Kidney microphysiological models for nephrotoxicity assessment

Abstract

Nephrotoxicity testing is an important step in preclinical development of new molecular entities (NMEs) and has traditionally been performed in 2-D cell culture systems and animal models. However, 2-D culture systems fail to replicate complex in vivo microenvironment and animal models face interspecies differences including the overexpression of drug transporters. In the last decade, 3-D microphysiological systems (MPS) have been developed to address these concerns. Here, we review recent advancements in kidney MPS and their application in drug-induced toxicity testing and kidney disease research. We find that current research is making significant progress addressing MPS limitations such as throughput, incorporating various regions of the nephron such as the glomerulus, and successfully modeling and predicting clinically relevant nephrotoxicity of current and new drugs.

The kidney is a major organ in both drug elimination and susceptibility to toxic exposure, thus researching mechanisms of renal toxicity is of high importance. More than 850 million people worldwide are affected by kidney diseases. Chronic kidney disease (CKD) is a major global health burden that in 2017, resulted in 5.2 million years of life lost due to premature mortality in the Americas alone and 35.8 million disability adjusted life years (DALYs) worldwide [1–6]. In addition to diseases that are risk factors for CKD such as diabetes and hypertension, acute kidney injury (AKI), frequently caused by drug-induced nephrotoxicity has been shown to account for 24% of hospitalized patients with renal failure [1].

Translatable nephrotoxicity assessment in vitro is an important part of understanding mechanisms of kidney diseases as well as for preclinical testing and evaluating new drugs. While 2-D cell culture as well as animal models have been used extensively in the past, challenges in replicating complex in vivo microenvironment and interspecies differences in microphysiology and transporter abundance create significant limitations in clinical translatability. This may contribute to the fact that only 10% of new molecular entities will progress from pre-clinical development to regulatory approval [7, 8]. In the last decade, significant progress has been made in developing 3-D kidney microphysiological systems (MPS) that recapitulate in-vivo aspects of kidney physiology in order to address this issue. Since 2016, a number of studies have started utilizing these MPS for nephrotoxicity testing. The goal of this article is to review progress in the development of kidney MPS in the last four years, with a focus on system innovation and translatable toxicological assessment.

Kidney MPS (kidney-on-a-chip) have been developed to model various segments of the nephron including the proximal and distal tubule, the glomerulus, and the collecting duct. The proximal tubule is of particular importance as it is the site of active renal secretion, reabsorption, and accumulation of drugs and xenobiotics, and hence toxicity is a concern [9–11]. Jang et al. were the first to publish a proximal tubule MPS system in 2013, utilizing a polydimethylsiloxane (PDMS) device with a pressure-pump-controlled fluidic channel compartment lined with collagen IV and seeded with primary human proximal tubule epithelial cells (hPTECs) [12]. Furthermore, the authors showed that this system can recapitulate certain in-vivo proximal tubule functions not seen in 2-D cultures such as increased alkaline phosphatase expression, secondary glucose active transport, and albumin uptake (Figure 1a) [12].

Figure 1.

A) Custom single channel proximal tubule chip developed by Jang et al. [12] B) Nortis single-channel chip used by the Kelly lab group[13–16, 18, 21, 22, 29, 38, 48, 50]. C) Vascularized proximal tubule MPS system on a Nortis platform used by Imaoka et al. and Chapron et al. [13, 14, 16, 48, 50]. The Creative Commons license for this image can be found at http://creativecommons.org/licenses/by/4.0/. D) Design process of a 3-D printed proximal tubule MPS system by Homan et al. The Creative Commons license can be found at http://creativecommons.org/licenses/by/4.0/. [31] E) Diagram of the OrganoPlate, commercially available from Mimetas BV[51], used by Tiesch et al and Vormann et al.[32, 34–37, 43] F) Custom proximal tubule MPS system developed by van der Made et al. [23] G) Custom glomerulus-on-a-chip developed by Schmieder et al. [41] H) Custom PDMS-polycarbonate glomerulus on a chip developed by Zhou et al. The Creative Commons license can be found athttp://creativecommons.org/licenses/by/4.0/. [42] I) Novel intestine-kidney on a chip developed by Li et al. [47] J) Linked Liver MPS and Kidney MPS Nortis systems used by Chang et al. [48]

