ORGAN-SPECIFIC MPS CONSIDERATIONS
A number of organ-specific MPS are in development and being utilized in biopharmaceutical research including liver, gastrointestinal tract, vascular system, kidney, and lung. These are further discussed below.
Liver
Drug-induced liver injury (DILI) is poorly predicted using animal models (Olson et al., 2000) and attrition of therapeutic candidates due to DILI in nonclinical and clinical development remains high despite advancements in predictive DILI model developments. This may be related to the fact that current approaches using immortalized cancer-derived cell lines and monoculture conditions fail to recapitulate the multicellular and complex structural environment found in liver tissue in vivo. For example, screening compounds in general cell viability assays along with experiments using liver-specific assays (such as BSEP inhibition) and primary human sandwich-cultured hepatocytes still fail to predict DILI potential in many cases including complex DILI mechanisms e.g., idiosyncratic DILI. Therefore, modeling hepatic responses with complex liver MPS models may help to close the gaps associated with poor prediction potential of DILI in humans.
Details on the recommended specifications for evaluating liver MPS models are discussed by Baudy and colleagues in their recent publication (Baudy et al., 2020). When selecting liver MPS, regardless of platform complexity, initial assessment of certain characteristics should be performed to assure the best functional performance of the model. As loss of normal hepatocyte function is very commonly observed in traditional 2D primary in vitro hepatocytes, one way to determine and monitor the overall hepatocyte health status in the MPS platform over time is to evaluate albumin and urea production. In addition to the evaluation of albumin and urea assessment, consideration of the baseline metabolic capacity of hepatocytes in the MPS platform is recommended. Gene/protein expression of key drug metabolizing phase I and phase II enzymes and transporters should be examined over a time course of at least 14 days but can be longer if required by individual liver MPS experiments. Important markers of liver viability, including alanine aminotransferase (ALT) and lactate dehydrogenase (LDH) activities, and transporter function and bile acid homeostasis should also be examined. Furthermore, inclusion of the relevant cell types in liver MPS (i.e. hepatocytes only vs. hepatocytes and non-parenchymal cells such as Kupffer cells and cholangiocytes) should be driven by the specific question and context of use. For example, if liver MPS is to be used for identifying a mechanism of immune-mediated DILI, inclusion of Kupffer cells might be necessary. For the study of bile acid homeostasis and biliary clearance of drugs, inclusion of polarized hepatocytes with adequate bile canaliculi in combination with bile duct epithelial cells (cholangiocytes) should ideally be integrated into liver MPS. Importantly, powerful microscopic imaging offering sufficient resolution of miniature liver MPS models may be necessary when traditional assays evaluating marker release into media lack relevant sensitivity. Examples of high-content imaging of hepatic spheroids following hepatotoxic insult and GFP transduction in liver microfluidic chips are presented in Fig. 2 and Fig. 3, respectively.
Overall, specific research questions being asked will dictate the composition and complexity of liver MPS platforms. As an example, albumin secretion, urea synthesis and cytochrome P450 expression were enhanced by coculturing primary human hepatocytes with stromal cells (Khetani and Bhatia, 2008) or with endothelial cells (Liu et al., 2014). This has been explored as liver-on-a-chip MPS using a conventional microfluidic bilayer chip used previously for lung-on-a-chip (Huh et al., 2010). Liver sinusoids were mimicked by sequentially seeding sinusoidal endothelial cells and Kupffer cells on one side of a porous membrane, while hepatic stellate cells were seeded on the other side and hepatocytes were cultured on the bottom surface of the channel (Du et al., 2017). In this model, flow-induced shear stress enhanced the secretion of hepatocyte growth factors and enzyme activity, and the co-culture enhanced albumin secretion and neutrophil recruitment. In another experimental design, liver spheroids made of HepG2/C3A cells were embedded with gelatin methacryloyl hydrogel in a polydimethylsiloxane/poly (methyl methacrylate) (PDMS/PMMA) chamber (Bhise et al., 2016). In that setting, long-term culture (30 days) with media flow rate of 200 μL/hr enabled sufficient oxygen supply (1.0 x 10−4 mol/m3) by perfusion culture. In a third example, conventional sandwich culture of rat primary hepatocytes and collagen was evaluated in a microfluidic chip, which was also exposed to perfusion flow (Hegde et al., 2014). Flow-enabled bile canaliculi formation resulted in higher albumin and urea secretions and induced cytochrome P450 1A1 activity in comparison with static culture (Hegde et al., 2014). These examples portray the importance of 3D co-culture in recapitulating adequate structure and function for evaluating physiologic changes in the liver MPS.
