Open Access Article
This Open Access Article is licensed under a Creative Commons Attribution-Non Commercial 3.0 Unported Licence

Developing in vitro assays to transform gastrointestinal safety assessment: potential for microphysiological systems

Matthew F. Peters *a, Allison L. Choy b, Carmen Pin c, Derek J. Leishman d, Annie Moisan e, Lorna Ewart a, Peggy J. Guzzie-Peck f, Radhakrishna Sura g, Douglas A. Keller h, Clay W Scott a and Kyle L. Kolaja i
aClinical Pharmacology and Safety Sciences, BioPharmaceuticals R&D, AstraZeneca, Boston, USA. E-mail:
bOncology Biometrics IP, Oncology R&D, AstraZeneca, Boston, USA
cClinical Pharmacology and Safety Sciences, BioPharmaceuticals R&D, AstraZeneca, Cambridge, UK
dEli Lilly, Lilly Corporate Center, Indianapolis, IN 46285, USA
eRoche Pharma Research and Early Development, Roche Innovation Center Basel, Grenzacherstrasse, Switzerland
fJanssen, 1400 McKean Rd, Spring House, PA 19477, USA
gAbbVie, Preclinical Safety, 1 Waukegan Rd, N Chicago, IL 60064, USA
hSanofi US, 55 Corporate Drive, Bridgewater, NJ 08807, USA
iCelgene, 556 Morris Avenue, Summit, NJ 07901, USA

Received 8th November 2019 , Accepted 20th February 2020

First published on 21st February 2020


Drug-induced gastrointestinal toxicities (DI-GITs) are among the most common adverse events in clinical trials. High prevalence of DI-GIT has persisted among new drugs due in part to the lack of robust experimental tools to allow early detection or to guide optimization of safer molecules. Developing in vitro assays for the leading GI toxicities (nausea, vomiting, diarrhoea, constipation, and abdominal pain) will likely involve recapitulating complex physiological properties that require contributions from diverse cell/tissue types including epithelial, immune, microbiome, nerve, and muscle. While this stipulation may be beyond traditional 2D monocultures of intestinal cell lines, emerging 3D GI microtissues capture interactions between diverse cell and tissue types. These interactions give rise to microphysiologies fundamental to gut biology. For GI microtissues, organoid technology was the breakthrough that introduced intestinal stem cells with the capability of differentiating into each of the epithelial cell types and that self-organize into a multi-cellular tissue proxy with villus- and crypt-like domains. Recently, GI microtissues generated using miniaturized devices with microfluidic flow and cyclic peristaltic strain were shown to induce Caco2 cells to spontaneously differentiate into each of the principle intestinal epithelial cell types. Second generation models comprised of epithelial organoids or microtissues co-cultured with non-epithelial cell types can successfully reproduce cross-‘tissue’ functional interactions broadening the potential of these models to accurately study drug-induced toxicities. A new paradigm in which in vitro assays become an early part of GI safety assessment could be realized if microphysiological systems (MPS) are developed in alignment with drug-discovery needs. Herein, approaches for assessing GI toxicity of pharmaceuticals are reviewed and gaps are compared with capabilities of emerging GI microtissues (e.g., organoids, organ-on-a-chip, transwell systems) in order to provide perspective on the assay features needed for MPS models to be adopted for DI-GIT assessment.


Drug-induced gastrointestinal toxicities (DI-GITs) are the most common category of adverse events (AEs) both during clinical trials1 and after drug approval.2 When incidence of drug-induced AEs is broken down by individual symptoms, five GI AEs (nausea, vomiting, constipation, diarrhoea, and abdominal pain) rank among the 12 most frequent.1 These AEs occur across all drug modalities and span all therapeutic areas.3 However, the consequences of GIT often differ from those in other organ systems with respect to timing of discovery, medical response, impact on drug development, and effects on patients. Specifically, GIT risks are frequently discovered late in development during clinical trials. Since GITs are generally non-life-threatening, the medical response tends to focus on dose reduction and/or palliative co-therapies (e.g., anti-diarrheal).4,5 Whilst such clinical management strategies avoid drug attrition, symptomatic relief is transient, and the underlying issues are not addressed. Hence, the existing safety assessment paradigm results in (i) drugs advancing into the clinic with GI liabilities, (ii) GI-related limitations on dosing/efficacy persisting throughout the drug life cycle, and (iii) patients bearing the burden of reduced quality of life, reduced drug compliance, and compromised efficacy.4

In an ideal scenario, DI-GIT would be identified early with assays suitable for screening out the liability during preclinical drug discovery. One recent review evaluated toxicity-related drug attrition by target organs during the preclinical and clinical phases. GI toxicity was the least likely cause of failure in the preclinical phase despite DI-GITs being the most common AEs in the clinical phase.5 One notable limitation on GI preclinical testing cascades is the lack of in vitro assays suitable for routine screening. New in vitro assays that can predict DI-GITs will therefore be positioned for significant impact. The demand will likely extend beyond a single assay; ultimately a suite of in vitro assays may be utilized such that each of the major GI AEs is supported with a frontline screening assay for hazard detection along with multiple investigative assays for mechanistic follow-up.

For many GI AEs, microphysiological systems appear uniquely suited to fill the void in in vitro assays. MPS are biomimetic devices recapitulating natural physiology of human or animal tissues in a microenvironment that induces in situ cell phenotypes (e.g., architecture, polarity, genetic expression, differentiation, response to mechanical factors such as stretch and perfusion, etc.). For the purposes of this manuscript, MPS are also defined as going beyond traditional 2D sandwich culture and could include several of the following design aspects: a multi-cellular environment within a biopolymer or tissue-derived matrix, a 3D structure that replicates essential anatomical architecture, mechanical factors such as stretch, shear force or perfusion, incorporate primary or stem cell-derived cells that express appropriate cell type-specific phenotypic markers and functional capabilities, and/or inclusion of immune system components. In contrast to traditional 2D assays, which are mostly comprised of monocultures of intestinal cell lines, 3D microtissues recapitulate the interactions among diverse cell and tissue types. Complex interactions between intestinal epithelium, immune cells, microbiome, nerves, and muscles are emerging as the essential biology underlying most DI-GITs. It follows that an array of diverse GI MPS models may be needed, each capturing a microphysiological interaction between the minimal set of tissues needed to emulate one particular GI AE in order to transform predictive accuracy for DI-GIT. To guide the design of GI-MPS assays to meet safety assessment needs, this review details a drug discovery perspective on assay features of high priority (Table 1). The topics discussed include: screening assays for early detection, investigative models for mechanistic elucidation, tool compounds for assessing predictive performance, gaps and opportunities emerging with epithelial-only GI microtissues, specific microphysiologies that need to be emulated to align with top GI AEs, kinetic data to support mathematical modelling, and guidance on assay endpoint selection.

Table 1 Safety assessment priorities for GI MPS models in drug discovery
Priority risks to be assessed
Pan-therapeutic area GI AEs Top 5 most common DI-GI AEs: nausea, vomiting, constipation, diarrhea, and abdominal pain, all functional toxicities
Oncology GI AEs The leading DI-GI AE in oncology is mucositis. Chemotherapeutics and targeted oncotherapeutics disrupt epithelial renewal by on-target cytostatic and cytotoxic mechanisms leading to compromised barrier function, infection, and diarrhea

Priority microphysiologies to be replicated
Enterocyte–microbiome–immune interaction Disturbances in the homeostatic interaction between epithelial renewal and maintenance of tight barrier function, bacterial penetration, and immune surveillance is fundamental to GI disease and toxicity. Replicating this microphysiology may transform in vivo predictivity of in vitro models
Enterochromaffin–neuronal interaction Enterochromaffin epithelial cells are proposed to act as epithelial chemosensors, transducing information via serotonergic synapses with enteric neurons. This microphysiology may yield breakthroughs enabling in vitro models to predict nausea, vomiting, or abdominal pain

Priority capabilities to be included
Functional endpoints Functional endpoints measuring mechanistically aligned microphysiologies are needed rather than cytotoxic endpoints which has been insufficient to predict organ-specific toxicity6
Moderate throughput for mechanistic studies Moderate throughput assays (e.g., 96-well) are needed for investigative studies aiming to elucidate mechanisms by pharmacological characterization
Genetic manipulation Capacity to manipulate gene expression is needed for mechanistic toxicity studies. Genetic inhibition needed to distinguish on- vs. off-target mechanisms of toxicity
Cross-species comparison Comparing toxicity responses across species used for in vivo safety assessment (rat, dog, non-human primate (NHP), human) is needed for translating preclinical animal study toxicities into clinical risk
Exposure control To match oral drugs, access to apical surface is needed to mimic the selectively high lumen exposure; drug washout capacity is needed to match pulsatile exposure
Response kinetics for PK–PD modeling The ability to mimic clinical drug exposure profiles and measure the resulting drug responses (i.e., pharmacokinetic/pharmacodynamic, PK/PD) is needed. In particular, matching sampling frequency/duration with clinical dose schedules and oral-like exposure (apical dosing with washout) is needed to inform PK–PD mathematical modeling to optimize clinical dose schedules
Cellular kinetics for mechanistic modeling Mechanistic kinetic endpoints need to be measured over a time scale that is clinically matched to drug exposure duration (washout) and schedule (weekly)
High-throughput for frontline screening 384-Well throughput is desired to enable routine frontline screening during chemistry optimization to assess and minimize DI-GIT risk

Frontline assays for GIT risk detection

Currently, there are no in vitro screening assays suitable for routine GIT risk assessment early in drug discovery. Rather, preclinical DI-GIT risk detection depends on testing in animal models. The performance record of the industry's in vivo-focused cascade has been quantitatively assessed in two surveys of multinational pharmaceutical companies. Retrospective evaluation of compounds with clinical GIT (100% prevalence) revealed that non-rodent studies had 83% clinical agreement whereas rodent studies had 46% agreement.7 Conversely, selecting compounds with preclinical GIT and prospectively assessing the diagnostic value of different species revealed a proportionate reduction in uncertainty of 87%, 28%, and 15%, NHP, dog, and rat respectively.8 Both studies confirm that the current in vivo testing cascade is highly dependent on data from higher-order species and delivers suitable predictive accuracy if non-human primate is included. However, due to both ethical and cost considerations, testing in non-rodent species is often appropriately limited to a few potential compounds that are evaluated in a species that may not necessarily be well-suited for human GI functionality. Thus, the standard preclinical in vivo safety assessment package supports detecting GIT just prior to entering clinical trials but inherently lacks the capacity to select/design drug molecules with improved safety.

