A high-throughput liver-kidney metabolic interaction chip for insights into the nephrotoxicity mechanisms of triptolide

Abstract

The kidney organ-on-a-chip (OoC) is a powerful tool for studying drug-induced nephrotoxicity, but its application is limited by the absence of liver metabolism and low throughput. Here, we developed a high-throughput liver-kidney OoC system (HLKOC) featuring microfluidics, plug-in biomimetic cups (MIMICups), scalable flow channel plates, precision-cut liver slices (PCLS), and 3D HK-2 cell spheroids. We first established a functional endothelial barrier by optimizing cell types, biomimetic blood flow rate, serum content, and membrane pore size. The structural and functional integrity of the PCLS and HK-2 spheroids within the MIMICups was then confirmed through histological staining, metabolic assays, and functional tests for viability, polarization, and transport. To evaluate the system's utility, we integrated the HLKOC with a single kidney OoC control and multidisciplinary techniques—including biochemical analysis, computational toxicology, molecular docking, metabolomics, and transcriptomics—to investigate the nephrotoxicity of triptolide (TPL) and its underlying mechanisms. Results showed that, compared to the single kidney OoC, the HLKOC exhibited higher levels of urea, total protein, and albumin in the biomimetic blood, confirming the robust biosynthetic capacity of the PCLS-based liver chip and its ability to better simulate in vivo conditions. Notably, a TPL-induced elevation in urea was observed only in the HLKOC, demonstrating the superior sensitivity of the liver-kidney co-culture. Multi-omics analysis revealed that TPL induced distinct metabolic and transcriptional responses in the HLKOC, involving pathways related to linoleic acid metabolism and vesicle-mediated processes, and led to the significant downregulation of transport proteins cubilin and GLUT1. These findings highlight the advantages of the HLKOC over single-organ systems for drug toxicity assessment and provide new insights into the mechanisms of TPL-induced nephrotoxicity.

---

1. Introduction

The kidneys are the primary organs for the clearance and detoxification of xenobiotics. Tubular epithelial cells, which are integral components of the kidney's functional units, are responsible for the active reabsorption and secretion of solutes. Proximal tubular epithelial cells, in particular, are enriched with membrane proteins that facilitate active transport functions, significantly elevating intracellular compound concentrations.¹ This active uptake and accumulation of substances are the primary mechanisms underlying drug-induced nephrotoxicity (DIN).² DIN represents a significant clinical challenge, contributing substantially to the acute kidney injury (AKI) and the development of chronic kidney disease.³ Moreover, the insidious nature of DIN hinders its early detection, posing a challenge for effective intervention and prevention.⁴ The effective detection of DIN and the exploration of its mechanisms in preclinical stages are thus crucial for mitigating drug-induced kidney injury, advancing new drug development, and devising detoxification strategies. However, current strategies, including computational prediction, animal testing, and in vitro models, have several drawbacks.⁴ For instance, the accuracy of computational prediction is constrained by the scarcity of reliable datasets,^5,6 while animal-based toxicity studies face growing scrutiny due to species differences and ethical concerns.⁷ Currently, widely used cell line-based in vitro models often lack the microenvironment and multi-organ interactions pivotal to drug toxicity manifestation.⁸ For example, orally administered drugs experience hepatic metabolism prior to renal excretion. This hepatic biotransformation is a crucial factor in a drug's overall therapeutic and toxicological profile, potentially leading to detoxification, excretion, or bioactivation.⁹ However, constructing complex in vitro models that can recapitulate multi-organ interactions to study drug metabolism and delivery remains a significant challenge.^10,11

Organ-on-a-chip (OoC) technology offers an opportunity to address this issue. OoC, also known as microphysiological systems, are a class of microdevices that contain cells and/or tissues and mimic the structure, function, and response to exogenous stimuli (including drugs) of organs.¹² By incorporating liver chip, kidney chip, and microfluidic-based biomimetic blood circulation with shear stress, OoC enables the creation of a physiological model simulating liver metabolism and kidney excretion. This facilitates the study of drug-induced renal toxicity and its underlying mechanisms under physiologically relevant conditions.¹ Given the substantial functional and structural variations across biological materials, the selection of tissues and/or cell types as well as the culturing strategy are key considerations in constructing OoC for drug metabolism and renal toxicity research. Several studies have constructed liver-kidney OoC models, each with its own focus on functionality. For instance, Huang and colleagues¹⁰ constructed a liver-kidney OoC system with a parallel circulation system, utilizing L-02 and LX-2 cells to mimic the hepatic lobules, and HK-2 cells to mimic the renal proximal tubules. This system prioritizes safety evaluation over metabolic capacity and features a lower throughput.¹⁰ Theobald et al.¹³ cultured HepG2 and RPTEC cells in interconnected compartments to mimic the liver and kidney, respectively, and evaluated the system's capability for the bioactivation of vitamin D3. Nevertheless, 2D cultures of single cell lines have been shown to exhibit only weak constitutive expression of metabolic enzymes, necessitating the use of activators to induce the expression of specific enzymes.¹⁴ Precision-cut liver slices (PCLS), which include the extracellular matrix, a variety of cell types, and retain cellular polarity and strong metabolic capabilities, are a good representation of intact liver tissue and has gained widespread recognition in liver pathological and toxicological research.^9,15 Although the functionality of PCLS may be compromised by tissue necrosis due to limited oxygen diffusion, the integration of a dynamic microfluidic perfusion system can provide a microenvironment that closely mimics in vivo conditions. This includes the continuous renewal of culture medium and removal of metabolic waste, thereby significantly enhancing the utility of PCLS.¹⁶ For the kidney-on-a-chip, the renal proximal tubule is the main toxic target of many drugs (as mentioned above).¹ As an immortalized proximal tubule cell line from normal adult human kidney, HK-2 is reported to retain transportation functions and sensitivity to toxin.¹⁷ However, under monolayer (2D) culture conditions, the HK-2 cell line exhibits negligible expression of polarity proteins, indicating an inability to establish basal-apical polarity. This dedifferentiated state is ameliorated in three-dimensional (3D) cell spheroid cultures, suggesting that 3D cultured HK-2 cell spheroids are an ideal material for constructing kidney chip models.¹⁸ Moreover, the sodium ion transport function is a critical performance indicator for in vitro models of proximal tubule epithelial cells. Sodium transport and its related processes are essential for maintaining in vivo acid–base balance and electrolyte homeostasis. Proximal tubule epithelial cell membranes contain sodium–glucose co-transporters, which transport sodium and glucose from the tubular fluid into the cells, and sodium–hydrogen exchangers, which exchange intracellular protons for extracellular sodium ions.¹⁹ Similarly, sodium ion transport functionality was demonstrated to be recapitulated under 3D culture conditions of HK-2, rather than monolayer culture.¹⁸ In addition to biological elements, throughput is a pivotal factor for the practical adoption of OoC systems, encompassing two key dimensions: (i) the capacity for cost-effective, large-scale fabrication, and (ii) the ability to host sufficient biomass to generate enough material for analyzing clinically relevant endpoints, such as biomarkers.²⁰

Triptolide (TPL) constitutes the principal active constituent of Tripterygium wilfordii Hook. f., a frequently utilized medicinal herb in traditional Chinese medicine. It exhibits a spectrum of pharmacological properties including anti-inflammatory, immunosuppressive, and antineoplastic activities.²¹ However, non-clinical studies and clinical case reports indicate that TPL is toxic to various organs, including the liver, kidneys, spleen, gastrointestinal tract, and heart.²² For instance, an animal study indicates that high-dose TPL induces severe kidney damage in SD rats, characterized by loss of brush border, tubular obstruction, and detachment of proximal tubular epithelial cells from the basement membrane.²³ However, in contrast to the relatively well-characterized mechanisms of hepatotoxicity, research into the nephrotoxic mechanisms of TPL is still in its infancy, hindering the development of targeted detoxification strategies.²⁴ As with many other orally administered drugs, TPL undergoes initial hepatic biotransformation in vivo, forming various metabolites such as hydroxylated compounds and glutathione conjugates.²⁵ These findings underscore the need for multi-organ chip systems to elucidate its nephrotoxic mechanisms. Additionally, drug toxicity often extends beyond a single target, exerting a systemic assault on interconnected biomolecular networks.²⁶ A deep understanding of the systemic response of tissues or organs to toxic compounds is of great significance for the early detection of toxicity in clinical settings and the development of targeted detoxification strategies. To capture the full spectrum of toxicity and its molecular underpinnings, an integration of interdisciplinary technologies centered on the OoC framework is indispensable. For example, data-driven computational toxicity predictions aid in the evaluation of a compound's ADMET (absorption, distribution, metabolism, excretion, and toxicity) profile. Among these, prediction models based on deep neural networks, such as ADMETlab, are emerging as some of the most promising tools in toxicology, enabling automated feature construction.⁶ Furthermore, molecular docking techniques can be utilized to ascertain a collection of potential targets by analyzing compound-target interactions and affinities.²⁷ Concurrently, multi-omics approaches, such as transcriptomics, provide a holistic view of molecular network dynamics, helping to elucidate the nephrotoxic mechanisms of TPL and its metabolites. In the present study, we first developed a high-throughput liver-kidney OoC device, which integrates biological elements such as endothelial barrier, PCLS, and 3D culture of renal HK-2 cells, and demonstrated its functionalities related to metabolism and transport. Following this, we investigated the renal toxicity and underlying mechanisms of TPL as manifested in HLKOC employing computational toxicology, molecular docking, biochemical analysis, metabolomics, and transcriptomics. We hypothesized that the HLKOC would better recapitulate in vivo dynamics than a single kidney OoC, and that TPL would exhibit distinct nephrotoxicity profiles and molecular targets in the HLKOC compared to the single-chip system.

2. Materials and methods

2.1 The structure and components of the high-throughput liver-kidney OoC (HLKOC) device

In addition to the culture medium reservoir and the peristaltic pump, the core components of the HLKOC device are scalable flow channel plates and insertable biomimetic cups (MIMICups) (Fig. 1A, B, and S1). To enhance fabrication throughput, the main bodies of both the flow channel plate and the MIMICup are made of polystyrene (PS), and their structural designs are also suitable for large-scale injection molding.

