Herein, we engineered an immunocompetent liver-on-a-chip platform integrating six distinct cell types (HepG2, LX-2, EA.hy926, U937, HuT-78, and HL-60 cell lines) to dissect cell-type-specific hepatotoxic profiles of pharmaceutical candidates. This chip builds upon our previous development of biomimetic liver sinusoid-on-a-chip platforms24,25,26, which reconstructed hepatic architecture with four hepatocyte subtypes and dual blood-bile circulation. The bioengineered system replicates chemokine-mediated immune recruitment and pathophysiological stress signaling, with computational fluid dynamics ensuring physiological hemodynamics. Inspired by gene-knockout cells (a powerful tool for identifying signaling pathways) and immune-compromised nude mice (widely used in immunology studies), we propose a targeted cellular depletion strategy to rapidly assess pharmaceutical candidates' hepatotoxic profiles within this biomimetic chip platform. This approach compares responses in the intact chip with those in the targeted cell-depleted chip to investigate the role of specific cell types. For proof-of-concept validation, we applied this strategy to resolve cell population-specific contributions to hepatotoxicity using four mechanistically diverse compounds: APAP, ethinyl estradiol (EE), sulfamethoxazole (SMX), and abacavir (ABC). Our results revealed distinct mechanisms. We further validated the detection of immune-dependent toxicity through targeted cellular depletion using allopurinol (ALP), a known immune-mediated DILI drug. This study establishes a targeted cell depletion strategy integrated within an immunocompetent liver-on-a-chip platform. This approach not only enables the rapid identification of hepatotoxic determinants but also represents a significant advancement in predictive toxicology systems, while providing strategic insights for subsequent pathway analysis.
Establishment of the immune- liver-on-a-chip platform
The immuno-liver chip was constructed by integrating immune components (T-cell and neutrophil systems) into a liver sinusoid chip optimized from our previous work. The core structure retains the original tri-layer Polydimethylsiloxane (PDMS) design with dual porous membranes and perfusion channels (Fig. 1A). To accommodate immune cells, two lateral chambers (1.5 mm × 10 mm) were added to the upper layer for culturing T lymphocytes (left) and neutrophils (right). These chambers are interconnected with the central region via 20-μm-wide microbarriers to enable cellular communication and chemotaxis. The functional microfluidic interfaces (Fig. 1B) feature three parallel chambers: a central blood-polarity channel (1.5 mm × 15 mm) matching our previous liver sinusoid chip, flanked by immune cell chambers. These side chambers connect to 1.6-mm-diameter inlet/outlet ports through 500-μm-wide channels, fabricated using a 200-μm-height template.
The device comprises a vertical assembly of three patterned PDMS layers (Fig. 1C). The top layer contains two lateral chambers (for HuT-78 and HL-60 cells) connected by 20-μm-wide microchannels (1 mm length, 50 μm spacing) and a central perfusion channel for artificial liver blood. The middle layer houses a chamber for HepG2 cells, while the bottom layer features a perfusion channel for artificial bile flow directed opposite to the overlying blood flow. A Polycarbonate (PC) porous membrane between the top and middle layers supports EA.hy926 and U937 cells on its upper surface and LX-2 cells on its lower surface. A second PC membrane between the middle and bottom layers supports HepG2 cells on its top surface. After cell seeding, lateral ports were sealed, and bidirectional perfusion (1 μL min) was applied to the blood and biliary channels for co-culture. The entire device measures 3 cm × 5 cm × 1.5 cm (Supplementary Fig. S1). Minimal adsorption by the chip was confirmed for drug delivery and detection applications, ensuring detection accuracy (Supplementary Fig. S2). This liver chip integrates an artificial liver sinusoid and an immune system incorporating U937 Kupffer cells, HuT-78 T cells, and HL-60 neutrophils. Under xenobiotic stimulation, HuT 78 and HL-60 cells from the lateral chambers can migrate into the central artificial sinusoid via chemotaxis as part of the hepatotoxicity process.
Biological characterization and dynamic immune cell chemotaxis of the immune- liver-on-a-chip
We first evaluated the chip's ability to simulate hepatic sinusoid physiology. In vivo, hepatocytes form complex polarized structures to facilitate substance transport between blood and bile. We characterized the polarization of 3D HepG2 clusters within the chip, focusing on multidrug resistance-associated protein (MRP) and bile salt export pump (BSEP) expressions. MRP2, localized at intercellular junctions with minor punctate cytoplasmic signals, while BSEP localized to the perinuclear membrane region, co-mediates toxic compound transport and bile secretion. Our analysis showed abundant MRP2 and BSEP expressions in the 3D HepG2 clusters (Fig. 2A). To assess bile acid dynamics, we perfused the upper channel with 2 μg mL 5-(and-6)-carboxy-2',7'-dichlorofluorescein diacetate (CDFDA) (1 μL/min for 2 h) followed by standard medium. Retrieved HepG2 clusters exhibited canaliculus-like fluorescence at cell junctions (Fig. 2B), indicating CDFDA metabolism and apical transport. We quantified active transport using cholyl-lysyl-fluorescein (CLF), observing significant higher (p < 0.0001) accumulation in HepG2(+) bile channels versus HepG2(-) controls (Fig. 2C). This process was inhibited by 250 μM benzbromarone, confirming HepG2-dependent transport. These results demonstrate that the 3D HepG2 clusters in the chip recapitulate in vivo-like tissue architecture and drug metabolism capabilities, with distinct polarization characteristics.