Frontline in vitro assays are a high priority need. Concordance with clinical events for a given GI AE is the primary criterion used to determine whether an assay is worthy of implementation. For DI-GIT, routine clinical assessment includes the range of signs and symptoms, duration, and severity scored based on a 5-grade system provided by the National Cancer Institute Common Toxicity Criteria (NCI CTC). GI toxicities are determined to be drug-related if the AE (i) coincides with drug treatment, (ii) resolves with removal, and (iii) is exacerbated by re-challenge. Since most GI toxicities resolve with drug removal, tissue biopsies are difficult to procure from the site of injury and thus for GI AEs discovered in clinical studies, histopathology is most likely not available to provide mechanistic insight, hence the toxicities are therefore defined functionally. The clinical data that is routinely and publicly available to assess the predictive performance of an in vitro assay consists of functional symptoms, frequency of occurrence, and total drug exposure in plasma (i.e., drug bound to serum proteins plus free/unbound drug).

Qualification of an in vitro assay with evidence establishing the predictive accuracy for a particular clinical AE9 is paramount for a new MPS assay to gain adoption in drug discovery. The confirmation of performance requires well-characterized positive and negative reference compounds to determine the suitability of a model for use in exploring a specific question. For broader toxicological applications, the compound set used should include diverse chemistry and cover multiple modalities (e.g., small molecules, biologics, nucleic acids, etc.), if possible. Alternatively, a set of compounds could be used to qualify an assay for narrow usage within a chemical series. To provide sets of drugs for evaluating GI MPS model predictive performance, clinical data were mined for marketed drugs known to cause nausea, vomiting, or constipation. Approved drugs with high or low incidence were selected and the total plasma exposures associated with the clinical findings were identified (Tables 2–4; for references see ESI Tables S1, S2, and S3, respectively). A similar set of compounds for diarrhoea (Table 5, see references in ref. 10) was tested in an in vitro human GI microtissue and revealed that barrier function measured by transepithelial electrical resistance (TEER) predicted clinical diarrhoea with 83% accuracy10 which would be suitable for drug screening.11 We propose these four well-annotated compound sets as standards to be used to benchmark predictive accuracy of future MPS assays. Drug concentrations tested in the in vitro systems should achieve multiples of the clinical plasma Cmax when feasible, as is often done for liver and other in vitro systems such as for the potassium channel hERG,12 to closely replicate local concentrations found under clinical use and in toxicology studies and to compare results across assays more effectively.

Table 2 Assay qualification compound set for nausea
Drug name Nausea drug incidence Nausea placebo incidence Clinical Cmax (total, M)
Carboplatin 89% Not determined 8.5 × 10−5
Oxycodone 67% 13% 4.9 × 10−8
Pomalidomide 36% Not determined 2.7 × 10−7
Cabozantinib 47% 21% 1.1 × 10−6
Bortezomib 64% Not determined 1.3 × 10−6
Raltitrexed 58% Not determined 2.4 × 10−6
Tapentadol 49% 13% 2.5 × 10−7
Bosutinib 46% Not determined 3.8 × 10−7
Ixabepilone 42% Not determined 5.0 × 10−7
Decitabine 42% 16% 3.2 × 10−7
Pemetrexed 39% Not determined 2.5 × 10−4
Lubiprostone 31% 5% 2.6 × 10−11
Varenicline 30% 10% 1.9 × 10−8
Ritonavir 30% 8% 1.6 × 10−5
Mycophenolic acid 29% Not determined 8.3 × 10−5
Docetaxel 25% Not determined 2.7 × 10−6
Tramadol (ER) 23% 8% 1.3 × 10−6
Duloxetine 22% 9% 1.3 × 10−7
Pentostatin 22% Not determined 1.3 × 10−6
Dofetilide 5% 4% 8.6 × 10−9
Febuxostat 1% 1% 8.7 × 10−6
Amlodipine 3% 2% 1.0 × 10−8
Verapamil 3% Not determined 9.9 × 10−8
Maprotiline 2% Not determined 2.3 × 10−7
Fesoterodine 1% 1% 8.4 × 10−9
Furosemide 1% Not determined 6.7 × 10−6
Nadolol 0.5% Not determined 4.3 × 10−7
Simvastatin 0.4% 0.6% 1.1 × 10−7
Finasteride 0.1% 0% 2.2 × 10−7
Haloperidol 0% 0% 2.7 × 10−8
Valsartan 2% 2% 6.8 × 10−6
Prednisone 1% 0% 5.9 × 10−8
Oxazepam ‘Rare’ Not determined 1.6 × 10−6

Table 3 Assay qualification compound set for vomiting
Drug name Vomiting drug incidence Vomiting placebo incidence Clinical Cmax (total, M)
Carboplatin 82% Not determined 8.5 × 10−5
Oxycodone 42% 3% 4.9 × 10−8
Raltitrexed 37% Not determined 2.4 × 10−6
Bosutinib 37% Not determined 3.8 × 10−7
Bortezomib 36% Not determined 1.3 × 10−6
Tapentadol 32% 3% 2.5 × 10−7
Ixabepilone 29% Not determined 5.0 × 10−7
Cinacalcet 27% 15% 1.1 × 10−7
Cabozantinib 26% 2% 1.1 × 10−6
Decitabine 25% 9% 3.2 × 10−7
Pemetrexed 25% Not determined 2.5 × 10−4
Mycophenolic acid 23% Not determined 8.3 × 10−5
Docetaxel 19% Not determined 2.7 × 10−6
Ritonavir 17% 4% 1.6 × 10−5
Varenicline 5% 2% 1.9 × 10−8
Duloxetine 5% 4% 1.3 × 10−7
Lubiprostone 5% 1% 2.6 × 10−11
Amlodipine 1% Not determined 1.0 × 10−8
Furosemide 1% Not determined 6.7 × 10−6
Nadolol 0.5% Not determined 4.3 × 10−7
Simvastatin 0.2% 0% 1.1 × 10−7
Prednisone 0% 0% 5.9 × 10−8
Verapamil Not noted Not determined 9.9 × 10−8
Valsartan <1% Not determined 6.8 × 10−6
Maprotiline ‘Rarely’ Not determined 2.3 × 10−7
Haloperidol 0% 0% 2.7 × 10−8
Guaifenesin Not noted Not determined 1.1 × 10−5
Oxazepam Not noted Not determined 1.6 × 10−6
Dofetilide Not noted Not determined 8.6 × 10−9
Fesoterodine 0% 0% 8.4 × 10−9
Febuxostat <1% Not determined 8.7 × 10−6

Table 4 Assay qualification compound set for constipation
Drug name Constipation incidence Constipation placebo Clinical Cmax (total, M)
Clomipramine 47% 11% 2.9 × 10−7
Bortezomib 43% Not determined 1.3 × 10−6
Pomalidomide 36% Not determined 2.7 × 10−7
Decitabine 35% 14% 3.2 × 10−7
Alosetron 29% 6% 1.7 × 10−8
Cabozantanib 28% 6% 1.1 × 10−6
Clozapine 25% Not determined 1.3 × 10−6
Hydromorphone 23% Not determined 1.7 × 10−8
Tramadol (ER) 17% 4% 1.3 × 10−6
Oxycodone 15% 1% 5.0 × 10−8
Ondansetron 14% 6% 1.7 × 10−7
Duloxetine 11% 3% 1.3 × 10−7
Morphine 11% Not determined 8.4 × 10−8
Cabergoline 10% 0% 1.1 × 10−10
Voriconazole <2% Not determined 8.6 × 10−6
Amlodipine ≤1 but >0.1% Not determined 1.0 × 10−8
Nadolol 0.5% Not determined 4.3 × 10−7
Lubiprostone 1% 1% 2.6 × 10−11
Simvastatin 0.3% 0.4% 1.1 × 10−7
Zolmitriptan <0.1% Not determined 3.4 × 10−8
Dofetilide Not noted Not determined 8.6 × 10−8
Prednisone 0% 0% 5.9 × 10−8
Fomepizole Not noted Not determined 2.1 × 10−4
Trimethoprim Not noted Not determined 8.9 × 10−6
Oxazepam Not noted Not determined 1.6 × 10−6
Ritonavir 0.2% 0.4% 1.6 × 10−5
Finasteride 0.1% 0% 2.2 × 10−7