Intestine
Gastrointestinal toxicity is one of the leading dose-limiting target organ toxicities predominantly identified in oncology programs. Drug-induced gastrointestinal toxicity hampers clinical development, frequently leading to dose reduction, dose delay and can even prevent compounds from achieving desired efficacy. The adverse effects associated with gastrointestinal toxicity can be broken down by individual symptoms and include nausea, vomiting, constipation, diarrhea and abdominal pain. It is important to stress that while in vivo modeling of these symptoms is heavily reliant on histopathological assessment of the intestine coupled with functional endpoints like fecal pellet count or motility imaging, recapitulation of these adverse effects in vitro has been very challenging.
The required cell types necessary to recapitulate relevant gut function will depend on the specific mechanistic question to be addressed but will generally require major epithelial cell type functions such as stem cell renewal, cell proliferation, Paneth cell support, goblet cell secretion and enterocyte barrier function. Moreover, self-organization into crypt and villus-like domains would provide spatial evidence of relevant cell-to-cell interactions, mimicking intestinal structure and physiology. Ideal gastrointestinal MPS would also carefully replicate the intestinal epithelial cell (enterocyte) interaction with commensal bacteria (microbiome) that is rarely evaluated in traditional gut epithelial monocultures. Finally, incorporation of mechanical factors mimicking peristalsis might further improve in vivo predictivity of intestinal MPS. Important considerations for MPS model specifications are discussed in detail by Peters and co-authors in their recent manuscript (Peters et al., 2020).
The existence of a well-established human intestinal epithelial cell line, Caco2, has greatly contributed to early evaluation of intestinal MPS models. Prior to the development of a gut-chip, Caco2 cells were traditionally cultured in a static Transwell format. For example, a typical assay was performed by seeding Caco2 on collagen I and a Matrigel-coated porous membrane (Kim et al., 2012). As discussed later in lung-on-a-chip case, the application of shear stress exerted by the force of flowing medium over cells along with cyclic mechanical strain imparted by a stretchable PDMS membrane led to the improvement of barrier function and paracellular permeability. The same model was applied to recapitulate 3D villus morphology and damage caused by pathogenic bacterial (enteroinvasive E. coli) overgrowth even without mechanical strain (Kim et al., 2016). Co-culture with vascular endothelial cells can be used to investigate polarized cytokine release in response to the addition of lipopolysaccharide in the “lumen” channel and human peripheral blood mononuclear cells in the “endothelium” channel. In another example, human intestinal crypts derived from biopsy samples were cultured with intestinal microvascular endothelium to recapitulate intestine-on-a-chip mimicking human duodenum (Kasendra et al., 2018). Compared with the Caco2 chip, intestine chips were found to have elevated sucrase activity and mucin 2 levels, demonstrating that key intestinal physiologic functions were present. Other studies recapitulated a hypoxia gradient across the endothelial-epithelial interface by introducing aerobic and anaerobic human gut microbiota onto an intestine-on-a-chip (Jalili-Firoozinezhad et al., 2019). In addition to intestine organ-chips, stem-cell derived intestinal organoids have also found routine applications in screening assays and biological modeling of tissues. Recent work described generation of patient-derived gastrointestinal cancer organoids via stem-cell aggregation in microcavity arrays that were used to evaluate efficacy of anti-cancer drugs (Brandenberg et al., 2020). The organoid array technology [e.g. using induced pluripotent stem cells (iPSCs)] is likely going to be applicable to other target tissues and organ systems to enable higher throughput scalable screening efforts with many downstream applications (Fig. 4).