We next evaluated immune responses through N-formylmethionyl-leucyl-phenylalanine (fMLP)-stimulated HL-60 migration. HL-60 cell migration through microbarriers was quantified following stimulation with the chemotactic peptide fMLP (Fig. 2D). Our analysis revealed time- and concentration-dependent increases in HL-60 cell accumulation within microbarriers, with 0 nM fMLP serving as the negative control. As demonstrated in Fig. 2E, a 6-hour exposure to 100 nM fMLP induced an average of 7 HL-60 cells per microbarrier, representing a 7-fold increase compared to the control group (1 cell/microbarrier). These findings suggest that immune cells within the chip exhibit chemotaxis-directed migration toward hepatic zones, effectively recapitulating in vivo immune cell infiltration patterns during inflammatory processes. The observed migratory behavior validates the platform's ability to model inflammation-mediated immune cell trafficking. Notably, the graded response to increasing fMLP concentrations and prolonged exposure durations mirrors physiological chemotaxis dynamics, thereby confirming the system's capacity to process differential chemotactic signals (Supplementary Fig. S3). To characterize single-cell migration kinetics, real-time tracking of HL-60 cells was implemented during 100 nM fMLP stimulation over a 10-minute interval (Supplementary Fig. S4). Quantitative trajectory analysis demonstrated polarized cell movement from microbarrier entry points into the central channel, establishing definitive chemotactic responsiveness.
Biomechanical analysis of fluid dynamics inside the liver chip
To better understand and quantify fluid flow within the liver-on-a-chip system, a Computational Fluid Dynamics (CFD) model was constructed based on the chip geometry for numerical simulations (Fig. 3A). Numerical results demonstrate that the flow remains steady, with fluid velocities in the upper and lower channels significantly exceeding those in the middle channel. Velocity profiles in the upper and lower channels exhibit parabolic distributions, while streamlines align parallel to the channel substrates, originating from the inlet and terminating at the outlet (Fig. 3B). In contrast, fluid motion in the middle channel is driven by flows from the upper or lower channels penetrating through the porous PC membranes. The larger viscous forces near the inlet regions of the upper and lower channels induce transverse flow within the middle channel toward the opposing side. This convective mechanism generates weak recirculation patterns in the central region of the middle microchannel (Fig. 3C). Velocity components along (x-direction) and perpendicular to (y-direction) the flow were calculated to characterize flow field evolution. At positions closer to the inlet, x-direction velocities peak and display axisymmetric profiles (Fig. 3C), indicating enhanced seepage velocity near the porous membranes at the channel periphery. Conversely, y-direction velocities follow analogous trends but evolve inversely (Fig. 3D). These analyses provide critical insights into the internal flow field topology of the chip, which is essential for understanding mass transfer under dynamic fluid conditions.
Subsequently, we calculated the shear stress distribution perpendicular to the flow interface (Fig. 3E). In the upper and lower microchannels, pronounced shear stress distributions were observed near the walls due to wall-induced resistance effects. However, in the middle channel, negligible shear stress was detected owing to the substantially lower flow velocities. A systematic quantification of the average stress across the membrane surfaces revealed a linear correlation with the channel flow velocity (Fig. 3F), providing critical guidance for selecting optimal operational parameters in practical applications.
Targeted cellular depletion strategy for detecting drug-induced hepatotoxicity heterogeneity
Building upon this immune-liver-chip platform, we developed a targeted cellular depletion strategy to rapidly identify drug-induced hepatotoxicity heterogeneity. Inspired by gene knockout technology, this approach was adapted for functional characterization of specific cellular components rather than genes or signaling pathways. Cell-type-specific liver chips were generated by omitting target cells during assembly, and toxicity changes were assessed by comparing depleted versus intact systems (Fig. 4A). Specifically, we sequentially depleted individual components of the immuno-liver-chip to generate four functionally impaired models: (i) bile transport-blockaded, (ii) stellate cell-depleted, (iii) Kupffer cell-depleted, and (iv) T lymphocyte-depleted. Using these modified platforms alongside the intact immune-competent chip as a control, we systematically investigated hepatotoxicity mechanisms of four clinical drugs: APAP, EE, SMX, and ABC. These compounds were selected based on their well-established liver injury mechanisms in clinical practice, which involve distinct processes including hepatocyte damage, cholestasis, and immune activation. Testing drugs with clearly defined toxicity mechanisms enabled direct evaluation of the chip's capability in toxicity assessment.