Table 5 Assay qualification compound set for diarrhoea
Drug name Diarrhoea drug incidence Diarrhoea placebo incidence Clinical Cmax (total, M)
Afatinib 96% Not determined 7.8 × 10−8
Colchicine 77% 14% 1.7 × 10−8
Idarubicin 73% Not determined 8.8 × 10−8
Tacrolimus 72% Not determined 8.3 × 10−8
Imatinib 60% Not determined 3.2 × 10−6
Capecitabine 55% Not determined 9.7 × 10−6
Axitinib 54% Not determined 5.6 × 10−8
Bortezomib 51% Not determined 1.3 × 10−6
Prostacyclin 50% Not determined 9.9 × 10−10
Crizotinib 49% Not determined 2.2 × 10−7
Sorafenib 43% 13% 6.2 × 10−6
Docetaxel 42% Not determined 3.7 × 10−6
Diacerein 41% 8% 1.3 × 10−5
Quinidine 40% Not determined 1.2 × 10−5
Miglustat 89% Not determined 6.1 × 10−6
Metformin HCl 53% 12% 8.9 × 10−6
Stavudine 50% Not determined 2.4 × 10−6
Mycophenolate mofetil 48% Not determined 5.7 × 10−5
Acetaminophen 1% 1.4 × 10−4
Verapamil 2% 9.9 × 10−8
Dofetilide 3% 8.6 × 10−9
Amiodarone <2% 4.7 × 10−6
Fondaparinux <3% 8.4 × 10−7
Alfuzosin Not noted 3.2 × 10−8
Triamcinolone Not noted 2.7 × 10−8
Ranolazine Not noted 1.4 × 10−5
Fomepizole Not noted 2.1 × 10−4
Haloperidol Not noted 2.7 × 10−8
Isoprenaline Not noted 2.7 × 10−9
Amlodipine Not noted 1.0 × 10−8
Dexamethasone Not noted 1.6 × 10−7
Finasteride Not noted 2.2 × 10−7
Nadolol Not noted 4.3 × 10−7
Nifedipine Not noted 4.6 × 10−7
Methoxsalen Not noted 1.5 × 10−6
Flecainide Not noted 9.3 × 10−7
Maprotiline Not noted 2.3 × 10−7
Dexmedetomidine Not noted 4.7 × 10−9
Furosemide Not noted 6.7 × 10−6

Investigative assays to elucidate GIT mechanisms

Current preclinical approaches to elucidate mechanisms of DI-GIT are generally dependent on in vivo or ex vivo models. A foundational component of mechanistic GI toxicological studies includes testing in animal models to characterize the histopathology associated with the functional symptoms. Follow-up studies rely on a suite of models specialized for three functionalities: motility, secretion, or nausea/emesis (Table 6 and reviewed in ref. 4). Motility is evaluated with organ bath, charcoal meal test, or fecal pellet counts. Newer approaches include the wireless SmartPill™ and Bravo capsule as well as imaging transit of nonabsorbable markers using radiography or ultrasonography. Wireless capsules can measure pH, thus offering a less invasive measure of gastric secretion compared to traditional monitoring using cannula inserted in the stomach, intestine, or bile duct. Emesis is assessed by counting instances of vomiting in ferrets or dogs while nausea can be assessed in rodents by quantifying pica behaviour, e.g., measuring consumption of non-nutritive material.13 Permeability and ion transport can be quantified with ex vivo intestinal tissues in Ussing chambers.14 Precision-cut intestinal slices are an alternative approach to defined MPS approaches.15,16 These in vivo and ex vivo tools are primarily descriptive in nature thereby leaving a barrier to elucidating mechanisms. In pharmaceutical companies, mechanistic studies need to confirm pharmacological mechanism often through genetic manipulation to determine target involvement. Critical endpoints include assessing barrier fidelity, immune surveillance, and epithelial cell renewal.
Table 6 Methods used to assess pharmaceuticals gastrointestinal safety
GI drug AEs In vivo models Ex vivo/In vitro models Microtissue models
• Nausea • Histopathology of GI • Not currently available • Not currently available
• Pica, consumption of kaolin (rat)
• Vomiting
• Emesis, retching count (ferret/dog)
• Diarrhea • Histopathology of GI • Ussing chamber • Organoid swelling (secretory diarrhea)
• Constipation • Fecal pellet count (rat) • GI segment in organ bath, assessment of smooth muscle function
• Meal motility tests with phenol red or charcoal (rat)
• Transwell μ-tissue TEER (diarrhea)
• Capsules (SmartPill™ & Bravo)
• Motility imaging (radiography, ultrasonography)
• Abdominal pain • Histopathology of GI • Not currently available • Not currently available

New GI MPS models capable of supporting mechanistic investigative toxicology is a need that rivals frontline screening assays for impact, but such models must be developed to meet different standards. Although the diversity of investigative questions/applications varies, the critical question to address is whether a DI-GIT is mediated by intended or unintended mechanisms. This distinction is critical for devising next steps in dealing with a GIT, whether to modify the structure of the drug molecule (to mitigate off-target activity) or abandon the drug target or modify dose schedule (to address on-target toxicity). Moderate throughput, more flexibility, and deep access to molecular events in the model are needed to support these types of investigative studies. Specifically, in vitro assays need to enable (i) controlled drug exposure levels and duration, (ii) testing of drugs with differing potency, selectivity, chemical structure, (iii) accurate reflection of in vivo molecular functionality of the target and (iv) genetic manipulation of target expression. In addition, kinetic data are valued for informing mathematical modelling used to discover novel mechanistic-based clinical mitigation strategies (see below: Enabling PK–PD and mechanistic mathematical modelling). Compared to the limitations of mechanistic animal models described above, MPS assays with meaningful biology for supporting investigative pharmacology and/or genetic manipulation have the potential to transform studies elucidating mechanisms of GIT.

One reoccurring investigative concern of critical importance for GITs is the need to resolve in vivo safety findings that differ between preclinical species and determine how these differences translate into human risk. Drug development regulatory guidance documents for conducting human clinical trials require safety assessment in one rodent and one non-rodent species. The most common species, rat and dog, have morphological and/or functional features of their intestinal tracts that complicate clinical translation. Rat intestines have a large cecum and almost half of the stomach surface area is a non-glandular region similar to the human esophagus, but such a region is not found in human stomachs. Rats are relatively resistant to diarrhoea and are incapable of vomiting. Since dogs are carnivores, their stomach acid has a lower pH and the overall length of the GI tract is shorter than in humans. Likely related to these morphologic and functional differences, dogs are more prone to diarrhoea and emesis.17 Having a GI MPS that can be used with multiple species (e.g., rat/dog/human) would be exceptionally valuable for assessing translational safety risk. If drug-induced toxicity is observed in a preclinical species, MPS from that same species would be used to determine if a suitable toxicity signal can be replicated with drug treatment using the in vitro assay. This positive response provides confidence in the ability to interpret the results from human MPS for predicting the effect in humans. However, without positive in vitroin vivo concordance in the preclinical MPS, interpreting a negative response in a human MPS (should it occur) is speculative at best.18 Assays with validated species translation and clinical predictivity will help reduce reliance on testing in animals.

Microtissues-current limitations, critical features, opportunities for MPS

Complex intestinal organoids and next-generation models are likely to be the first in vitro assays implemented for routine GIT screening. Since organoid technology has been aptly reviewed elsewhere,19,20 comments here are focused on their applications and emerging features relevant to developing assays for drug-induced GIT. For existing organoid models, this review highlights (i) the currently available data on assay predictive performance and (ii) the limitations for drug screening. For next-generation models, this review highlights (iii) the breakthrough in interactions between diverse cell types, and (iv) the emerging differentiation cues available to induce the more specialized microphysiologies aligned with GI AEs.

Perhaps the most straightforward and direct application is using GI organoids to screen for drug-induced cytotoxicity and/or reduced cell growth (see caveat on this endpoint in: Selecting assay endpoints for DI-GIT). Although concordance has been seen within small sets of 3 to 4 oncology drugs,21,22 studies using larger, diverse compound sets such as in Tables 2–5 will be required to determine which in vitro endpoints yield sufficient predictivity for particular clinical GI AEs. Independent of those findings, there will be broad demand to extend organoid technology with second generation platforms tailored for DI-GIT applications.

The capacity to control drug exposure is essential for most drug discovery applications but is limited with current organoid technology. Organoids grow embedded in hydrogel which complicates the ability to control/determine the free drug concentration at the organoid exterior. Control of exposure in the luminal space of traditional (apical-in) organoids is limited by the spherical 3D configuration combined with the tendency of epithelial cells to form tight-junctions which may effectively seal these cysts to penetration by some drugs. For orally administered drugs, exposure in the intestinal lumen can be particularly high. In vitro assays that replicate selectively high exposure at the apical surface of enterocytes will be needed. In addition, organoids shed senescent enterocytes into the organoid interior as part of the normal life cycle. The resulting cellular debris accumulates in the organoid core (vs. being cleared via the GI tract in vivo). Consequently, organoids are typically physically fragmented and passaged on a weekly basis to remove the cellular debris. This limits the time course for a drug treatment experiment and complicates studies where longer-term exposure is required, e.g. to compare drug dosing schedules. The recently developed reversed polarity (apical-out) organoids are an important advancement that addresses some of these challenges.23 Replating organoid-derived cultures into transwells or microdevices may be required to allow dual exposure control at both apical and basal surfaces (e.g., ref. 24). Techniques for automated microinjection of organoids with drugs and microbes may be valuable particularly for processes involving highly oxygen sensitive anaerobic microbiota.25,26 Alternatively, 2-dimensional culture platforms yield microtissues with access to both apical and basal surfaces.10,27,28

The breakthrough feature of GI microtissues may be delivering complex epithelial cultures that capture interactions between diverse intestinal cell types. Previous in vitro models based on monocultures of intestinal cell lines (e.g., Caco2) were successfully used as frontline assays for intestinal drug absorption/permeability but were limited as a routine screen for intestinal toxicity. The minimal unit necessary to recapitulate GI epithelial biology/toxicity will vary based on mechanism of action but at a broad level will require major epithelial cell type functions including: stem cell renewal, Paneth cell support, transient amplifying cell proliferation, goblet cell secretion, and enterocyte barrier function. Rare intestinal cell types may be important contributors particularly for replicating drug pharmacology that is reflected in these cells, since expression of unique combinations of G-protein coupled receptors has recently been used to distinguish previously unknown intestinal cell types.29,30 Beyond replicating the presence of primary and rare epithelial cell types, organoids capture the interactions between diverse cell-types. Organoids self-organize into crypt and villus-like domains which is spatial evidence of coordinated interactions between diverse cell-types – the definition of a microphysiology. The interactions between these diverse cellular types are the foundation of GI microphysiologies and may be tailored by carefully combining differentiation cues and specifically designed microdevices to induce biology relevant for GIT.

