Integrated Microphysiological Systems: Transferable Organ Models and Recirculating Flow

Abstract

Studying and understanding of tissue and disease mechanisms largely depend on the availability of suitable and representative biological model systems. These model systems should be carefully engineered and faithfully reproduce the biological system of interest to understand physiological effects, pharmacokinetics, and toxicity to better identify new drug compounds. By relying on microfluidics, microphysiological systems (MPSs) enable the precise control of culturing conditions and connections of advanced in vitro 3D organ models that better reproduce in vivo environments. This review focuses on transferable in vitro organ models and integrated MPSs that host these transferable biological units and enable interactions between different tissue types. Interchangeable and transferrable in vitro organ models allow for independent quality control of the biological model before system assembly and building MPS assays on demand. Due to the complexity and different maturation times of individual in vitro tissues, off-chip production and quality control entail improved stability and reproducibility of the systems and results, which is important for large-scale adoption of the technology. Lastly, the technical and biological challenges and open issues for realizing and implementing integrated MPSs with transferable in vitro organ models are discussed.

1. Introduction

Animal models are prevailingly used to obtain comprehensive and systemic data on the effects of new pharmaceutical compounds. The use of animal models, however, becomes more and more questionable, as animal models do not necessarily represent all aspects of the human organism, and as obtained results proved to be of low relevance.^[1,2] Moreover, there are ethical and political initiatives to reduce or completely avoid the use of animals for drug and compound testing.

Advanced 3D in vitro models (organoids or tissue cultures) may constitute an alternative test system for initial screening and characterization of lead compounds in early development stages. 3D tissue structures have been proven to more closely reproduce behavior and characteristics of a living organism in comparison to traditional 2D cell cultures in dishes.^[3,4] For such advanced in vitro models, a large range of automated test systems and analysis methods would be potentially available, so that testing procedures could be parallelized to increase throughput. This prospect has fueled rapid progress in developing 3D in vitro models of individual organs of the human body for pharmaceutical and toxicological research. Such model systems need to be carefully engineered to faithfully reproduce physiological effects, pharmacokinetics, and toxicity that may occur within a human body in order to enable successful identification of new drug compounds and the corresponding mechanisms of action.

To realize advanced in vitro models, not only the 3D nature of a human body and its tissues needs to be considered, but also the potential interplay of organs. More recently, microphysiological systems (MPSs) emerged in an effort to better recapitulate in vivo conditions of human physiology and to better control in vitro cell and tissue culturing conditions.^[5–10] MPSs are in vitro platforms designed to model the spatial, chemical, structural, and physiological elements of in vivo cellular environments and include, in most cases, a combination of advanced cell-culture or tissue models and microfluidic technology. Important features of MPSs include the use of more complex multicellular or tissue structures and the possibility to expose cells to cues that they also experience in their native environment in the human body, such as mechanical cues in the form of flow-induced shear stress,^[11,12] stretching,^[13,14] or biochemical stimuli.^[15–17] Key features of MPSs that are usually combined in a single system include: a) advanced 3D tissue or organ models, b) the presence of culture medium flow or perfusion, and c) fluidic interconnection and potential interaction of different organ models through the liquid phase:

• a)3D tissue models feature increased functionality and a more organotypic cellular microenvironment in comparison to conventional monolayer cultures.^[18] In many instances, e.g., for liver microtissues (MTs), several different cell types (hepatocytes, Kupffer cells, endothelial and stellate cells) that are needed for organ functions (e.g., metabolic functions) are combined in the right proportions to obtain a representative and functional 3D tissue model.^[4]
• b)Medium flow or perfusion is necessary to culture these 3D in vitro models in long-term (weeks) experiments and keep them viable. Perfusion or medium flow can be used to maintain a constant supply of nutrients and to dose compounds of interest.^[19,20] Flow is also needed to emulate and investigate more complex tissue processes in healthy and diseased states.^[21–23]
• c)Interconnection and the possibility to realize interactions of different organ models through the liquid phase are enabled by the use of microfluidic technology. Organ models can be arranged and interconnected in a physiologically relevant order, so that secreted and signaling molecules are exchanged between the different tissue types. Such arrangements enable to realize elaborate systemic in vitro models that yield results that can typically only be obtained through in vivo experi-ments.^[24–27] Dedicated fluid-routing schemes enable to recapitulate in vivo conditions as close as possible.^[28–31]

1.1. Transferable Organ Models

The formation and maturation of the different organ models in a single MPS is a major challenge. Culturing protocols will be different—more or less complex and specific—for various tissue types. This challenge can be overcome by using transferrable organ models that can be individually generated and matured off-chip, and subsequently be loaded into the platform (Figure 1A). The use of such transferrable organ models has several advantages including a) the possibility to mature delicate organ models that require specific medium compositions off-chip,^[32,33] b) the possibility to introduce quality control steps before the transfer to improve the stability of the system and reproducibility of the results,^[34–36] and c) the possibility to configure MPS assays with different organ combinations on demand.^[9,37,38]

Figure 1. Concept of an integrated recirculating microphysiological system (MPS) that can hold transferable biological cell models.

A) Transferable organ models, such as cell layers on transwells (TWs), spheroids, and organoids, can be individually formed and matured off-chip according to their specific maturation times. B) The open concept of the MPS allows for subsequent assembly of the final assay platform.

1.2. Integrated MPSs

To realize MPSs with transferrable organ models, a microfluidic platform needs to have open or accessible compartments for tissue loading and harvesting (transfer step from Figure 1A to Figure 1B). Functional units or modules of the MPS, such as channel structures, tissue compartments, and pumps for culture medium perfusion and circulation need to be integrated into an overall system design, which is suitable for use with different tissue models.

1.3. Recirculating Flow

Recirculation of culturing medium within the MPS enables the usage of less culture medium volume, which entails a lesser dilution of secreted metabolites and cytokines for system-wide tissue-tissue interaction (Figure 1B).

