state, the medium was replaced, and the medium volume of the intestinal chamber was made up to 100 μL.

For the gut-on-chips model cultured on an IBAC M1 chip beneath a transparent membrane, the chambers were pretreated with D-PBS for at least 24 h. Tissue culture was treated with 300 μg mL⁻¹ Matrigengel on an interval IBAC rocker (switching between +20° and −20° inclinations for 1 circle per min) at 37 °C for at least 4 h. It was washed once with D-Hanks, and basic electrical resistance was measured. Then, 15 μL of cell suspension (1.86 × 10⁶ cells per mL) was added to the bottom chamber before 20 μL of complete medium was added to the opposite chamber. The chip was turned over to allow the cells to attach to the surface of the membrane for about 4 h in the static state. Fresh medium was added to both sides of the chambers before putting the chip on the rocker for dynamic cultivation.

IBAC M1 chips were put on the interval rocker for dynamic cultivation at 37 °C. The rocker was switched between +20° and −20° inclinations for 1 circle per min, thus allowing bi-directional flow. The medium was refreshed every day from day 1 to day 6 and then twice a day.

2.4 Establishment of 3D (liver) equivalents

Based on previous research,¹² HepG2, HUVEC-T1, THP-1, and HHSC cell lines were collected following digestion, and the cell suspensions were mixed in a ratio of 60 : 19 : 15 : 6. For equivalent formation, 100 cells were seeded per well into a round-bottom U-shaped low-adherence 96-well plate (Liver Biotechnology (Shenzhen) Co., Ltd., LV-ULA002-96UW) following dilution. After 3 days of culture, the spheroids were collected and loaded on the IBAC M1 chips. Spheroid morphology was observed using a high-content cell imaging analyzer (PerkinElmer, Operetta CLS), and the fluorescent signals and the morphology of the spheroids were measured with ImageJ.

2.5 Establishment of gut–liver-on-chips

On day 14 after establishment of the intestinal model, 20 000 HepG2 cells or eight liver equivalents were loaded onto the other side of the chips to form the intestinal-liver-on-chip models. After attachment, the medium of the liver chamber was replaced and made up to a volume of 200 μL. Chips were put on the interval rocker for dynamic cultivation at 37 °C. The rocker was switched between +20° and −20° inclinations for 1 circle per min, thus allowing bi-directional flow. The medium was collected daily, and at the end of the 7 days of co-culture, the biomarkers were analyzed by immunofluorescence. The lactate dehydrogenase (LDH) of the two chambers, and the aspartate aminotransferase (AST), alanine aminotransferase (ALT), and albumin (ALB) in the liver chamber were detected. The intestinal apparent permeability coefficient (Papp) of fluorescein sodium was detected on day 6, before the immunofluorescence of occludin and ZO-1 was detected in the intestinal equivalents of gut-on-chips, gut–2D liver-on-chips, and gut–3D liver-on-chips. Live/dead staining was carried out in the liver equivalents of U-bottom plates and gut–3D liver-on-chips to indicate the signals from dead cells, and CYP3A4, MRP2, PGP, Ki67, ZO-1, BSEP, and DAPI were detected in the liver equivalents of gut–3D liver-on-chips and 3D liver-on-chips to profile liver function. Villin and DAPI were detected in intestinal slides of gut–3D liver-on-chips.

2.6 Intestinal integrity

Transendothelial electrical resistance (TEER) and Papp of fluorescein sodium were measured for assessment of the integrity of intestinal equivalents (SI 1).

2.7 Liver function assessment

A sensitive two-color fluorescence cell viability assay was selected for discrimination between live and dead cells with calcein acetoxymethyl ester (Calcein AM), a cell-permeant dye as the live cell indicator, and propidium iodide (PI) as the dead cell indicator in liver models. The live cell component produces an intense, uniform green fluorescence in live cells (Ex/Em 488 nm/515 nm), and the dead cell component produces a predominantly nuclear red fluorescence (Ex/Em 570 nm/602 nm). Functional biomarkers (AST, ALT, LDH, and ALB) were observed. The CYP3A4, MRP2, and PGP fluorescent signals were detected on day 5.

2.8 Morphology and immunostaining

Hematoxylin and eosin (HE) and periodic acid-Schiff (PAS) staining: sections were fixed with paraformaldehyde and dehydrated with sucrose solution. Frozen sections were stained with HE or PAS. Bright-field morphology was observed with a high-content cell imaging analyzer or microscope (Olympus IX71).

The polarity of the gut barrier was tested with rhodamine 123 (Rho123) transportation, and transport direction was tested on day 14 in gut–liver-on-chips. Rho123 was administered from the gut chamber and liver chamber, respectively, and the medium was collected from lateral chambers to detect the concentration of Rho123. The Papp of apical to basolateral (A to B) and basolateral to apical (B to A) was calculated and compared with a two-tailed Student's t-test.

Immunostaining: the samples were fixed and dehydrated before being sliced into frozen sections. For in situ immunofluorescence staining of frozen sections, the slices were washed with PBS three times, and then permeabilized in PBS containing 0.5% Triton X-100 (Solarbio) for 10 min and blocked in goat serum (Beyotime Biotechnology) for 30 min at room temperature. Primary antibodies were applied to each well at a certain dilution with reference to the instructions and incubated overnight at 4 °C. The primary antibodies are listed as follows: anti-occludin (91131S, CST, 1 : 400 dilution), anti-ZO-1 (ab221547, Abcam, 1 : 100 dilution), anti-ZO-1 (339194, Invitrogen, 1 : 50), anti-ZO-1 (sc-33725, Santa Cruz, 1 : 200), anti-PGP (13342S, CST, 1 : 250 dilution), anti-MRP2 (Invitrogen, MA1-26536, 1 : 50), anti-cleaved caspase-3 (9664S, CST, 1 : 400 dilution), and anti-H2AFX (Phospho-Ser139) (K001453M, Solarbio, 1 : 200 dilution). The cells were washed and incubated with the corresponding fluorescent conjugated secondary antibodies and co-localization with DAPI (Beyotime Biotechnology). The slices were observed using a high-content cell imaging analyzer or microscope. The secondary antibodies are listed as follows: goat anti-rabbit IgG H&L (Alexa Fluor® 488) (ab150077, Abcam, 1 : 1000 dilution), goat anti-rat IgG (H + L) (Alexa Fluor® 555 conjugate) (4417, CST, 1 : 1000 dilution), goat anti-mouse IgG H&L (Alexa Fluor® 488) (ab150113, Abcam, 1 : 1000 dilution), and donkey anti-rabbit IgG H&L (Alexa Fluor® 647). They were evaluated using a high content analyzer (PerkinElmer, Operetta CLS, or Molecular Devices, ImageXpress Micro Confocal) and a fluorescence microscope.

2.9 Oral administration simulation of APAP in a co-culture chip

We constructed 6 gut-on-chips, 9 gut–3D liver-on-chips and 6 3D liver-on-chips. Gut-on-chips and 3D liver-on-chips were divided into 2 groups: control and APAP p.o. groups, which we gave APAP (11.1 mM) from the chamber above the membrane (representing the apical side) to mimic the oral administration process. Gut–liver-on-chips were divided into 3 groups: control, APAP p.o. and APAP i.v. group. For APAP i.v. group, we gave APAP into both sides of the chamber to mimic the intravenous injection process, which drug concentration quickly got balance after administration. Drug treatment continued for 48 h. The chamber above the medium was given 180 μL of medium, and 320 μL was given to the lateral side. 10 μL of supernatants were collected at 1, 2, 4, 8, 24, 32, and 48 h for liquid chromatography quantification. After drug administration for 48 h, TEER was evaluated, and supernatants for AST, ALT, ALB, and tumor necrosis factor-α (TNF-α) were collected for measurement. Gut equivalents and liver equivalents in gut–liver-on-chips were stained for detection of cell mitochondrial function with Rho123 and MitoTracker, and an apoptosis assay through detection of cleaved caspase-3, as described in SI 1. For gut-on-chips, cell viability was determined with CellTiter-Glo, and the medium in the gut chamber was collected to determine the LDH level. For liver-on-chips, the medium in the liver chamber was collected to determine the relative ALB level to illustrate liver injury.

Then, another 9 gut–3D liver-on-chips were constructed for fluorescent detection, and the chips were divided into 3 groups: control, APAP p.o., and APAP i.v.. After 48 h of treatment, gut equivalents and liver equivalents on gut–liver-on-chips were stained for oxidative stress measurement with CellROX, as described in SI 1.

2.10 Detection of time-related APAP-induced injury

First, we established 9 gut–3D liver-on-chips, and the chips were divided into 2 groups: control and APAP p.o. For observation of ALT and ALB, 90 μL of supernatants were collected at 24, 32, and 48 h before detection of ALT and ALB at the endpoint.

Then, 27 gut–3D liver-on-chips were constructed and were divided into 3 groups: control, APAP p.o. and APAP i.v. At different time-points (24, 32, and 48 h), the samples were stained for cell mitochondrial function detection with TMRE, oxidative stress measurement with CellROX, and apoptosis assay through detection of cleaved caspase-3, as described in SI 1.

2.11 CYP450-mediated DDI

First, we tested the CYP450-dependent APAP-induced liver injury in 3D spheroids. CYP3A4 enzyme activity was detected by a P450-Glo CYP3A4 assay (Promega) according to a previous report to illustrate the inhibition effect of ABT.¹² For ABT-pretreated groups, the spheroids were incubated with ABT (1 mM) for 1 h, after which we changed the medium to reduce the ABT concentration below 10 μM; then, APAP was added to each well at final concentrations of 0 and 4 mM. CellTiter-Glo (Promega) was used for ATP detection. Cell survival was calculated and compared through multi-way ANOVA (SPSS). The spheroids were stained for detection of cell mitochondrial function with TMRE, for oxidative stress measurement with CellROX, and for apoptosis assay through detection of cleaved caspase-3, and fluorescent signals were compared among control, APAP, ABT, and ABT + APAP groups.

Then, 18 gut–3D liver-on-chips were constructed and divided into 6 groups: control, APAP p.o., APAP i.v., ABT, ABT + APAP p.o. group, and ABT + APAP i.v. group. The control group was treated with medium alone. In the APAP group, APAP (11.1 mM) was administered from the chamber above the medium. The ABT and ABT + APAP groups were pretreated with 1 mM ABT for 1 h before treatment with medium and APAP (11.1 mM), respectively. The gut chamber was given 180 μL of medium, and 320 μL was given to the lateral side. After drug administration for 48 h, supernatants were collected, and AST, ALT and ALB were measured, while TEER was evaluated. Liver equivalents on gut–liver-on-chips were stained to assess the cell mitochondrial function using TMRE, oxidative stress using CellROX, and apoptosis by detecting cleaved caspase-3.