Cardiovascular effects including electrophysiological and hemodynamic changes, as well as histopathological findings contribute to as much as 18% of compound terminations in the late discovery phase (Fabre et al., 2020). While Fabre and others acknowledge that it will be challenging for in vitro MPS to reproduce all potential toxicities, understanding the capabilities and limitations of each system will help to determine the appropriate questions to ask of each system. Among histopathologic changes, drug-induced vascular injury (DIVI) remains a recurring source of nonclinical toxicity-related attrition. In our experience, due to the low incidence and variable location of vascular changes, exploratory animal toxicology studies have not been able to detect DIVI on a consistent basis possibly as a result of low number of animals used in these screening studies. However, in pivotal toxicity studies, DIVI accounted for ~8% of attrition of new therapeutic candidates (Pfizer’s internal data). Our efforts to identify an in vitro screen for DIVI using static 2D preparations tended to lack sensitivity when compared with in vivo exposure ranges.
Human umbilical vein endothelial cells (HUVECs) and organ-specific primary human endothelial cells show promise in modeling vascular injury in vitro with potentially better clinical translation. It is challenging to create an organ-specific architecture requiring multiple cell types including endothelial cells, pericytes and/or smooth muscle cells, and immune system components and even more difficult to include organ-specific shear flow to mimic physiologic conditions. Many model systems have early promise to overcome one or more of these issues, but few systems have the capability to overcome all of them.
Due to the importance of the vasculature in all tissues in vivo, recapitulating vascular connections in a chip format is an important consideration. In recent years, one of the main research directions in the MPS area is to integrate multiple organs-on-a-chip to realize a human/body-on-a-chip. Considering each organ-on-a-chip as a unit component, it is possible to interconnect several components by engineering means, i.e. PDMS microfluidic channels (Schimek et al., 2020), PMMA modular devices (Esch et al., 2016) and three polymer layers (Edington et al., 2018). However, what remains challenging is the incorporation of vessels between individual MPS organs. Given the long history in vascular biology, fundamental studies on angiogenesis and vasculogenesis have been conducted using on-chip vessels even prior to the establishment of MPS technologies.
There are two major vessel bioengineering methods, which include the “pre-designed method”, where a hollow structure for the vascular network is prefabricated and the “self-organizing method”, which relies on spontaneous formation of endothelial cells in tubes as occurs with vasculogenesis or angiogenesis (Miura and Yokokawa, 2016). Prefabrication requires seeding of endothelial cells in extracellular matrix (ECM) or polymer materials using photolithography (Galie et al., 2014; Song et al., 2012; Song and Munn, 2011), needles (Chrobak et al., 2006; Sadr et al., 2011; Usuba et al., 2019; Yoshida et al., 2013) or sacrificial molding (Golden and Tien, 2007; Miller et al., 2012). ECM structures enable facile embedding of endothelial cells (Chrobak et al., 2006; Galie et al., 2014; Golden and Tien, 2007; Usuba et al., 2019) when polymer channels are coated with ECM prior to cell seeding (Miller et al., 2012; Sadr et al., 2011; Song et al., 2012). Once endothelial cells, HUVECs in most cases, are cultured inside of the structures, effects of fluid shear stress (Galie et al., 2014; Song and Munn, 2011) and growth factors (Usuba et al., 2019) on angiogenesis can be studied. The “pre-designed method” is also used for co-culture with organ-specific cells for tissue engineering applications (Zhang et al., 2016). Owing to the stable and robust features of the method, vascular networks can be pre-designed to interconnect multiple organ components to create multi-organ MPS. Moreover, it is possible to quantitatively evaluate the behavior of endothelial cells at the single-cell level in response to biomechanical and biochemical cues, because the method precisely defines the structure of the vascular network, flow rate and gradient of growth factors in the microfluidic environment. The drawbacks of this approach are poor permeability of vascular walls and inability for vascular remodeling, both very important for the evaluation of DIVI in vitro.