Seeing that broad multi-tissue MPSs and their applications have been covered in recent review articles,^[6,39,40] we will focus on integrated MPSs with recirculating flow, which are most suitable to study organ or tissue interactions and signaling processes. Moreover, we will focus on systems that allow to accommodate transferable organ models. Combining integrated MPSs with transferable organ models and recirculating flow supports the development of standardized human-on-chip platforms to address biological or biomedical questions^[41,42] and paves the way toward scalable assays and personalized-medicine applications.

This review includes three sections. In the first section, we will cover in vitro organ models that can be formed and matured off-chip before transfer into MPSs (see Figure 1A). Then, we will discuss integrated MPSs that allow for hosting transferable organ models and facilitate fluidic interconnection to enable interorgan crosstalk (Figure 1B). In the last section, we will discuss open technological and biological issues and future perspectives associated with integrated MPSs and recirculating flow.

2. Transferable In Vitro Organ Models

In the field of pharmaceutical and toxicological research, the selection of cells from an appropriate source is critical. a) In many laboratories, immortalized or transformed cell lines remain the standard choice due to their availability, transfectability, and ease of culturing.^[43] However, such cells accumulate genetic mutations over time, which may result in phenotypic and metabolic abnormalities, which may lead to poor recapitulation of in vivo physiology.^[43] b) More advanced cell culturing methods have enabled the use of patient-derived primary cells or adult stem cells (ASCs) to more faithfully represent healthy and diseased stages of the corresponding organs/tissues in vitro.^[44] Nevertheless, the arduous culturing and the limited possibility to expand such cells hitherto limit scientific and clinical applications.^[44] c) Alternatively, induced pluripotent stem cells (iPSCs)—cells, which are directly reprogrammed to a stem cell-like state from somatic donor tissues— open another route to overcome some of these drawbacks.^[45] Ultimately, the selection of the cell nature and source depends mostly on the research question that needs to be addressed and requires a compromise between biological authenticity, experimental reproducibility, maturation time, availability, cost, and throughput.

To use organ models with microfluidic systems in a “plug-and-play” manner, these organ models have to be transferrable from their formation platform to the MPS device (Figure 1A). In this review, we will focus on the few existing biological models that feature this transferability: a) biological barriers grown in transwell cell culture inserts and b) microtissues in the form of spheroids or organoids.

2.1. Transwell Models

The introduction of transwell cell culture inserts by Grob-stein^[46] led to the establishment of representative models of important endothelial barriers, e.g., skin,^[47–51] placenta,^[52,53] and the blood-brain barrier (BBB),^[54–56] as well as epithelial barriers, e.g., intestine,^[57–59] kidney,^[60,61] and lung.^[62–65] The porous membranes of transwell structures mimic the basal membranes and may assume a supportive function in cell polarization (Figure 2). The porosity of these membranes enables cells, that are cultured in the apical (ACs) and basolateral (BCs) compartments of the transwell, to communicate via soluble signaling molecules.^[66] The transferable format of the transwell unit enables access to apical and basolateral compartments and enables easy handling of the units that host the cell models for downstream analyses. These features make transwell models good candidates for MPS technology.

Figure 2.

Transwell-based cell models in a schematic view (upper panel) and in representative images (lower panels): (left) Hematoxylin Eosin (HE) staining of a diseased skin model with a multilayered human epidermal equivalent after stimulation with Th1 cytokines (scale bar = 100 µm). Reproduced under the terms and conditions of the Creative Commons Attribution 4.0 International License.^[47] Copyright 2017, Springer Nature; (middle) immunofluorescence staining of a human airway (lung) model with polarized cell layers consisting of ciliated cells (AcTub) that have been differentiated from human primary airway basal cells (CK5; scale bar = 20 µm). Reproduced under the terms and conditions of teh Creative Commons Attribution 4.0 International License.^[65] Copyright 2017, Springer Nature; (right) immunofluorescence staining of a biopsy-derived human intestine model with a polarized cell layer consisting of different cellular lineages. CD14 macrophages (red) and Escherichia coli ETEC (arrow heads) were added in basolateral (BCs) and apical compartments (ACs) to mimic a bacterial infection (scale bar = 20 pm). Reproduced under the terms and conditions of teh Creative Commons Attribution 4.0 International License.^[59] Copyright 2017, Springer Nature. Both skin and lung models were matured at an air-liquid interface.

2.1.1. Endothelial Barriers

Skin

In vitro 3D human skin or epidermal equivalents, cultured in transwell systems, have been extensively applied in investigative dermatology to study the physiology and pathophysiology of human skin (Figure 2, left).^[47–51] Different cell types, ranging from immortalized cell lines^[47,51] to primary cells,^[48–50] were selected to recapitulate relevant key parameters. As an example, the response to topical remedies of NativeSkin^[67] skin explants—originating from full-thickness skin biopsies—has been shown to reproduce that of human skin, enabling its use as a predictive model to test preclinical dermatocosmetic prod-ucts.^[49] Several skin models hold promise for drug screening, as they have been shown to develop inflammatory phenotypes when dosed with artificial cytokines,^[47] or when cocultivated with immune cells^[48,50] or bacterial pathogens.^[51]

Placenta

Human primary trophoblasts monocultures^[52] or human placental choriocarcinoma cells, cocultured with human placental venous endothelial cells^[53] in transwell systems, provided insights into placental glucose transport and translocation of model compounds and nanoparticles. Incorporation of these models into interconnected systems may offer an alternative method to predict the translocation of therapeutic compounds and may ultimately lead to the replacement of the “gold standard” method, a laborious and low-throughput ex vivo perfusion of human term placenta.^[53]

Blood–Brain Barrier

Another important endothelial barrier for both, academic research and pharmaceutical industry, is the blood-brain barrier.^[54] Initially, cell-line-derived endothelial monocultures were employed to elucidate transport properties of nonsteroidal anti-inflammatory drugs across the BBB.^[55] Further research introduced cocultures, where culture models consisting of primary brain microvascular endothelial cells, normal human astrocytes, and human blood monocytes were employed to inspect monocyte transmigration through the barrier in response to chemokines.^[56] As a next step, human iPSC-derived BBB cell types were introduced in monocultures or cocultures of three or four cell types and may be instrumental in establishing isogenic cultures or patient-cell cultures in the context of precision and personalized medicine in the future.^[54]