Another 18 gut–3D liver-on-chips were constructed, and the group division was the same as previously described. Drug treatment lasted for 48 h, during which 10 μL of supernatants were collected at 2, 4, 8, 24, 32, and 48 h for liquid chromatography quantification. Gut–liver-on-chips were treated for live/dead staining and apoptosis assay through the detection of H2AFX (Phospho-Ser139).

2.12 Detection of transporter-mediated DDI

First, 21 gut–3D liver-on-chips were constructed and divided into 7 groups: control, emodin p.o., cyclosporine, emodin p.o. + cyclosporine, emodin i.v., berberine p.o., and berberine p.o. + cyclosporine. The control group was treated with medium alone, and the cyclosporine group was treated with cyclosporine (1 μM) on both sides. The emodin p.o. group was given emodin 63.3 μM to the gut chamber, while medium was given to the liver chamber. As for the emodin p.o. + cyclosporine group, emodin 63.3 μM and cyclosporine (1 μM) were given to the gut chamber, while medium containing cyclosporine (1 μM) was given to the liver chamber. For the emodin i.v. group, emodin (30 μM) was given to both sides. For the berberine p.o. group, 20 μM berberine was given to the gut chamber, whereas for the berberine p.o. + cyclosporine group, berberine 20 μM and cyclosporine (1 μM) were given to the gut chamber, while medium containing cyclosporine (1 μM) was given to the liver chamber. The gut chamber was given 180 μL of medium, and 200 μL was given to the lateral side. During treatment, 10 μL of supernatant was collected at 4, 8, 24, 32, and 48 h for quantification. After drug administration for 48 h, the supernatants of the liver chamber were collected for ALB measurements, and the supernatants from the gut chamber were collected for quantification before TEER measurement.

2.13 Sample quantification with high-performance liquid chromatography (HPLC)

With reference to previous reports,¹³ the concentrations of APAP were measured using HPLC (SHIMADZU, LC-2040C 3D). At each time point, 10 μL of the samples was taken, as previously described. Methanol was selected for protein precipitation, and we removed the sediment by centrifuging (13 000 rpm, 4 °C, 15 min). An InertSustain C18 (150 mm length, I.D. 4.6 mm, particle size 5 μm, SHIMADZU) was selected as the chromatographic column. Two mobile phases were used: solvent A (methanol) and solvent B (distilled water) at a flow rate of 0.8 mL min⁻¹. The solvent composition was 5% A (0–4 min), 5–100% A (4.01–10 min), 5% A (10.01–15 min). The sample injection volume was 10 μL, and the analyte APAP was detected using a photo-diode array (PDA) detector (wavelength 280 nm). The concentrations of APAP were calculated from the standard curve.

A UPLC-MS was used to examine the signals of emodin, emodin-O-glucuronide, emodin-O-sulfate, and berberine. A reference solution of emodin and berberine standard (National Institutes for Food and Drug Control, China) was prepared by dissolving them in acetonitrile at a final concentration of 1 μg mL⁻¹. The emodin, berberine, and the formation of emodin-O-glucuronide and emodin-O-sulfate were analyzed at 4, 8, 24, 32, and 48 h following drug administration using a UPLC-MS with an AB SCIEX LC AC system with a Triple Quad 6600+ instrument. Acetonitrile was selected for protein precipitation, and we removed the sediment by centrifuging (13 000 rpm, 4 °C, 15 min). Chromatographic separation was achieved using a Waters ACQUITY UPLC HSS T3 column (2.1 mm × 100 mm, 1.8 μm). The mobile phases were (A) 0.1% (v/v) formic acid in water and (B) acetonitrile, with a gradient elution program used for separation, as shown in Table S1. The UPLC flow rate was 0.3 mL min⁻¹, and UPLC-MS analysis was performed using 10 μL samples. The characteristics of ion pairs corresponding to the declustering potential, collision energy, curtain gas, and temperature were m/z 269.04 (−80, −5, 35, and 500) for emodin, m/z 445.07/269.04 (−80, −5, 35, and 500) for emodin-O-glucuronide, and m/z 349.00/269.04 (−80, −5, 35, and 500) for emodin-O-sulfate, while the characteristics of ion pairs corresponding to the declustering potential, collision energy, collision energy spread, and temperature were m/z 336.12 (80, 15, 0, and 500) for berberine.

2.14 Real-time quantitative polymerase chain reaction (RT-qPCR)

To detect the influence of co-culture on the intestinal and liver equivalents and of dynamic fluid on liver equivalents, we constructed gut-on-chips, gut–liver-on-chips, liver-on-chips, and spheroids on U-bottom plates. For a comparison of intestinal function with monoculture and co-culture cultivation, intestinal cells in gut-on-chips and gut–liver-on-chips were gathered and recorded as single and co-cultures. For comparison of liver function with monoculture and co-culture cultivation, as well as static-culture and dynamic-culture, liver cells in gut–liver-on-chips, liver-on-chips, and spheroids on U-bottom plates were gathered and recorded as co-culture, single, and static. Except for spheroids on U-bottom plates, all the chips were placed on the rocker, which means all the liver chambers were supplied with dynamic fluid forces. On day 7 of co-culture, we gathered all the cells and prepared them for RT-qPCR to compare the structural, enzyme, and transporter expression.

Cells were gathered after digestion with TrypLE (ThermoFisher) at 37 °C on the rocker for 10 min, before we stopped digestion and gathered the cells by centrifugation at 1000 rpm for 5 min. Total RNAs were isolated using Trizol reagent (ThermoFisher), during which procedure, 10 μg of RNase-free glycogen (Beyotime) was added as a carrier to the water phase, with reference to the protocol. RNA was reverse transcribed using a FastKing RT kit (TIANGEN, KR116-02) according to the kit protocol. RT-qPCR was performed using a QuantStudio Dx Real-Time PCR instrument (ThermoFisher) with SuperReal PreMix Plus (SYBR Green) (TIANGEN). The qPCR reactions were performed as follows: denaturation at 95 °C for 15 min, followed by 45 cycles of amplification (95 °C for 10 s and 60 °C for 20 s), following a melting/dissociation curve stage. The primer sequences were designed with reference to primerbank (https://pga.mgh.harvard.edu/primerbank/) and are listed in Table S2. Gapdh was used as a reference gene, and the relative expression ratios were calculated using the 2^−ΔΔCt method.

For comparison of influences on intestinal cells, a two-tailed Student's t-test was applied to investigate variations between two independent samples. For a comparison of the influence on liver cells, multi-way ANOVA was first applied to determine the key genes influenced by different cultivation factors, following a one-way ANOVA to describe the differences among groups. If a difference was detected with analysis of variance, Bonferroni post hoc testing was performed. Data analysis was completed using SPSS version 25 (IBM Corporation). P < 0.05 was considered significant.

2.15 Statistical analysis

Continuous variables are shown as mean ± standard deviation (SD). A two-tailed Student's t-test was applied to investigate variations between two independent samples, and ANOVA was applied to investigate variations among multiple groups. If a difference was detected with analysis of variance, Bonferroni post hoc testing was performed. If data transformation failed to generate data with a normal distribution, Dunnett-T3 tests were performed. Data analysis was completed using SPSS version 25 (IBM Corporation). Linear regression models were performed using Prism. P < 0.05 was considered significant.

3. Results

3.1 Establishment of gut-on-chip model

We constructed and reprofiled a gut-on-chip model on an IBAC M1 chip. Intestinal cells were seeded above or beneath the membrane, and multiple integrated layers were formed after cell attachment to the membrane (Fig. 2A and S2A). The TEER kept increasing over 4 weeks (Fig. 2B and C). Then, we assessed the integrity of the intestinal barrier by determining the Papp of fluorescein sodium. A Papp of less than 5 × 10⁻⁷ cm s⁻¹ was considered to indicate low permeability,¹⁴ and all the Papps in the gut-on-chip models met this requirement (Fig. 2D and E). Both construction methods confirmed the integrity of the intestinal barrier. We hoped that the liver equivalents could be maintained in a dynamic environment; thus, we seeded the membrane with intestinal cells in subsequent experiments.

Fig. 2 The establishment of gut-on-chips based on IBAC M1. (A) The morphology on days 7, 14, 20, and 30 after seeding above or beneath the membrane. (B and C) The TEER profile of gut-on-chip after seeding above (B) or beneath the membrane (C). (D and E) The Papp of fluorescein sodium of gut-on-chip after seeding above (D) or beneath the membrane (E). (F) HE staining of slides on days 14 and 21. (G) Tight junction ZO-1 (orange), occludin (green) and nuclei (blue) of the gut membrane on days 14 and 21 (10×) (bar = 100 μm). (H) Rho123 transport on day 14. (I) Tight junction ZO-1 (orange) and occludin (green) staining on day (40×). (J) Transporter PGP, MRP2 and nuclei (blue) staining on day (40×). n = 3–6, data are shown as mean ± SD.

The intestinal barrier retained layer integrity over the culture period, whereas Caco-2/HT29-MTX-E12, originally seeded as a monolayer (Fig. 2A), proliferated and formed a multilayer (Fig. 2F). Next, to further assess the integrity of the intestinal barrier, we detected the formation of intracellular tight junction (TJ) ZO-1 and occludin. Intestinal cells expressed ZO-1 and occludin continuously on day 14, and the expression remained stable until day 21 (Fig. 2G, I, S1B and C). Furthermore, transporter PGP and MRP2 were distributed mainly on the apical side of the barrier (Fig. 2J, S1D and Video S1). Rho123, a substrate of PGP, was effluxed from the intestinal barrier (Fig. 2H), indicating the polarity of the intestinal barrier. All these results indicate that a polar intestinal barrier can be formed on the IBAC M1 chip, and the barrier remains stable from day 14 to day 21.