Fig. 1 The components of HLKOC and the condition optimization for endothelial barrier establishment. A. Besides the culture medium reservoir and the peristaltic pump, the core components of the HLKOC device are a scalable flow channel plate and an insertable biomimetic cup (MIMICup). The MIMICup serves as the culture chamber within the chip. The assembly of a MIMICup containing either liver- or kidney-related biomaterials and the flow channel plate is referred to as a “liver chip” or “kidney chip”, respectively. The channel beneath the chip was pod-shaped to ensure the consistency of shear stress across different locations. B. Photographs and structural characteristics of the scalable flow channel plate (upper panel) and MIMICup (lower panel). The upper panel shows the flow channel plate, which features 24 wells (6 arrays of 4 well each, with each array sharing a single flow channel). Arrays 1–3 (top to bottom) are left empty, while the wells in arrays 4–6 are all fitted with MIMICups to better illustrate the device's composition. Longitudinal structure of the insertable MIMICup for liver chip (C) and kidney chip (D) and E. The X-ray microscope (XRM) image showed the structure of the PET membrane at the bottom of the MIMICup. The 8 μm pores are uniformly distributed across the membrane. F. The PET membrane with an 8 μm pore size exhibited significantly better macromolecular permeability compared to the PET membrane with a 0.4 μm pore size (n = 3 for each group; one-way ANOVA; ***, P < 0.001). G. Hematoxylin and eosin (H&E) staining shows the endothelial barrier formed by EA.hy926 cells adhering to the underside of the MIMICup under different flow rates and serum concentrations on days 1, 3, 5, and 7. H. Immunofluorescence staining demonstrated the expression of ZO-1 (Zona occludens-1) protein (150 μl min⁻¹, 20% FBS, D5).

Scalable flow channel plate. The flow channel plate holds the MIMICup and provides underlying medium flow channels. It is available in single- or multi-well designs to accommodate various experimental requirements (Fig. 1A, B, and S1). The sides of the plate feature openings designed to accommodate standard Luer connectors (with a 6% taper), allowing for medium inflow and outflow. On the plate's underside, directly beneath the culture cells, a podiform cavity (1 mm height) is designed, connecting the two side openings. The underside of the plate is sealed with a highly transparent PMMA (polymethyl methacrylate) film using an acrylate adhesive. The space between the podiform cavity and the PMMA film constitutes the 1 mm-high flow channel.

Biomimetic cup (MIMICup). The MIMICup functions as the culture chamber within the chip. It features laterally protruding flanges on the top for easy handling and is open at both the top and bottom. The outer surface of the MIMICup's body features two circumferential grooves for securing O-rings. When inserted into the flow channel plate, these two O-rings create a seal. The bottom opening of MIMICup is sealed with a porous PET (polyethylene terephthalate) membrane, which has a diameter matching the outer diameter of the MIMICup and is attached via thermal bonding. This porous PET membrane allows for substance exchange between the culture chamber and the flow channel, while its underside provides a substrate for endothelial cells to adhere, grow, and form an endothelial barrier.

Selection of PET membrane pore size. The permeability of different pore-sized PET membranes (0.4 μm and 8 μm) to biomolecules was evaluated using the permeation level of dextran (40 kDa). The 40 kDa dextran probe was chosen to evaluate the PET membrane, which must serve a dual function: providing an attachment substrate for endothelial cells (e.g., EA. hy926) while permitting the transport of vital substances. This transport must include key growth factors and nutrients from the culture medium—such as basic fibroblast growth factor (bFGF, ∼16.5 kDa), insulin-like growth factor (IGF, ∼7.5 kDa), platelet-derived growth factor (PDGF, ∼30 kDa), and transforming growth factor β1 (TGF-β1, ∼44 kDa)—that are critical for maintaining the physiological activity of PCLS and renal HK-2 cell spheroids. The three-dimensional structure and pore size distribution of the PET membrane were characterized using 3D X-ray microscopy (XRM).

2.2 The biological elements in the HLKOC device

The biological components in the HLKOC include an endothelial barrier, formed by endothelial cells growing to confluence on the underside of the PET membrane, and either liver- or kidney-related biomaterials cultured within the MIMICup. The assembly of a MIMICup containing liver- or kidney-related biomaterials with the flow channel plate is referred to as a “liver chip” or “kidney chip”, respectively.

Endothelial barrier. A selectively permeable endothelial barrier is crucial for recapitulating in vivo functions in an organ-on-a-chip. We performed two rounds of optimization for cell type, shear stress, and medium composition (e.g., serum concentration), as these factors are critical for endothelial cell growth, confluence, and barrier formation.²⁸ First, we evaluated the ability of two commonly used endothelial cell models—primary human umbilical vein endothelial cells (HUVEC) and the immortalized EA.hy926 cell line—to grow and form a barrier under a gradient of flow rates, ranging from low shear stress (0.0165 dyne per cm² at 15 μL min⁻¹) to high shear stress (0.825 dyne per cm² at 750 μL min⁻¹). The EA.hy926 cell line was included because previous studies have shown it retained most of the characteristics of primary HUVECs and that their phenotypic responses to drugs (e.g., doxorubicin) are highly similar.^29,30 HUVECs and EA.hy926 cells were purchased from National Collection of Authenticated Cell Cultures (code: SCSP-5330) and Shanghai Fuhe Biotechnology Co., Ltd (code: FH0279),³¹ respectively, and authenticated by STR profiling. Prior to seeding, the underside of the PET membrane (8 μm) at the bottom of the MIMICup was pretreated with 200 μL of fibronectin (100 μg mL⁻¹) for 12 h under aseptic conditions. Then, 200 μL of HUVEC or EA.hy926 cell suspension (2.5 × 10⁵ cells mL⁻¹) was seeded onto the fibronectin-coated surface. After 12 hours, the MIMICup was inverted and inserted into the wells of the flow channel plate, which was pre-assembled with a PMMA bottom film. The establishment of tight junctions and the structural and functional integrity of the endothelial barrier were evaluated over time by assessing ZO-1 expression using immunofluorescence staining. The EA.hy926 cell line, which demonstrated the ability to form a robust endothelial barrier under higher shear stress, was selected for subsequent optimization of the culture medium. Briefly, MIMICups seeded with EA.hy926 cells were cultured for 1, 3, 5, and 7 days under two flow rates (50 and 150 μL min⁻¹) and two FBS concentrations (10% and 20%). Endothelial barrier establishment was assessed using hematoxylin and eosin (H&E) and immunofluorescence (ZO-1) staining.

Liver chip. Rat-derived precision liver slices (PCLS) were selected as the biological element for the liver chip. Female SD rats (120–180 g) were purchased from Beijing Viton Lihua Laboratory Animal Technology Co., Ltd (Beijing, China). Following a 3–5 day acclimatization period in a temperature-controlled room (23 ± 3 °C), 40–70% relative humidity, and a 14 h : 10 h light/dark cycle, the rats were euthanized, and their livers were excised under sterile conditions. The livers were washed three times with cold, oxygen-pre-saturated DMEM and continuously bubbled with a 95% O₂/5% CO₂ gas mixture. Under sterile conditions, the livers were embedded in sterilized, cooled 2% agarose and placed at 4 °C for 7 minutes to solidify. After solidification, 300 μm-thick PCLS were prepared using a vibratome and then trimmed into 6 mm diameter circular pieces using a biopsy punch.³² The entire preparation was conducted rapidly (≤30 min) under strict sterility conditions. After preparation, an initial volume of 180 μL of complete DMEM was introduced into the MIMICup with endothelial barrier, after which the PCLS were carefully placed within the medium. All animal procedures were performed in accordance with the “Regulations on the Management of Laboratory Animal Welfare of the Institute of Chinese Materia Medica, China Academy of Chinese Medical Sciences (ICMM-CACMS)” and approved by the Animal Ethics Committee of ICMM-CACMS (approval no. 2023B162 and 2023B191).

Kidney chip. The biological element of the kidney chip is a 3D-bioprinted HK-2 cell spheroid. The HK-2 cell line was purchased from BeNa Culture Co., Ltd (code: BNCC339833) and authenticated by STR profiling. Within in the MIMICup featuring an endothelial barrier, 50 μL of 70% Matrigel (Corning Co., Ltd.) was introduced and allowed to solidify for 30 minutes at 37 °C, serving as the growth matrix for HK-2 cell spheroids. An aseptic print nozzle was affixed to a high-precision bioprinter (BP4000, Shanghai AureFluidics Technology Co., Ltd.) and primed sequentially with 75%, 50%, and 25% ethanol, followed by sterile water and PBS to saturate the microfluidic channels. A HK-2 cell suspension (3 × 10⁶ cells mL⁻¹ in PBS) was used as the bio-ink. Using the bioprinter, four deposition sites were printed in each MIMICup, with four iterations per site, resulting in approximately 10 000 cells per MIMICup. The MIMICups were then incubated at 37 °C for 1 hour before the addition of 150 μL of complete DMEM supplemented with 10% Matrigel. Finally, the MIMICups were integrated into a kidney-on-a-chip platform. The viability of the HK-2 cell spheroids was assessed using calcein-AM/PI staining.³³

2.3 Assembly of the HLKOC device

A liver chip array (4 wells, housing 4 MIMICups) and two kidney chips (housing 2 MIMICups) were interconnected via TPE (thermoplastic elastomer) tubing to a common media reservoir, forming a closed-loop circulatory system. A peristaltic pump drives the culture medium from the reservoir through the liver chip and then through the kidney chip before returning to the reservoir. The selected number and ratio of organ chips (four liver and two kidney chips) were determined to ensure sufficient metabolic capacity of the hepatic module and enable the collection of adequate biological samples. Specifically, the dimensions of MIMICup were constrained (inner diameter: 6.3 mm) to facilitate efficient substance exchange and minimize edge-to-center consistency. This design limited the cell payload per cup. Our preliminary experiments demonstrated that a configuration of four liver chips was necessary to generate sufficient drug metabolites. For the renal module, two kidney chips were connected to mimic bilateral kidneys.