2.1.2. Epithelial Barriers

Intestine

Among many tissue models for ADMET (absorption, distribution, metabolism, elimination, and toxicology) research, the intestinal barrier—in the form of epithelial cell monolayers—has become widely used as a tool to assess intestinal permeability of orally administered drug candidates. Carcinoma cell lines (Caco-2, HT-29, etc.) are the most commonly used cell source for such models, owing to their large growth rate and easy maintenance. However, in many instances, these cell lines, grown in monocultures, do not reliably reflect in vivo permeability for hydrophilic compounds or compounds transported by carrier-mediated absorption.^[57] Cocultures of human enterocyte-like cells, mucus-secreting cells, and lymphocytes show a better correlation of insulin permeation characteristics with respect to their in vivo counterpart (Figure 2, right).^[57] Homeostatic and inflamed states of cell-line-derived intestinal barriers could be reconstituted using enterocyte-like-cell and macrophage cocultures in transwells.^[58] More recently, an ex vivo human “mini” intestine model was established from primary intestinal crypts and monocyte-derived macrophages. This system was shown to offer more faithful recapitulation of a) the human intestine, b) epithelial cell-macrophage interactions, and c) innate immune responses to enteric pathogens in comparison to immortalized cell-based systems.^[59]

Kidney

While intestinal models are mainly used in the context of drug absorption studies, kidney cell models can provide valuable information on drug-induced toxicity and excretion. Although few models currently exist, 3D kidney tissue constructs can be formed in transwells using patient-derived, immortalized human renal cortical epithelial cells,^[60] or human primary proximal tubular cells.^[61] Both transwell models were proven to recapitulate in vivo renal function better than existing 2D cell models.

Airway/Lung

Transwell systems offer the possibility to simulate the air-liquid interface of lung barrier tissues at a fairly broad range of complexity depending on the different combinations of cell types (Figure 2, middle). A coculture of bronchial epithelial cells, microvascular endothelial cells, and macrophages—all obtained through cell lines—could be used to reproduce the barrier properties for studying nanoparticle translocation across the alveolo-capillary barrier.^[62] A standardized set of culture conditions was proposed to reproducibly recapitulate alveolar epithelial morphology, phenotype, as well as fluid and ion transport by using cell lines.^[63] Under well-controlled culture conditions, transwell models including airway-derived cell lines also support studies of cell signaling pathways, basic cellular responses, and even pathomechanisms of airway diseases.^[64] Improvements have been implemented to establish transwell models for coculturing primary cells with innate immune cells, which offers a physiologically relevant platform to study neutrophil transmigration behavior in inflamed airways.^[65]

2.2. Microtissue Models

Microtissues are 3D cellular structures, which have been shown to be representative and functional in vitro organ models.^[68] MTs can be broadly divided into two distinct groups:^[18] a) Spheroids, which are cell aggregates, formed from a cell suspensions (Figure 3A) including single or multiple cell types, with cell lines, primary cells, or iPSCs as cell source, and b) organoids, which are self-organizing, highly complex structures (Figure 3B) that are formed from proliferating cells, mostly stem cells, that differentiate and grow.^[69] A main advantage of spheroid models is the highly uniform and reproducible, rapid generation through several established methods, such as the use of ultralow-adhesion plates, hanging drops, bioreactors,^[70] or bioprinting,^[71] which renders spheroid production scalable for high-throughput applications. Organoids, on the other hand, are promising models to study organ development, diseases, and the impact of compounds on cellular or tissue processes.^{[18,69,72,73]} While the generation of organoids, as reviewed elsewhere,^[74] can take several weeks and yields less uniform tissues, organoids represent highly organized structures with detailed organ-specific architectures.

Figure 3. Microtissue organ models for MPS applications in schematic views (upper panels) and in representative images (lower panels).

A) Multicellular spheroid models: (Left) The liver spheroid consists of only hepatocytes, which are stained for cytochrome P450 3A4 (green) and cell nuclei (blue; scale bar = 50 µm). Reproduced under the terms and conditions of the Creative Commons Attributions 4.0 International License.^[75] Copyright 2018, The Authors, Springer Nature. (Right) The heart spheroid has a mixed population of cardiomyocytes (green), fibroblasts (orange), and endothelial cells (red; scale bar = 50 µm). Reproduced with permission.^[76] Copyright 2017, Oxford University Press. B) Organoid models: (Left) A luminal intestinal organoid has been stained for cell nuclei (blue) and proliferating cells (red; scale bar = 50 pm). Reproduced with permission.^[77] Copyright 2009, Springer Nature. (Right) A section of a cerebral organoid with stainings for neural progenitor cells (red), neurons (green), and cell nuclei (blue; scale bar = 200 pm). Reproduced with permission.^[78] Copyright 2013, Springer Nature.

During the last decades, a wide range of microtissue models has been developed and used in several applications. In the following, we will present the most popular microtissue models and some of their applications.

Liver

The liver is the main metabolizing organ in the human body and, therefore, also of high potential interest to assess pharmaceutical side effects. The establishment and use of a representative in vitro liver model to assess the biotransformation of substances and their liver toxicity are of particular interest. The most widely used model consists of spheroids reaggregated from multiple liver-resident cell types and has been demonstrated to maintain hepatic functionality over several weeks in culture.^[79–81] Popular applications include a) compound metabolism studies,^[81] b) the onset and rescue of liver damage associated with metabolic disorders,^[75,82] or c) drug-induced liver injury (DILI).^[83–85] Recently, human liver organoids, formed from mature hepatocytes, were presented. These functional organoids allowed for long-term maintenance and expansion of primary hepatocytes. Proliferation of the hepatocytes was artificially triggered by liver damage signaling cues.^[86] Suitable and functional in vitro liver models could initiate a paradigm shift in drug development and regenerative medicine.