Weber et al. utilized a similar MPS system based on hardware developed by Nortis Inc. and conducted a comprehensive evaluation of proximal tubule morphology and function (Figure 1b). Confocal microscopy showed localization of tight junction protein ZO-1 to the apical side of the hPTEC tubule and Na/K-ATPase to the basal and lateral membranes, recapitulating correct in-vivo polarization of the tubule. Additionally, a number of in-vivo proximal tubule functions were assessed and validated including y-glutamyl transpeptidase activity, ammonia secretion in response a drop in luminal pH, vitamin D uptake and intracellular bioactivation, and activity of basolateral uptake (OAT1/3) and apical efflux (MRP2/4) transporters, all of which were not feasible in 2-D Transwell systems [13]. Furthermore, recent advances in this MPS platform created vascularized human proximal tubule MPS (VPT-MPS). The Nortis dual channel chip allows for establishment of a vascular endothelial channel with seeding of human umbilical vein endothelial cells, modeling peritubular capillaries next to an hPETC - proximal tubule channel. Chapron et al. was able to use this device to model active secretion of p-aminohippuric acid (PAH), and with in vitro to in vivo scaling of their secretory clearance data was able to reasonably approximate previously reported in vivo PAH renal clearance [14]. Recently, Imaoka et al. successfully used a parent-metabolite full body physiologically-based pharmacokinetic (PBPK) model of morphine and morphine-6-glucoronide to scale VPT-MPS experiment data to predict opioid clearance in both healthy patients and CKD patients (Figure 1c). Similarly, Maass et al. incorporated a quantitative systems pharmacology (QSP) model for in vitro to in vivo translation of hPTEC-MPS data to predict a clinical dosing regimen for rifampicin [15]. These approaches allow for estimation of clinically relevant renal clearance values for compounds regardless of transport mechanism, overcoming previous limitations in 2-D systems that attribute active secretion to a single mechanism such as OATs [15, 16]. These studies demonstrate that the microenvironment created by these 3-D MPS models replicates in vivo functionalities not seen in 2-D systems, allowing more reliable transport and toxicity studies to be conducted.

With the establishment of reliable proximal tubule MPS platforms, there have been many studies utilizing these systems to model human kidney disease states and drug-induced toxicity. Nephrotoxicity caused by the chemotherapeutic drug cisplatin is one of its major dose-limiting adverse events, where it accumulates in hPTECs by transporter activity. There have been several studies incorporating MPS systems to model this accumulation [17]. An early MPS study showed reduced baseline injury compared to 2-D hPTEC control culture, and importantly, that cisplatin-induced injury could be attenuated with co-administration of OCT2-inhibitor cimetidine, indicating correct polarization of transporters [12]. This study used a supra-therapeutic cisplatin concentration (100 uM), however Nieskens et al. recently published a comprehensive study using human-derived renal PTECs (HRPTECs) in a Nortis dual channel MPS confirming the polarization-dependent toxicity of cisplatin at 25 uM, closer to the reported maximum in vivo concentration of 14.4 uM [18, 19]. Confocal fluorescence microscopy and gene expression analysis confirmed basolateral localization of OCT2 and apical localization of lotus lectin and acetylated a-tubulin (primary cilia). Significant toxicity was observed when cisplatin was administered from the basolateral compartment, but no toxicity was observed when exposed via the apical compartment, demonstrating for the first time a sensitive system with correct polarized localization of drug transporters used for translational nephrotoxicity testing [18]. In a subsequent study, Nieskens et al. demonstrated the translational power of these systems by modeling nephrotoxicity of an antisense oligonucleotide drug SPC5001 that failed in a phase-1 clinical trial due to nephrotoxicity. In mice and non-human primate preclinical studies for this drug, no adverse hepatic and renal studies were observed, however using a panel of six urinary biomarkers recently qualified by the FDA, Nieskens’ group was able to reproduce clinically relevant levels of these biomarkers in their MPS systems [20]. This was the first study to clearly demonstrate the translational advantages of MPS systems over animal models to study drug-induced nephrotoxicity. Similarly, in 2018 for the first time, new drug entities NAB739 and NAB741, which are analogs of polymyxin B (PMB) were used in MPS-based preclinical testing, and the researchers were also able to demonstrate cholesterol biosynthesis as a novel mechanism of PMB-induced toxicity [21].

In recent years, MPS systems have been used to assess nephrotoxicity of a number of different drugs, environmental toxins, and disease states, indicating a growing acceptance and a transition from proof-of-concept studies to various applications in preclinical testing and disease research. Imaoka et al. recently published a study elucidating a possible mechanism of ochratoxin-A-dependent nephrotoxicity in the context of chronic kidney disease of unknown etiology (CKDu). hPTEC-MPS system was used to investigate the bioactivation and metabolism of OTA, and it was found that in MPS systems, OTA nephrotoxicity could be attenuated with co-administration of P450 pan-inhibitor ABT but significant nephropathy was seen with co-administration of GST-inhibitor NBDHEX. The authors hypothesized the role of OTA-induced NRF-2 downregulation (and thus GST downregulation) as a possible mechanism of OTA-mediated nephrotoxicity [22].