The “self-organizing method” overcomes some of the disadvantages of the “pre-designed method”, as it relies on spontaneous pattern formation of endothelial cells, as seen with vasculogenesis or angiogenesis, which is induced using Matrigel (tube formation assay) (Lawley and Kubota, 1989), fibroblast-coated fibrin beads (Nakatsu and Hughes, 2008) or direct application of recombinant VEGFs (Gerhardt et al., 2003). These culture dish-based methods are easily accessible in conventional drug discovery settings; however, the method is limited in the ability to evaluate intraluminal flow that is essential for assessing the efficacy of compounds. The advantage of microfluidic devices is to provide access to the intravascular space to reconstitute 3D perfusable vascular networks in vitro. In early models, human lung fibroblasts were co-cultured with HUVECs in microfluidic channels and VEGF and other angiogenic factors were introduced in a fibrin-collagen gel, in which HUVECs autonomously invaded the gel structure to create a hollow lumen, while leaving an opening at the interface of microfluidic channels (Kim et al., 2013; Yeon et al., 2012). Advantages of the “self-organizing method” include in vivo relevance such as vascular morphology, permeability and dynamic morphogenetics. A significant number of studies with co-culturing target cells in the vascular network have been reported for evaluating cancer cell extravasation (Boussommier-Calleja et al., 2019; Chen et al., 2017; Jeon et al., 2015) and permeability of the blood-brain barrier (Bang et al., 2017; Campisi et al., 2018). More recently, in vivo relevance of the method promoted not only the co-culture with target cells but also the connection of vessels with 3D tissues including spheroids (Nashimoto et al., 2017; Nashimoto et al., 2020) (Fig. 5) and organoids (Park et al., 2019). The self-organizing model also provides a platform to evaluate the effect of blood flow on drug efficacy (Nashimoto et al., 2020). One major drawback of the method is the design capability of the structure of the vascular network, which might make it challenging to interconnect multiple on-chip organs to create a human-on-a-chip.
Cell-derived microparticles or microvesicles are a type of extracellular vesicle (typically between 100 and 1000 nm in diameter) which have been reported to have roles in intercellular communication in addition to other biologic functions such as immunity and tissue regeneration (Hromada et al., 2017). In vascular MPS models with a microporous membrane separating two microfluidic channels (e.g. an endothelial channel and pericyte channel), when the membrane pores are 1-3 µm in diameter, cell-derived microparticles (< 1 µm) could travel from one channel to the other. Flow cytometry for pericyte-specific (NG2) or endothelial-specific (CD31) markers might have utility in exploring the movement of microparticles to allow investigation of intercellular communication in a co-culture system (Fig. 6). In-house flow cytometry analysis of media samples from a vascular microfluidic device (PREDICT-96 platform described in Tan et al. (2019)) demonstrated presence of cellular microparticles in both channels under high shear conditions (Fig. 6). While we cannot exclude the possibility that small cell fragments were also captured in the analysis (gating captured all particles < 1 µm), the preliminary results indicate that the majority of identified microparticles were found in the channel consistent with their cell of origin. Further exploration of high shear effects and TGF-β activation on the number of cell-derived microparticles in a vascular-chip is ongoing. In our experience, multichannel co-culture systems that incorporate pressurized fluid flow are capable of measuring the response to perturbations in a complex system such as cellular damage, barrier disruption, cytokine stimulation, and fluid shear stress. While initial results indicate improved robustness of these co-culture models, additional qualification for context of use is required. When considering qualification for context of use, specific criteria must be utilized to ensure adequate accuracy and reproducibility.
Drug-induced nephrotoxicity is among the top 5 most frequently observed adverse effects reported during drug development. The development of in vitro models to successfully predict drug-induced kidney toxicity requires solid understanding of the specific cellular targets, mode of action of nephrotoxicants and reliable biomarkers of nephrotoxicity. In standard in vitro models, kidney epithelial cells exhibit poor apical-basal polarization and have low or no expression of transporter proteins, resulting in poor predictivity for kidney toxicity screening. The majority of kidney MPS models utilize the proximal tubules, as this region is considered the major site of drug-induced toxicity, although other systems have also emerged that focus on different regions of the kidney. To this end, prediction of glomerular injury remains incredibly difficult as glomerular physiology has yet to be accurately represented in vitro. Moreover, models incorporating distal tubule or collecting ducts may be developed to capture renal segment-specific toxicants. Ideally, a future kidney MPS would integrate an entire nephron-on-a-chip containing all relevant units in the proper succession to allow screening and characterization of all renal toxicants in a single platform. Since it is recognized that this is a high bar to reach, current kidney models require proper characterization to support specific context of use for generation of decision-making data.