Heart

In vitro heart models are important, as the heart is another common organ suffering from off-target toxicity. Cardiac spheroids are commonly formed by differentiating iPSCs into mature cardiomyocytes that feature typical, spontaneous beating behavior. Models formed using iPSCs alone,^[87] or iPSCs in combination with endothelial cells, have recently been presented.^[88,89] In some cases, these models are also suitable for high-throughput production and applications.^[76,90]

Pancreas

3D models of the pancreas are less frequently reported. Nevertheless, the organ plays a crucial role in glucose homeostasis and, therefore, is involved in metabolic disorders such as diabetes. Efforts have been made to establish protocols for the uniform generation of functional insulin-secreting pancreatic islet spheroids using primary cells.^[91,92]

Adipose

Due to its central metabolic role in the regulation of energy homeostasis, adipose tissue is gaining more and more interest as an in vitro model in the context of metabolic disorders.^[93] Spheroid models, capable of accumulating and releasing fatty acids, have been formed with established cell lines^[94,95] and primary adipose-derived stem cells.^[95–97]

Embryo

Embryoid bodies (EBs) are spheroids formed with embryonic stem cells, which recapitulate many lineage-specific differentiation programs that are also found in the embryo.^[98] Thus, EBs are an established in vitro model to understand early lineage development^[99] and to assess embryo- and developmental toxicity.^[100] EB models are also suitable for high-throughput screening.^[101]

Brain

Depending on the application, two methods to form brain models exist: a) brain organoids have been established to study brain development and diseases;^[78,102,103] and b) reaggregated spheroids with spontaneous electrical activity have been established to assess neurotoxicity,^[104] to perform research in the field of neurodegenerative diseases,^[105] and to perform drug screening with diseased model systems.^[106,107]

Intestine

To study developmental processes and mechanisms of disease, intestine models have been mostly formed as organoids, since complex tissue structures are crucial for the organotypic function of the intestine.^[108] Mature organoids feature an enclosed luminal phenotype, with the polarized intestinal endothelial cell layer forming characteristic villi structures. The presence of tissue-specific stem cells in the adult organ enables the generation of these organoids directly from ASCs^[77] or iPSCs.^[109]

Kidney

Kidney organoids were presented by several research groups in the context of regenerative and personalized medi-cine,^[110] disease modeling,^[111] and developmental biology.^[112] Recently, vascularization could be introduced into a kidney model using a perfused setup, which promoted maturation of the organoids.^[113]

Cancer

Tumor spheroids are widely established and commonly used to screen compounds for their efficacy to kill cancer cells selectively. The drug response of tumor spheroids from several cell lines has been characterized.^[114] Tumor spheroids have been used to mimic solid^[114] and metastasizing tissue.^[115] In addition, efforts have been made to move toward patientspecific drug sensitivity testing by using patient-derived tumor tissue from dissociated cells^[116–118] or microdissections.^[119,120]

Less Established Microtissue Models

In addition to these well-established in vitro models, there are a number of less frequently reported transferrable organ models. Skin spheroids could be successfully formed and are suitable for high-throughput screening applications.^[121] Further, bone marrow spheroids could be formed from mesenchymal stem cells and are mostly used for regenerative medicine approaches.^[122,123]

3. Enabling Technology for Integrated Platforms

As mentioned in the introduction, crosstalk between in vitro organ models is mostly based on secreted soluble molecules.^[124] Therefore, the physiological microenvironment of individual biological tissue models and the fluidic connection between those are of paramount importance in any multi-organ system.^[125] There are numerous approaches to realize fluidic connections to achieve tissue-tissue communication, which can be categorized as a) static, b) single-pass, and c) recirculating approaches:

• a)Static systems feature a culture of multiple tissue models in proximity to each other, usually in the same well or compartment.^[40,126] However, this approach i) relies solely on diffusive transport of nutrients and metabolites, and ii) often leads to unwanted tissue fusion.
• b)Single-pass systems rely on active and passive transport of molecules of interest through controlled perfusion in fluidic channels.^{[12,127–130]} However, this approach usually i) limits tissue interaction to one direction, downstream, and ii) entails that analytes, produced by the tissue models, are continuously flushed out.
• c)Recirculating systems enable transport of metabolites back to the initial organ model (Figure 1B) and can be used to achieve continuous medium circulation. This approach also entails upconcentration of analytes and secreted factors over the duration of an experiment. Furthermore, the flow path can be adapted to the needs of the biological models.

Since the exchange of secreted molecules that facilitates interaction between organ models is an important feature of MPSs, single-pass flow-through systems are less popular, as they feature compound and molecule wash out over time. Therefore, we focus, within this review, on recirculating-flow architectures. Moreover, as described in the introduction, we center this review on integrated MPSs that host transferable cellular models described in the previous section as such systems are most suitable to study organ or tissue interactions and signaling processes.

We categorize the integrated recirculating systems according to the way in which recirculation is achieved: a) gravity-driven and b) peristaltic flow. The latter can be further split up according to usage of a) on-chip or b) off-chip pumping systems. Figure 4 shows existing integrated recirculating MPSs that host microtissues or transwells or both.

Figure 4. Integrated microphysiological systems (MPSs) hosting transferable cell models. The method to achieve flow recirculation has been indicated as gravity-driven (*), peristaltic pump off-chip (†), and peristaltic pump on-chip (‡).