3.2 Cell viability and function after gut–liver platform integration

We detected the morphology and the physiological function of liver spheroids over 2 weeks. The diameter of the spheroids increased rapidly over 7 days and then remained stable at around 400 μm (Fig. S2A), consistent with the increasing tendency of the cell number (Fig. S2B). HE staining was performed to observe the necrotic core, and the results suggested that there was a non-observed necrotic core in the HE stain on day 7 (Fig. S2C), but a mild necrotic core or some necrocytosis on day 10. With an extension of the culture time, we observed more dead cells, as shown by live/dead staining (Fig. S2D), especially on day 14 compared with days 4, 7 (P < 0.01) and 10 (P < 0.05). PAS staining showed glycogen deposition (Fig. S2C), and the tendency of ALB and ALB per cell to increase showed an increasing anabolic function after longer cultivation (Fig. S2E and F). To assess the damage to liver cells, the secretion of AST (Fig. S2G), ALT (Fig. S2H), and LDH (Fig. S2I) was detected, and AST and ALT levels increased on day 10. We detected the expression of CYP3A4 on the inner side of the spheroids and the expression of MRP2 and PGP at the cell membrane of the spheroid edge (Fig. S2G). The results indicated the basal level of the biomarkers in 2 weeks of cultivation.

On the 14th day of the establishment of the intestinal model, liver equivalents were loaded on the basolateral side to form gut–liver-on-chips according to the time schedule (Fig. 3A). The 3D equivalents can also be mildly attached to the bottom of the dynamic chamber and undergo sustainable growth, even in the fluid passageway (Fig. 3B). As 2D and 3D were both common methods for the cultivation of liver equivalents, we maintained the two-organ systems for 7 days and compared the integrity of the gut barrier and the function/dysfunction of liver equivalents in gut–2D liver-on-chips and gut–3D liver-on-chips. After the liver equivalents had been loaded, gut–liver-on-chips sustained increasing TEER values (Fig. 3C), consistent with gut-on-chip models, indicating that the tight junction of the intestinal equivalents kept forming. The average Papp of gut-on-chips and gut–3D liver-on-chips of fluorescein sodium on day 6 satisfied our requirement (Fig. 3D), indicating a barrier with integrity after 6 days of co-culture. Also, the fluorescence staining showed uninterrupted contiguity for tight junction ZO-1 and Occludin in gut–liver-on-chips (Fig. 3G and S3D), and the LDH level in the gut chamber was sustained (Fig. S3C). All the above results indicate that the gut model may maintain its integrity regardless of the presence or absence of the liver model and the type of liver equivalents used in cultivation. Then, the liver function/dysfunction parameters were used to determine the difference between gut–2D liver-on-chips and gut–3D liver-on-chips in terms of liver function and homeostasis. No significant difference was detected in UREA (Fig. S3A), AST (Fig. S3B), or ALT (Fig. 3E) in the liver chamber; however, the ALB level was higher from day 2 to day 4 (Fig. 3E) in gut–3D liver-on-chips than in gut–2D liver-on-chips. The inoculated number of liver cells of the 2D liver model in gut–2D liver-on-chips was a little higher than that in gut–3D liver-on-chips, and our data are in line with previous results,¹² which reported that the 3D model showed better liver function than the 2D model. Although mature and stable gut–liver-on-chip models can be constructed by both methods, we preferred 3D liver models to construct co-cultured systems in subsequent experiments.

Fig. 3 Establishment, detection of cell viability, and organic function after gut–liver platform integration. (A) Time schedule of establishment of gut–liver-on-chip. (B) Bright-field observation of gut–3D liver-on-chip on day 7 (bar = 2 mm). (C) TEER profile of gut-on-chip, gut–2D-liver-on-chip, and gut–3D liver-on-chip. (D) The Papp of gut-on-chips and gut–3D liver-on-chips of fluorescein sodium on day 6. (E) The levels of ALB and ALT in the liver chamber of gut-on-chips, gut–2D liver-on-chips, and gut–3D liver-on-chips. (F) Live/dead staining of equivalents cultured in the U-bottom plates and gut–3D liver-on-chips. (G) Signals of intestinal occludin (green), ZO-1 (orange), and DAPI (blue) in gut-on-chip and gut–3D liver-on-chip on day 21 (bar = 100 μm). (H and I) Signals of CYP3A4 (green), MRP2 (yellow), PGP (red), Ki67 (green), ZO-1 (yellow), BSEP (red), and DAPI (blue) of liver equivalents in gut–3D liver-on-chips (G) and 3D liver-on-chips (I) (bar = 100 μm). Data are mean ± SD. *P < 0.05 compared with gut–2D liver-on-chip (n = 3).

After the liver equivalents had been loaded, we further maintained the two-organ system for 7 days and detected morphology, cell viability, and system homeostasis, as well as transport and metabolic activity. Fig. 3B shows the bright-field morphology after the co-cultured system was kept for 7 days, and Fig. 3F shows the distribution of live/dead cells on days 3, 5, and 7 after co-culture. Then, we compared enzyme and transporter in liver spheroids in gut–liver-on-chips, liver-on-chips and U-bottom plates. The results show that static/dynamic cultivation greatly affected the expression of CYP2D6 (P < 0.05) and ABCC2 (P < 0.05), and mono/co-cultivation influenced the expression of ABCC1 (P < 0.01), ABCC2 (P < 0.05), and ABCC3 (P < 0.05). No significant difference was observed between the integrity of the intestinal barrier in gut-on-chips and gut–liver-on-chips according to the evidence of TEER (Fig. 2A), fluorescence signals (Fig. 2G and S3D) and the relative expression (Fig. S4A) of ZO-1 and occludin. Finally, we detected the expression of CYP3A4, MRP2, PGP, Ki67, ZO-1 and BSEP (Fig. 3H and I) in the liver chamber, indicating the expression of these functional proteins.

3.3 Absorption and toxicity of APAP on gut–liver-on-chips

In the gut-on-chips and gut–liver-on-chips, the drug was administered via the intestinal apical side to mimic the oral route. To examine the influence of intestinal equivalents on the toxic effect of APAP, we initially measured the parameter of intestinal integrity and the liver injury biomarkers. First, the gut-on-chips were given a single dose of APAP (11.1 mM) from the apical side. The intestinal cell survival rate exhibited more than 90% viability (Fig. 4A) and the LDH level in the gut chamber showed no severe cell damage (Fig. 4B). The TEER remained stable (Fig. 4C), indicating the integrity of the barrier after treatment. These results indicated that non-obvious intestinal damage occurred after treatment. Then, we tested the variation in TEER in the gut–liver-on-chips (Fig. 4D). In accordance with the results from gut-on-chips, no obvious decline in TEER was observed in gut–liver-on-chips, and after 48 h of treatment, we detected a non-significant difference in TEER between the control group and APAP-treated groups. Similarly, no significant differences were detected in the signals of MitoTracker, Rho123, CellROX, or cleaved caspase-3 (Fig. S5). These results indicate that, during the drug treatment, the gut equivalents may sustain integrity, and APAP is absorbed through the intestine before secretion into the liver chamber.

Fig. 4 Toxic effect after a single-dose APAP administration for 48 h. (A–C) Evaluation of cell viability (A), LDH levels in the intestinal chamber (B), and TEER changes (C) in gut-on-chip models. (D) TEER changes in gut–liver-on-chip models. (E–H) Measurement of AST (E), ALT (F), ALB (G), and TNF-α (H) levels in gut–liver-on-chip models. (I) Measurement of ALB levels in liver-on-chip models. (J) Hepatotoxicity assessment at the endpoint after single-dose administration: Rho123, MitoTracker, CellROX, and cleaved caspase-3 signals (bar = 100 μm). (K and M) APAP concentration profile in non-inoculated membrane (K), gut-on-chips (L), and gut–3D-liver-on-chips (M). (N and O) Change in APAP concentration in gut chambers (N) and liver chambers (O). n = 3. Data are shown as mean ± SD. *P < 0.05.

For the detection of liver toxicity, we detected AST, ALT, ALB, and inflammatory cytokine TNF-α levels at the final time-point in gut–liver-on-chips. APAP induced an increasing trend in AST and ALT levels, which indicates a potential liver injury (Fig. 4E and F). The ALB (Fig. 4G) level decreased after APAP treatment, suggesting the liver synthetic function has been hampered. As expected, compared with the APAP p.o. group, an increasing trend in ALT as well as a decreasing trend in ALB can be observed in the APAP i.v. group. Then, an increasing level of TNF-α was detected, and compared with the APAP p.o. group, the level of TNF-α increased more in the APAP i.v. group (Fig. 4H). We attribute the difference to the administration route. A slow absorption process likely prevents the liver spheroids from being exposed to the drug, eliminating the liver injury caused by APAP. In liver-on-chips, we observed a decreasing level of ALB (Fig. 4I) similar to that of the APAP i.v. group, but more severe than that of the APAP p.o. group, which we also attributed to the enhanced exposure to the drug in the liver chamber.

To investigate APAP-induced liver injury, mitochondrial dysfunction, oxidant stress, and cell apoptosis were detected after 48 h of exposure (Fig. 4J). The dysfunction of mitochondrial function was tested with Rho123 and MitoTracker. Compared with the control group, the intensity of the Rho123 signal decreases, illustrating that the mitochondrial membrane potential decreased after APAP treatment. The decreasing trend and tendency were reinforced as exposure time and concentration increased in the APAP i.v. group. Then, the signal of the mitochondria was detected by MitoTracker, and we detected no differences in distribution. Then, oxidant stress was examined, as indicated by the CellROX signal, and stronger oxidant stress was detected after APAP treatment. Compared with the control groups, cleaved caspase-3, a pro-apoptotic protein signal, was observed in the APAP groups, indicating the process of apoptosis. The signal intensity was stronger in the APAP i.v. group than in the APAP p.o. group. These results elucidate that in gut–liver-on-chips, APAP-induced liver injury occurred, and the injury was aggravated as the drug exposure increased in the APAP i.v. group, which to some extent hints at the existence of a gut barrier, delaying the contact between the liver and xenobiotics.

To further confirm our assumption that the difference in hepatotoxic effects was related to drug exposure, we measured the absorption profile after administration of 11.1 mM APAP from the apical side. After administration of APAP from the apical compartment, a rapid increase in APAP content in the liver chamber could be observed in the non-inoculated membrane (Fig. 4K), in which the APAP concentration reached equilibrium within 24 h in the IBAC M1 system (Fig. 4K). However, in gut-on-chips, we observed slow absorption (Fig. 4L). Fig. 4L and M show the drug absorption process in gut–liver-on-chips, and a similar absorption profile can be seen in gut-on-chips and gut–liver-on-chips (Fig. 4N and O). Compared with the non-inoculated membrane and liver-on-chips, the extra-intestinal membrane significantly slowed down the transition of APAP from the apical side (Table S3) and reduced the exposure of the liver spheroids to the drug, in accordance with previous results where intestinal equivalents alleviated hepatotoxicity on liver spheroids.