The cues that direct microtissue differentiation/specialization are being discovered at a rapid pace and merit brief review here. In the original organoid manuscript,31 the combination of cues included defined soluble growth factors (Wnt/β-catenin signalling pathway agonist R-spondin 1, epidermal growth factor, Noggin) and extracellular matrix. Although less appreciated, non-spherical GI microtissues have been developed by combining stimuli consistent with a microtiter plate format such as transwell culturing, air-liquid interface, and undefined growth factors from a feeder layer of mesenchymal cells.32–35 In the future, the growth factor cues may be switched from proteins to small molecules in order to improve consistency, reduce cost, and refine pharmacology. To date, small molecule inhibitors of GSK3 (CHIR-99021) and the BMP type I receptor (LDN-193189) have been shown to maintain intestinal stem cells.36 Novel small molecule factors may be discovered using platforms like the microraft array which allows high-throughput screening of factors that replicate stem cell niche conditions.37 In addition to growth factor/chemical support, 3D structure and 3D gradients can provide important differentiation cues. Collagen scaffolds microengineered to replicate the 3D crypt–villus-like enfolding promoted crypt compartmentalization of stem cells and lineage differentiation along the villus. 3D gradients of growth factors further promoted specialization zones including both crypt-to-villus gradients of growth factors38 and villus-to-crypt gradients of the bacterial metabolites, e.g., butyrate.39 The power of gradients has been harnessed in 2D cultures with microholes distributed in a planar array. Above the microholes, where concentrations of growth factors were the highest, stem/proliferative zones formed and as these cells migrated radially, they became more differentiated.40 Finally, physical and mechanical factors add another means for inducing differentiation. Subjecting Caco2 cells to fluid flow (shear stress 0.02 dyne per cm2) induced a 6-fold increase in cell height yielding a columnar shape similar to healthy human intestinal epithelium.41 Adding cyclic peristaltic strain has minimal effect on cellular architecture but increases changes in paracellular permeability and aminopeptidase activity.41 Models with both microfluidic flow and cyclic strain, induced Caco2 and primary human intestinal cells to differentiate into diverse nature epithelial cells (absorptive, goblet, enteroendocrine and Paneth) that assemble with villus-like morphology.42,43 Taken together, the success of growth factors, 3D structure, chemical gradients, and physical/mechanical factors in stimulating epithelial morphogenesis affords GI MPS model developers a diverse toolbox to induce specialized GI microtissues (reviewed in ref. 35 and 44). Hopefully, innovative combinations of these factors can be discovered that coax microtissues into emulating microphysiologies aligned with each GI AE.

A priority microphysiology to replicate: enterocyte/immune/microbiome interaction

Intestinal microbes, enterocytes, and immune cells engage in a complex interaction that are unique to symbiotic gut biology and is a high priority for replication in MPS models. The commensal microbiome is essential for digestion of dietary fibers, proteins, and peptides, yet presents a constant threat of infection and disease pathology.45 Enterocytes form a barrier to prevent microbial invasion, while still allowing nutrient absorption. Immune cells detect and respond to harmful microbial penetration yet are not chronically activated by commensal microbes. Disturbances in the homeostatic balance between enterocyte/immune/microbiome tissues underpin many GI diseases and toxicities. This interactive nature of GI diseases is evident in inflammatory bowel diseases (IBD) such as food allergy and celiac disease, which have intestinal epithelial barrier dysfunction as a primary contributing factor. Hence, IBD are often termed leaky gut diseases (reviewed in ref. 46 and 47). Conversely, immune cell responses can alter enterocyte barrier function and disease. Release of cytokines TNF-α and IL-13 contributes to barrier dysfunction in inflammatory diseases such as Crohn's and ulcerative colitis. Inhibiting TNF-α dramatically reduces gut inflammation and largely restores the gut barrier in Crohn's disease (reviewed in ref. 48 and 49). Disturbances in the enterocyte/immune/microbiome interaction can also be central to toxicological responses. For example, approximately 50% of all patients treated with chemotherapy will suffer from treatment-related gastrointestinal mucositis. The pathogenesis of mucositis involves changes in intestinal epithelial renewal, barrier function, and immune response (reviewed in ref. 50). Immune checkpoint inhibitors can induce immune-related GI adverse events, with 10–30% frequency being reported in clinical trials and covering a wide spectrum of symptoms (reviewed in ref. 51). Finally, intestinal microbiota can contribute to drug metabolism and formation of reactive metabolites that can adversely affect both efficacy and toxicity.52

Achieving the goal of including components of the microbiome in MPS models poses a significant technical hurdle and offers a powerful tool to drive GI microtissue specialization. Since most gut bacteria are obligate or facultative anaerobes, co-culturing with oxygen-requiring enterocytes requires specialized microdevices. Technical feasibility has been demonstrated with several in vitro platforms using Caco2-derived epithelial cells and anaerobic gut bacteria (reviewed in ref. 53). Selecting the appropriate constituents in these microdevices present challenges for both the epithelia and microbiota. The epithelial barrier will likely need to be comprised of more cell types than enterocytes forming monolayer with tight junctions. Goblet cells will be needed to secrete the protective mucus layer (50–450 μm thick in the small intestine) that physically separates microbes from the epithelium. Tuft cells may be needed to support the defence against protozoa and helminth infections (reviewed in ref. 44). Selecting microbiome components (bacterial strains, viruses, and eukaryotes) will need to be defined at a granular level to reduce risk of experimental inconsistency across systems and between laboratories. However, an opportunity to study patient-specific microbiomes could replicate patient/microbiome-specific GI disease and toxicity. At the in vivo level, germ-free animals lacking commensal bacteria exhibit thinner villi, shallower crypts, and reduced cellular proliferation.54 Reductionist in vitro studies reveal that microbial metabolites, like short-chain fatty acids butyrate and acetate, enhance epithelial tight junction protein expression and barrier function in vitro.55,56 There is broad evidence that exposure to biodiverse microbiota correlates with improved immune function and immune-reduced disease,57–59 leading to the hypothesis that diverse microbial exposure is necessary to educate and shape host immunity.60–63 Thus, for models able to surmount these technical difficulties, adding a microbiome component is expected to deliver a powerful cue for enterocyte and immune cell development.

Recent GI MPS microdevices have integrated microbiome/enterocyte/immune microphysiology.64,65 The mechanical cues of cyclic peristaltic strain and microfluidic flow were used in an MPS model to coax Caco2 cells to differentiate into diverse epithelial cells that assemble with villi-like morphology. This model was then inoculated with a microbiome and unlike prior co-cultures where bacteria overgrowth prevented sustained co-culture, bacteria in this MPS microtissue reached a steady state that was maintained for more than a week. Cessation of the cyclic peristaltic-like stretch triggered bacterial overgrowth, which is consistent with post-surgical complication of bacterial overgrowth observed in some patients that lose ileal peristaltic tone.64 Further experiments revealed that co-culturing pathogenic bacteria in the microtissues induced villus atrophy, which could be prevented with probiotic and antibiotic treatments. The addition of immune cells derived from peripheral blood (PBMCs) along with lipopolysaccharide endotoxins, stimulated epithelial cells to produce proinflammatory cytokines (IL-8, IL-6, IL-1β, and TNF-α) sufficient to induce villus injury and compromise intestinal barrier function. This breakthrough MPS model replicating enterocyte/immune/microbiome microphysiology in vitro has profound potential for clinical translation. Opportunities for future iterations of improved GI-MPS models include finding novel methods that expand the mucosal immune features by including resident immune cells, capturing innate and adaptive immune responses, and replicating antigen sampling and activation functions.

Epithelial-neuronal microphysiology – modelling nausea, vomiting, abdominal pain, and constipation

Integrating sensory epithelial cells and signal transducing enteric neurons is a priority for developing in vitro models of the most common DI-GITs: nausea, vomiting, abdominal pain, and constipation. Enterochromaffin (EC) cells act as epithelial chemosensors, transducing information via serotonergic synapses with enteric neurons. Alterations in EC cell-released serotonin have been implicated in nausea, vomiting, GI dysmotility, visceral hypersensitivity disorders, and inappetence.66,67 Serotonin receptor 5-HT3 targeted therapeutics are used to treat chemotherapy-induced emesis, chronic nausea, and visceral pain disorders. EC cells in organoids, when co-cultured with neurons, form functional 5HT3R synapses capable of sensing and transducing noxious stimuli from distinct sources including ingested chemicals, commensal organisms, and endogenous regulatory pathways.68

The enteric nervous system (ENS) of the small intestine and colon has autonomous reflex pathways controlling intestinal contractile activity, local blood flow and transmucosal movement of fluids69 and therefore compounds that disrupt ENS function likely elicit diarrhoea or constipation depending on the affected pathway. Emerging in vitro models have paired stem cells from human intestinal organoids with neural crest cells differentiated into neurons. When these cells are implanted in vivo they form neuroglial structures similar to a myenteric and submucosal plexus.70 Implanting these co-cultures in vivo is an emerging approach but offers potential for building epithelial-neuronal models. To support development of in vitro models capable of assessing drugs for the risk of constipation, we have generated a constipation validation test set in (Table 4) which includes drugs with primary effects on neurons and smooth muscle.