Comparative analysis of Lactate Dehydrogenase (LDH) release levels across the five models revealed the critical role of specific cellular components in drug-specific hepatotoxicity. As shown in Fig. 4B, dose-dependent LDH secretion in response to APAP, EE, SMX, and ABC was observed across the multi cell-depleted liver chips, consistent with established dose-response relationships for drug-induced liver injury. Divergent hepatotoxicity profiles across experimental platforms were also observed. Notably, toxicity results for each drug consistently differed between deficient models and intact immune-competent chips, though the magnitude of changes varied. This confirms that diverse hepatic cell populations collectively mediate toxicological outcomes in the immune-competent liver chip. Furthermore, due to the relatively weak immune activity of HuT 78 and HL-60 cells, and because LDH release does not comprehensively reflect hepatotoxicity, we conducted parallel experiments using human peripheral blood mononuclear cells (PBMCs) instead of the chip's immune components (Supplementary Fig. S5). As expected, alanine aminotransferase (ALT) secretion levels in all chip models also showed dose-dependent increases. This strategy of using deficient liver chip models not only enables rapid identification of key hepatotoxicity factors but also provides critical guidance for subsequent mechanistic elucidation.
Cell-type-specific heterogeneity of drug-induced hepatotoxicity
We then analyzed the test results of these four drugs across different liver-on-a-chip models to reveal their differences in liver toxicity. First, we plotted heat maps of hepatotoxicity for these four drugs in different models (Fig. 5A). We found that the four drugs showed significant consistency in the release of toxicity indicators LDH and ALT, indicating the feasibility of using LDH and ALT as liver toxicity detection indicators. Interestingly, we found that SMX and ABC have more similar toxicity profiles, while APAP shows significant toxicity differences from other drugs. This might relate to the fact that hepatotoxicity of SMX and ABC tends to be immune-mediated, whereas APAP is mainly mediated by hepatocyte damage.
To further identify the hepatotoxic determinants of each drug, we plotted model-specific toxicity histograms. From these (Fig. 5B and C for LDH and ALT, respectively): For APAP, we observed no significant differences in dose-dependent LDH release curves between the four depleted models and the intact immune-competent liver-on-a-chip under identical exposure conditions. This indicates that Kupffer cell, stellate cell, or T lymphocyte depletion, as well as bile duct obstruction, did not markedly affect APAP-induced hepatotoxicity, consistent with its known mechanism where reactive metabolites directly damage hepatocytes. For EE, the bile transport-blockaded model exhibited substantially elevated hepatotoxicity (14.8% increase in LDH (p < 0.0001) and 24.2% increase in ALT (p < 0.0001) versus the intact model), while other depleted models (T lymphocyte-, stellate cell-, or Kupffer cell-depleted) showed negligible variation. This suggests a synergistic toxic interaction between bile accumulation from cholestasis and EE, though we acknowledge the precise mechanism remains undetermined due to current limitations in cholestatic hepatotoxicity models. While we cannot conclusively determine whether bile components exacerbate EE toxicity or vice versa, our findings align with clinical evidence of EE-induced hepatotoxicity through cholestasis. For SMX, we detected markedly reduced hepatotoxicity in Kupffer cell-depleted models (LDH: ↓16.9%, p < 0.0001; ALT: ↓8.2%, p < 0.01) and T lymphocyte-depleted models (LDH: ↓13.1%, p < 0.0001; ALT: ↓8.3%, p < 0.01) compared to the intact model, with minimal changes (p > 0.05) in stellate cell-depleted or bile transport-blockaded models. These results support reported mechanisms of SMX-induced immune-mediated hepatotoxicity. For ABC, while sharing similarities with SMX, we identified distinct hepatotoxic mechanisms. Our data demonstrate that T lymphocyte depletion significantly attenuated ABC-induced hepatotoxicity (LDH: ↓7.7%, p < 0.001; ALT: ↓21.8%, p < 0.001), whereas Kupffer cell depletion had negligible effects. This corresponds to ABC's mechanism involving covalent binding to hepatocyte HLA domains that triggers T cell-mediated immune attacks, conclusively establishing its T cell-dependent hepatotoxicity.
To further validate the resolution of immune-dependent toxicity through targeted cellular depletion within our immune-liver-chip platform, we tested ALP (a known immune-mediated DILI drug) (Supplementary Fig. S6). As expected, allopurinol-induced toxicity was significantly attenuated in the Kupffer cell-depleted configuration. This reduction aligns with clinically observed immune-mediated liver toxicity and demonstrates our system's capability to discern immune-specific mechanisms, thereby strengthening the mechanistic validation of our approach.