3.4 Observation of time-dependent hepatotoxic events in gut–liver-on-chips

In the gut–liver platform, APAP-induced cell damage appeared in a time-dependent and dose-dependent manner. There appeared to be no difference in the initial levels of ALT and ALB (Fig. 5A and D). Examination of the ALT level (Fig. 5B and C) over time revealed that ALT exhibited an observable increasing trend within 32 h, and the tendency progressed thereafter. Hampered ALB (Fig. 5E and F) secretion occurred within 24 h, and we detected a significant decrease in the ALB level in a time-dependent manner. The fluorescence signal revealed an evolving hepatotoxic event with increasing concentration and duration of drug exposure. An increasing ROS signal, impaired mitochondrial membrane potential, and a more cleaved caspase-3 signal can be detected after APAP administration for a longer time (Fig. 5G). At 24 h, an evident decrease in the TMRE signal was observed in the APAP i.v. group, which is found later in the APAP p.o. group at 32 h, consistent with the tendency of ROS and cleaved caspase-3 signals, illustrating that the presence of an intestinal barrier postponed the hepatotoxic events of APAP. Consistent with the time–concentration curve we measured before (Table S2), these results illustrate the influence of both time and the intestinal barrier on drug exposure.

Fig. 5 Observation of dose-time-dependent hepatotoxic events in gut–liver-on-chips. (A) Baseline of ALT level at 0 h. (B and C) ALT level (B) and relative ALT level (C) at 24, 32, and 48 h. (D) Baseline of ALB level at 0 h. (E and F) ALB level (E) and relative ALB level (F) at 24, 32, and 48 h. (G) Fluorescent signals of CellROX, TMRE, and cleaved caspase-3 of control, APAP p.o. group, and APAP i.v. group at 24, 32, and 48 h (bar = 100 μm). n = 3. Data are shown as mean ± SD. *P < 0.05.

3.5 Gut–liver-on-chips for DDI study

3.5.1 Alleviation of APAP-induced liver injury after ABT pretreatment. For CYP-mediated DILI research, ABT is selected as a CYP enzyme inhibitor to study APAP-induced liver injury after CYP inhibition. To elucidate the potential effect of the CYP enzyme on alleviation of APAP-induced liver injury, four groups were studied in 3D spheroids: an untreated control group, APAP 4 mM, ABT pretreated 1 hour, and ABT pretreated 1 hour before APAP (4 mM) administration. The cell viability of the APAP 4 mM group decreased to approximately 67.6%, whereas ABT pretreatment alleviates this decline, maintaining viability at 86.2% on average (Fig. S6B). After ABT treatment, CYP3A4 activity decreased to an average of 12.5% (Fig. S6A), consistent with previous results showing that CYP3A4 activity was partially inhibited.¹⁵ Dysfunction of the mitochondrial function was detected with TMRE (Fig. S6C). Compared with the control group, the TMRE signal showed that the mitochondrial membrane potential decreased after APAP treatment and the decreasing trend was partially inhibited by ABT. In the following oxidative stress examination, a CellROX signal in the spheroid was detected (Fig. S6C). Specifically, the CellROX signal weakened after ABT disturbance, suggesting that ABT prevents ROS generation. Compared with the other groups, cleaved caspase-3, a pro-apoptotic protein signal, can be observed in the APAP group, indicating that the process of apoptosis has taken place (Fig. S6C), which can also be partially disturbed after ABT pretreatment. These results show that ABT temporarily inhibits CYP3A4 activity and in addition alleviates liver injury induced by APAP in 48 h.

In gut-liver-on-chips, ABT was pretreated for 1 h before APAP given to the co-cultured system (Fig. 6A). Concentration of APAP in liver chamber was supervised and there was no significant difference in APAP exposure after ABT disturbed (Fig. 6B). Even if no significant difference was observed in AST change fold after ABT disturbed (Fig. 6C), ALT change fold and ALB levels were slightly ameliorated (Fig. 6D and E) after ABT pretreatment. Fluorescent staining suggests mitigated liver injury after ABT pretreated (Fig. 6F). PI, CellROX and H2AFX signals were slightly decreased after ABT treatment in ABT+APAP i.v. group compared with APAP i.v. group, while ameliorated TMRE and cleaved caspase-3 signals can also be observed. The change fold of CellROX, TMRE and cleaved caspase-3 slightly decreased compared with that without ABT treatment (Fig. 6G–I). These results indicate the ABT alleviates liver injury induced by APAP without relation to APAP concentration.

Fig. 6 APAP toxicity in ABT pretreated gut–liver-on-chips. (A) Time schedule. (B) Change in APAP concentration after ABT interference. (C) AST change-fold of control after ABT interference. (D) ALT change-fold of control after ABT interference. (E) ALB change-fold of control after ABT interference. (F) Toxicity of APAP indicated by fluorescence signals after ABT interference (bar = 100 μm). (G–I) Area of CellROX, TMRE, and cleaved caspase-3 signals after ABT interference. Data are mean ± SD. n = 3. *P < 0.05. **P < 0.05.

3.5.2 Influence of cyclosporine on the absorption and toxicity of emodin and berberine. To study the effects of DDI from the perspective of ADME, emodin, a hepatotoxic component of herbal medicine, was selected as the test compound, and PGP and MRP2 inhibitors of cyclosporine were used to interfere with emodin transport and toxic effects. As seen in previous reports, emodin showed toxicity to HepG2 under 25 μM treatment,¹⁶ but had little toxic effect on Caco-2,¹⁷ which met the requirement for use of the content of gut–liver-on-chips for hepatotoxic research. After determining the toxic concentration of emodin and cyclosporine, we measured the toxicity of emodin in Caco-2 and liver spheroids (Fig. S7A and B). Finally, 30 μM was selected as the test dosage of emodin, and 1 μM was selected as the intervention concentration of cyclosporine. In gut–liver-on-chips, both emodin and cyclosporine could cause minor damage to the intestinal barrier function, as shown in the change in TEER 48 h after administration (Fig. 7B), but the bright-field observation showed that the intestinal barrier was completely retained (data not shown). The ALB level (Fig. 7C) decreased by 22.57% and 42.13% compared with the blank group, respectively, indicating hepatotoxicity in the emodin p.o. and emodin i.v. groups, while cyclosporine did not show significant hepatotoxicity after treatment (Fig. 7D), but cyclosporine slightly enhanced the hepatotoxicity of emodin (Fig. 7D). Then, emodin and its metabolites were detected to observe drug exposure in the chips. Fig. 7A and E show that emodin experienced severe first-pass effects in the intestinal barrier, undergoing rapid phase II metabolism to form emodin-O-glucuronide and emodin-O-sulfate. We first detected the signals of emodin, emodin-O-glucuronide,¹⁸ and emodin-O-sulfate¹⁹ (Fig. S8) before recording the exposure–time curve in the liver chamber (Fig. 7F) and the concentration or intensity in the gut chamber (Fig. 7H) 48 h after administration. Before administration, the emodin concentrations in the emodin p.o. group and emodin p.o. + CsA group were 87.04 μM and 86.09 μM, respectively. At 48 hours post administration, the final emodin concentrations in the liver chamber were 8.45 ± 1.00 μM and 14.03 ± 1.99 μM, respectively, showing a significant difference (P < 0.05) (Fig. 7Fa), and the Papp of emodin also increased significantly (P < 0.05) (Fig. 7Ia). All the above results demonstrate that intervention by cyclosporine significantly enhanced the intestinal absorption of emodin. Compared with the emodin p.o. group, the concentration (Fig. 7Fb) and proportion (Fig. 7Jb) of emodin-O-glucuronide, a metabolite of emodin by UGT, significantly increased in the emodin p.o. + CsA group in the liver chamber. However, compared with the emodin p.o. group, the emodin concentration in the emodin p.o. + CsA group increased significantly (P < 0.05) (Fig. 7Ia), and the level of emodin-O-glucuronide in the gut chamber decreased slightly (Fig. 7Hb). After cyclosporine interference, both the eliminated amount of emodin (Fig. 7Ib), and the total production of emodin-O-glucuronide (Fig. 7Ja) decreased in the gut–liver-on-chips. All the above results indicate that emodin is an ideal test compound for gut–liver-on-chips, in which we observed the toxic effect, absorption, transport and first-pass effects of emodin before interference from cyclosporine. Cyclosporine enhanced the absorption of emodin and reduced the first-pass effects in the intestinal barrier. For a negative control, a herbal ingredient with low oral absorption, berberine, a substrate of PGP, was selected. As an inhibitor of PGP with non-obvious DILI effects, cyclosporine may theoretically enhance the absorption of berberine. The initial concentrations of berberine in the berberine p.o. group and berberine p.o. + CsA group were 17.77 and 16.48 μM; at these concentrations, a limited toxic effect on Caco-2 and liver spheroids (Fig. S7C) was observed. In gut–liver-on-chips, limited liver toxicity (Fig. S9A) and intestinal toxicity (Fig. S9B) of berberine were observed. We detected the signal for berberine in the samples (Fig. S10), and the results indicated that, compared with the berberine p.o. group, the berberine concentration in the liver chamber in the berberine p.o. + CsA group showed an increasing trend (Fig. S9C), while the concentration in the gut chamber decreased significantly (P < 0.05) (Fig. S9D). These results indicate that the first-pass effect is mitigated by the barrier function in the gut–liver-on-chips. The co-culture system can be used to study transporter-related DDI, subsequent toxic events, and change in drug exposure.

Fig. 7 Influence of cyclosporine on oral administration of emodin at toxic dosage. (A) Absorption and main metabolism process of emodin in gut–liver-on-chip. (B) Change in TEER. (C and D) Change in ALB level. (E) Metabolism of emodin by phase II enzyme. (F) Level of emodin a), emodin-O-glucuronide b), and emodin-O-sulfate c) at the basolateral side of gut–liver-on-chips. (G) Diagram of emodin transport and metabolism after cyclosporine interference. (H) Level of emodin a), emodin-O-glucuronide b), and emodin-O-sulfate b) at the apical side of gut–liver-on-chips at 48 h. (I) Relative Papp level a) and eliminated amount b) of emodin in gut–liver-on-chips. (J) Total production intensity of emodin-O-glucuronide in gut–liver-on-chips a) and percentage of emodin-O-glucuronide in liver chamber b). n = 3. Data are shown as mean ± SD. *P < 0.05. *, P < 0.05.