Enabling PK–PD and mechanistic mathematical modelling

In instances where GITs are mediated by on-target mechanisms, such as for many oncotherapeutics, it is illogical to expect that optimizing chemistry and reducing off-target pharmacology will mitigate the liability. In such cases, an alternative strategy is to optimize the clinical dose and schedule to maximize therapeutic margin. Historically, effective and tolerated treatment schedules have been identified by extensively testing different schedules in clinical trials, an approach that is empirical and costly. Recent work has demonstrated that dosing schedules can be rationally designed using mathematical models derived by relating drug exposure pharmacokinetics to efficacy and toxicity pharmacodynamics. For example, PK–PD modelling of capecitabine GIT in animal and clinical studies was used to predict a clinical dose and schedule with improved efficacy and reduced on-target toxicity.71 Application of modelling is often limited by access to robust kinetic data. Hence MPS models able to deliver kinetic data to inform modelling have the potential to assist in clinical management of DI-GIT.

PK–PD models are empirical and can resolve the relationship between concentration and effect over time. The iterative rounds of data collection required for model predictions to be tested and refined are often lacking and have limited implementation of this approach. In early clinical trials focused on safety, patient numbers and sampling frequency are limited, while switching to preclinical animal studies introduces cross-species differences. Furthermore, determination of GIT in animal studies depends on terminal histological evaluation, which limits sampling frequency and is an endpoint that can be an ‘indirect’ reflection of toxicity, since the tissue damage is several steps downstream from the primary toxicity target effect. By contrast, human derived MPS models offer a variety of endpoints to quantify response dynamics, with automated high frequency sampling and live real-time observation to feed mathematical modelling. For example, TEER data from GI microtissues were used to model drug-induced intestinal barrier damage and recovery dynamics.10 For the particular drug investigated, modelling revealed complex epithelial responses comprising damage, recovery, and sensitization. However, due to the slow recovery kinetics, all predicted dosing schedules include extended drug holidays, that are too long to maintain the target inhibition needed for efficacy.10 Although a viable clinical schedule could not be ascertained as achieved with capecitabine,71 both examples demonstrate the important value of utilizing PK–PD models in GIT clinical plans.

In contrast to PK–PD models, mechanistic mathematical modelling accounts for the underlying biology and provides a causality between inputs and outputs. Mechanistic models can be expanded to include multiple compounds affecting the same underlying physiology processes. Models have been developed to describe the temporal72,73 or spatiotemporal cell dynamics74–77 in the GI epithelium and to elucidate epithelial organization and renewal. Emerging multiscale approaches78 utilize a comprehensive understanding of the main biological processes to enable the prediction of toxicity and recovery at multiple temporal and spatial (e.g., cells, organ, organism) scales. As an example, development of intestinal tumorigenesis has been modelled based on Wnt-activating mutations.79 MPS platforms capable of providing data with enough resolution to parametrize these models and/or test hypotheses will be applicable for developing mechanistic models of DI-GIT affecting epithelial renewal. Mechanistic-inspired modelling has been applied to DI-GIT and successfully predicted disruption of epithelial proliferation and recovery following oncology therapies.80 Parameters fitted included normal rates of doubling, arrest, death, and shedding for each epithelial cell-type. For those MPS platforms capable of providing these data, this mechanistic model should be applicable for DI-GIT affecting epithelial renewal.

Selecting assay endpoints for DI-GIT

Experience with in vitro toxicity screening across diverse organ systems suggests several general principles for selecting endpoints for DI-GIT detection in MPS. Cytotoxicity-related endpoints often fail to accurately predict organ-specific toxicities even when the cytotoxicity is assessed in organ-specific cultures (i.e., hepatic-, cardiac-, or kidney-derived).6,81 Cytotoxicity endpoints do add to interpretations of other endpoints but may not be a useful measure alone. Second, Valentin and colleagues observed that most GI AEs are functional toxicities thereby indicating that functional screening assays should be prioritized.4 Third, strategies for managing GIT in the clinic have exploited dose scheduling, thereby suggesting that kinetic data should be generated, if possible, to inform mathematical modelling and guide discovery of optimized clinical schedules (discussed in Enabling PK–PD and mechanistic mathematical modelling). Selecting assay endpoints that measure GI-specific functions in a non-destructive, re-readable manner will likely enhance predictivity and expand utility.

Examples of MPS suitable endpoints that capture GI-specific functions and their relationships to drug toxicity include the following:

(i) Intestinal barrier function is a core role of the GI epithelia and its disruption is associated with drug-induced diarrhoea and inflammatory bowel diseases. Robust barrier data have been achieved with GI microtissues using TEER in transwells10 and microfluidic devices.64 Another electrically-based method, impedance, delivers robust barrier data but requires electrodes on solid surfaces.82 TEER and impedance measurements are non-destructive, label-free and can be repeated to generate kinetic data for modelling. Measuring barrier changes by dye diffusion eliminates the need for electrodes but lacks the exquisite sensitivity of electrical measures83 and is an endpoint assay.

(ii) Immune function is central to GI homeostasis, disease and DI-GIT, thus quantifying immune changes resulting as a primary drug response or secondary to barrier disruption and microbial-related stimulation will be fundamental. Inflammatory responses have been quantified in MPS models by measuring proinflammatory cytokines released into the culture media from GI microtissues cultured in microfluidic devices.64,65

(iii) EC cell chemosensory signalling is a GI function of particular interest due to linkage with nausea, vomiting and visceral pain. Endpoints possibly as simple as measuring serotonin release also warrant exploration.

(iv) Continual renewal of epithelial cells is a core GI physiological process encompassing the maintenance of quiescent reserve stem cells within the intestinal crypts through to enterocyte senescence at the tip of villi. Endpoints that quantify diverse epithelial cell types, ideally with high-content capacity to distinguish cell types over time will allow assessment of cell type-specific drug toxicity and adaptive population shifts. A prototypic example for this application would be the drug-induced differentiation of proliferative cells into goblet cells by Notch/γ-secretase inhibitors.84

(v) Mucus secreted by goblet cells is an essential frontline protective barrier for the GI epithelium. Causal linkage between changes in mucus secretion and GIT have not been widely reported however it is not routinely examined.85 Exploratory assessment of mucus secretion in an MPS model suggests this endpoint has potential for drug toxicity testing.65

(vi) Finally, secretory diarrhoea can occur as a consequence of disrupted electrolyte homeostasis. Although not a common mechanism for drug-induced diarrhoea, it can be quantified by measuring drug-induced GI organoid swelling as exemplified by an assay designed to detect modulators of CFTR mutations to treat cystic fibrosis.86

Preclinical biomarkers of GIT are endpoints that may prove valuable for some MPS models. Potential biomarkers include L-citrulline, which is associated with reduced enterocyte mass, calprotectin, increases of which are associated with IBD, and miR-194, which has been shown to correlate well with histopathology and may detect transient GI damage that could be repaired. Other biomarkers used in vivo that could be useful in vitro include diamine oxidase for small intestinal mucosal lesions, gastrin for injury to the duodenum and gastric antrum, pepsinogen for inflammation of the gastric mucosa, and I-FABP for cell damage of enterocytes. An advantage of many MPS models is that the material used is transparent or sufficiently translucent to allow imaging of the cells, providing the opportunity to correlate biochemical biomarkers with cell morphology.

Novel biomarkers for GIT are a high priority need and could be revealed by MPS models that include endpoints for unbiased biomarker discovery. While the existing biomarkers of GIT have shown fit-for-purpose as pre-clinical tools, they suffer from significant inter-species variability and lack demonstrable clinical translation. For many clinicians, the best biomarker of GIT is still diarrhoea (or bloody diarrhoea indicating colitis) (reviewed in ref. 87). As an example of the expected impact, the cleaved ectodomain of KIM1 has emerged as a circulating biomarker of renal proximal tubule injury and is widely used to facilitate the early diagnosis of renal toxicity/disease. For GIT, MPS in vitro models may be sufficiently reductionist to enable elucidation of novel biomarkers. Once MPS models are established for a particular GIT, multi-omics screening for biomarkers should be considered.

Assay qualification and context of use

While assay qualification is an important step in the eventual use of an MPS as a predictive tool in drug discovery, for MPS assays to guide clinical studies (such as an in vitro biomarker or an MPS-driven mathematical model) the standards are higher, and guidance set by the NIH and FDA related to biomarkers and clinical assessments should be appreciated.88 Qualification is driven from clearly defining the context of use (CoU) which is “a statement that fully and clearly describes the way the medical product development tool is to be used and the medical product development-related purpose of the use”.88 This includes both the precise statement of the purpose and the conditions upon which the assay is to be evaluated. In the case of a GI MPS model, the CoU would describe the segment of the GI tract being modelled, along with the toxicity/disease of interest, the expected outcomes, and how the data will be used. As an example, ‘the small intestine MPS can be used to assess toxicity of drugs during exposures of up to two weeks. Epithelial barrier function and biomarkers such as L-citrulline and diamine oxidase can be assessed to demonstrate toxicity. Data can be used to discriminate between drug candidates or illustrate mechanisms of toxicity’. When working with a CoU statement, the proposed use guides the qualification studies and the data obtained can feed back to modify the CoU statement.

The boundaries of utility within the CoU framework can be defined by characterization and comparison of the in vitro model to in vivo responses and GI functions being modelled. Characterization can range from gene and protein expression, morphology, cell function, and pharmacologic and toxicologic responses. The layering of characterization efforts with effects from well characterized molecules is the foundation of sound model development and ultimately can be used as a rationale to expand CoU (see above: Frontline assays for GIT risk detection). Examples include a colon cancer line, LS180, which was qualified in a narrow CoU approach on mycophenolic acid (MPA) induced GI injury. The authors demonstrated key MPA metabolism genes and consistent cytotoxicity during line passaging.89 A second example is a transwell human GI microtissue model where TEER IC15 values were used to predict drug-induced diarrhoea. Across a set of 30 compounds, human microtissue TEER data were a better predictor of diarrhoea than data derived from Caco2 cells.10 This study is a de facto blueprint for model development and establishment of a CoU for a screen to identify drug-induced diarrhoea.