In the context of CKD, a hallmark of kidney failure is accumulation of indoxyl sulfate (IxS) in the blood due to reduced renal clearance. Van der Made et al. utilized an MPS system seeded with human conditionally immortalized PTECs (ciPTECs) overexpressing OATs (Figure 1e) to investigate the role of altered albumin-facilitated OAT transport of IxS, and using a PBPK model was able to predict clinically observed renal clearances, highlighting the role of this process in IxS accumulation [23]. In these cases, proximal tubule MPS systems were used to elucidate poorly understood mechanisms of renal disease and toxicity.

A drawback of MPS systems, highlighted by Low et al. and other reviews, is the low throughput [24–29]. Most PDMS-based MPS systems contain one to three channels with relatively limited cell numbers, and experiments which provide high content information often are limited in sample number [12, 13, 30]. This is a significant limitation for the adoption of MPS systems as an early-state screening method in toxicology and preclinical drug research. One solution may come from 3-D printing. Homan et al. developed a 3-D bioprinted convoluted proximal tubule (CPT) MPS using hPTECs which recapitulates in vivo CPT structure (Figure 1d), whereas other proximal tubule MPS have linear tubule structure. This advanced model incorporates engineered extracellular matrices, allowing for the formation of a basement membrane on the basal side of the tubule and a microvilli brush border on the apical side. Hundreds of thousands of cells could be analyzed from this system, and multiple tubules can be printed alongside each other [31].

Another alternative is a moderately high-throughput system based on a 384-well microtiter plate containing 40 three-lane microchambers, in which 3-D cell cultures can be used for a number of applications [32]. Recent work has utilized this platform, now commercially developed by Mimetas as OrganoPlates, to develop high-throughput proximal tubule MPS systems (Figure 1d). Like PDMS-based MPS systems, plates are coated with extracellular matrix gel composed of collagen 1 and then seeded with hPTECs. Fluidics in these systems rely on gravity to produce flow and shear stress, contrary to previous systems with mechanical or pneumatic perfusion. Vormann et al. calculated the shear stress in this system and determined that physiologically relevant flow rates and shear force can be achieved (2.02 uL/min and 0.5–2.0 dyne/cm², respectively), in addition to other physiological validations such as correct membrane polarization, tight junction formation, and barrier integrity [33]. Dose-dependent cisplatin nephrotoxicity was assessed through fluorescent dextran leakage, LDH release, DNA damage assay and WST-8 viability measurement, all yielding agreeable results, indicating robustness and reliability of the model [34]. In addition, fluorescence-based transporter assays were used to assess activity and inhibition of several transporters such as P-glycoprotein (P-gp), multidrug resistance protein (MRP), and glucose transporter SGLT2 [35]. In a recent publication, a multi-laboratory collaboration led by Vormann et al. expanded this platform and developed the Nephroscreen, an OrganoPlate-based assay using several readouts including formazan, LDH release, extracellular miRNA detection, HMOX1 and NGAL expression, epithelial barrier integrity and transporter function [36, 37]. ciPTEC-OAT1 cells were utilized for toxicity/transporter assays, and pseudo-immortalized renal PTECs (RPTECs) were used for barrier integrity assays. Toxicity was assessed in not only four known nephrotoxicants (cisplatin, tenofovir, cyclosporin A, and tobramycin), but also for eight potentially toxic test substances provided by several pharmaceutical companies in a blinded manner. Toxicity and drug-interaction data for modeled nephrotoxicants and the eight potential test substances agreed with previous results and clinical/preclinical data provided by the sponsors after experimentation and analysis had been finalized. For most compounds, 48-hour exposure in the OrganoPlate was sufficient to detect nephrotoxicity, but a subsequent experiment also was successful in modeling long-term (11 days) damage induced by cefepime. While it should be noted that side-by-side comparison with 2-D cultures were not done with this experiment, and non-toxic substances were not tested, this platform has potential as a reliable, advanced, low cost, multiparametric high-throughput 3-D screening tool for toxicity assessments [34–37].