A) A seven-organ MPS by Griffith and co-workers hosting transwell (TW) models, perfused by applying sophisticated physiological routing schemes and pneumatic on-chip pumps (scale bar ≈ 20 mm). Reproduced under the terms and conditions of the Creative Commons Attributions 4.0 International License.^[24] Copyright 2018, The Authors, published by Springer Nature. B) Plug-and-play organ system hosting TW models on electronically actuated micropumps have been developed by DRAPER (scale bar ≈ 30 mm). Reproduced with permission.^[37] Copyright 2017, The Royal Society of Chemistry. C) Hybrid system by Philip Morris hosting microtissues (MTs) and TW models that are connected to a reservoir plate and that can be perfused by a peristaltic pump (scale bar ≈ 20 mm). Reproduced with permission.^[131] Copyright 2018, The Royal Society of Chemistry. D) The TissUse hybrid MPS relies on pneumatic on-chip pumps and features four-cell model compartments (intestine (1), liver (2), skin (3), and kidney (4) tissue) in two connected recirculation circuits (yellow and magenta; scale bar ≈ 20 mm). Reproduced with permission.^[132] Copyright 2015, The Royal Society of Chemistry. E) MT platforms, developed at ETH, leverage gravity-driven flow for perfusion of ten individual MT standing-drop compartments (scale bar ≈ 10 mm). Reproduced with permission.^[34] Copyright 2018, SAGE Publications, Inc. F) Hanging-drop networks can be equipped with pneumatic pump drops that allow to realize perfusion on-chip. The MTs are then exposed to recirculating flow at the air-liquid interface (scale bar ≈ 20 mm). Reproduced with permission.^[133] Copyright 2015, The Royal Society of Chemistry.

3.1. Gravity-Driven Flow

Gravity-driven flow relies on hydrostatic pressure induced by a height difference between an inlet and an outlet reservoir of a device. In many cases, recirculating gravity-driven flow is achieved by tilting a device back and forth to induce bidirectional flow between two reservoirs that are connected by a channel. By changing the tilting angle or the aspect ratio and flow resistance of the channel, it is possible to modulate the flow rate.

Several approaches make use of gravity-driven flow to achieve multi-tissue interactions with different tissue configurations and flow paths.^{[16,134–137]} However, these systems do not allow for the use of transferable cellular tissue models. The integration of loading ports into gravity-driven flow systems enabled easy loading of preformed microtissues or organoids into the system and subsequent harvesting for endpoint analysis.^[138,139] Coculturing of liver and colon cancer spheroids in such an MPS, fabricated from polydimethylsiloxane (PDMS), demonstrated the cytotoxic effect of cyclophosphamide on cancer spheroids upon bio-activation in the liver.^[139] A second generation of these chips with ten microtissue standing-drop ports per channel was produced in polystyrene using injection molding to alleviate absorption of small molecules in the device material—a common problem with devices fabricated with PDMS (Figure 4E).^[34] Parameter modifications of the gravity-driven flow system enabled the realization of different physiologically relevant dosing curves.^1140] Integration of transferable transwell models into gravity-driven flow systems has not been shown, but could be achieved by modifying current closed platforms^[134] to host transwell systems instead of fixed membranes.

3.2. Peristaltic Flow

Generally, peristaltic flow relies on the coordinated contraction of subsequent segments of a fluidic channel to generate unidirectional flow.^[141] Recirculating peristaltic flow can be achieved either by using off-chip pumps—through a closed-loop connection with an external peristaltic or pneumatic pump—or by realizing integrated pumps or valve systems on-chip.

3.2.1. Off-Chip Pump

Mahler et al. introduced one of the first silicon-based MPSs, which combined intestinal models in transwell inserts with 2D liver cultures to model acetaminophen metabolism.^[142] Using an external peristaltic pump, the apical to basolateral flow through the intestinal module was connected to a liver module and a reservoir/debubbler via tubing to achieve medium recirculation. A more recent example of MPS that uses off-chip peristaltic flow included coculturing of liver microtissues with lung tissue in a transwell configuration (Figure 4C).^[131] The transwell was only perfused on the basolateral side to create an air-liquid interface on the apical side of the transwell, which was needed for full maturation of the lung model.

3.2.2. On-Chip Pump

Microtissue-hosting hanging-drop networks^[129]—initially single-pass systems—can be rendered recirculating systems by integrating hanging-drop pneumatic pumps.^[133] Pneumatic chambers integrated with dedicated pumping drops expand and increase the Laplace pressure in the drop, which subsequently induces flow (Figure 4F). Engineering of a soft plug that increases the hydraulic resistance on one side of the pumping drops generated unidirectional, recirculating flow.^[133]

On-chip pumping with a pneumatic pump that allowed for use of transferrable tissue models in transwell inserts was developed by the Marx group.^[143] The initial application focused on the connection of liver microtissues and skin biopsies. Using the transwell insert for the skin model enabled to include an air-liquid interface into MPS so that liver-skin interaction over 4 weeks could be demonstrated.^[35,143] Furthermore, Marx and co-workers used this PDMS-based platform to investigate interactions between a) liver and intestine tissue under chronic trogl itazone exposure,^[35] b) liver and pancreatic islets to model type II diabetes,^[144] and c) liver and neurospheres for safety and efficacy applications.^[145] Another version of the chip enabled the cultivation of four different organ models on-chip (Figure 4D).^[132] Two recirculation circuits enabled the implementation of an excretory pathway featuring kidney tissue connected to the systemic circuit, which hosted a) intestine, liver, and skin,^[132] or b) liver, pancreatic islets, and skeletal muscle.^[146] The function and viability of the individual tissues were maintained over a period of 28 days. Therefore, these long-term coculture systems enable long-term investigations to study ADME profiling and systemic toxicity upon repeated compound dosing.