Challenges for GI MPS models specific to assessing DI-GIT

In addition to the challenges of recapitulating normal GI physiology (reviewed in ref. 19 and 90), MPS models for DI-GIT screening will encounter several application-specific challenges. The next major advance requires supplementing epithelial microtissues with additional tissue types. Among the leading GITs, the diverse mechanisms are proposed to include contributions from immune cells, microbiome, nerve, smooth muscle, and vasculature. Some of these tissue functions require complex organization such as enteric neural networks, circular versus longitudinal muscle layers, or circulating versus resident immune cell concentrated into discrete domains (gut-associated lymphoid tissue) and in the extreme case, some instances of nausea and vomiting that are mediated by the CNS. A central challenge for MPS model development will be assembling the minimal set of necessary tissues with appropriate 3D structure to recapitulate the targeted toxicity.

Two additional challenges unique to assessing DI-GIT arise in use-specific context. GI MPS can be used in drug discovery in either a proactive mode (to predict GI toxicity in advance of in vivo testing) or reactive mode (responding to a preclinical in vivo or clinical finding). The prerequisite to deploy an assay for proactive screening is establishing accuracy of clinical translation. Concordance requires testing drugs in vitro at concentrations that reflect the toxic concentration. A key challenge for this will require knowledge of site of action and local concentration. It is difficult to determine if toxicity is a response to acute high concentration of drug that occurs in the intestinal lumen after oral dosing or reflects secondary systemic exposures. Intraluminal drug exposures can be several orders of magnitude higher than plasma,91 which makes it difficult to know whether toxicity is an on-target response or is due to modulation of lower affinity off-targets at these high concentrations.

After a toxicity has been identified, the next goal is typically to identify an underlying mechanism for toxicity and then determine if chemical modifications of the lead compound can improve tolerability. However, the mechanisms and pathogenesis underlying DI-GITs are often unknown. Although diarrhoea has been considered the best indicator for GIT, it gives little to no information on the underlying mechanism or the target cells for toxicity.87 Thus, a challenge for intestinal MPS models will be to reveal mechanistic insights of DI-GIT with histology, cell type analysis, or similar high-content tissue-level data. Mechanistic insight at the molecular level is equally important as this enables preclinical to clinical translation assessment. Elucidating causal linkage between target pharmacology and toxicity is a reoccurring need requiring molecular approaches such as gene knockout.

GI microtissues typically require a source of intestinal stem cells. Fetal-derived stem cells carry ethical concerns that severely restrict their use within pharmaceutical companies. Sourcing stems cells from biopsied adult intestinal tissue is a minimal requirement. Assays capable of utilizing induced pluripotent stem cells (iPSC)-derived stem cells will be important for comparing healthy and diseased patient populations without the need for biopsy.70


From a drug discovery perspective, there is high demand for in vitro systems that can accurately detect drug-induced GITs. Given the complexity of cell types, their anatomic distribution and biochemical mechanisms that regulate GI homeostasis and function, approaches based on simple cell lines in 2D assays will remain limited. New strategies are needed. Fortunately, the ability to generate multi-cell type intestinal microtissues using either primary stem cells or iPSCs are providing greater opportunities for success in the near-term. Incubating these physiologically-relevant microtissues in transwell, organoid or bioprinted scaffolds with additional tissue-relevant parameters such as pulsatile stretch, fluid flow, and an anaerobic microbial luminal environment may enable predictive in vitro systems for drug-induced GI toxicity.

As with any assay, the balance between complexity and utility is dependent on CoU. For example, moderate throughput screens for a pharmacological endpoint require testing at multiple drug concentrations and replicate samples to generate quantitative values for structure–activity relations and therapeutic index insight. Assays of this type tend to minimize biocomplexity, utilize an economy of scale and platform endpoints to enable cost-effective testing. In contrast, assays designed to provide mechanistic understanding of toxicity or disease typically require increased complexity and cell types combined with bespoke endpoints where low throughput is acceptable. By definition, MPS models attempt to recapitulate tissue biology rather than single cell-type biology and therefore begin at the more complex end of the assay spectrum. However, they need not be restricted to low throughput environments since many of the detection endpoints described in detail (see Selecting assay endpoints for DI-GIT) are compatible with higher throughput screening platforms. Indeed, microtiter plate-based 3D microtissues have been developed83,92 demonstrating that moderate throughput MPS screening is possible. Testing clinically-defined positive and negative control compounds for particular GITs (such as those described in: Frontline assays for GIT risk detection) in an MPS will help establish their suitability for use as a predictive screen for human DI-GIT and to investigate underlying mechanisms of action.

Remarkable progress has been made in the past decade in the development of complex GI microtissues and MPS with the goal of emulating normal intestinal features/function. These efforts provide confidence that we are making headway towards the goal of using in vitro systems in preclinical drug discovery to identify and mitigate DI-GITs. The rate of progress will depend, in part, on continued investment in MPS technologies, intestinal cell biology research and identification/application of endpoints and biomarkers of drug-induced GI dysfunction.

Conflicts of interest

There are no conflicts of interest to declare. All authors are/were current employees of their respective institutions with substantive contributions to the manuscript in the form of conceptualization, analysis, and writing (original drafting, reviewing, and editing).


The authors thank Claire Sadler for comments on the manuscript. This manuscript was developed by members of the IQ MPS Affiliate with the support of the International Consortium for Innovation and Quality in Pharmaceutical Development (IQ, IQ is a not-for-profit organization of pharmaceutical and biotechnology companies with a mission of advancing science and technology to augment the capability of member companies to develop transformational solutions that benefit patients, regulators and the broader research and development community. The IQ MPS Affiliate was launched within IQ in June 2018 to provide a venue for appropriate cross-pharma collaboration and data sharing to facilitate industry implementation and qualification of MPS models. This manuscript is a part of a series of organ-specific manuscripts outlining contemporary pharmaceutical industry perspectives and considerations for developing, evaluating and characterizing various MPS models to support drug discovery and development.