When selecting a kidney MPS, assessment of metabolizing enzymes and renal transporters that play a key role in disposition of drugs and toxicity is necessary. For example, in renal proximal tubule cells (RPTECs), there are multiple basolateral transporters necessary for uptake of metabolic products and drugs into the proximal tubule epithelium (e.g. OAT1, OAT3, OCT2) (Nieskens et al., 2016), as well as efflux transporters localized on the apical membrane (e.g. MRP2, MRP4). Examples of fluorescent staining and brightfield images of proximal tubule-on-a-chip are shown in Fig. 7. The epithelial (RPTEC) and endothelial (HUVEC) sides are identified by EpCAM and CD31 markers, respectively. Frequently, the cell lines and even primary cells used in kidney toxicity screening lack the appropriate transporter expression, and engineered cells overexpressing certain transporters are linked with dramatic increase in screening sensitivity. Given these considerations, careful cell sourcing for kidney MPS models is crucial. In addition to enzyme and transporter evaluation at the gene or protein level (and/or via immunohistochemistry), structural and functional readouts, like assessment of active transport capabilities is necessary for kidney MPS characterization. Finally, the ability to detect biomarkers in culture media (e.g. kidney injury molecule-1 [KIM-1], clusterin and neutrophil gelatinase-associated lipocalin [NGAL]) using commercial ELISA assays would further strengthen the performance and effectiveness of kidney MPS models in screening efforts. Additional kidney MPS qualifications and current gaps are discussed in detail by Philips and co-authors in their recent manuscript (Phillips et al., 2020).
In one of the reported examples of the glomerulus-on-a-chip which aimed to recapitulate glomerular barrier function, chips were created by seeding rat podocytes onto an ECM in a microfluidic device (Wang et al., 2017). In that model, authors demonstrated an increase in filtered albumin which was linked with increased glucose concentrations (from 5.5 mM to 35.5 mM), recapitulating a model of diabetic nephropathy. Other types of human podocytes, including primary, immortalized and amniotic fluid-derived podocytes, have been evaluated in a commercialized glomerulus-chip (Petrosyan et al., 2019). In that setting, immortalized podocytes did not deposit enough components of glomerular basement membrane such as collagens and laminins (specifically COL4A3 and LAMA5), resulting in less retention of albumin compared with primary and amniotic fluid-derived podocytes. In another study, glomerulus-chip was reconstituted with human iPSCs, which were differentiated to podocytes and expressed relevant phenotype markers (e.g. WT-1 and NPHS1) (Musah et al., 2017). Evaluation of adriamycin-induced toxicity demonstrated an increase in albumin filtration and podocyte damage, effects consistent with in vivo nephrotoxicity. These examples highlight models that are moving towards demonstration of barrier function and toxicity; however, glomerular physiology has yet to be fully reconstituted.
Significant efforts have been made in proximal tubule-on-a-chip in the MPS community owing to the above-mentioned importance of nephrotoxicity and transporter assays which are already being employed in biopharmaceutical research. When primary human proximal tubule epithelial cells were cultured under a perfusion flow, ZO-1 expression and albumin/glucose reabsorption increased compared with the static culture condition (Jang et al., 2013). In addition, P-glycoprotein ATP-binding cassette membrane transporter activity was elevated by the perfusion culture condition and the inhibition of its activity was demonstrated with verapamil. Co-culture with normal human microvascular endothelial cells resulted in enhanced proliferation of proximal tubule cells and mitochondrial activity compared with the epithelial monolayer culture condition. Another commercial chip recreated 3D proximal tubules (Tourovskaia et al., 2014; Weber et al., 2016), in which expression of γ-glutamyl transpeptidase and sodium/glucose cotransporter 2, ammoniagenesis and lower levels of KIM-1 were assessed as markers of cellular function. Importantly, secretory transport of the prototypical organic anion, para-aminohippurate, was measured with and without an inhibitor, probenecid, which typically cannot be assessed using a Transwell system. Long-term culture produced in vivo-like morphological characteristics of RPTECs such as physiologically relevant cell height, microvilli length and density (Homan et al., 2016), for which a sacrificial patterning method of tubular structure was developed using 3D printing. The result was applied to a co-culture model with an adjacent channel for endothelial cells (Lin et al., 2019). Glucose reabsorption was enhanced 5- to 10-fold in comparison with Transwell experiments. Owing to the capability of long-term culture, the MPS models described have the potential to be applied for evaluation of glucose filtration and reabsorption.