4. Discussion

This study has focused mainly on the application of gut–liver-on-chips in the study of hepatotoxicity and DDI. After a pump-less gut–liver-on-chip was established, it functioned well for 7 days, and applications of the co-cultivations were validated by proof-of-concept for the relationships between drug exposure and toxicity, resolving the temporal progression of toxicity, and modeling DDI, demonstrating its transformative potential for mechanistic toxicology.

Gut-on-chip models were first established to ensure a drug-permeable intestinal barrier with integrity. A polarized intestinal barrier model was established through the bilayer co-culture of Caco-2 and HT29-MTX-E12, reconstituting key physiological transport characteristics. TEER and fluorescence transport showed the formation of an intact barrier, consistent with the standards of intestinal-on-chip,²⁰ where TEER values exceed 100 Ω cm² in co-cultures of Caco-2 and HT-29-MTX cells, and the Papp of sodium fluorescein (molecular weight 376.27) is lower than that of 4 kDa dextran (1 × 10⁻⁶ per hour (ref. 20)) on day 14. Furthermore, to keep liver spheroids in the dynamic chamber, the intestine was constructed above the membrane. Histology slides revealed multi-cellular layer formation above the membrane by day 14 and day 21. Sustained tight junction integrity (days 14–21) coupled with transport of Rho123 confirmed polarized epithelium maturation and bidirectional efflux capacity in the gut-on-chip. Confirmation of barrier polarization integrity and the absence of cellular migration confirm the ability of the IBAC M1 to sustain intestinal barrier functions.

Organ-on-chip models developed from microfluidic devices replicate the complicated structure and physiological functions of human organs.^21,22 To simulate the processes of drug absorption, distribution and hepatotoxicity, gut–liver-on-chips were constructed based on the gut-on-chip model. In previous reports,¹² the quadruple-cell 3D spheroid system demonstrated superior metabolic functionality, and the cell components could be maintained for 10 days. Therefore, liver spheroids were adopted for subsequent investigations. The cellular composition, phenotypic stability, and functional fate of each cell type with the polarized intestinal barrier and quadruple-cell co-cultured liver spheroids may change over time in the dynamic co-cultured chip system. Although we have previously characterized these cells under static conditions (data not shown), their long-term fate, persistence, and functional maintenance in the current platform were not systematically assessed, which should be clarified in future work through comprehensive type-specific cell characterization. On day 14 of the construction of the gut-on-chip, 3D liver equivalents were integrated into gut-on-chip models. The function of gut–liver-on-chips was profiled through assessment of intestinal barrier integrity and hepatic metabolic capacity, with both organs retaining homeostasis throughout the 7-day cultivation period. Further, to support the biological interaction of gut and liver, the expressions of metabolic enzymes and transporters were detected on the gut–liver-on-chips. Consistent with a previous report,²³ gut–liver-on-chips preserved a physiological tight junction architecture and metabolic profiling compared with monocultures, with only minor changes in expression in specific transporters (CES2, P < 0.05), suggesting that co-cultivation seldom has an impact on the expression of intestinal protein. For liver spheroids, the chip may offer a low level of shear force of less than 0.45 dyn cm⁻², at a level which may avoid hepatocyte damage or dysfunction caused by mechanical obstruction.²⁴ Moreover, the level of shear force is aligned with the predicted physiological range (0.1 to 0.5 dyn cm⁻²).^25,26 Physiologically relevant fluid shear stress in the hepatic model enhanced metabolic and transporter expression, in agreement with previous reports in which low fluid shear stress improved polarization, liver-specific functions, and metabolic activity.^27–29 Furthermore, the co-cultured system increased the expression of liver transporters compared with mono-cultivation in our study, in agreement with a previous report in which co-culturing with intestinal cells improved hepatic function.²³ These results suggest that cultivated liver equivalents with shear stress and co-culturing with an intestinal barrier is a strategy to promote liver function, which can be easily offered by organ-on-chips.

Simultaneous drug exposure and detection of toxicity after oral administration of a drug was assessed with APAP, a hepatotoxic prototypical model compound³⁰ with a revealed toxic mechanism,³¹ as a test compound. After rapid absorption and metabolism in the liver, excess APAP is oxidized by CYP, leading to disturbance of the mitochondrial electron transport chain and subsequent production of reactive oxygen species (ROS), which triggers a cascade of stress signaling and ultimately leads to cell necrosis and apoptosis.^32,33 Specific toxic events occur in APAP-induced liver injury,³⁴ providing abundant parameters to illustrate the stage of hepatotoxicity. Assessment of APAP cytotoxicity revealed preserved intestinal viability and enhanced barrier integrity, demonstrating intact intestinal barrier function during evaluation of toxicity. In gut–liver-on-chips, hepatotoxicity was evidenced by elevated hepatic injury biomarkers AST and ALT, while decreasing ALB synthesis reflected compromised synthetic function. These findings align with clinical DILI progression patterns^35–38 and in vitro hepatocyte models demonstrating a decrease in APAP-induced ALB.³⁹ Moreover, APAP-induced hepatotoxicity involves sterile inflammation triggered by necrotic cell death.³⁴ Elevation of TNF-α serves as a complementary injury biomarker, reinforced by preclinical evidence demonstrating APAP-induced inflammatory cascades.^34,40,41 The gut–liver-on-chips replicated this procedure, confirming its predictive capacity for inflammation-associated DILI mechanisms. Mitochondrial dysfunction,⁴² extensive oxidative stress,⁴³ and later apoptosis or necrosis,^32–34 contribute to APAP-induced liver injury and are commonly used as parameters for an assessment of the therapeutic effect in DILI treatment in accordance with previous reports. Increased hepatotoxicity markers (ALT, ALB, and TNF-α) and apoptosis (cleaved caspase-3) in the APAP i.v. group were observed compared with the APAP p.o. group. Comparative analysis of APAP exposure in liver-on-chips, gut-on-chips, and gut–liver-on-chips revealed that the gut barrier modulates APAP absorption dynamics, decreased hepatic exposure and consequent toxicity, consistent with the principles of concentration-dependent toxicity.⁴⁴ The absorption profile explained the corresponding relationship between drug toxicity and drug exposure, demonstrating the utilization of the gut–liver-on-chips for simulating absorption-dependent pharmacological outcomes.

Conventional toxicological in vitro testing relies primarily on an evaluation of a single, late time-point, which is insufficient for a comprehensive risk assessment,³ while the gut–liver-on-chip tracks drug exposure and hepatotoxic events over time. The gut–liver platform addresses a key challenge in toxicology by integrating intestinal modulation of drug exposure⁴⁵ with consequent hepatic effects—essential for mechanistic risk assessment⁴⁴—and improved in vitro–in vivo extrapolation.¹ The system reliably detected hepatotoxicity through the elevation of ALT. As indicated in vivo, serum ALT levels increased significantly 12–24 h after APAP administration,⁴⁶ while our model recapitulated the injury trend with delayed progression, showing increases in ALT emerging at 32 h. However, as for ALB, a highly sensitive toxicity biomarker for APAP in liver-on-chips, compared to ATP, α-GST, or microRNA-122, demonstrated in Foster's research,⁴⁷ showed a decreasing trend at 24 h in our study, in line with Ullrich's study.³⁹ These variations in biomarker response^47,48 highlight methodology-dependent sensitivity profiles in the assessment of liver injury. Therefore, it is necessary to consider and explain these differences when comparing hepatotoxicity outcomes.

For the detection of hepatotoxic signals at different time-points, an evident decrease in TMRE signal was observed in the APAP i.v. group at 24 h, which agreed with an increasing tendency of ROS and cleaved caspase-3 signals at 32 h. This phenomenon can be attributed to the different pathological phases of APAP-induced liver injury, as Jaeschke summarized, wherein a mitochondrial permeability transition is characteristically observed during the early injury phase following administration of APAP.³⁴ APAP induces mitochondrial abnormalities, leading successively to pro-apoptotic morphological changes and apoptosis,⁴⁹ consistent with the observed cleaved caspase-3 signal at 32 h in the APAP i.v. group. An overdose of APAP is generally accepted to cause necrosis in the human liver, isolated hepatocytes, human hepatoma cell lines like HepaRG, animal models, and patients. However, hepatoma cell lines such as HepG2 and Huh-7 may undergo apoptosis when exposed to high levels of APAP for prolonged periods,^33,34 due to the lack of CYP enzyme activity in these cell lines.³³ Although mitochondrial superoxide formation occurs in the early phase after APAP has been administered,³⁴ we observed the CellROX signal late at 32 h in the APAP i.v. group. In human bodies, the main mechanism of APAP-induced liver injury requires bioactivation to N-acetyl-p-benzoquinone imine (NAPQI) by CYP enzymes. The high level of NAPQI binds to cellular proteins and depletes GSH, leading to the accumulation of reactive oxygen species (ROS) and subsequent oxidative cell death.⁵⁰ The delay in the CellROX signal may be attributed to the lack of CYP enzyme activity, highlighting the limitation of alternative models constructed with hepatoma cell lines, and suggesting that APAP-induced liver injury may follow different mechanisms.