2.4 Validation of transport and metabolism-related functionalities of the HLKOC

Metabolic capability of PCLS. The physiological structure of the PCLS after 24 hours of perfusion culture was observed using H&E staining. To assess the activity of key CYP enzymes in PCLS, a quantitative analysis was conducted by co-incubating PCLS with specific substrates and detecting the products via mass spectrometry. Specifically, nifedipine is metabolized by rat CYP3A1/2 (corresponding to human CYP3A4) into dehydronifedipine; while chlorzoxazone is metabolized by rat CYP2E1 (corresponding to human CYP2E1) into 6-hydroxychlorzoxazone.³⁴ Therefore, PCLS were co-incubated with nifedipine and chlorzoxazone (20 μM) under perfusion culture for 2, 6, 12, and 24 hours, respectively. The culture medium and a cold acetonitrile rinse (3-fold volume) from the MIMICups were collected. After drying with a nitrogen blower, 500 μL of methanol was added for extraction. The resulting supernatant was analyzed by UPLC-QQQ-MS/MS to determine the concentration of the corresponding products.

Polarity protein expression in the kidney OoC (ezrin/integrinβ). HK-2 cell spheroids (perfusion for 3 days) were rinsed three times with PBS and then treated with 4% paraformaldehyde and 0.5% Triton X-100 for 30 minutes each, with three PBS rinses following each treatment. They were then blocked with 3% BSA for 1 hour. The primary antibody for integrinβ1 was diluted 1 : 50 in 1% BSA and incubated with the spheroids at room temperature for 2 hours. The primary antibody for ezrin was also diluted 1 : 50 in 1% BSA and incubated with the spheroids at 4 °C overnight. After three PBS rinses, a mixture of the two secondary antibodies—Alexa Fluor 488-conjugated goat anti-rabbit IgG (H + L) diluted 1 : 250 and Alexa Fluor 594-conjugated goat anti-mouse IgG (H + L) diluted 1 : 100—was incubated with the spheroids at room temperature for 1.5 hours. The secondary antibodies were then removed, and the spheroids were incubated with PBS-diluted Hoechst (1 : 100) at room temperature, protected from light, for 10 minutes. After a final wash, the spheroids were aspirated with a Pasteur pipette, transferred to a slide, and sealed with 50% glycerin. Images were acquired using an ultra-high resolution laser confocal microscope.¹⁸ For comparison, HK-2 cells cultured in 2D (i.e., cultured in a monolayer in well plates) underwent the same treatment and observation.

Transport and excretion capabilities of the kidney OoC. To assess Na⁺/K⁺ pump activity, HK-2 cell spheroids (perfused for 4 days) were transferred to a 96-well black wall, clear bottom plate and rinsed three times with PBS. A series of concentrations of ouabain octahydrate in PBS were prepared as Na⁺/K⁺ pump inhibitors. Two hundred microliters of each ouabain dilution were added to the spheroids and co-incubated at 37 °C for 1 hour. The spheroids were then rinsed three times with PBS and incubated with a PBS solution containing 5 μM CoroNa™ Green AM and 10 μg mL⁻¹ Hoechst at 37 °C for 1 hour. After incubation, the spheroids were rinsed three times with PBS. Images were captured using a high-content cell imaging analysis system. To assess the excretory capacity for PAH (4-aminohippuric acid), the perfusion system was supplied with DMEM containing 0.3 mg mL⁻¹ PAH. After 48 hours, the circulating fluid (outer, representing biomimetic blood) and the fluid within the cups (inner, representing biomimetic urine) were collected to measure PAH content. Excretory capacity was evaluated using the inner/outer (I/O) ratio, where higher values indicate stronger excretion. To assess the reabsorption capacity for glucose, the system was supplied with PBS containing 10% FBS as the circulating fluid, while 150 μL of pure serum was placed inside the MIMICups. After 24 hours, the outer and inner fluids were collected to measure glucose concentration. Reabsorption capacity was evaluated using the I/O ratio, where lower values indicate stronger reabsorption. Cell-free MIMICups and MIMICups with 2D-cultured HK-2 cells served as controls in the PAH and glucose experiments.

2.5 Computational toxicology prediction of TPL and its metabolites

Structures of TPL metabolites. Using liquid chromatography-tandem mass spectrometry (LC-MS/MS) and nuclear magnetic resonance (NMR), Liu et al. identified a series of major triptolide (TPL) metabolites in rats, which were classified into three types: hydroxylated TPL, hydroxyl-sulphates, and glutathione adducts.²⁵ For our computational toxicology analysis, we selected six structurally defined metabolites representing these three categories. Among them, TPL_M1-M4 were identified as abundant mono-hydroxylated TPL metabolites at different positions. TPL_M22 was a sulphate of a mono-hydroxy TPL metabolite, and TPL_GSH was the glutathione adduct of TPL. Furthermore, to validate our model, the biomimetic blood from the TPL-supplemented HLKOC system was qualitatively analyzed by LC-MS/MS, with a specific focus on detecting these six metabolites.²⁵ The TPL addition and biomimetic blood sampling procedures are described below.

Computational toxicology prediction. The potential metabolic characteristics, toxicity endpoints, and targets of TPL and its six metabolites were then compared. The pharmacokinetic characteristics and toxic endpoints were predicted using ADMETlab 3.0.³⁵

2.6 Molecular docking

Target selection. To identify potential targets associated with nephrotoxicity, we searched several databases, including OMIM, Open Targets, and GeneCards, using nephrotoxicity-related keywords. This approach was necessary as the SwissTargetPrediction tool was unable to forecast targets for these compounds. Specifically, targets from GeneCards with a relevance score >20 were selected using the keywords “nephrotoxicity” and “drug-induced kidney injury”. The OMIM database was searched using “nephrotoxicity”, “acute kidney injury”, and “kidney disease”, while the Open Targets database was queried with “nephrotoxicity” and “proximal renal tubular acidosis”. Additionally, kidney-specific proteins from the Human Protein Atlas were extracted to supplement the target set.

Docking and affinity comparison. Molecular docking was employed to analyze the intermolecular interactions between the proteins in the target set and TPL and its metabolites. The crystal structures of the proteins were downloaded from the RCSB Protein Data Bank (PDB). For proteins lacking experimentally verified structures, predictions were performed using Alphafold3.³⁶ Water molecules and original ligands were removed from the protein structures using PyMOL. The proteins were then prepared for docking using AutoDock v4.2.6 for hydrogenation, charge calculation, and the addition of non-polar hydrogens. Molecular docking was subsequently performed using AutoDock Vina v1.1.2. Based on the binding affinity, the targets were categorized into two groups: high-affinity targets for TPL and high-affinity targets for TPL metabolites. The latter group was further subdivided into two subgroups: targets with stronger affinity for any metabolite than for TPL (“Any”), and targets with stronger affinity for all metabolites than for TPL (“All”). Gene Ontology (GO) enrichment analysis was conducted for each category using the clusterProfiler package in R v4.3.1.³⁷

2.7 Organ toxicity evaluation and mechanism exploration using HLKOC and kidney OoC

Biochemical analysis. The experiment design for renal toxicity evaluation using the HLKOC and kidney OoC is shown in Fig. 4A. On day 8, the circulating fluid (biomimetic blood) and the fluid within the kidney MIMICup (biomimetic urine) were collected. For biomimetic blood, the concentrations of urea (UREA), creatine/creatinine (CREA), total protein (TP), and albumin (ALB), as well as the activity of N-acetyl-β-D-glucosaminidase (NAG), were measured using an automatic biochemical analyzer (Canon). For biomimetic urine, the concentrations of UREA, CREA, TP, microalbumin (MALB), and NAG activity were also determined. The expression of kidney injury molecule-1 (Kim-1) in biomimetic blood and urine samples was quantified using a Kim-1 ELISA kit and a microplate reader. Inter-group differences were tested using one-way ANOVA followed by an LSD post hoc test. All indicators had three replicates per group, except for the urinary Kim-1, which was combined into two samples due to insufficient volume.

Metabolomic analysis of biomimetic blood. After freeze-drying, the biomimetic blood underwent ultrasonic extraction with methanol. The extract was filtered through a 0.22 μm membrane, and the resulting supernatant was analyzed by UPLC-Q-TOF-MS/MS (Waters, Xevo G2) for untargeted metabolomics. Raw MS data were processed using Progenesis QI (v3.0) for peak alignment, extraction, and relative quantification. Compounds were identified using the Human Metabolome Database (HMDB). Inter-group differences in metabolic profiles were visualized by unsupervised principal component analysis (PCA) using the vegan package in R. metabolites were clustered based on abundance patterns using the k-means method.³⁸ The optimal number of clusters was determined with the factoextra package, and clustering was performed and visualized using the Mfuzz package. KEGG enrichment analysis was conducted using the clusterProfiler package. A t-test was employed to compare metabolite abundance between the control and TPL groups in the kidney OoC and HLKOC systems, respectively. Metabolites with |log 2FC| > 1 and P < 0.01 were identified as differential abundant metabolites. Each group contained three biological replicates.