Lung
Modeling lung responses in vitro is met with multiple challenges due to the complex architecture and diverse functions of the lung. Desirable characteristics for lung models should at minimum include alveolar epithelium with air interface as well as pulmonary endothelium with access to a media/blood compartment to best mimic in vivo-like lung functions. Lung was the first organ-on-a-chip model reported in 2010 that led to the development of the MPS community (Huh et al., 2010). As with any MPS system, sources of pulmonary cell types and their functional capabilities were closely examined before selecting a relevant model. Multiple lung in vitro models have been described to date for drug discovery applications, including lung organoids, air-liquid interface cultures using Transwell systems (Harney et al., 2021), bioprinted human lung tissues and lung-chip devices, with the latter exhibiting the greatest complexity. PDMS microfluidic devices allow for the cyclic stretch to be applied to reproduce physiological breathing movements. In that model, human alveolar epithelial cells and pulmonary microvascular endothelial cells can be seeded in the top and bottom channel of the device, respectively. Air and media flow can be independently set for each channel with cyclic vacuum applied to the sides of the channels to imitate breathing.
Earlier versions of the lung-chip model have been utilized in several experiments including drug toxicity-induced pulmonary edema (Huh et al., 2012), evaluation of metabolic activity and cytokine secretion by physical stimuli (Stucki et al., 2015), asthma and chronic obstructive pulmonary disease models (Benam et al., 2016b), and effects of smoking including e-cigarettes (Benam et al., 2016a). These applications demonstrated applicability of the lung-on-a-chip to disease models and to other organs as discussed in previous sections. Additional details regarding desirable characteristics of lung MPS models have been carefully reviewed by Ainslie et al. (2019).
The respiratory viral pandemic caused by SARS-CoV-2 (COVID-19) makes the lung-on-a-chip an attractive platform to evaluate the efficacy of existing drugs. In a publication preprint, an airway model with human lung epithelial cells expressing ACE2 and TMPRSS2 and endothelium were prepared for evaluation of FDA-approved drugs, demonstrating the significant decrease of viral entry by amodiaquine, toremifene, and clomiphene (Si et al., 2020). Although the study employed a pseudotyped SARS-CoV-2 virus, the model may provide potential value for the current pandemic. Another publication preprint effort recently reported that the native SARS-CoV-2 infection induced lung injury and an immune response (Zhang et al., 2020). The infection on the chip was demonstrated by immunostaining of viral Spike protein and evaluation by transmission electron microscopy. Further, in vivo infection was modeled by circulating immune cells caused by removal of endothelial cells and an increase of inflammatory cytokines. Suppression of viral replication and damage of the alveolar-capillary barrier was significant with the use of remdesivir, despite the clinical findings reporting mixed evidence (Eastman et al., 2020; Singh et al., 2020). The acute demand during the pandemic significantly accelerates the use of and need for MPS models in nonclinical evaluation not only of existing drugs but also for the evaluation of new drugs in the future. MPS will likely enable understanding of the infection mechanism by monitoring infection and immune responses using a route of infection model from upper to lower airway and alveoli-on-a-chip. Manifestation of COVID-19 symptoms depends on the age, race, and concomitant diseases and additionally on potential mutation of the SARS-CoV-2 proteins. Such diversification of COVID-19 infection can likely be recapitulated using MPS models with a variety of viruses and patient-derived iPSCs in contrast to the cost- and time-intensive development of new humanized mouse models.