Drug metabolizing enzyme activity is a potential risk factor for DILI.⁵¹ The classic view of the pathogenesis of APAP-induced liver injury is that the parent compounds lead to hepatotoxicity through metabolism by CYP, that is, CYP2E1, as well as several other CYP iso-enzymes, notably CYP1A2, 2D6, and 3A4.⁵² CYP-dependent liver injury from APAP is evidently ameliorated by CYP inhibitors either in animal models or in “in vitro alternatives”, such as ABT,⁵³ metyrapone,⁵⁴ micro-RNA⁵⁵ and some other natural products.⁵² ABT, a nonselective irreversible CYP inhibitor⁵⁶ that inactivates ∼80% of human CYP activity with 30 min of pretreatment,⁵⁷ was selected to confirm CYP-dependent toxicity, to prove the relative contribution of oxidative metabolism.¹⁵ In a previous report, it could largely inhibit CYP activity remaining in hepatocytes following 60 min of pre-incubation, but significant CYP inhibition remained even for the following 26 h.¹⁵ Herein, to investigate CYP450-dependent hepatotoxicity in gut–liver-on-chips, 3D hepatic spheroids were dynamically cultured to establish a two-organ chip. Even if weakened or absent expression of CYPs is commonly accepted in the HepG2 cell line compared to normal hepatocytes,⁵⁸ stronger CYP3A4 enzyme activity can be detected in the spheroids¹² than in the 2D HepG2 cell line, which was the pre-condition for CYP enzyme-related DDI being tested in a quadruple-cell co-cultured spheroid. Although APAP is metabolized primarily by CYP2E1, CYP3A4 is also an important isoenzyme, and the detection method for CYP3A4 enzyme activity is well established. Therefore, we chose changes in CYP3A4 enzyme activity to illustrate the inhibitory effect of ABT on CYP enzymes. Our quadruple-culture spheroids effectively modeled ABT-mediated CYP inhibition, confirming their suitability for DDI research. Although ABT delays oral absorption through gastric emptying effects,^59–61 exclusion of gastric components in gut–liver-on-chips adversely affected APAP concentrations in the liver chamber. Accordingly, ABT pretreatment showed no changes in hepatic APAP concentration in this system, especially in CellROX intensity. APAP is metabolized by CYP enzymes into the toxic metabolite NAPQI. When CYP enzyme activity is inhibited, a decrease in NAPQI levels would theoretically be expected. However, due to the instability of NAPQI,^62–64 with reference to other literature, ROS signals were selected as a substitute to indirectly indicate the production of NAPQI.^65,66 We used CellROX to detect ROS signals, and a reduction in ROS signal intensity was observed, suggesting that ABT pretreatment decreased ROS generation and indirectly indicated reduced production of NAPQI. Additionally, compared to the APAP i.v. group, we found that after ABT intervention, APAP-induced liver injury events, including mitochondrial damage, dead cell signals, and apoptosis signals, were reduced in the ABT + APAP i.v. group. Consistent with CYP inhibition by ABT,^53,67 the observed attenuation of toxicity suggests residual enzyme activity, although the exact in vitro mechanisms of ABT remain debated.^53,68 The system provides an example of a co-cultured organ-on-chip platform for CYP-mediated DDI and hepatotoxicity studies.

To explore drug transporter-related DDI, emodin and berberine were selected as victims and cyclosporine was selected as the perpetrator to alter the activities of transporters. The levels of prototype and metabolites were detected through UPLC-MS to investigate the changes in ADME of the victims caused by the perpetrator. Emodin, an active component in Rheum palmatum and Pleuropterus multiflorus, has a hepatoprotective effect at low concentration and a hepatotoxic effect at high concentration. Through detection of TEER and ALB levels, we observed limited gut toxicity and significant hepatotoxic effect, respectively, under the test dosage in the emodin p.o. group and the emodin i.v. group. Liver toxicity increased with increased exposure, as shown by the 25% lower ALB level in the emodin i.v. group than in the emodin p.o. group. Emodin showed limited absorption in gut–liver-on-chips, with oral exposure significantly lower than for intravenous administration. The tendency is consistent with a previous report where emodin at 20–40 μM induced detectable hepatocyte toxicity within 24 h.^69,70 Otherwise, the dose in our study fell between the blood concentration and hepatotoxic dose in rat models, corresponding to an observation of mild hepatic toxicity.⁷¹ As for first-pass effects, rapid intestinal glucuronidation⁷² explains the observation of low systemic parent compound levels and poor oral bioavailability. In the oral administration group, we observed substantial levels of emodin-O-glucuronide and minimal emodin-O-sulfate, both phase II enzyme metabolites. The results demonstrate metabolism of emodin in the chip system, leading to the low concentration of prototype in the liver chamber. As shown in a previous study, mono-glucuronide is the major metabolite of emodin in rodents, primates, and humans.⁷³ In Teng's study,¹⁹ approximately 22.5% of emodin appeared on the vascular side in a rat small intestine perfusion model. Moreover, the proportions of free emodin, emodin-O-glucuronide, and emodin-O-sulfate were 12.01%, 8.69%, and 1.84%, respectively, indicating a severe first-pass effect and low oral bioavailability. Notably, emodin-O-glucuronide and emodin-O-sulfate constituted 5.23% and 1.08% on the luminal side.¹⁹ In animal models, a severe first-pass effect after emodin administration was also observed, as well as a low concentration of prototype in the serum.^74,75 In this study, gut–liver-on-chips have been used to simulate the hepatotoxicity of emodin and reproduce the first-pass effect and low bioavailability of emodin in vivo. In particular, two first-pass metabolites were detected, and the signal intensity ratio of the metabolites was similar to that of the ex vivo study.¹⁹

Later, with cyclosporine selected as the perpetrator, we detected changes in the toxicity and ADME of emodin after interference. Cyclosporine is widely accepted to be an inhibitor of PGP, OATPs, and MRP2.⁷⁶ As for emodin, with a substrate of PGP and MRP2,^77,78 it is a liposoluble molecule, which can be rapidly absorbed through cell membranes and then reach a peak intracellular concentration, after which the prototype is metabolized or pumped out of the cell by a transporter, resulting in a decrease in the concentration of intracellular emodin.⁷⁹ A previous study reported that upregulation of the absorption and transport of emodin was observed after treatment with probenecid (an MRP2 and UGT inhibitor),⁸⁰ verapamil and cyclosporine.⁷⁸ The gut–liver-on-chip showed increasing hepatotoxicity and revealed increased concentration of hepatic emodin after intervention by CsA, consistent with the tendency for an increase in concentration in blood in vivo,⁸⁰ demonstrating its predictive absorption modeling ability. Emodin is metabolized through phase II enzymes in the intestine, and is transformed mainly into emodin-O-glucuronide by UGT,⁷³ which is inhibited by stilbene glucoside (a UGT inhibitor),⁸¹ probenecid (an inhibitor of MRP2 and UGT),⁸⁰ and piperine (a UGT inhibitor).⁸⁰ We observed the total inhibition of the production of emodin-O-glucuronide and the metabolism of emodin by cyclosporine. The effect can probably be attributed to the competitive inhibition of UGT enzymes by cyclosporine and emodin. Specifically, cyclosporine is an inhibitor of UGT1A1 and UGT1A4 (ref. 82) and is also a substrate of UGT1A and UGT2B,^83,84 suggesting potential competition with emodin for UGT-mediated glucuronidation. In a previous report, emodin-O-glucuronide was also a substrate of MRP2, and LTC4 and MK-571 (MRP2 inhibitors) can significantly inhibit the transport of emodin-O-glucuronide in the B to A direction in Caco-2 cells.⁸⁵ Similarly, we also observed a relative increase in emodin-O-glucuronide on the basal side after interference by cyclosporine in gut–liver-on-chips. A similar tendency to increase by berberine, a substrate of PGP, was observed in the liver chamber after interference by CsA, which is in agreement with that in vivo.⁸⁶ In this section, we selected cyclosporine and the substrates of transporters from herbal compounds to further explore the possibility of organ-on-chips to study transporter-related DDI in gut–liver-on-chips, in which the effect of the perpetrator on the victim transport, metabolism, and toxicity was observed.

The experiment was conducted in a pump-less organ-on-chip with porous membranes. While its structure is similar to that described in previous studies,^87,88 the chip used in this study had miniaturized chambers and channels to integrate multiple chip units onto a platform of the size of a standard 96-well plate. This design enables high-throughput screening and facilitates rapid detection using high-content imaging analysis systems. The chip can also be used for co-cultivation of 3D cellular aggregates, a key component of organ-on-a-chip systems. However, the miniaturized chip design makes it difficult to extract samples from specific chambers. In situ fluorescence staining and fluorescent probe detection were selected as alternative methods, enabling direct observation and analysis of cellular responses. Given that the chip design shares similarities with previously reported platforms,^87,88 the present study primarily emphasizes the application of our platform for mechanistic toxicity assessment and drug–drug interaction studies, which represents its key contribution.

In drug discovery, most gut-on-chips are focused on absorption and metabolism, rather than on the assessment of toxicity. This is largely because the intestine is regarded primarily as an absorption barrier, whereas the liver is considered the important target for drug toxicity. By integrating the two organs, we provide a greater understanding of the oral administration process, from absorption and transport to toxicity in both organs. On the other hand, although many advanced liver-on-chips have been developed, studies applying gut–liver-on-chips to drug toxicity remain limited. While liver-on-chips provide a tool for drug screening and hepatotoxic mechanisms, gut–liver-on-chips allow for the simultaneous investigation of pharmacokinetics and hepatotoxicity. In this study, with an IBAC M1 chip platform, we not only replicated key physiological features of the gut and liver but also applied the system to mechanistic risk assessment of hepatotoxic drugs and DDI. Furthermore, we provided data ranging from toxic effects to drug exposure, demonstrating the potential of gut-over-on-chips in preclinical hepatotoxicity assessment.

The limitations of the research may be listed as follows: 1) although the absorption process by the human body has been simulated, there is still a difference between the absorption profile of drugs and that of the human body, as gut absorption and liver metabolism do not occur at a sufficiently high rate in the gut–liver chip compared to those in the human body. Therefore, referring to Lee's research,⁸⁹ modification of the chip design parameters to compensate for these shortcomings will be able to improve the PK profile and improve the transformation of the data in vitro and in vivo. 2) Human primary liver cells are still the gold standard for in vitro studies. Although it has stronger CYP enzyme activity than HepG2, the quadruple-cell line formed liver spheroids are not the optimal solution for studying drug toxicity as their CYP enzyme activity is still lower than that of primary hepatocytes, and the model may be improved by constructing it with primary liver cells. 3) Although the association between drug exposure and liver toxicity can be described in the intestinal and liver co-culture model, the two-organ culture model also complicates the study, and researchers need to carefully consider the use of content. 4) Quantitative analysis of fluorescence signals, such as those for tight junction and transporter localization, and high-magnification imaging was not performed in the current study, which would have further validated the functional and morphological integrity of the model. This limitation will be addressed in future work.

Conclusions

This study demonstrates the successful application of a pump-less gut–liver-on-chip system for comprehensive hepatotoxicity assessment of exposure-dependent, time-dependent, enzyme/transporter-mediated drug–drug interactions, and first-pass effects. By combining Caco-2 and HT29-MTX-E12 intestinal barriers with quadruple-culture liver spheroids, we demonstrated the ability of the system to capture the complex interplay between drug exposure and toxicity responses. The intestinal barrier was shown to significantly modulate APAP pharmacokinetics and subsequent hepatotoxic effects, as evidenced by dynamic changes in ALT and ALB levels. Then, time-dependent experiments revealed the progressive development of toxicity, while CYP450 inhibition studies confirmed the expected patterns of ROS signaling, mitochondrial dysfunction, and caspase activation without altering APAP concentrations. Importantly, the platform successfully reproduced transporter-mediated interactions and the first-pass effect, with cyclosporine producing distinct effects on the distribution patterns of emodin (at toxic doses) and berberine (at non-toxic doses). These collective findings confirm the capacity of the system to recapitulate critical drug metabolism and toxicity pathways, including first-pass effects and enzyme/transporter-mediated interactions, providing a physiologically relevant alternative to conventional hepatotoxicity testing methods.