Transcriptomic analysis of HK-2 cell spheroids. Total RNA was extracted from HK-2 cell spheroids using the PureLink RNA Micro Scale Kit (Thermo). RNA quality was assessed using Agilent 2100 Bioanalyzer. After mRNA enrichment using mRNA capture beads (Vazyme), the mRNA was reverse transcribed into cDNA. The cDNA library was then constructed according to the manufacturer's instructions (Illumina). Raw sequencing data underwent quality control and filtering using fastp to obtain clean reads.³⁹ The hisat2 and featureCounts pipeline was used for genome alignment (GRCh38) and transcript quantification. PCA was used to visualize differences in transcriptional profiles among samples. Differential expression genes (DEGs) were identified using the DESeq2 package, with thresholds of P < 0.05 and |log 2FC| > 1. GO enrichment analysis of DEGs was performed using the clusterProfiler package. Highly confident interactions among DEGs were extracted from the STRING database (combined score >0.7). Network analysis was conducted using the igraph package with the fast-greedy algorithm for module identification, and the network was visualized using Gephi.⁴⁰

Expression quantitative analysis of target proteins and genes. The transcriptional and protein expression levels of multiple metabolic enzyme-encoding genes (CYP1A2, CYP2E1, CYP2C9, and CYP3A1) in PCLS were quantified by RT-qPCR and Western blot, respectively. For RT-qPCR, the Actb gene (encoding β-actin) was used as the internal reference, and primer sequences were synthesized by Sangon Biotech (Table S1). The 2^−▲▲Ct method was used to quantify relative gene expression.⁴¹ For Western blot, GAPDH (36 kDa) served as the loading control, and all antibodies were purchased from Thermo Fisher Scientific. Protein expression levels were quantified by measuring grayscale values with ImageJ.⁴² For HK-2 cell spheroids, the expression of GLUT1 and CUBN were quantified by immunofluorescent staining and visualized with an ultra-high resolution laser confocal microscope as described above. Five images were acquired from the center of each spheroid, and the relative fluorescence intensity was quantified using ImageJ. Inter-group differences were tested using one-way ANOVA followed by an LSD post hoc test.

3. Results

3.1 HLKOC device fabrication and endothelial barrier optimization

The HLKOC device comprises a culture medium reservoir, a perfusion system (peristaltic pump), scalable flow channel plates, and insertable biomimetic cups (MIMICups) (Fig. 1A, B, and S1). The assembly of MIMICups containing endothelial barriers and precision-cut liver slices (PCLS) with the flow channel plate is defined as the liver chip. Similarly, the assembly of a MIMICup containing an endothelial barrier and HK-2 cell spheroids with the flow channel plate is defined as the kidney chip (Fig. 1A–D). The liver chip features a 24-well plate, with wells arranged in groups of four per channel to enhance metabolic flux. The kidney chip consists of a single-well plate, with each channel connected to two such plates to mimic bilateral kidneys (Fig. 1A and S1). The underlying structure of the MIMICups is uniform for both liver and kidney applications, employing a three-layered configuration to replicate the endothelial barrier (Fig. 1C–D). An 8 μm polyethylene terephthalate (PET) membrane is first affixed to the bottom of the cup to provide a physical substrate for cell adhesion. Its pore size facilitates the exchange of biomolecules between the cup's interior and exterior, as demonstrated by a dextran permeability assay (Fig. 1E and F). The underside of the PET membrane was coated with 200 μL of 100 μg mL⁻¹ fibronectin to serve as an adhesive layer, followed by the seeding of endothelial cells, which form a dense endothelial barrier upon culture (Fig. 1C and D). Within the liver MIMICup, 180 μL of complete DMEM is first introduced, followed by the careful placement of 6 mm diameter PCLS (Fig. 1C). In the kidney MIMICup, a 50 μL layer of 70% Matrigel is first applied. A bioprinter is then used for the precise deposition of HK-2 cells, with four printing locations per cup. After bioprinting, 150 μL of complete DMEM supplemented with 10% Matrigel is added; the resulting liquid inside the cup mimics biological urine (biomimetic urine) (Fig. 1D). The circulatory system is powered by a peristaltic pump, with a 15 mL centrifuge tube serving as a reservoir for 3 mL of complete DMEM (Fig. 1A, B, and S1). The culture medium (biomimetic blood) circulates through microchannels (1 mm in height) at the bottom of the flow channel plates, forming a perfusion loop that supplies nutrients to and removes metabolic waste from the biological elements within the MIMICups. The podiform structure of the channel ensures uniform shear stress across the cup base, as depicted in the shear stress heatmap in Fig. 1B.

To establish a robust endothelial barrier, crucial for simulating biological activities and organ crosstalk, we first evaluated endothelial cell models (EA.hy926 and HUVECs) and flow rates. When cultured to confluence in static 6-well plates, both cell types exhibited high expression of the tight junction protein ZO-1 (Fig. S2). However, when seeded on the bottom of PET membranes and subjected to perfusion with culture medium (DMEM with 10% FBS) at various flow rates (15–750 μL min⁻¹), only EA.hy926 cells maintained a robust endothelia barrier at flow rates of 50–150 μL min⁻¹ (0.055–0.165 dyne per cm²), whereas HUVECs failed to reach confluence (Fig. S3). Therefore, the EA.hy926 cell line was selected for further optimization of serum concentration and perfusion rate. H&E staining demonstrated that at a flow rate of 150 μL min⁻¹ in the presence of 20% FBS, EA. hy926 cells sustained dense proliferation through day five (Fig. 1G). Immunofluorescent labeling for ZO-1 further confirmed that under these conditions, the EA. hy926 cells formed effective tight junctions, thereby establishing endothelial barrier integrity (Fig. 1H).

3.2 The HLKOC components possess metabolic, transport, and excretory functions

The structural integrity and metabolic enzyme activity of the PCLS underpin the metabolic capacity of HLKOC. H&E staining revealed that PCLS maintained superior structural integrity after 24 hours of dynamic culture, with no significant alterations in cellular morphology (Fig. 2A). we assessed the activities of CYP3A1/2 (corresponding to human CYP3A4) and CYP2E1 (corresponding to human CYP2E1) by co-incubating PCLS with enzyme-specific substrates and quantifying the resulting metabolites via UPLC-MS/MS. The former is a pivotal metabolic enzyme for TPL. The concentration of dehydronifedipine, a metabolite of nifedipine, rapidly increased within 12 hours before decreasing at 24 hours, indicating robust CYP3A4 activity. A similar trend was observed for 6-hydroxychlorzoxazone (a metabolite of chlorzoxazone), signifying considerable CYP2E1 activity (Fig. 2B).

Fig. 2 Validation of the structure and functions of bionic liver and kidney in HLKOC. A. H&E staining indicated that precision-cut liver slices (PCLS) maintained intact structure after 24 hours of perfusion culture. B. Time course of the metabolism of specific substrates by CYPs in PCLS. Each substrate had two biological replicates, with each replicate containing four PCLS. The value for each biological replicate was represented by the mean of two technical replicates. C. Bright-field images of bio-printed HK-2 cell spheroids at different culture days. D. Calcein-AM/PI staining indicated that the HK-2 cell spheroids maintained good viability during up to 7 days of culture. E. Immunofluorescence staining for polarity proteins indicated the polarization of HK-2 cell spheroids (red: integrin β; green: ezrin). The images were captured using a confocal laser scanning microscope. F. High-content microscopy demonstrated Na⁺ retention within HK-2 cell spheroids at various concentrations of ouabain, confirming the Na⁺ transport capability of cell spheroids. Evaluation of the renal OoC's capacity for PAH excretion (G) and glucose reabsorption (H). For each treatment, three biological replicates (each consisting of two MIMICups) were analyzed. The value for each biological replicate was calculated as the mean of the two MIMICups. Nocell: no cells in the MIMICup; 2D: paved HK-2 cells in the MIMICup; 3D: HK-2 cell spheroids in the MIMICup. One-way ANOVA. *, P < 0.05; ***, P < 0.001.

For the kidney OoC, HK-2 cells condensed into spheroids on Matrigel the day after bioprinting and continued to grow and densify from days 2 to 7 (Fig. 2C). Calcein AM-PI staining indicated that the HK-2 cell spheroids maintained high viability throughout the 7-day culture period (Fig. 2D). Immunofluorescent staining revealed distinct polarized protein expression within the spheroids, with ezrin as an apical marker and integrinβ as a basal marker (Fig. 2E). In contrast, HK-2 cells in 2D culture expressed only ezrin, with no detectable integrinβ (Fig. S4). These results indicate the establishment of polarity in the 3D culture of HK-2 spheroids. We further assessed the transport capabilities of the spheroids. Na⁺ transport function was evaluated using the Na⁺ channel inhibitor ouabain and the membrane-permeant fluorescent indicator Coro-Na green. The observation of increased intracellular Na⁺ retention with higher inhibitor concentrations confirmed the presence of functional Na⁺ transport (Fig. 2F). The excretory capacity for 4-aminohippuric acid (PAH) was assessed by adding PAH to the biomimetic blood and measuring the inner/outer (I/O) concentration ratio. The I/O values for both 2D- and 3D-cultured HK-2 cells were significantly higher than the cell-free control, demonstrating effective PAH excretion (Fig. 2G). Finally, glucose reabsorption from biomimetic urine (in-cup) to biomimetic blood (flow channel) was evaluated. After 24 hours, the I/O value for HK-2 spheroids was significantly lower than for cell-free and 2D cultures, indicating the spheroids' capacity to reabsorb glucose into biomimetic blood (Fig. 2H). These findings demonstrate that the liver and kidney chips possess robust metabolic, transport, and excretory functions, rendering the HLKOC an ideal platform for investigating the impact of hepatic drug metabolism on renal function.

3.3 TPL and metabolites exhibited potentially distinct pharmacokinetic characteristics and renal toxicity targets

Prior to investigating TPL toxicity with the HLKOC, we employed computational toxicology and molecular docking to assess the pharmacokinetic properties, toxicity endpoints, and potential renal toxicity targets of TPL and its metabolites. We selected six structurally characterized rat hepatic metabolites of TPL, which were also detected in the biomimetic blood of the TPL-supplemented HLKOC system by LC-MS/MS (Fig. 3A and S5). Pharmacokinetic predictions indicated that the five hydroxylated metabolites possess a favorable lipophilicity-hydrophilicity balance (indicated by log D and log P values). The T_1/2 and CL_plasma data suggested that the clearance times of the metabolites were approximately equal to or longer than that of TPL. Additionally, TPL_M3/M4/GSH exhibited high microsomal stability and were less prone to further metabolism (Fig. 3B). Toxicity endpoint predictions indicated a high probability of renal toxicity for both TPL and its metabolites. With the exception of TPL_GSH, the remaining metabolites also showed a high probability of hematotoxicity, genotoxicity, ototoxicity, and neurotoxicity (Fig. 3C).