Griffith and co-workers presented integrated systems with on-chip pneumatic pumps and transferrable liver tissues several years ago.^[147] By machining the chips out of polysulfone (PSF), drug absorption in the system was minimized, while an open platform configuration enabled to host transferrable organ modules and allowed for online media sampling. Multi-tissue configurations were achieved by connecting the liver compartment with an intestinal transwell insert to investigate liver-intestine crosstalk.^[25,148] Subsequent versions with four, seven, and ten organs were developed that allowed to precisely control intra- and intertissue flow as well as drug distribution in the system (Figure 4A).^[24] The four-organ chip included liver, intestine, endometrium, and lung that were maintained for 2 weeks as evidenced through phenotypic markers. Pancreas, heart, and brain were added for the seven-organ platform and muscle, skin, and kidney for the ten-organ platform. Viability and phenotypic functions were demonstrated during 3 weeks (seven-organ system) or 4 weeks (ten-organ system). In addition, the seven- and ten-organ platforms were used for pharmacokinetic analysis of diclofenac metabolism using mathematical models.^[149,150]

An alternative to pneumatic pumps are electrically actuated micropumps that can be integrated on chip.^[37] A sophisticated fluidic control scheme, developed by DRAPER, enabled flexible plug-and-play systems to be assembled on the platform (Figure 4B). Perfusion was used to realize recirculation within individual tissue chambers, but also to interconnect different tissue models. Liver and airway models were cultured together on a single board, and their functionality was preserved over a period of 2 weeks.^[37] Furthermore, the female reproductive tract and three surrounding organs (ovary, fallopian tube, uterus, cervix, and liver) were integrated.^[148] This in vitro tool enabled the study of hormonal signaling in the menstrual cycle and the realization of a pregnancy-like endocrine loop.

4. Challenges and Open Issues

4.1. Technical Implementation Challenges

Even though the advantages of integrated recirculating MPSs hosting transferable organ models over organ model monocultures are evident, we found that only few currently published or commercialized devices or systems exhibit such features (Table 1). The research groups working to develop flexible integrated MPSs have found ways to deal with technical implementation challenges concerning: a) culturing chamber accessibility, b) recirculation methods, and c) compromising between various technical aspects.

Table 1.

Overview of integrated recirculating MPSs that allow to host transferable cell models. The mode of action for recirculation in the MPS is indicated by a star for gravity-driven (*) and cross or double-cross for off-chip (†) and on-chip (‡) peristaltic flow.

Platform	MPS Material	Cell Model Format	Readout	Ref.
MIT‡ a)	Polysulfone	Transwell	Supernatant-based assays; optical (only off-chip); cell removal for fixing or lysis-based assays	[24,25,36]
DRAPER‡ b)	Polyetherimide	Transwell	Supernatant-based assays; optical (only off-chip); cell removal for fixing or lysis-based assays	[37,148]
Philip Morris†	Polyetherether-ketone	Microtissue and Transwell	Supernatant-based assays; optical (only off-chip); cell removal for fixing or lysis-based assays	[131]
TU Berlin/TissUse‡ c)	PDMS and polycarbonate	Microtissue and Transwell	Supernatant-based assays; optical; cell removal for fixing or lysis-based assays	[35,132,143–145]
ETH* d)	PDMS or polystyrene	Microtissue	Supernatant-based assays; optical; cell removal for fixing or lysis-based assays	[34,133,139,151]

Platforms are commercially available from

^a)CN Bio Innovation Ltd.

^b)Charles Stark Draper Laboratory, Inc.

^c)TissUse GmbH

^d)InSphero AG.

4.1.1. Culturing Chamber Accessibility

Accessibility to the culturing chambers is the main technical hurdle when developing new integrated recirculating MPSs that allow for accommodating transferable cellular models. The obvious method used to guarantee such access is open fluidics (Figure 5). Access ports can be small—leading to standing (Figure 5A) or hanging drops (Figure 5B)—or large—enabling transwell compatibility (Figure 5C). However, this increased accessibility with an air-liquid interface comes at a price: a) higher risk of bacterial and fungal contamination, b) devicespecific drop instability, and c) increased evaporation.

Figure 5. Culturing chamber accessibility to enable the use of transferable organ models.

If the access port to the culturing chamber is small, it leads to A) a standing drop or B) a hanging drop by flipping the device. C) If the access port is large, it enables the use of transwells. A pipette tip, used for microtissue transfer, is drawn to scale. D) Top left: The air-liquid interface at the access port is unstable if gravitational forces (red arrow) are greater than surface tension forces (black arrow). Bottom left: This interface can also be unstable if hydrodynamic forces (blue arrow) due to pumping exceed the combined gravitational and surface tension forces. The interface stability can be ensured by carefully designing access ports in such a way that their surface tension counteracts gravity (top right) or by using lower flow rates (bottom right).

• a)Higher risk of contamination is due to increased chamber access. Bacterial and fungal contaminants have an increased chance to enter either passively with airborne contaminants or actively with the biological model transfer steps. To solve this problem, it is necessary to manipulate the transferable biological models and the MPSs carefully in a sterile environment, while minimizing the number of transfer steps.
• b)Device-specific drop instability (Figure 5D) is due to liquid surface tension being overcome by gravitational or hydrodynamic forces that may lead to device leakage. To solve this problem, it is possible to increase the surface tension forces on drops by carefully designing access ports to cope with the effect of gravity or flow.^[152]
• c)Increased evaporation is due to the high surface-to-volume ratio associated with open microfluidics. Excessive evaporation leads to molecule upconcentration (e.g., nutrients, metabolites, compounds) and drying out of the device.^[153] Researchers have proposed pragmatic solutions to alleviate this problem by sealing the device externally to reduce air exchange at air-liquid interfaces^[34] or by incorporating humidity sources i) integrated in the device or its holder,^[24] ii) in liquid-filled well-plates stacked under the measurement device,^[34] or iii) in liquid-soaked hydrogels or pads.^[154]

4.1.2. Recirculation Techniques

Another technical challenge is the way in which and the means with which medium recirculation is achieved. Two approaches have been reported: a bidirectional gravity-driven flow or a unidirectional peristaltic flow.

The main challenges related to gravity-driven flow result from the bidirectionality of the flow. a) Bidirectional flow means that there is no control over the order in which organ models, hosted in the MPS, are exposed to compounds or metabolites.^[125] Organ order can be crucial for certain applications like pharmacokinetic modelling,^[22,155] compound first-pass assays,^[21,26,127] or devices mimicking placenta-body interactions.^[53] b) Moreover, problems arise when trying to culture organ models that require a unidirectional flow to achieve cell polarization (e.g., intestinal epithelium).^[156]

The main challenges related to peristaltic flow result from the bulkiness of the pumping unit. In order to actuate pneumatic pumps, an external pressure or vacuum source is required.^[36,37,143] Additionally, an external controller is required to control the pumping scheme and applied pressures, which adds to the setup complexity.