Abbreviations

2D	Two-dimensional
3D	Three-dimensional
ABT	1-Aminobenzotriazole
ADME	Absorption, distribution, metabolism, and excretion
ADME-T	Absorption, distribution, metabolism, excretion, and toxicity
ALB	Albumin
ALT	Alanine aminotransferase
APAP	Acetaminophen
AST	Aspartate aminotransferase
A to B	Apical to basolateral
B to A	Basolateral to apical
Calcein AM	Calcein acetoxymethyl ester
CYP	Cytochrome enzyme P450 families
DDI	Drug–drug interaction
DILI	Drug-induced liver injury
HE	Hematoxylin and eosin
HHSC	Human hepatic stellate cells
HPLC	High-performance liquid chromatography
HUVEC-T1	Human umbilical vein endothelial cell line
LC	Liquid chromatography
LDH	Lactate dehydrogenase
ICH	The International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use
OoCs	Organs-on-chips
Papp	Apparent permeability coefficient
PAS	Periodic acid-Schiff
PDA	Photo-diode array
PI	Propidium iodide
Rho123	Rhodamine 123
ROS	Reactive oxygen species
RT-qPCR	Real-time quantitative polymerase chain reaction
UPLC-MS	Ultra-performance liquid chromatography–tandem mass spectrometry
SD	Standard deviation
TEER	Transendothelial electrical resistance
TNF-α	Tumor necrosis factor-α

Author contributions

Yue Yu: conceptualization, methodology, investigation, data curation, formal analysis, visualization, writing – original draft; Tian Lin: methodology and investigation; Xiao Ye: methodology and investigation of content analysis; Yupeng Wang: methodology and investigation of emodin part; Rongrong Xiao: methodology (offered chip operation technical); Baiyang Sun: methodology and investigation of spheroid construction; Manman Zhao: writing; Jie Song: investigation; Bo Li: conceptualization, funding acquisition; Xiaobing Zhou: conceptualization, funding acquisition, project administration, supervision, writing – review & editing.

Conflicts of interest

The authors declare that they have no conflicts of interest.

Data availability

The data supporting this article have been included as part of the Results and supplementary information (SI).

Supplementary information: part of the method is available in SI Method. UPLC gradient conditions for emodin, berberine, and metabolites analysis in Table S1, RT-qPCR primers in Table S2, and APAP concentration is available in Table S3. See DOI: https://doi.org/10.1039/d5lc01094b.

Acknowledgements

This work was financially supported by the National Key Scientific Instrument and Equipment Development Projects of China during the 14th Five-year Plan Period (Grant No. 2022YFF0711100), State Key Laboratory Project (2023SKLDRS0129) and Beijing Science and Technology Planning Project (Z231100007223001). The interval IBAC rocker (MR100110) and resistance meter were kindly provided by Beijing Daxiang Biotech Co., Ltd. Fig. 3H, I, S3G and Video S1 were recorded with ImageXpress Micro Confocal (Molecular Devices), and these services were kindly supplied by the trial center of Molecular Devices. During the preparation of this work, we used Deepseek to improve readability and language. After using this tool/service, we reviewed and edited the content as needed and take full responsibility for the content of the publication.