Fig. 3 Exploring the toxicological characteristics and potential target set of triptolide and its liver metabolites by computational toxicology and molecular docking. A. The structure of triptolide (TPL) and its hepatic metabolites identified by LC-MS/MS and NMR.²⁵ Pharmacokinetic (B) and toxicity endpoint (C) predictions of TPL and its metabolites by ADMETLab 3.0. D. Venn diagram showed the overlap of potential nephrotoxicity targets from different databases and kidney-specific proteins. These proteins were used for molecular docking with TPL and its metabolites. E. The heatmap displayed the docking score (free energy score) for proteins interacting with TPL and its metabolites. Pink indicates low free energy scores (high affinity), while grey represents high free energy scores (low affinity). F. GO enrichment analysis for protein sets with high affinity with TPL and its metabolites, respectively. Blue font indicated that the term was enriched specifically by TPL high-affinity proteins, while yellow font indicated that the term was enriched specifically by high-affinity proteins for TPL metabolites. P values (hypergeometric test) were adjusted using FDR.

As definitive renal toxicity targets for TPL and its metabolites were not established, and the SwissTargetPrediction tool was unable to forecast them, we compiled a list of database-derived targets associated with renal toxicity and injury. This was achieved by searching databases including OMIM, Open Targets, and GeneCards with keywords such as “nephrotoxicity” and “drug-induced kidney injury”. We further integrated kidney-specific targets from The Human Protein Atlas to create a compendium of 547 potential targets (Fig. 3D and Table S2). Notably, both TPL and its metabolites exhibited high affinity for the majority of these targets (Fig. 3E). Based on binding affinity comparisons, targets were categorized into two groups: high-affinity targets for TPL and high-affinity targets for its metabolites. The latter group was further subdivided into targets with stronger affinity for any metabolite than for TPL (“Any”) and those with stronger affinity for all metabolites than for TPL (“All”) (Fig. 3E). For instance, MET, CUBN, SLC12A3, and PPARGC1A exhibited higher affinity for all TPL metabolites compared to TPL itself (Fig. 3E). GO enrichment analysis of these genes revealed distinct pathways associated with TPL and its metabolites. Pathways specific to high-affinity proteins for TPL included potassium ion import across plasma membrane, monoatomic anion homeostasis, chloride ion homeostasis, basolateral plasma membrane, and active transmembrane transporter activity. In contrast, pathways specific to high-affinity proteins for TPL metabolites included regulation of oxidoreductase activity, response to reactive oxygen species, nephron epithelium development, nephron development, and cellular response to chemical stress. These findings suggest that TPL and its metabolites possess distinct pharmacokinetic characteristics and target profiles, providing a foundation for subsequent HLKOC-based research into the renal toxicity and mechanisms of TPL.

3.4 TPL induced significant urea elevation in the biomimetic blood of HLKOC but not in single kidney chips

We then employed the HLKOC to evaluate the nephrotoxicity and underlying mechanisms of TPL following hepatic metabolism, using a single kidney chip (without a connected liver chip) as a control. The administered concentration of TPL was 25 ng ml⁻¹, which resulted in a 30% inhibition rate for HK-2 cells (Fig. S6). In the HLKOC system, HK-2 cell spheroids were first bioprinted and dynamically cultured for four days. Subsequently, the liver chip was connected, and the culture medium was replaced with medium containing either TPL or vehicle. For the single kidney chip, the medium was replaced on day four without connecting the liver chip. Samples were collected 24 hours after the medium replacement for analysis (Fig. 4A).

Fig. 4 TPL toxicity evaluation experiments based on HLKOC and single-kidney OoCs as well as biochemical analysis. A. Schematic diagram of the experimental procedure. K: single-kidney OoC; LK: HLKOC. Comparison of biochemical indicators in biomimetic blood (B) and biomimetic urine (C) among four groups. K-Control: single-kidney OoC without TPL treatment; K-TPL: single-kidney OoC with TPL treatment; LK-Control: HLKOC without TPL treatment; LK-TPL: HLKOC with TPL treatment. Each group consisted of three biological replicates (i.e., three HLKOC channels, each containing four liver chips and two kidney chips). The only exception was Kim-1 in the biomimetic urine, which was pooled into two samples (two biological replicates) due to insufficient sample volume. One-way ANOVA followed by LSD post hoc test. *, P < 0.05; **, P < 0.01; ***, P < 0.001. UREA: urea; CREA: creatine/creatinine; Kim-1: kidney injury molecule 1; NAG: the activity of N-acetyl-β-D-glucosaminidase; TP: total protein; ALB: albumin; MALB: microalbumin. (D) RT-qPCR quantification of the four key hepatic CYP enzyme-encoding genes. Actb was used as the internal reference. All P values for the comparisons were greater than 0.05. (E) Quantitative results of Western blotting for the four key hepatic CYP enzymes. GADPH was used as the internal reference. All P values for the comparisons were greater than 0.05 (one-way ANOVA). For RT-qPCR and Western blotting, each group consisted of four biological replicates, and each replicate had its corresponding internal reference.

Multiple biochemical indicators and injury biomarkers were measured in the biomimetic blood (circulating fluid) and biomimetic urine (fluid within the kidney chip wells), such as urea (UREA), creatine/creatinine (CREA), Kim-1, NAG, total protein (TP), and albumin (ALB/MALB). Notably, compared to the single kidney OoC, the biomimetic blood in the HLKOC exhibited higher levels of urea, total protein, and albumin, with the difference in urea being statistically significant (P < 0.05) (Fig. 4B). Urea, the primary end product of protein metabolism in humans, is synthesized by the liver and excreted by the kidneys. Similarly, albumin is synthesized exclusively by hepatocytes in the liver. These results confirmed the robust synthetic function of the liver chip within the HLKOC. TPL treatment significantly elevated urea levels in the biomimetic blood of the HLKOC but not in that of the single kidney chip, suggesting that TPL may impair the urea excretion function of the kidney within the HLKOC system (Fig. 4B and C). Furthermore, there were no significant differences in creatine/creatinine levels in the biomimetic blood, which is consistent with the absence of a glomerular filtration function in our OoC systems. In the biomimetic urine, creatine/creatinine levels were significantly higher in the control group of the HLKOC than in the single kidney chip (P < 0.001), consistent with the liver being the primary site for creatine synthesis (Fig. 4C). However, TPL significantly reduced creatine/creatinine levels in the biomimetic urine of the HLKOC, suggesting that TPL may also inhibit the excretion of creatine/creatinine by the kidney chip (P < 0.001) (Fig. 4C). The levels of Kim-1 and NAG in the biomimetic urine showed no statistically significant changes among the groups.

We further explored the effect of TPL addition on the expression of metabolism-related genes in the liver chip. Quantification of four key hepatic CYPs (CYP1A2, CYP2C9, CYP2E1, CYP3A1) at both the transcriptional and protein levels indicated that TPL did not impair the metabolic potential of the PCLS (Fig. 4D and E). In summary, these results indicate that the incorporation of the liver OoC enhances the biomimetic capability of the HLKOC compared to the single kidney OoC, and that TPL may impair the secretion function of metabolic waste by the HK-2 cell spheroids within the HLKOC.

3.5 Metabolomic analysis of biomimetic blood reveals distinct response patterns to TPL in HLKOC and single kidney chips

We performed comprehensive metabolic profiling of the biomimetic blood using high-resolution mass spectrometry and the HMDB database to investigate TPL-induced metabolic alterations in the OoC systems (Table S3). In total, 6247 compounds were identified in the biomimetic blood samples (n = 12). PCA revealed significant separation among the four experimental groups (Fig. 5A). K-means clustering categorized all compounds into six clusters (C1–C6) based on their abundance patterns. Cluster C3 was enriched in the TPL-free single kidney chip group, C1 in the TPL-treated single kidney chip group, C2 in the TPL-free HLKOC group, and C5 and C6 in the TPL-treated HLKOC group (Fig. 5B). KEGG enrichment analysis of each cluster identified significantly overrepresented metabolic pathways. For instance, metabolites in C1 and C3 were associated with sphingolipid metabolism, sphingolipid signaling pathway, and efferocytosis. Pathways specifically enriched in C2 were related to amino sugar and nucleotide sugar metabolism, whereas pathways enriched in C5 and C6 included linoleic acid metabolism, arachidonic acid metabolism, and fatty acid biosynthesis (Fig. 5C).

Fig. 5 Metabolomic analysis of biomimetic blood from different systems and treatments. A. Principal-component analysis (PCA) showing distinct metabolic profiles among the four experimental groups. B. K-means clustering of all metabolites into six groups (C1–C6). The optimal number of clusters was determined using gap statistics. C. KEGG enrichment analysis of metabolites in each cluster, showing significantly overrepresented pathways (P < 0.05). Text color corresponds to the P value. Volcano plots of differential metabolites between TPL-treated and control groups for the single kidney OoC (D) and HLKOC (E), identified by Wilcoxon rank-sum test. Metabolites with |log 2FC| > 1 and P < 0.01 were considered significant. The text below each volcano plot presents the results of KEGG enrichment analysis (P < 0.05, hypergeometric test), where text color corresponds to the P value.

We then identified differential metabolites in the biomimetic blood of the HLKOC and single kidney chip following TPL treatment (P < 0.01, |log 2FC| > 1). The number of differential metabolites in the HLKOC system (n = 435) was significantly greater than that in the single kidney chip (n = 187). In the single kidney chip, metabolites upregulated by TPL were primarily associated with tyrosine metabolism. In contrast, in the HLKOC, upregulated metabolites were linked to processes such as linoleic acid metabolism, drug metabolism, and tyrosine metabolism (Fig. 5D and E). Furthermore, TPL treatment decreased the levels of metabolites associated with pathways like retrograde endocannabinoid signaling and diabetic cardiomyopathy in the biomimetic blood of the HLKOC (Fig. 5E). These results indicate that TPL treatment induced more profound metabonomic changes in the HLKOC system compared to the single kidney chip.