By breaking the bidirectional flow symmetry of the microfluidic channels during tilting, a novel gravity-driven system has achieved unidirectional flow with a rocking motion.^[156,157] Flow symmetry was broken by staggering two channels perpendicularly to the tilting axis at both device reservoirs and by exploiting surface tension in a way that liquid flows through one of the channels upon tilting the device to one side and back through the other channel upon tilting the device back to the other side. However, this approach has yet to show compatibility with transferable cellular models. Nevertheless, if investigated, this could be an option to combine the advantages of both recirculation approaches in a single experimental setup.

4.1.3. MPSs: A World of Compromise

Developing new MPSs or deciding which MPS to use for your application will lead you to realize that the perfect in vitro device will not exist and that a balance of compromises is required. Table 2 establishes the advantages and disadvantages that result from key MPS design choices. Ultimately, knowing the caveats and features of the specific MPS that was selected is important to accurately interpret the results of biological experiments.

Table 2. MPS design compromises and their associated advantages and disadvantages.

Choices		Advantages	Disadvantages
Material	Flexible	Enables the realization of pneumatic pumps and mechanical stretching Allows ample oxygenation[13,133]	Usually absorbs small molecules[13,133,139]
	Rigid	Helps to preserve molecule concentrations in the liquid phase[34,131]	No passive oxygenation[34]
Organ model chamber	General-purpose design	Flexible and hosts most simplified organ models (subject of this review)	Unable to sustain more complex or sensitive organ models
	Specific design	Elaborate chamber structures tailored to a specific organ model	Single-purpose chambers
Fluidic interconnection	Single channel	Straightforward design that enables sequential organ model connection[34]	No multiplexing
	Channel network	Enables complex organ interconnection and multiplexing[24,37,134]	Elaborate design requiring extensive understanding of microfluidics
Flow actuation	Gravity-driven	Enables stacking of multiple well-plates for parallelization[34]	No single-chamber control—gravity acts indiscriminately on the entire stack
	Pneumatic pumps	Enables individual flow control of specific chambers or subunits[24,27]	Bulky setup requiring external pressure sources and controllers[24,27,132]

4.2. Biological Workflow Challenges

Combining multiple organ models on a single integrated microphysiological platform is not a trivial task. The main steps—each involving a varying set of challenges—include:

• a)Establishing of transferable biological models: It entails finding a way to emulate organ function with a single microtissue or transwell model.
• b)Implementing new organ models: It entails finding optimal protocols to reach model maturity and keep it viable.
• c)Synchronizing maturity protocols: It entails establishing proper timing to conduct a multi-organ experiment.
• d)Culturing of multiple organs in a common medium: It entails finding the right culture environment to sustain several different tissue types in a multi-organ experiment over weeks.

4.2.1. Establishing Models

Considering the body-on-a-chip field is in its infancy, many of its critics challenge the physiological relevance and benefit of using such microscale organ models.^[158] Therefore, when establishing transferable biological models, it is critical to ensure their organotypic behavior, particularly when compared to established cell culture models. However, achieving all organotypic functions within an organ model is a major challenge.^[27] Additionally, some models may be unstable over long periods, which makes their long-term culturing impossible.^[159] Two alternative strategies share the goal to recapitulate key in vivo organ functions and architecture while preserving model transferability: a) reproducing in vivo cell and tissue arrangement and heterogeneity with coatings or layering in a transwell support,^[4,160] or b) reproducing in vivo passively generated nutrient gradients and cell arrangement within 3D mono- or cocultures of cells in the form of microtissues.^[161–163]

With regard to the second approach, the passive gradients in microtissues are sometimes used as a point of criticism against their use. However, we argue that they are necessary to accurately recapitulate in vivo conditions of certain cells where the distance to the nearest nutrient source—e.g., blood vessel—can exceed 150 µm.^[164]

Ultimately, although many organ models have already been established, some—bone, skeletal muscle, vasculature, and reproductive glands—are yet to be recapitulated in a microtissue or transwell format. As they do not yet exist as transferable organ models, they are not compatible with the integrated MPSs, the subject of this review, so that there would be research opportunities to introduce them.

4.2.2. Implementing Models

When implementing transferable biological models, the set of challenges varies drastically depending on the type of model— either barrier models in transwells or microtissues such as spheroids and organoids. Cell sourcing for these tissue models is crucial: we distinguish between mono- and cocultures using a) cell lines, b) primary cells, and c) iPSCs. Primary cells can be further broken down as being derived from either healthy donors and patients, or adult stem cells. Table 3 presents the implementation challenges that result from the choice of cell sourcing.

Table 3. Challenges arising from various cell sources: cell lines, healthy and patient-derived primary cells, adult stem cells (ASCs), or induced pluripotent stem cells (iPSCs).

Cell lines	Primary cells		iPSC-derived cells
	Donor and patient-derived	Adult stem cells (ASC)
Overdependence: Cell lines are widely used for compound validation but their biological relevance is rarely questioned.[165] Authenticity: Genetic instability of the cells in long-term culture[166]	Limited availability Expensive Bad reproducibility: Donor-to-donor variability[167] Authenticity: Instability of genetic expression in long-term cultures[117]	Limited to certain organs: Intestine, skin, bone marrow Similar challenges as iPSC-derived cells and other primary cell lines	Generation: Low efficiency[168] Maintenance: Complicated protocols Authenticity: Questionable due to genetic or epigenetic modifications required for iPSC reprogramming[169] Diseased models: Generation of diseased iPSCs is challenging from technical point of view[170,171] Heterogeneity: Difficult to achieve strict monocultures when differentiating[172] Maturity: Most iPSC-derived cells maintain embryonic phenotype in culture[172]

4.2.3. Synchronizing Maturity Protocols

One of the most attractive features of integrated recirculating MPSs is the versatility in applying tissue maturation protocols (Figure 1A). Organ models can be formed and matured either in the MPSs themselves,^[13] or brought to specific maturation stages outside of the MPSs and subsequently transferred into it.^[24] Once key protocols enabling organ model preparation have been defined, maturity timings must be considered carefully. This maturity synchronization to initiate experiments is a challenge that is more difficult to tune; the more organ models are combined.