References

1. P. A. Walker, S. Ryder, A. Lavado, C. Dilworth and R. J. Riley, Arch. Toxicol., 2020, 94, 2559–2585.
2. D. G. Brown and H. J. Wobst, J. Med. Chem., 2021, 64, 2312–2338.
3. S. Ramirez-Hincapie, B. Birk, P. Ternes, V. Giri, F. M. Zickgraf, V. Haake, M. Herold, H. Kamp, P. Driemert, R. Landsiedel, E. Richling, D. Funk-Weyer and B. van Ravenzwaay, Arch. Toxicol., 2023, 97, 2903–2917.
4. J. Yang, S. Imamura, Y. Hirai, T. Tsuchiya, O. Tabata and K.-I. Kamei, Biomicrofluidics, 2022, 16, 044113.
5. Y. Guo, X. Chen, P. Gong, G. Li, W. Yao and W. Yang, Int. J. Mol. Sci., 2023, 24, 4089.
6. M. Jie, H. Lin, Z. He, H. Liu, H. Li and J.-M. Lin, Sci. China:Chem., 2017, 61, 236–242.
7. J. W. Jeon, S. H. Lee, D. Kim and J. H. Sung, Biotechnol. Prog., 2021, 37, e3121.
8. M. Lucchetti, K. O. Aina, L. Grandmougin, C. Jäger, P. Pérez Escriva, E. Letellier, A. S. Mosig and P. Wilmes, Adv. Healthcare Mater., 2024, 13, e2303943.
9. T. Prueksaritanont, X. Chu, C. Gibson, D. Cui, K. L. Yee, J. Ballard, T. Cabalu and J. Hochman, AAPS J., 2013, 15, 629–645.
10. J. Yu, I. D. Petrie, R. H. Levy and I. Ragueneau-Majlessi, Drug Metab. Dispos., 2019, 47, 135–144.
11. A. Jala, S. Ponneganti, D. S. Vishnubhatla, G. Bhuvanam, P. R. Mekala, B. Varghese, P. Radhakrishnanand, R. Adela, U. S. Murty and R. M. Borkar, Drug Metab. Rev., 2021, 53, 285–320.
12. B. Sun, Z. Liang, Y. Wang, Y. Yu, X. Zhou, X. Geng and B. Li, Toxicology, 2024, 505, 153829.
13. T. M. Marin, N. de Carvalho Indolfo, S. A. Rocco, F. L. Basei, M. de Carvalho, K. de Almeida Gonçalves and E. Pagani, Chem.-Biol. Interact., 2019, 299, 59–76.
14. M. Kus, I. Ibragimow and H. Piotrowska-Kempisty, Pharmaceutics, 2023, 15, 2523.
15. K. E. Parrish, J. Mao, J. Chen, A. Jaochico, J. Ly, Q. Ho, S. Mukadam and M. Wright, Biopharm. Drug Dispos., 2016, 37, 200–211.
16. D. Chen, J. Yin, Z. Yang, W. Qin, J. Huo, J. Huang, J. Sun and W. Piao, Nutrients, 2022, 14, 2768.
17. I. V. Lebedeva, H. Min, M. Niu, W. Zhang, J. Yan, J. Li, X. Tan, B. Li, M. Su, B. Di and F. Yan, PLoS One, 2017, 12, e0187971.
18. W. Liu, L. Tang, L. Ye, Z. Cai, B. Xia, J. Zhang, M. Hu and Z. Liu, AAPS J., 2010, 12, 424–436.
19. Z.-H. Teng, S.-Y. Zhou, R.-T. Yang, X.-Y. Liu, R.-W. Liu, X. Yang, B.-L. Zhang, J.-Y. Yang, D.-Y. Cao and Q.-B. Mei, Biol. Pharm. Bull., 2007, 30, 1628–1633.
20. H. Liu, Y. Wang, X. Zhang, M. Zhang, P. Wang, J. Shang, Z. Li, L. Gong, J. Guo, W. Sun, J. Pi, X. Li, W. Ding, D. Wang, Z. Li, J. Zhang, L. Wang, X. Geng, R. Yang, P. Zhou, W. Tang, X. E. Zhang, C. Chen, S. Yang and J. Qin, Cell Regener., 2024, 13, 16.
21. B. Zhang, A. Korolj, B. F. L. Lai and M. Radisic, Nat. Rev. Mater., 2018, 3, 257–278.
22. A. Herland, B. M. Maoz, D. Das, M. R. Somayaji, R. Prantil-Baun, R. Novak, M. Cronce, T. Huffstater, S. S. F. Jeanty, M. Ingram, A. Chalkiadaki, D. Benson Chou, S. Marquez, A. Delahanty, S. Jalili-Firoozinezhad, Y. Milton, A. Sontheimer-Phelps, B. Swenor, O. Levy, K. K. Parker, A. Przekwas and D. E. Ingber, Nat. Biomed. Eng., 2020, 4, 421–436.
23. M. Shinohara, H. Arakawa, Y. Oda, N. Shiraki, S. Sugiura, T. Nishiuchi, T. Satoh, K. Iino, S. Leo, Y. Kato, K. Araya, T. Kawanishi, T. Nakatsuji, M. Mitsuta, K. Inamura, T. Goto, K. Shinha, W. Nihei, K. Komori, M. Nishikawa, S. Kume, Y. Kato, T. Kanamori, Y. Sakai and H. Kimura, Sci. Rep., 2021, 11, 5437.
24. W. Li, P. Li, N. Li, Y. Du, S. Lü, D. Elad and M. Long, Am. J. Physiol., 2021, 320, G272–G282.
25. P. D. Menezes, N. Gadegaard, R. M. N. Jorge and S. I. S. Pinto, Int. J. Numer. Methods Biomed. Eng., 2021, 37, e3445.
26. Y. Du, N. Li, H. Yang, C. Luo, Y. Gong, C. Tong, Y. Gao, S. Lü and M. Long, Lab Chip, 2017, 17, 782–794.
27. L.-D. Ma, Y.-T. Wang, J.-R. Wang, J.-L. Wu, X.-S. Meng, P. Hu, X. Mu, Q.-L. Liang and G.-A. Luo, Lab Chip, 2018, 18, 2547–2562.
28. H. Rashidi, S. Alhaque, D. Szkolnicka, O. Flint and D. C. Hay, Arch. Toxicol., 2016, 90, 1757–1761.
29. S. Khodabakhshaghdam, A. B. Khoshfetrat and R. Rahbarghazi, J. Biol. Eng., 2021, 15, 14.
30. S. Stern, H. Wang and N. Sadrieh, Cell, 2023, 12, 1476.
31. M. J. Toes, A. L. Jones and L. Prescott, Am. J. Ther., 2005, 12, 56–66.
32. D. S. Umbaugh and H. Jaeschke, Expert Rev. Gastroenterol. Hepatol., 2020, 15, 363–375.
33. H. Jaeschke and A. Ramachandran, J. Clin. Transl. Res., 2020, 6, 36–47.
34. H. Jaeschke, O. B. Adelusi, J. Y. Akakpo, N. T. Nguyen, G. Sanchez-Guerrero, D. S. Umbaugh, W.-X. Ding and A. Ramachandran, Acta Pharm. Sin. B, 2021, 11, 3740–3755.
35. G. Liss, S. Rattan and J. H. Lewis, Expert Opin. Drug Metab. Toxicol., 2010, 6, 1047–1061.
36. A. Wong and A. Graudins, Clin. Toxicol., 2017, 55, 879–892.
37. M. A. Kalas, L. Chavez, M. Leon, P. T. Taweesedt and S. Surani, World J. Hepatol., 2021, 13, 1688–1698.
38. L. Meunier and D. Larrey, Front. Pharmacol., 2019, 10, 1482.
39. A. Ullrich, C. Berg, J. G. Hengstler and D. Runge, Altex, 2007, 24, 35–40.
40. Z. Wang, W. Hao, J. Hu, X. Mi, Y. Han, S. Ren, S. Jiang, Y. Wang, X. Li and W. Li, Antioxidants, 2019, 8, 395.
41. S. Sriphoosanaphan, P. Rattanachaisit, K. Somanawat, N. Wanpiyarat, P. Komolmit and D. Werawatganon, Biomedicines, 2023, 11, 1534.
42. D. S. Umbaugh, N. T. Nguyen, H. Jaeschke and A. Ramachandran, Toxicol. Sci., 2021, 180, 186–195.
43. K. Du, A. Ramachandran and H. Jaeschke, Redox Biol., 2016, 10, 148–156.
44. H. A. Tennekes and F. Sánchez-Bayo, Toxicology, 2013, 309, 39–51.
45. C. M. Arian, T. Imaoka, J. Yang, E. J. Kelly and K. E. Thummel, Pharmacol. Ther., 2022, 230, 107962.
46. C. S. Bruells, P. Duschner, G. Marx, G. Gayan-Ramirez, N. Frank, T. Breuer, O. Krenkel, F. Tacke and J. C. Mossanen, Sci. Rep., 2021, 11, 6302.
47. A. J. Foster, B. Chouhan, S. L. Regan, H. Rollison, S. Amberntsson, L. C. Andersson, A. Srivastava, M. Darnell, J. Cairns, S. E. Lazic, K.-J. Jang, D. B. Petropolis, K. Kodella, J. E. Rubins, D. Williams, G. A. Hamilton, L. Ewart and P. Morgan, Arch. Toxicol., 2019, 93, 1021–1037.
48. M. Nautiyal, R. J. Qasem, J. K. Fallon, K. K. Wolf, J. Liu, D. Dixon, P. C. Smith and M. Mosedale, Toxicol. In Vitro, 2021, 70, 105010.
49. Y. A. Jagtap, P. Kumar, A. R. Dubey, S. Kinger, A. Choudhary, S. Karmakar, G. Lal, A. Kumar, A. Kumar, A. Prasad and A. Mishra, Life Sci., 2024, 349, 122732.
50. S. Sanoh, M. Santoh, M. Takagi, T. Kanayama, K. Sugihara, Y. Kotake, Y. Ejiri, T. Horie, S. Kitamura and S. Ohta, Toxicol. In Vitro, 2014, 28, 1176–1182.
51. D. Utkarsh, C. Loretz and A. P. Li, Chem.-Biol. Interact., 2016, 255, 12–22.
52. A. J. Langhammer and O. G. Nilsen, Phytother. Res., 2014, 28, 1573–1576.
53. K. Miyakawa, R. Albee, L. G. Letzig, A. F. Lehner, M. A. Scott, J. P. Buchweitz, L. P. James, P. E. Ganey and R. A. Roth, J. Pharmacol. Exp. Ther., 2015, 354, 230–237.
54. E. B. Nelson, M. Montes and M. Goldstein, Toxicology, 1980, 17, 73–81.
55. P. Gill, S. Bhattacharyya, S. McCullough, L. Letzig, P. J. Mishra, C. Luo, H. Dweep and L. James, Sci. Rep., 2017, 7, 12331.
56. S. Capello, L. Henderson, F. DeGrazia, D. Liberato, W. Garland and C. Town, Drug Metab. Dispos., 1990, 18, 190–196.
57. C. D. Linder, N. A. Renaud and J. M. Hutzler, Drug Metab. Dispos., 2009, 37, 10–13.
58. V. A. Arzumanian, O. I. Kiseleva and E. V. Poverennaya, Int. J. Mol. Sci., 2021, 22, 13135.
59. R. Stringer, V. Cordier, C. Afatsawo, P. Arabin, S. Desrayaud, L. Hoffmann, D. Lehmann, P. J. Lowe, F. Risser, J. Thiel, T. Widmer, P. Wipfli and M. Bigaud, Xenobiotica, 2018, 49, 13–21.
60. R. A. Stringer, E. Weber, B. Tigani, P. Lavan, S. Medhurst and B. Sohal, Drug Metab. Dispos., 2014, 42, 1117–1124.
61. S. Padmanabhan, H. Kaur, A. Rao, A. Saxena, Y. K. Gupta, T. T. Mariappan and V. K. Holenarsipur, Xenobiotica, 2018, 49, 646–654.
62. X. Zhang, R. Li, W. Hu, J. Zeng, X. Jiang and L. Wang, Biomed. Chromatogr., 2018, 32, e4331.
63. H. Jaeschke, Y. Xie and M. R. McGill, J. Clin. Transl. Hepatol., 2014, 2, 153–161.
64. M. R. McGill, M. Lebofsky, H.-R. K. Norris, M. H. Slawson, M. L. Bajt, Y. Xie, C. D. Williams, D. G. Wilkins, D. E. Rollins and H. Jaeschke, Toxicol. Appl. Pharmacol., 2013, 269, 240–249.
65. E. Leclerc, J. Hamon, I. Claude, R. Jellali, M. Naudot and F. Bois, Cell Biol. Toxicol., 2015, 31, 173–185.
66. T. Geib, C. Lento, V. Marensi, M. Thulasingam, J. Z. Haeggström, M. Olsson, D. J. Wilson, E. M. Leslie and L. Sleno, Anal. Sci. Adv., 2021, 2, 263–271.
67. M. Santoh, S. Sanoh, Y. Ohtsuki, Y. Ejiri, Y. Kotake and S. Ohta, Biochem. Biophys. Res. Commun., 2017, 486, 639–644.
68. L. A. Livoti, R. Sison-Young, D. Reddyhoff, C. P. Fisher, I. Gardner, R. Diaz-Nieto, C. E. Goldring and I. M. Copple, Toxicol. Sci., 2025, 203, 35–40.
69. L.-L. Jiang, Y. Jiang, D.-S. Zhao, Y.-X. Fan, Q. Yu, P. Li and H.-J. Li, Chem. Res. Toxicol., 2018, 31, 1052–1060.
70. M. Yang, Z. Xiong, H. Deng, X. Chen, Q. Lai, H. Wang and Y. Leng, J. Physiol. Pharmacol., 2023, 74 DOI:10.26402/jpp.2023.4.04.
71. D. Li, Q. Song, X. Ji, Y. Lyu, Y. S. Lai and Z. Zuo, Toxicol. Lett., 2022, 365, 74–85.
72. X. Dong, J. Fu, X. Yin, S. Cao, X. Li, L. Lin and J. Ni, Phytother. Res., 2016, 30, 1207–1218.
73. W. Liu, Q. Feng, Y. Li, L. Ye, M. Hu and Z. Liu, Toxicol. Appl. Pharmacol., 2012, 265, 316–324.
74. C.-S. Shia, Y.-C. Hou, S.-Y. Tsai, P.-H. Huieh, Y.-L. Leu and P.-D. L. Chao, J. Pharm. Sci., 2010, 99, 2185–2195.
75. W. Liu, Z. Zheng, X. Liu, S. Gao, L. Ye, Z. Yang, M. Hu and Z. Liu, J. Pharm. Biomed. Anal., 2011, 54, 1157–1162.
76. FDA, Drug Development and Drug Interactions | Table of Substrates, Inhibitors and Inducers.
77. J. Huang, L. Guo, R. Tan, M. Wei, J. Zhang, Y. Zhao, L. Gong, Z. Huang and X. Qiu, Front. Pharmacol., 2018, 9, 646.
78. Z.-H. Teng, S.-Y. Zhou, Y.-H. Ran, X.-Y. Liu, R.-T. Yang, X. Yang, C.-J. Yuan and Q.-B. Mei, Biosci., Biotechnol., Biochem., 2014, 71, 1636–1643.
79. S. Rao, J. Biomed. Opt., 2010, 15, 027005.
80. X. Di, X. Wang, X. Di and Y. Liu, J. Pharm. Biomed. Anal., 2015, 115, 144–149.
81. J. Ma, L. Zheng, T. Deng, C.-L. Li, Y.-S. He, H.-J. Li and P. Li, J. Ethnopharmacol., 2013, 147, 335–340.
82. Y. Liu, M. She, Z. Wu and R. Dai, J. Enzyme Inhib. Med. Chem., 2010, 26, 386–393.
83. C. P. Strassburg, A. Barut, P. Obermayer-Straub, Q. Li, N. Nguyen, R. H. Tukey and M. P. Manns, J. Hepatol., 2001, 34, 865–872.
84. R. Dupuis, A. Yuen and F. Innocenti, Clin. Chim. Acta, 2012, 413, 1318–1325.
85. Y. Wang, M. Zhao, B. Li and X. Geng, Heliyon, 2024, 10, e33631.
86. D. Colombo, L. Lunardon and G. Bellia, J. Toxicol., 2014, 2014, 1–6.
87. Y. I. Wang, H. E. Abaci and M. L. Shuler, Biotechnol. Bioeng., 2016, 114, 184–194.
88. C. Beaurivage, E. Naumovska, Y. Chang, E. Elstak, A. Nicolas, H. Wouters, G. van Moolenbroek, H. Lanz, S. Trietsch, J. Joore, P. Vulto, R. Janssen, K. Erdmann, J. Stallen and D. Kurek, Int. J. Mol. Sci., 2019, 20, 5661.
89. D. W. Lee, S. K. Ha, I. Choi and J. H. Sung, Biomed. Microdevices, 2017, 19, 100.

---

Footnote

† Joint first authors

---