  1. C. Federer, M. Yoo and A. C. Tan, Assay Drug Dev. Technol., 2016, 14, 557–566 CrossRef CAS PubMed .
  2. C. Tamaki, T. Nagayama, M. Hashiba, M. Fujiyoshi, M. Hizue, H. Kodaira, M. Nishida, K. Suzuki, Y. Takashima, Y. Ogino, D. Yasugi, Y. Yoneta, S. Hisada, T. Ohkura and K. Nakamura, J. Toxicol. Sci., 2013, 38, 581–598 CrossRef PubMed .
  3. J. Luo and R. A. Cisler, Cancer Inf., 2016, 15, 211–217 Search PubMed .
  4. A. Al-Saffar, A. Nogueira da Costa, A. Delaunois, D. J. Leishman, L. Marks, M. L. Rosseels and J. P. Valentin, Handb. Exp. Pharmacol., 2015, 229, 291–321 Search PubMed .
  5. D. Cook, D. Brown, R. Alexander, R. March, P. Morgan, G. Satterthwaite and M. N. Pangalos, Nat. Rev. Drug Discovery, 2014, 13, 419–431 CrossRef CAS PubMed .
  6. Z. Lin and Y. Will, Toxicol. Sci., 2012, 126, 114–127 CrossRef CAS PubMed .
  7. H. Olson, G. Betton, D. Robinson, K. Thomas, A. Monro, G. Kolaja, P. Lilly, J. Sanders, G. Sipes, W. Bracken, M. Dorato, K. Van Deun, P. Smith, B. Berger and A. Heller, Regul. Toxicol. Pharmacol., 2000, 32, 56–67 CrossRef CAS PubMed .
  8. T. M. Monticello, T. W. Jones, D. M. Dambach, D. M. Potter, M. W. Bolt, M. Liu, D. A. Keller, T. K. Hart and V. J. Kadambi, Toxicol. Appl. Pharmacol., 2017, 334, 100–109 CrossRef CAS PubMed .
  9. J. A. Wagner, Annu. Rev. Pharmacol. Toxicol., 2008, 48, 631–651 CrossRef CAS .
  10. M. F. Peters, T. Landry, C. Pin, K. Maratea, C. Dick, M. P. Wagoner, A. L. Choy, H. Barthlow, D. Snow, Z. Stevens, A. Armento, C. W. Scott and S. Ayehunie, Toxicol. Sci., 2019, 168, 3–17 CrossRef CAS .
  11. E. Genschow, H. Spielmann, G. Scholz, A. Seiler, N. Brown, A. Piersma, M. Brady, N. Clemann, H. Huuskonen, F. Paillard, S. Bremer and K. Becker, ATLA, Altern. Lab. Anim., 2002, 30, 151–176 CrossRef CAS PubMed .
  12. W. S. Redfern, L. Carlsson, A. S. Davis, W. G. Lynch, I. MacKenzie, S. Palethorpe, P. K. Siegl, I. Strang, A. T. Sullivan, R. Wallis, A. J. Camm and T. G. Hammond, Cardiovasc. Res., 2003, 58, 32–45 CrossRef CAS .
  13. C. Grant, L. Ewart, D. Muthas, D. Deavall, S. A. Smith, G. Clack and P. Newham, Toxicol. Appl. Pharmacol., 2016, 296, 10–18 CrossRef CAS PubMed .
  14. L. L. Clarke, Am. J. Physiol., 2009, 296, G1151–G1166 CrossRef CAS PubMed .
  15. I. A. de Graaf, P. Olinga, M. H. de Jager, M. T. Merema, R. de Kanter, E. G. van de Kerkhof and G. M. Groothuis, Nat. Protoc., 2010, 5, 1540–1551 CrossRef CAS PubMed .
  16. M. Li, I. A. de Graaf and G. M. Groothuis, Expert Opin. Drug Metab. Toxicol., 2016, 12, 175–190 CrossRef CAS PubMed .
  17. A. M. Holmes, J. A. Rudd, F. D. Tattersall, Q. Aziz and P. L. Andrews, Br. J. Pharmacol., 2009, 157, 865–880 CrossRef CAS PubMed .
  18. T. R. Van Vleet, M. J. Liguori, J. J. Lynch, 3rd, M. Rao and S. Warder, SLAS Discovery, 2019, 24, 1–24 CAS .
  19. S. E. Blutt, J. R. Broughman, W. Zou, X. L. Zeng, U. C. Karandikar, J. In, N. C. Zachos, O. Kovbasnjuk, M. Donowitz and M. K. Estes, Exp. Biol. Med., 2017, 242, 1633–1642 CrossRef CAS PubMed .
  20. S. Rahmani, N. M. Breyner, H. M. Su, E. F. Verdu and T. F. Didar, Biomaterials, 2019, 194, 195–214 CrossRef CAS PubMed .
  21. T. Grabinger, L. Luks, F. Kostadinova, C. Zimberlin, J. P. Medema, M. Leist and T. Brunner, Cell Death Dis., 2014, 5, e1228 CrossRef CAS PubMed .
  22. S. Hoyle, A. M. Bonavita, A. Murdoch, M. Brown and C. Booth, presented in part at the European Organisation for Research and Treatment of Cancer, 2016.
  23. J. Y. Co, M. Margalef-Catala, X. Li, A. T. Mah, C. J. Kuo, D. M. Monack and M. R. Amieva, Cell Rep., 2019, 26, 2509–2520 CrossRef CAS PubMed  , e2504.
  24. Y. Wang, R. Kim, S. J. Hwang, J. Dutton, C. E. Sims and N. L. Allbritton, Anal. Chem., 2018, 90, 11523–11530 CrossRef CAS PubMed .
  25. S. Bartfeld, T. Bayram, M. van de Wetering, M. Huch, H. Begthel, P. Kujala, R. Vries, P. J. Peters and H. Clevers, Gastroenterology, 2015, 148, 126–136 CrossRef PubMed  , e126.
  26. I. A. Williamson, J. W. Arnold, L. A. Samsa, L. Gaynor, M. DiSalvo, J. L. Cocchiaro, I. Carroll, M. A. Azcarate-Peril, J. F. Rawls, N. L. Allbritton and S. T. Magness, Cell. Mol. Gastroenterol. Hepatol., 2018, 6, 301–319 CrossRef PubMed .
  27. Y. Wang, I. L. Chiang, T. E. Ohara, S. Fujii, J. Cheng, B. D. Muegge, A. Ver Heul, N. D. Han, Q. Lu, S. Xiong, F. Chen, C. W. Lai, H. Janova, R. Wu, C. E. Whitehurst, K. L. VanDussen, T. C. Liu, J. I. Gordon, L. D. Sibley and T. S. Stappenbeck, Cell, 2019, 179, 1144–1159 CrossRef CAS PubMed  , e1115.
  28. Y. Wang, M. DiSalvo, D. B. Gunasekara, J. Dutton, A. Proctor, M. S. Lebhar, I. A. Williamson, J. Speer, R. L. Howard, N. M. Smiddy, S. J. Bultman, C. E. Sims, S. T. Magness and N. L. Allbritton, Cell. Mol. Gastroenterol. Hepatol., 2017, 4, 165–182 CrossRef PubMed  , e167.
  29. D. Grun, A. Lyubimova, L. Kester, K. Wiebrands, O. Basak, N. Sasaki, H. Clevers and A. van Oudenaarden, Nature, 2015, 525, 251–255 CrossRef PubMed .
  30. A. L. Haber, M. Biton, N. Rogel, R. H. Herbst, K. Shekhar, C. Smillie, G. Burgin, T. M. Delorey, M. R. Howitt, Y. Katz, I. Tirosh, S. Beyaz, D. Dionne, M. Zhang, R. Raychowdhury, W. S. Garrett, O. Rozenblatt-Rosen, H. N. Shi, O. Yilmaz, R. J. Xavier and A. Regev, Nature, 2017, 551, 333–339 CrossRef CAS PubMed .
  31. T. Sato, R. G. Vries, H. J. Snippert, M. van de Wetering, N. Barker, D. E. Stange, J. H. van Es, A. Abo, P. Kujala, P. J. Peters and H. Clevers, Nature, 2009, 459, 262–265 CrossRef CAS PubMed .
  32. F. Boccellato, S. Woelffling, A. Imai-Matsushima, G. Sanchez, C. Goosmann, M. Schmid, H. Berger, P. Morey, C. Denecke, J. Ordemann and T. F. Meyer, Gut, 2019, 68, 400–413 CrossRef CAS PubMed .
  33. A. Ootani, X. Li, E. Sangiorgi, Q. T. Ho, H. Ueno, S. Toda, H. Sugihara, K. Fujimoto, I. L. Weissman, M. R. Capecchi and C. J. Kuo, Nat. Med., 2009, 15, 701–706 CrossRef CAS PubMed .
  34. X. Wang, Y. Yamamoto, L. H. Wilson, T. Zhang, B. E. Howitt, M. A. Farrow, F. Kern, G. Ning, Y. Hong, C. C. Khor, B. Chevalier, D. Bertrand, L. Wu, N. Nagarajan, F. A. Sylvester, J. S. Hyams, T. Devers, R. Bronson, D. B. Lacy, K. Y. Ho, C. P. Crum, F. McKeon and W. Xian, Nature, 2015, 522, 173–178 CrossRef CAS PubMed .
  35. Y. Wang, R. Kim, S. S. Hinman, B. Zwarycz, S. T. Magness and N. L. Allbritton, Cell. Mol. Gastroenterol. Hepatol., 2018, 5, 440–453 CrossRef PubMed  , e441.
  36. Y. Li, Y. Liu, B. Liu, J. Wang, S. Wei, Z. Qi, S. Wang, W. Fu and Y. G. Chen, Cell Discovery, 2018, 4, 49 CrossRef PubMed .
  37. A. D. Gracz, I. A. Williamson, K. C. Roche, M. J. Johnston, F. Wang, Y. Wang, P. J. Attayek, J. Balowski, X. F. Liu, R. J. Laurenza, L. T. Gaynor, C. E. Sims, J. A. Galanko, L. Li, N. L. Allbritton and S. T. Magness, Nat. Cell Biol., 2015, 17, 340–349 CrossRef CAS PubMed .
  38. Y. Wang, D. B. Gunasekara, M. I. Reed, M. DiSalvo, S. J. Bultman, C. E. Sims, S. T. Magness and N. L. Allbritton, Biomaterials, 2017, 128, 44–55 CrossRef CAS PubMed .
  39. Y. Wang, R. Kim, D. B. Gunasekara, M. I. Reed, M. DiSalvo, D. L. Nguyen, S. J. Bultman, C. E. Sims, S. T. Magness and N. L. Allbritton, Cell. Mol. Gastroenterol. Hepatol., 2018, 5, 113–130 CrossRef PubMed .
  40. R. Kim, Y. Wang, S. J. Hwang, P. J. Attayek, N. M. Smiddy, M. I. Reed, C. E. Sims and N. L. Allbritton, Lab Chip, 2018, 18, 2202–2213 RSC .
  41. H. J. Kim, D. Huh, G. Hamilton and D. E. Ingber, Lab Chip, 2012, 12, 2165–2174 RSC .
  42. M. Kasendra, A. Tovaglieri, A. Sontheimer-Phelps, S. Jalili-Firoozinezhad, A. Bein, A. Chalkiadaki, W. Scholl, C. Zhang, H. Rickner, C. A. Richmond, H. Li, D. T. Breault and D. E. Ingber, Sci. Rep., 2018, 8, 2871 CrossRef PubMed .
  43. H. J. Kim and D. E. Ingber, Integr. Biol., 2013, 5, 1130–1140 CrossRef CAS PubMed .
  44. J. S. Dutton, S. S. Hinman, R. Kim, Y. Wang and N. L. Allbritton, Trends Biotechnol., 2019, 37, 744–760 CrossRef CAS PubMed .
  45. A. Koh, F. De Vadder, P. Kovatcheva-Datchary and F. Backhed, Cell, 2016, 165, 1332–1345 CrossRef CAS PubMed .