3.6 Transcriptomic analysis reveals the molecular network of HK-2 spheroids in response to TPL within the HLKOC

PCA analysis indicated significant transcriptional alterations in HK-2 spheroids within both the HLKOC and the single kidney chip following TPL treatment (Fig. 6A). Significant differences in gene expression were also observed between HK-2 spheroids in the HLKOC and the single kidney chip under TPL-free conditions (Fig. S7). A total of 341 genes were significantly differentially expressed (P < 0.05, |log 2FC| > 1) between the control groups of the HLKOC and the single kidney chip, with associations to functions such as mitochondrial intermembrane space and CD40 receptor complex (Fig. S7). This indicates that connecting to the liver chip significantly alters the baseline gene expression of HK-2 spheroids.

Fig. 6 Transcriptomic analysis of HK-2 cell spheroids under different systems and treatments. A. Principal-component analysis (PCA) showing differences in transcriptional profiles among the four groups. B. Volcano diagram illustrating the differential expression genes (DEGs) between TPL-treated and control groups in the HLKOC and single kidney OoC systems. DEGs were identified using the DESeq2 package with thresholds of |Log2FC| > 1 and P < 0.05. C. GO enrichment analysis of DEG subgroups. All DEGs were separated into three categories: DEGs identified in both systems (overlap), DEGs specific to the HLKOC, and DEGs specific to the single kidney OoC. D. Protein–protein interaction (PPI) networks of the DEGs. Nodes represent proteins and edges represent high-confidence interactions. The left network is colored by DEG subgroups, while the right network is colored by modules identified using the fast-greedy algorithm. Significantly overrepresented GO terms for proteins in each module are displayed. The gene with the highest degree within each module is shown in brackets.

Regarding the effects of TPL treatment, 623 genes were significantly differentially expressed in response to TPL in the HLKOC system, whereas only 266 were identified in the single kidney chip (Fig. 6B, Tables S4 and S5). There was an overlap of only 59 differentially expressed genes between the two systems. GO enrichment analysis revealed that these 59 overlapping genes were primarily associated with functions like elastic fiber assembly and nucleotide-sugar transmembrane transport. HLKOC-specific differentially expressed genes were associated with functions such as vesicle tethering and double-strand break repair, while kidney chip-specific genes were related to processes including neuron apoptotic process and intrinsic apoptotic signaling. A protein–protein interaction (PPI) network of these differentially expressed genes was constructed (Fig. 6D). Among the HLKOC-specific genes, STAT1 had the highest connectivity, while TP53 was the most highly connected node among the kidney chip-specific genes, and IRF7 had the highest connectivity among the overlapping genes. The HLKOC- and kidney chip-specific differentially expressed genes did not form separate modules (fast-greedy algorithm) but were instead evenly distributed across the eight main modules of the PPI network (Fig. 6D).

An integrated analysis of the molecular docking results and the transcriptomic data revealed that among the proteins with higher affinity for all TPL metabolites (“All”), the expression of the encoding genes for cubilin (CUBN) was significantly downregulated in the HLKOC system upon TPL treatment. This identifies CUBN as a potential direct target through which TPL metabolites may affect the function of renal HK-2 spheroids (Fig. S8, Tables S2 and S4). Furthermore, the PPI network indicated that proteins with high affinity for TPL metabolites had more interactions with the differentially expressed genes in the HLKOC compared to proteins with high affinity for TPL itself. This suggests a strong concordance between the molecular docking predictions and the transcriptome data (Fig. S8). Overall, these results provide insights into the TPL-induced gene expression changes in HK-2 spheroids within the HLKOC system and identify potential toxicological targets of TPL metabolites.

We further analyzed the expression of two transport-related proteins, cubilin and GLUT1, which were significantly downregulated in the HLKOC system following TPL treatment, using immunofluorescence staining. The GLUT1 was included as this protein is responsible for basal glucose uptake by proximal tubule, and metabolomic analysis identified changes in pathways related to glucose metabolism in the HLKOC system following TPL treatment. The results of relative fluorescence quantification indicated that the two proteins exhibited consistent trends in expression changes. In both the kidney OoC and HLKOC systems, TPL treatment reduced the expression of cubilin and GLUT1 significantly (P < 0.001). However, the expression levels of both proteins in the HLKOC system were lower than those in the kidney OoC system, especially after TPL treatment (Fig. 7A–C).

Fig. 7 Relative quantification of protein expression in HK-2 cell spheroids based on immunofluorescence staining. The immunofluorescence staining images of cubilin (A) and GLUT1 (B) obtained by laser scanning confocal microscope. Blue: DAPI; red: target protein. Relative quantification results of the expression of Cubilin (C) and GLUT1 (D) based on grayscale. The relative fluorescence intensity of five images acquired from the center of each spheroid were displayed and compared. One-way ANOVA followed by LSD post hoc test. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

4. Discussion

The development of organ-on-a-Chip (OoC) technology has significantly advanced preclinical drug toxicity investigation. However, limitations persist in simulating multi-organ physiology, particularly the interplay between the liver and other organs, and in generating sufficient biomaterials for clinical biomarker monitoring.⁴³ In this study, we constructed a HLKOC device with high fabrication and biological throughput, featuring injection-molding-compatible and scalable flow channel plates and insertable MIMICups, alongside a peristaltic pump and a medium reservoir. By integrating precision-cut liver slices (the liver chip) and a bioprinting-based 3D HK-2 cell spheroids (the kidney chip), this platform enables the investigation of drug-induced nephrotoxicity within the context of liver-kidney crosstalk. The system preserves key hepatic metabolic functions and renal transport capabilities, and facilitates metabolic communication between the two organs. We further employed the model drug triptolide (TPL) to demonstrate the efficacy of integrating the HLKOC with multi-disciplinary techniques—including computational toxicology, large-scale molecular docking, biochemical analysis, metabolomics, and transcriptomics—for investigating drug-induced kidney toxicity and its underlying mechanisms, using a single-kidney OoC as a control (Fig. 8). Initially, computational toxicology and molecular docking revealed that TPL and its hepatic metabolites possess distinct pharmacokinetic profiles and potential target sets. Biochemical analysis highlighted the contribution of the liver chip to urea biosynthesis and uncovered a significant TPL-induced elevation in urea level in the biomimetic blood, an effect detectable only in the HLKOC system. Metabolomic and transcriptomic analyses further revealed distinct systemic responses to TPL treatment in the HLKOC versus the kidney OoC. In summary, these findings underscore the critical role of hepato-renal chemical communication in detecting drug-induced nephrotoxicity and elucidating its mechanisms, positioning the HLKOC as a powerful tool for the in-depth investigation of drug nephrotoxicity.

Fig. 8 Paradigm of the present study.

Previous pioneering work has yielded a variety of multi-organ OoC systems, yet they are often unsuitable for mass production or lack sufficient biomass for downstream analyses like RNA-Seq and protein expression quantification.²⁰ For example, Edington and colleagues⁴⁴ developed a scalable multi-organ chip to investigate pharmacokinetic properties, but its design is exceedingly complex, requiring an external 36-channel controller that complicates large-scale manufacturing and application, and its individual chips have limited biomass loading capacity. In a landmark study, Ronaldson-Bouchard et al.⁴⁵ engineered an integrated multi-organ chip to study interactions among the heart, liver, bone, and skin. However, the main body of this device, made of polysulfone, required high-precision, low-throughput, and costly CNC machining to achieve a reliable seal between components (including O-ring and glass slide)—a process less efficient than mode-based injecting molding. Furthermore, the device's multi-part assembly, which includes inserting an endothelial cell-containing nylon mesh, poses a risk of contamination. In contrast, the core components of the HLKOC—the plug-and-play MIMICups and scalable flow channel plates—are engineered for injection molding. This design incorporates features like a double O-ring to compensate for manufacturing tolerances and microchannels defined by a PMMA base. Moreover, inspired by Ronaldson-Bouchard et al.,⁴⁵ the modular design of the MIMICups significantly enhances their versatility, enabling the accommodation of diverse biological components tailored to specific experimental objectives. The integrated and wing-like design of the MIMICups also avoid complex assembly procedures that could introduce contamination.

A key advantage of OoC technology is the potential to construct microphysiological systems using solely human cells, thereby mitigating interspecies differences.¹¹ Given that the kidney is the target organ of interest and that renal transporters exhibit significant differences between humans and rodents,⁴⁶ we utilized human proximal tubule epithelial cells (HK-2) to construct the kidney chip. Although an immortalized cell line, HK-2 cells have been widely used in drug toxicity studies and have demonstrated good in vivo consistency.⁴⁷ We acknowledge, however, that the kidney possesses a wide variety of cell types, complex circulatory systems, and intricate cellular microenvironments. Consequently, an immortalized cell line such as this cannot fully replace in vivo studies. In the future, hiPSC-induced organoids with well-defined structures and differentiation may represent a superior alternative.⁴⁸ While species variations also exist in hepatic CYP450 enzymes,⁴⁹ this study prioritized the reliable reproduction of liver metabolic capacity to assess its influence on drug effects in a second organ (the kidney). PCLS are advantageous because they contain diverse liver cell types (hepatocytes, stellate, Kupffer cells) and maintain intact hepatic architecture (bile ducts, central veins, space of Disse), exhibiting significantly higher metabolic activity than HepG2 cells.⁹ Consequently, PCLS were employed as the culture substrate for the liver chip. The superiority of this choice was demonstrated by our biochemical analysis. Urea, total protein, and albumin in serum function as an indicator of hepatic biosynthetic capacity.⁵⁰ In this study, all three indicators exhibited elevated levels in the liver-kidney system compared to the single kidney system, suggesting that the liver chip (PCLS) possesses robust biosynthetic capacity and that its incorporation allows for better simulation of in vivo conditions. However, we also recognize that this hybrid species system is a compromise under current technical conditions. As organoid technology advances, human liver organoids with biomimetic profiles of metabolic and synthetic enzymes are poised to become crucial components for future liver-centric multi-organ OoC systems.