Extreme examples, going from 4- to 13-organ MPSs, demonstrate the tediousness of having to preculture the various tissues to attain organ maturity so that they can be transferred into the interconnected MPS.^[24,132,173] In the extreme case of a ten-organ body-on-a-chip system, maturity times varied from 48 h to 21 days.^[24]

4.2.4. Culturing of Multiple Organs in a Common Medium

Once transferable organ models have been established and are ready to be implemented, and once maturing schemes have been optimized, the conditions under which the organ models are cultured in a common medium over long periods become crucial. Many researchers have noted the challenge of developing new common medium formulations—also known as blood surrogates^[41]—for microphysiological systems.^[27,174,175] However, few practical solutions are available to date. A big challenge arises from the fact that the optimal combination of nutrition, growth factors, serum, and drugs that sustains one organ model may alter the functionality of another.^[27,176] Examples include:

• a)liver culture medium contains insulin that needs to be removed upon coculturing liver microtissues with pancreatic-islet microtissues^[144]
• b)high glucose concentrations in cancer microtissue medium perturb pancreatic islet insulin release^[177]
• c)we have observed in our laboratory that fetal bovine serum (FBS) is necessary for proper differentiation of stem cells in cocultures of liver microtissues and embryonic-stem-cell-derived microtissues but does not support liver function in the long term when added in large proportions (J. Boos et al., publication in preparation).

As more organ models are interconnected, medium composition needs to be carefully revisited. Decades of research have led to viable options,^[178–182] including: a) mixing multiple media types at different ratios,^[25,183] b) gradually introducing a common medium in monoculture medium,^[24,184] or c) elaborating new media from scratch.^[185,186] However, the methods to find optimal media are not standardized and associated efforts and success rates are rarely reported on.

Another option is to interconnect organ models indirectly through a secondary channel system. Each organ model hosted in a culturing chamber can be sustained with its specific monoculture medium through an apical flow or periodic medium replenishment. Molecules secreted by an organ model can then diffuse through a semipermeable membrane placed between the culturing chamber and the secondary channel. These molecules are subsequently transported to another organ model via a basal medium flow in the secondary channel.^{[24,41,176,187,188]}

Moreover, as the liver generates many secreted factors necessary to sustain other organ types, adding it to an MPS is beneficial. In turn, adding the liver enables the use of simplified medium formulations.^[189,190]

Ultimately, proponents of the “human-on-a-chip” paradigm envision that whole blood could one day be used in lieu of cell culture medium.^[42,191]

5. Conclusion

The use of integrated recirculating MPS—as a tool to realize “body-on-a-chip” configurations—entails a paradigm shift in how researchers approach in vitro models. Historically, the focus in fundamental biological research has been on singleorgan models sustained with specialized systems and specific medium formulations. The thrust to combine various organ models in a single system leads to challenges, because every organ and organ combination may require a specialized system.

A suitable option to cope with some of the issues is to assemble transferable organ models—i.e., microtissue and transwell inserts—in open microfluidic systems. This allows for off-chip formation and maturation of organ models as well as for their recovery from the system for endpoint analysis, an aspect, which is especially important with the use of more complex cell systems like organoids sourced from primary cells or iPSCs. Furthermore, the use of transferable models enables quality controls, which entail a higher reproducibility, standardization, and scalability of the assays.^[38,124]

To develop personalized MPSs for precision medicine, microdissected tissues from biopsies^[119] or donor/patient cells, reprogrammed into iPSCs,^{[45,192–194]} could be used as a starting point to build patient-derived and patient-specific tissue and disease models of interest. Using patient samples as the base for organ models allows for investigating human diseases in a more systemic way. Such diagnostic MPSs could be used to a) find drug dosing regimens, b) test compound efficacy and safety, or c) investigate disease mechanisms and conduct “clinical trials” on chip.^[195–197]

Real added value is generated only if obtained results can also be translated into the clinic. Therefore, investing in sophisticated in vitro-in vivo translational tools is of paramount importance.^{[15,124,149,150,174,198]} These translational tools rely mostly on data gained from supernatant-based assays. Thus, developing platforms with integrated sensors^[199] leverages the full potential of MPSs by enabling real-time monitoring of individual microtissues or organoids—generating data that could be used for translational models. Efforts to combine the development of flexible multipurpose MPSs with extensive modeling to extrapolate to conditions in the human body are underway and will hopefully facilitate the translation from bench to bedside.

Biographies

Kasper Renggli received his B.Sc. and M.Sc. in nano-sciences from the University of Basel, Switzerland, and in 2013, he defended his Ph.D. thesis in the Department of Chemistry. Subsequently, he received a Swiss National Science Foundation postdoctoral fellowship to join the DARPA human microphysiological organ systems program at the Massachusetts Institute of Technology (USA) within the group of Linda Griffith. Since 2016, he has been holding a junior group leader position at ETH Zürich supervised by Andreas Hierlemann, developing microphysiological systems to investigate tissue and disease models for personalized medicine approaches.

Andreas Hierlemann completed his college education in chemistry at the University of Tübingen, Germany, and was awarded a Ph.D. degree in 1996. He then held postdoctoral positions at Texas A&M University, College Station, TX, USA, in 1997, and at Sandia National Laboratories, Albuquerque, NM, USA, in 1998. In 1999, he joined the Department of Physics, ETH Zurich, Switzerland, where he was appointed Associate Professor in June 2004. In April 2008, he became a Full Professor in the Department of Biosystems Science and Engineering (BSSE), ETH Zurich, Basel. His research interests include the development and application of microsensor, microfluidic, and microelectronic technologies to address questions in biology and medicine with applications in the fields of systems biology, drug testing, personalized medicine, and neuroscience. For details, see https://www.bsse.ethz.ch/bel/.