  46. K. R. Groschwitz and S. P. Hogan, J. Allergy Clin. Immunol., 2009, 124, 3–20 CrossRef CAS PubMed  , quiz 21–22.
  47. A. Michielan and R. D'Inca, Mediators Inflammation, 2015, 2015, 628157 Search PubMed .
  48. A. Buckley and J. R. Turner, Cold Spring Harbor Perspect. Biol., 2018, 10, a029314 CrossRef PubMed .
  49. P. Suenaert, V. Bulteel, L. Lemmens, M. Noman, B. Geypens, G. Van Assche, K. Geboes, J. L. Ceuppens and P. Rutgeerts, Am. J. Gastroenterol., 2002, 97, 2000–2004 CrossRef CAS .
  50. Y. Touchefeu, E. Montassier, K. Nieman, T. Gastinne, G. Potel, S. B. des Varannes, F. Le Vacon and M. F. de La Cochetiere, Aliment. Pharmacol. Ther., 2014, 40, 409–421 CAS .
  51. M. A. Samaan, P. Pavlidis, S. Papa, N. Powell and P. M. Irving, Nat. Rev. Gastroenterol. Hepatol., 2018, 15, 222–234 CrossRef CAS PubMed .
  52. H. Li, J. He and W. Jia, Expert Opin. Drug Metab. Toxicol., 2016, 12, 31–40 CrossRef CAS PubMed .
  53. J. Z. H. von Martels, M. Sadaghian Sadabad, A. R. Bourgonje, T. Blokzijl, G. Dijkstra, K. N. Faber and H. J. M. Harmsen, Anaerobe, 2017, 44, 3–12 CrossRef PubMed .
  54. A. Parker, M. A. E. Lawson, L. Vaux and C. Pin, Environ. Microbiol., 2018, 20, 2337–2353 CrossRef .
  55. C. Y. Hsieh, T. Osaka, E. Moriyama, Y. Date, J. Kikuchi and S. Tsuneda, Physiol. Rep., 2015, 3(3), e12327 CrossRef PubMed .
  56. H. B. Wang, P. Y. Wang, X. Wang, Y. L. Wan and Y. C. Liu, Dig. Dis. Sci., 2012, 57, 3126–3135 CrossRef CAS PubMed .
  57. T. Haahtela, S. Holgate, R. Pawankar, C. A. Akdis, S. Benjaponpitak, L. Caraballo, J. Demain, J. Portnoy, L. von Hertzen and W. A. O. S. C. O. C. Change and Biodiversity, World Allergy Organ. J., 2013, 6, 3 CrossRef .
  58. I. Hanski, L. von Hertzen, N. Fyhrquist, K. Koskinen, K. Torppa, T. Laatikainen, P. Karisola, P. Auvinen, L. Paulin, M. J. Makela, E. Vartiainen, T. U. Kosunen, H. Alenius and T. Haahtela, Proc. Natl. Acad. Sci. U. S. A., 2012, 109, 8334–8339 CrossRef CAS PubMed .
  59. B. N. Lambrecht and H. Hammad, Nat. Immunol., 2017, 18, 1076–1083 CrossRef CAS PubMed .
  60. L. V. Hooper, D. R. Littman and A. J. Macpherson, Science, 2012, 336, 1268–1273 CrossRef CAS PubMed .
  61. C. L. Maynard, C. O. Elson, R. D. Hatton and C. T. Weaver, Nature, 2012, 489, 231–241 CrossRef CAS PubMed .
  62. L. W. Peterson and D. Artis, Nat. Rev. Immunol., 2014, 14, 141–153 CrossRef CAS PubMed .
  63. H. Sokol, B. Pigneur, L. Watterlot, O. Lakhdari, L. G. Bermudez-Humaran, J. J. Gratadoux, S. Blugeon, C. Bridonneau, J. P. Furet, G. Corthier, C. Grangette, N. Vasquez, P. Pochart, G. Trugnan, G. Thomas, H. M. Blottiere, J. Dore, P. Marteau, P. Seksik and P. Langella, Proc. Natl. Acad. Sci. U. S. A., 2008, 105, 16731–16736 CrossRef CAS PubMed .
  64. H. J. Kim, H. Li, J. J. Collins and D. E. Ingber, Proc. Natl. Acad. Sci. U. S. A., 2016, 113, E7–E15 CrossRef CAS PubMed .
  65. W. Shin and H. J. Kim, Proc. Natl. Acad. Sci. U. S. A., 2018, 115, E10539–E10547 CrossRef CAS PubMed .
  66. M. D. Gershon, Curr. Opin. Endocrinol., Diabetes Obes., 2013, 20, 14–21 CrossRef CAS PubMed .
  67. G. M. Mawe and J. M. Hoffman, Nat. Rev. Gastroenterol. Hepatol., 2013, 10, 473–486 CrossRef CAS PubMed .
  68. N. W. Bellono, J. R. Bayrer, D. B. Leitch, J. Castro, C. Zhang, T. A. O'Donnell, S. M. Brierley, H. A. Ingraham and D. Julius, Cell, 2017, 170, 185–198 CrossRef CAS PubMed  , e116.
  69. J. B. Furness, Nat. Rev. Gastroenterol. Hepatol., 2012, 9, 286–294 CrossRef CAS PubMed .
  70. M. J. Workman, M. M. Mahe, S. Trisno, H. M. Poling, C. L. Watson, N. Sundaram, C. F. Chang, J. Schiesser, P. Aubert, E. G. Stanley, A. G. Elefanty, Y. Miyaoka, M. A. Mandegar, B. R. Conklin, M. Neunlist, S. A. Brugmann, M. A. Helmrath and J. M. Wells, Nat. Med., 2017, 23, 49–59 CrossRef CAS PubMed .
  71. K. A. Cadoo, D. Gajria, E. Suh, S. Patil, M. Theodoulou, L. Norton, C. A. Hudis and T. A. Traina, npj Breast Cancer, 2016, 2, 16006 CrossRef PubMed .
  72. M. D. Johnston, C. M. Edwards, W. F. Bodmer, P. K. Maini and S. J. Chapman, Proc. Natl. Acad. Sci. U. S. A., 2007, 104, 4008–4013 CrossRef CAS PubMed .
  73. A. Parker, L. Vaux, A. M. Patterson, A. Modasia, D. Muraro, A. G. Fletcher, H. M. Byrne, P. K. Maini, A. J. M. Watson and C. Pin, Cell Death Dis., 2019, 10, 108 CrossRef PubMed .
  74. P. Buske, J. Galle, N. Barker, G. Aust, H. Clevers and M. Loeffler, PLoS Comput. Biol., 2011, 7, e1001045 CrossRef CAS PubMed .
  75. F. A. Meineke, C. S. Potten and M. Loeffler, Cell Proliferation, 2001, 34, 253–266 CrossRef CAS PubMed .
  76. G. R. Mirams, C. J. Arthurs, M. O. Bernabeu, R. Bordas, J. Cooper, A. Corrias, Y. Davit, S. J. Dunn, A. G. Fletcher, D. G. Harvey, M. E. Marsh, J. M. Osborne, P. Pathmanathan, J. Pitt-Francis, J. Southern, N. Zemzemi and D. J. Gavaghan, PLoS Comput. Biol., 2013, 9, e1002970 CrossRef CAS PubMed .
  77. C. Pin, A. J. Watson and S. R. Carding, PLoS One, 2012, 7, e37115 CrossRef CAS PubMed .
  78. D. Groen, J. Borgdorff, C. Bona-Casas, J. Hetherington, R. W. Nash, S. J. Zasada, I. Saverchenko, M. Mamonski, K. Kurowski, M. O. Bernabeu, A. G. Hoekstra and P. V. Coveney, Interface Focus, 2013, 3, 20120087 CrossRef CAS PubMed .
  79. T. Thalheim, P. Buske, J. Przybilla, K. Rother, M. Loeffler and J. Galle, J. R. Soc., Interface, 2016, 13, 20160218 CrossRef PubMed .
  80. H. Shankaran, A. Cronin, J. Barnes, P. Sharma, J. Tolsma, P. Jasper and J. T. Mettetal, CPT: Pharmacometrics Syst. Pharmacol., 2018, 7, 26–33 CAS .
  81. R. Clothier, M. J. Gomez-Lechon, A. Kinsner-Ovaskainen, A. Kopp-Schneider, J. E. O'Connor, P. Prieto and S. Stanzel, Toxicol. In Vitro, 2013, 27, 1347–1356 CrossRef CAS PubMed .
  82. R. J. Haines, R. S. Beard, Jr., R. A. Eitner, L. Chen and M. H. Wu, PLoS One, 2016, 11, e0154351 CrossRef PubMed .
  83. S. J. Trietsch, E. Naumovska, D. Kurek, M. C. Setyawati, M. K. Vormann, K. J. Wilschut, H. L. Lanz, A. Nicolas, C. P. Ng, J. Joore, S. Kustermann, A. Roth, T. Hankemeier, A. Moisan and P. Vulto, Nat. Commun., 2017, 8, 262 CrossRef PubMed .
  84. J. H. van Es, M. E. van Gijn, O. Riccio, M. van den Born, M. Vooijs, H. Begthel, M. Cozijnsen, S. Robine, D. J. Winton, F. Radtke and H. Clevers, Nature, 2005, 435, 959–963 CrossRef CAS .
  85. H. Yamamoto, K. Ishihara, Y. Takeda, W. Koizumi and T. Ichikawa, BioMed Res. Int., 2013, 2013, 276186 Search PubMed .
  86. J. F. Dekkers, C. L. Wiegerinck, H. R. de Jonge, I. Bronsveld, H. M. Janssens, K. M. de Winter-de Groot, A. M. Brandsma, N. W. de Jong, M. J. Bijvelds, B. J. Scholte, E. E. Nieuwenhuis, S. van den Brink, H. Clevers, C. K. van der Ent, S. Middendorp and J. M. Beekman, Nat. Med., 2013, 19, 939–945 CrossRef CAS PubMed .
  87. D. F. Carr, S. Ayehunie, A. Davies, C. A. Duckworth, S. French, N. Hall, S. Hussain, H. R. Mellor, A. Norris, B. K. Park, A. Penrose, D. M. Pritchard, C. S. Probert, S. Ramaiah, C. Sadler, M. Schmitt, A. Shaw, J. E. Sidaway, R. G. Vries, M. Wagoner and M. Pirmohamed, Pharmacol. Ther., 2017, 172, 181–194 CrossRef CAS .
  88. F.-N. B. W. Group, BEST (Biomarkers, EndpointS, and other Tools) Resource Silver Spring (MD), 2016 Search PubMed.
  89. S. Heischmann and U. Christians, Xenobiotica, 2018, 48, 433–441 CrossRef CAS PubMed .
  90. H. Yu, N. M. Hasan, J. G. In, M. K. Estes, O. Kovbasnjuk, N. C. Zachos and M. Donowitz, Annu. Rev. Physiol., 2017, 79, 291–312 CrossRef CAS .
  91. R. Cristofoletti, N. Patel and J. B. Dressman, J. Pharm. Sci., 2016, 105, 2712–2722 CrossRef CAS PubMed .
  92. D. T. T. Phan, X. Wang, B. M. Craver, A. Sobrino, D. Zhao, J. C. Chen, L. Y. N. Lee, S. C. George, A. P. Lee and C. C. W. Hughes, Lab Chip, 2017, 17, 511–520 RSC .


Electronic supplementary information (ESI) available. See DOI: 10.1039/c9lc01107b
Current: Emulate Inc., 27 Dry Dock Avenue, Boston, MA, 02210, USA

This journal is © The Royal Society of Chemistry 2020