Serum urea levels are a traditional indicator for assessing kidney injury.⁵¹ Using triptolide (TPL), a model drug for liver and kidney injury, we observed that TPL significantly increased urea concentration in the biomimetic blood of the HLKOC. This effect may be attributed to TPL impairing the excretory function of proximal tubule epithelial cells, leading to urea accumulation. Notably, this TPL-induced elevation in urea was not observed in the single-kidney system, indicating that the liver-kidney co-culture demonstrates greater sensitivity than the single-organ system. Furthermore, in the HLKOC system, TPL treatment significantly decreased the creatine/creatinine levels in the biomimetic urine, further supporting the inhibitory effect of TPL on the excretory function of proximal tubule epithelial cells. This effect was also not significant in the single kidney OoC, which may be attributed to the lack of hepatic creatine/creatinine synthesis. These findings underscore the pressing need for advanced in vitro biomimetic models that possess high physiological relevance and sensitivity, faithfully recapitulating in vivo structural organization and inter-organ interactions, particularly for investigating drug metabolism, transport, and excretion. While the present study focused on the nephrotoxic effects of hepatic drug metabolites, the HLKOC device also shows promise for investigating the effects of drugs on hepatic functions within the context of liver-kidney crosstalk. For instance, the consistent expression of hepatic CYP enzyme-encoding genes between control and TPL-treatment groups indicates that TPL did not compromise the liver's metabolic capacity.

Many animal cells, including the HK-2 line, form a monolayer when adhering to a plastic dish, a morphology that differs from their in vivo structure.⁵² Studies on the cellular microenvironment indicate that cell–cell and cell-extracellular matrix (ECM) interactions are crucial factors in maintaining tissue morphology, structure, and physiological function.⁵³ ECM-based 3D spheroid culture techniques have been widely employed to construct in vitro physiological models for toxicological or pharmacological research. In this study, 3D-cultured HK-2 spheroids exhibited the expression of polarized proteins, in stark contrast to 2D cultures, consistent with previous findings on HK-2 three-dimensional cultures.¹⁸ This polarization endows HK-2 spheroids with enhanced excretory and reabsorptive capabilities, thereby better simulating in vivo physiology. Beyond the application of ECM matrices, the high-throughput and stable production of 3D spheroids has been greatly facilitated by bioprinting technology (Rahmani Dabbagh et al., 2022). The conventional construction of numerous 3D cultures is time-consuming, and variations in cell state during this process can affect spheroid consistency and experimental reproducibility. Nozzle-based bioprinting significantly reduces process time through pre-programming and rapid, targeted printing, thereby enhancing system stability, which is crucial for obtaining reliable system-level response data.⁵⁴ Furthermore, the shear stress stimulus introduced via microfluidics is another critical factor contributing to the enhanced physiological functionality of both the liver and kidney chips in this study. Previous studies have shown that fluid-induced stimulation promotes renal tubular epithelial cell polarization, primary cilium formation, and increased and stabilized liver metabolic biomarkers such as albumin, urea, and CYPs.⁵⁵

The response of organisms to exogenous stimuli is typically systemic and modular, and understanding these response pathways is key to developing biomarkers for early toxicity detection and detoxification strategies. This study found that, compared to the kidney OoC, the HLKOC system induced unique responses in the biomimetic blood metabolome and the transcriptome of HK-2 spheroids following TPL treatment. For instance, metabolomic analysis showed that TPL treatment induced the specific upregulation of linoleic acid metabolism in the HLKOC system. The metabolism of linoleic acid is associated with hepatic and renal CYP450s such as CYP1A2, CYP2E1, and CYP2J2.⁵⁶ Its metabolic products, including monoepoxides and corresponding diols, has been shown to be toxic to renal proximal tubules at pathologically relevant concentrations, a process related to mitochondrial dysfunction and cell death.^56,57 This result suggests that TPL exposure might lead to alterations in other metabolic pathways associated with nephrotoxicity. At the transcriptomic level, TPL-specific response genes in the HLKOC system are related to many vesicle-associated pathways. Extracellular vesicles (EVs) are important mediators of intercellular signal transduction; they contain RNAs, proteins, and lipids that can be transferred between cells.⁵⁸ When exposed to exogenous stimuli such as TPL metabolites, cells may reshape EV secretion and alter EV composition, thereby affecting their regulatory functions. These abnormal EVs could, in turn, cause physiological dysfunction in other cells, indirectly leading to toxicity.⁵⁹ This finding may indicate that EV-related mechanisms contribute to the nephrotoxicity of TPL in vivo. Furthermore, we found that the expression of cubilin (CUBN) and GLUT1, two important transport-related proteins, was significantly downregulated in both HLKOC and kidney OoC systems upon TPL treatment. Cubilin is a large, glycosylated extracellular protein that acts as an endocytic receptor, with its CUB domains serving as ligand-binding sites for a series of carrier proteins and drugs;⁶⁰ while GLUT1 is responsible for glucose uptake in renal tubules.⁶¹ The downregulation of these transport-related proteins may be a significant factor contributing to TPL-induced renal dysfunction, kidney injury, and overall systemic toxicity in humans.

Author contributions

Conceptualization: GZ, TL. Formal analysis: SL, YY, YFY, TL, GZ. Funding acquisition: TL, GZ. Methodology: SL, YY, GW, JL, ZY, YL, ZW. Project administration: TL. Resource: SL, YY. Software: SL, YY, GZ. Supervision: TL. Validation: SL, YY, YFY. Visualization: GZ, SL, YY. Writing – original draft: TL, GZ, SL, YY, GW, JL. Writing – reviewing & editing: TL, GZ, SL, YY, LZ.

Conflicts of interest

There are no conflicts to declare.

Data availability

The raw sequencing data reported in this paper has been deposited in the National Genomics Data Center (accession no. PRJCA039861) and can be obtained from https://ngdc.cncb.ac.cn/gsub/submit/bioproject/subPRO058635. Other data supporting this article have been included as part of the supplementary information (SI).

Supplementary information is available. See DOI: https://doi.org/10.1039/d5lc01051a.

Acknowledgements

This work was supported by Beijing Natural Science Foundation (7232301), National Natural Science Foundation of China (82574715), Scientific and Technological Innovation Project of China Academy of Chinese Medical Sciences (CI2023E002, CI2024E003), the Fundamental Research Funds for the Central Public Welfare Research Institutes (ZZ16-XRZ-072, ZZ17-YQ-025, ZXKT22052, ZZ14-YQ-025, ZXKT21009, and ZXKT22018), and the National Key Research and Development Program (2022YFC3501803). We thank the Shanghai AureFluidics Technology Co., Ltd. for their technical support in microfluidics, the GENE DENOVO Co., Ltd for their help in high-throughput RNA-Seq, and Professor Bingcheng Lin from Dalian Institute of Chemical Physics, Chinses Academy of Scineces, for his guidance and help during the revision of this manuscript.