Most kidney MPS have focused on modeling proximal tubules to the exclusion of other nephron segments [28]. To address this, recent work has been done to develop 3-D systems to model other portions of the kidney, specifically the glomerulus. Some studies use a passive filtration barrier along with computer-controlled flow rates to simulate glomerular filtration in their MPS systems [38–40], however several studies have been done on modeling glomerular filtration using podocyte MPS. Shmieder et al. developed a ZEBRA-Chip (Figure 1f) consisting of a heart-like micro pump, blood circuit, and a two-well chamber separated by a synthetic membrane in which immortalized podocytes and human blood-outgrown endothelial cells can be seeded to form an artificial glomerular filtration barrier (GFB). Immunostaining, transepithelial electrical resistance, and albumin penetration studies confirmed in vivo physiology of the GFB and recapitulation of pressure-dependent increase of substance filtration in healthy and CKD models [41]. A similar model was employed by Zhou et al. using mice podocytes to model hypertensive nephropathy (Figure 1g). Small, medium, and high molecular weight protein permeability and cytoskeletal integrity markers were assessed in their 3-D model as well as in urine analysis of hypertensive nephropathic rats. Similar hypertension-induced injury markers were observed in both groups, indicating in-vivo translatability [42]. Petrosyan et al. recently utilized the OrganoPlate platform to create a glomerulus-on-a-chip. Sera from individuals with membranous nephropathy (MN) was administered to the chips, which was able to recapitulate a loss of permselectivity and albumin leakage seen in clinical data measured in the same patients. Furthermore, proteinuria was found to be attenuated by administering a-melanocortin stimulating hormone, a clinical therapy for MN [43]. Other regions of the kidney, such as the collecting duct and distal tubules, have been the focus of few microphysiological models [44, 45]. However, Wang et al. recently utilized a similar custom hardware platform to Jang et al. to create a canine distal tubule-on-a-chip to study renal effects of pseudorabies virus, finding for the first time a novel mechanism of pathogenesis involving disruption of sodium reabsorption [44]. While immortalized mouse PC-like mpkCCD cells have been used to create a cortical collecting duct (CCD) MPS of polarized, sodium absorbing principal cells, this system has not been replicated with human cells or utilized for toxicological applications [45].

Replicating complex biological processes and microenvironments are major advantages of MPS. Development of linked MPS systems connecting kidney-on-a-chip platforms to another organ-on-a-chip, such as liver chips or intestine chips, take this a step further which could prove to be a powerful tool to elucidate underlying crosstalk in complex diseases such as CKD [46]. Two recent studies have utilized these systems in the context of nephrotoxicity assessment. In a multi-interface, lithography-printed PDMS device, a custom intestine-kidney (Figure 1h) was developed to examine the role of altered intestinal absorption in drug nephrotoxicity using digoxin along with P-gp inhibitors verapamil and colestyramine [47]. Chang et al. directly coupled individual kidney hPTEC-MPS and liver primary human hepatocyte MPS platforms (Figure 1i) to study mechanisms of aristolochic acid 1 (AA-1) nephrotoxicity, specifically hepatocyte-based bioactivation of AA-1, identifying a sulfate conjugate of an aristolactam metabolite as the nephrotoxic form of AA-1 [48]. Cohen et al. recently was able to introduce proximal tubule spheroids into a chip system, and replicate cisplatin and cyclosporine-A induced nephrotoxicity [49]. It is important to note, again, that with increased complexity of these systems, throughput becomes a drawback.

In conclusion, the last decade has seen significant development and application of kidney microphysiological systems that have started from proof-of-concept studies and advanced to complex drug-induced toxicity and disease research. These MPS range in complexity, from gravitational-flow controlled three channel OrganoPlates to custom lithography-printed multi-organ systems, with computer-controlled flow systems and readouts. Figure 2 provides a summary of the workflow of MPS experiments, including drug and disease conditions and analysis techniques referenced in this review. In conjunction with quantitative translational models, clinically relevant toxicology and clearance data have been produced in several studies. From industrial and clinical perspectives, throughput limitations, cell sourcing and certain fabrication factors, such as PDMS uptake of solutes and synthetic membrane barriers still have room for improvement to better model in vivo physiology. Primary cells like hPTECs, which were used in many studies mentioned in this paper, have limited availability. RPTECs may be an alternative as they do not overexpress any relevant transporters and offer greater reproducibility but are still immortalized [24, 50]. While animal MPS of distal tubule and collecting duct have been developed, human counterparts still need to be established as certain drug classes like aminoglycosides may cause toxicity in these regions. Furthermore, with the increasing development of biologics in industry, incorporation of an immune system compartment into kidney MPS will become a greater necessity to address immunogenicity in kidneys. It is important to note that, at least currently, no single MPS can be applied for every context of drug toxicity and disease research. Researchers will need to carefully consider the context of their use and determine the level of complexity and reproducibility that are fit-for-purpose. Nonetheless, kidney microphysiological models have demonstrated remarkable recapitulation of complex biological processes not achievable in 2-D systems and have become a powerful tool for preclinical toxicity testing and disease research.

Figure 2.

Basic workflow of kidney microphysiological system experiments, including a summary of different drug treatments, diseases, and analytical assays used by lab groups referenced in this article.