References

1. S. Y. Chang, E. J. Weber, K. V. Ness, D. L. Eaton and E. J. Kelly, Clin. Pharmacol. Ther., 2016, 100, 464–478.
2. E. J. Weber, J. Himmelfarb and E. J. Kelly, Curr. Opin. Toxicol., 2017, 4, 16–21.
3. H. Mody, V. Ramakrishnan, M. Chaar, J. Lezeau, A. Rump, K. Taha, L. Lesko and S. Ait-Oudhia, Clin. Pharmacol. Drug Dev., 2020, 9, 896–909.
4. E. Kwiatkowska, L. Domanski, V. Dziedziejko, A. Kajdy, K. Stefanska and S. Kwiatkowski, Int. J. Mol. Sci., 2021, 22, 6109.
5. A. M. B. Amorim, L. F. Piochi, A. T. Gaspar, A. J. Preto, N. Rosario-Ferreira and I. S. Moreira, Chem. Res. Toxicol., 2024, 37, 827–849.
6. N. Wang, X. Li, J. Xiao, S. Liu and D. Cao, Drug Discovery Today, 2024, 29, 104195.
7. S. Ramm, P. Todorov, V. Chandrasekaran, A. Dohlman, M. B. Monteiro, M. Pavkovic, J. Muhlich, H. Shankaran, W. W. Chen, J. T. Mettetal and V. S. Vaidya, Toxicol. Sci., 2019, 169, 54–69.
8. I. Maschmeyer, A. K. Lorenz, K. Schimek, T. Hasenberg, A. P. Ramme, J. Hubner, M. Lindner, C. Drewell, S. Bauer, A. Thomas, N. S. Sambo, F. Sonntag, R. Lauster and U. Marx, Lab Chip, 2015, 15, 2688–2699.
9. E. F. Brandon, C. D. Raap, I. Meijerman, J. H. Beijnen and J. H. Schellens, Toxicol. Appl. Pharmacol., 2003, 189, 233–246.
10. Q. Huang, T. Yang, Y. Song, W. Sun, J. Xu, Y. Cheng, R. Yin, L. Zhu, M. Zhang, L. Ma, H. Li and H. Zhang, Lab Chip, 2024, 24, 1715–1726.
11. H. Kimura, M. Nishikawa, N. Kutsuzawa, F. Tokito, T. Kobayashi, D. A. Kurniawan, H. Shioda, W. Cao, K. Shinha, H. Nakamura, K. Doi and Y. Sakai, Drug Metab. Pharmacokinet., 2025, 60, 101046.
12. L. A. Low, C. Mummery, B. R. Berridge, C. P. Austin and D. A. Tagle, Nat. Rev. Drug Discovery, 2021, 20, 345–361.
13. J. Theobald, M. A. Abu El Maaty, N. Kusterer, B. Wetterauer, M. Wink, X. Cheng and S. Wolfl, Sci. Rep., 2019, 9, 4616.
14. W. M. Westerink and W. G. Schoonen, Toxicol. In Vitro, 2007, 21, 1581–1591.
15. R. E. H. Karsten, K. Gier, J. S. H. Mulder, M. Grajewski, P. Olinga and E. Verpoorte, Anal. Chem., 2024, 96, 15871–15879.
16. J. v. Delft, K. Mathijs, J. Polman, M. Coonen, E. Szalowska, G. R. Verheyen, F. v. Goethem, M. Driessen, L. v. d. Ven, S. Ramaiahgari and L. S. Price, in Toxicogenomics-Based Cellular Models, ed. K. J, Academic Press, San Diego, 2014, ch. 5, pp. 193–212,  DOI:10.1016/B978-0-12-397862-2.00010-3.
17. B. Shu, W. Duan, J. Yao, J. Huang, Z. Jiang and L. Zhang, Toxicol. In Vitro, 2009, 23, 598–602.
18. K. Mizuguchi, H. Aoki, M. Aoyama, Y. Kawaguchi, Y. Waguri-Nagaya, N. Ohte and K. Asai, Exp. Cell Res., 2021, 404, 112630.
19. X. Wang, I. Armando, K. Upadhyay, A. Pascua and P. A. Jose, Curr. Opin. Nephrol. Hypertens., 2009, 18, 412–420.
20. C. M. Leung, P. de Haan, K. Ronaldson-Bouchard, G.-A. Kim, J. Ko, H. S. Rho, Z. Chen, P. Habibovic, N. L. Jeon, S. Takayama, M. L. Shuler, G. Vunjak-Novakovic, O. Frey, E. Verpoorte and Y.-C. Toh, Nat. Rev. Methods Primers, 2022, 2, 33.
21. W. Hou, B. Liu and H. Xu, Eur. J. Med. Chem., 2019, 176, 378–392.
22. C. Xi, S. Peng, Z. Wu, Q. Zhou and J. Zhou, Biomed. Pharmacother., 2017, 90, 531–541.
23. F. Yang, L. Zhuo, S. Ananda, T. Sun, S. Li and L. Liu, Huazhong Univ. Sci. Technol., Med. Sci., 2011, 31, 335–341.
24. Y. M. Luo, S. D. Yang, M. Y. Wen, B. Wang, J. H. Liu, S. T. Li, Y. Y. Li, H. Cheng, L. L. Zhao, S. M. Li and J. J. Jiang, Front. Plant Sci., 2023, 14, 1144583.
25. J. Liu, L. Li, X. Zhou, X. Chen, H. Huang, S. Zhao, X. Li and D. Zhong, J. Chromatogr. B:Anal. Technol. Biomed. Life Sci., 2013, 939, 51–58.
26. S. Huang, Sci. Total Environ., 2023, 905, 167904.
27. L. Pinzi and G. Rastelli, Int. J. Mol. Sci., 2019, 20, 4331.
28. C. K. Cheng, N. Wang, L. Wang and Y. Huang, Circ. Res., 2025, 136, 752–772.
29. M. Boerma, G. R. Burton, J. Wang, L. M. Fink, R. E. McGehee, Jr. and M. Hauer-Jensen, Blood Coagulation Fibrinolysis, 2006, 17, 173–180.
30. I. Y. Abdelgawad, K. Agostinucci, S. G. Ismail, M. K. O. Grant and B. N. Zordoky, Cell, 2022, 11, 1992.
31. D. Wang, Z. Chen, A. Wai Kan Yeung and A. G. Atanasov, Curr. Res. Biotechnol., 2021, 3, 135–145.
32. China Pat., CN114854585A, 2022.
33. K. Ning, J. Li, X. Yang, Y. Xie, R. Pan, W. Sun and L. Yu, Microchem. J., 2024, 206, 111458.
34. M. Martignoni, G. M. Groothuis and R. de Kanter, Expert Opin. Drug Metab. Toxicol., 2006, 2, 875–894.
35. L. Fu, S. Shi, J. Yi, N. Wang, Y. He, Z. Wu, J. Peng, Y. Deng, W. Wang, C. Wu, A. Lyu, X. Zeng, W. Zhao, T. Hou and D. Cao, Nucleic Acids Res., 2024, 52, W422–W431.
36. J. Abramson, J. Adler, J. Dunger, R. Evans, T. Green, A. Pritzel, O. Ronneberger, L. Willmore, A. J. Ballard, J. Bambrick, S. W. Bodenstein, D. A. Evans, C. C. Hung, M. O'Neill, D. Reiman, K. Tunyasuvunakool, Z. Wu, A. Zemgulyte, E. Arvaniti, C. Beattie, O. Bertolli, A. Bridgland, A. Cherepanov, M. Congreve, A. I. Cowen-Rivers, A. Cowie, M. Figurnov, F. B. Fuchs, H. Gladman, R. Jain, Y. A. Khan, C. M. R. Low, K. Perlin, A. Potapenko, P. Savy, S. Singh, A. Stecula, A. Thillaisundaram, C. Tong, S. Yakneen, E. D. Zhong, M. Zielinski, A. Zidek, V. Bapst, P. Kohli, M. Jaderberg, D. Hassabis and J. M. Jumper, Nature, 2024, 630, 493–500.
37. T. Wu, E. Hu, S. Xu, M. Chen, P. Guo, Z. Dai, T. Feng, L. Zhou, W. Tang, L. Zhan, X. Fu, S. Liu, X. Bo and G. Yu, Innovation, 2021, 2, 100141.
38. M. E. Futschik and B. Carlisle, J. Bioinf. Comput. Biol., 2005, 3, 965–988.
39. S. Chen, iMeta, 2023, 2, e107.
40. M. Bastian, S. Heymann and M. Jacomy, Proceedings of the International AAAI Conference on Web and Social Media, 2009, 8, 361–362.
41. K. J. Livak and T. D. Schmittgen, Methods, 2001, 25, 402–408.
42. E. C. Jensen, Anat. Rec., 2013, 296, 378–381.
43. D. E. Ingber, Nat. Rev. Genet., 2022, 23, 467–491.
44. C. D. Edington, W. L. K. Chen, E. Geishecker, T. Kassis, L. R. Soenksen, B. M. Bhushan, D. Freake, J. Kirschner, C. Maass, N. Tsamandouras, J. Valdez, C. D. Cook, T. Parent, S. Snyder, J. Yu, E. Suter, M. Shockley, J. Velazquez, J. J. Velazquez, L. Stockdale, J. P. Papps, I. Lee, N. Vann, M. Gamboa, M. E. LaBarge, Z. Zhong, X. Wang, L. A. Boyer, D. A. Lauffenburger, R. L. Carrier, C. Communal, S. R. Tannenbaum, C. L. Stokes, D. J. Hughes, G. Rohatgi, D. L. Trumper, M. Cirit and L. G. Griffith, Sci. Rep., 2018, 8, 4530.
45. K. Ronaldson-Bouchard, D. Teles, K. Yeager, D. N. Tavakol, Y. Zhao, A. Chramiec, S. Tagore, M. Summers, S. Stylianos, M. Tamargo, B. M. Lee, S. P. Halligan, E. H. Abaci, Z. Guo, J. Jackow, A. Pappalardo, J. Shih, R. K. Soni, S. Sonar, C. German, A. M. Christiano, A. Califano, K. K. Hirschi, C. S. Chen, A. Przekwas and G. Vunjak-Novakovic, Nat. Biomed. Eng., 2022, 6, 351–371.
46. N. H. Nguyen, Z. Song, S. T. Bates, S. Branco, L. Tedersoo, J. Menke, J. S. Schilling and P. G. Kennedy, Fungal Ecol., 2016, 20, 241–248.
47. X. Qiu, X. Zhou, Y. Miao and B. Li, Toxicol. Res., 2018, 7, 1205–1213.
48. Q. Liu, L. Yue, J. Deng, Y. Tan and C. Wu, Biochem. Biophys. Rep., 2024, 39, 101736.
49. J. Liu, Y. Du, X. Xiao, D. Tan, Y. He and L. Qin, J. Geophys. Res. Planets, 2024, 23, 33.
50. H. Vilstrup, Gut, 1980, 21, 990–995.
51. M. S. Al-Naimi, H. A. Rasheed, N. R. Hussien, H. M. Al-Kuraishy and A. I. Al-Gareeb, J. Adv. Pharm. Technol. Res., 2019, 10, 95–99.
52. S. Y. Lee, I. S. Koo, H. J. Hwang and D. W. Lee, SLAS Discovery, 2023, 28, 119–137.
53. M. Sun, A. Liu, X. Yang, J. Gong, M. Yu, X. Yao, H. Wang and Y. He, Adv. NanoBiomed Res., 2021, 1, 2000066.
54. M. Nooranidoost, D. Izbassarov, S. Tasoglu and M. Muradoglu, Phys. Fluids, 2019, 31, 081901.
55. J. Theobald, A. Ghanem, P. Wallisch, A. A. Banaeiyan, M. A. Andrade-Navarro, K. Taskova, M. Haltmeier, A. Kurtz, H. Becker, S. Reuter, R. Mrowka, X. Cheng and S. Wolfl, ACS Biomater. Sci. Eng., 2018, 4, 78–89.
56. J. H. Moran, L. A. Mitchell, J. A. Bradbury, W. Qu, D. C. Zeldin, R. G. Schnellmann and D. F. Grant, Toxicol. Appl. Pharmacol., 2000, 168, 268–279.
57. J. H. Moran, R. Weise, R. G. Schnellmann, J. P. Freeman and D. F. Grant, Toxicol. Appl. Pharmacol., 1997, 146, 53–59.
58. C. Grange and B. Bussolati, Nat. Rev. Nephrol., 2022, 18, 499–513.
59. Y. Qin, L. Long and Q. Huang, J. Appl. Toxicol., 2020, 40, 1166–1182.
60. J. Yang, Y. Xu, L. Deng, L. Zhou, L. Qiu, Y. Zhang and J. Zhou, BMC Nephrol., 2022, 23, 15.
61. A. Sedzikowska and L. Szablewski, Int. J. Mol. Sci., 2021, 22, 13522.

---

Footnote

† These authors contributed equally to this work.

---