Research ArticleDRUG TESTING

Multi-organ system for the evaluation of efficacy and off-target toxicity of anticancer therapeutics

See allHide authors and affiliations

Science Translational Medicine  19 Jun 2019:
Vol. 11, Issue 497, eaav1386
DOI: 10.1126/scitranslmed.aav1386

Testing for toxicity

Standard methods of in vitro preclinical drug testing can be useful to detect cytotoxicity and evaluate safety pharmacology but fail to capture effects of drug metabolites (including functional effects) that become apparent at the organ systems level. McAleer et al. constructed an in vitro system using multiple human cancer and healthy cell types from different organs and recirculating serum-free medium. Within the system, anticancer drugs including tamoxifen, diclofenac, imatinib, and verapamil demonstrated variable on-target and off-target effects, some of which were dependent on drug metabolism by liver cells. This study supports the use of simple yet versatile multi-organ cell-based systems for efficient preclinical drug testing.

Abstract

A pumpless, reconfigurable, multi-organ–on–a–chip system containing recirculating serum-free medium can be used to predict preclinical on-target efficacy, metabolic conversion, and measurement of off-target toxicity of drugs using functional biological microelectromechanical systems. In the first configuration of the system, primary human hepatocytes were cultured with two cancer-derived human bone marrow cell lines for antileukemia drug analysis in which diclofenac and imatinib demonstrated a cytostatic effect on bone marrow cancer proliferation. Liver viability was not affected by imatinib; however, diclofenac reduced liver viability by 30%. The second configuration housed a multidrug-resistant vulva cancer line, a non–multidrug-resistant breast cancer line, primary hepatocytes, and induced pluripotent stem cell–derived cardiomyocytes. Tamoxifen reduced viability of the breast cancer cells only after metabolite generation but did not affect the vulva cancer cells except when coadministered with verapamil, a permeability glycoprotein inhibitor. Both tamoxifen alone and coadministration with verapamil produced off-target cardiac effects as indicated by a reduction of contractile force, beat frequency, and conduction velocity but did not affect viability. These systems demonstrate the utility of a human cell–based in vitro culture system to evaluate both on-target efficacy and off-target toxicity for parent drugs and their metabolites; these systems can augment and reduce the use of animals and increase the efficiency of drug evaluations in preclinical studies.

INTRODUCTION

The current drug development process is inefficient, taking years from target compound identification to marketable drug, and costs up to $2 billion during the process. This inefficiency stems primarily from a very high failure rate of novel therapeutics during animal testing and clinical trials (13) in addition to ever-increasing complexity regarding regulatory requirements, high competition, and “exhaustion” of conventional drug target space. The considerable attrition rate of drug candidates at all stages of development arises from the poor predictive nature of preclinical models for efficacy and toxicity, especially the inability to translate efficacy between preclinical and clinical situations. Current human-based in vitro toxicity studies are relatively efficient at screening out drug-specific end points such as genotoxicity, safety pharmacology, and cytotoxicity to organ cells. However, they may not cover end points that depend on several organs such as the bioactivation of a compound to elicit pharmacological or toxic effects on another tissue. Specifically for cardiomyocytes, current in vitro systems have limited capacity to predict functional changes (4).

Advances in biomedical engineering over the last decade have led to the development of body-on-a-chip (BoaC) or human-on-a-chip multi-organ systems. These devices aim to integrate engineering principles with cell biology, allowing researchers to study compound bioactivation and efficacy, a precursor to a fully integrated pharmacokinetic (PK) and pharmacodynamic (PD) profile, in multiple organs simultaneously (5, 6). Recent BoaC systems with physiological readouts attempt to model the PK/PD relationship using human cells, potentially reducing some of the inefficiencies of drug compound development (7, 8); for example, Edington et al. (9) has interconnected up to 10 organs. Although this preliminary work builds a foundation for interconnected systems, they are limited and do not measure compound efficacy and toxicity simultaneously using functional measurements. The integration of multi-organ systems with cancer models could enable the determination of efficacy and off-target toxicity effects in the same system, for both parent drug and its metabolites, to determine therapeutic index.

Cancer encompasses a large group of diseases involving dysregulated cell growth and proliferation and can affect any body tissue. Complicating the field are the diverse mechanisms tumorigenic cells use to drive uncontrolled growth. Further complicating the issues are an array of cancers that express multidrug resistance protein 1 (MDR1), also known as permeability glycoprotein 1 (Pgp1). This protein is important in translocating foreign, often toxic, substances out of the cell. In normal tissues such as the liver, kidney, and blood-brain barrier, Pgp1 pumps xenobiotics back into the blood stream: in the intestine, for example, from the epithelium into the gut lumen. When expressed in cancer cells, it may reduce a therapeutic compound’s effectiveness by translocating chemotherapeutics out of the cell, contributing to multidrug resistance.

Initially, organ-on-a-chip systems were designed for specific applications with limited ability for reconfiguration and typically with cells from a single organ. Consequently, each system required a complicated and time-consuming development process to optimize both the culture conditions and engineering parameters. To address these issues, a reconfigurable BoaC system was developed with the capacity to house multiple organ-like tissue constructs grown on an array of biological microelectromechanical systems (bioMEMS) modules in a single recirculating serum-free medium (7). The system combines mature organ chips and functional organ modules that were initially cultured independently under medium conditions promoting optimal development for each module and then assembled in the system when functionally mature.

Consequently, adapting the system is as simple as inserting the organ chips of interest in different compartments in an integrated system. Here, we show that our system supported organ-organ communication for liver-generated drug metabolites with target tissues through the gravity-driven, rocker-based, pumpless recirculating medium, and these systems can be modeled for bioactivation and efficacy. Gravity-driven rocker platforms offer distinct advantages over pump-driven systems including (i) lower medium volume requirements, which allows for the generation of appreciable concentrations of metabolites to observe real-time effects on other organs in the system; (ii) reduced cavitation-induced bubble formation generated by pumps that can block microfluidic flow; and (iii) increased scalability in those multiple systems (up to ~20 can be stacked on a single rocker within an incubator). In system configuration 1, two bone marrow components were incorporated with liver tissue to measure the cytostatic effects of two anticancer drugs on bone marrow–derived cells and off-target effects on the liver. In system configuration 2, one multidrug-resistant (MDR+) vulva cancer cell line and one non–multidrug-resistent (MDR) breast cancer cell line were incorporated into the system with a liver compartment to determine metabolic effects and with functional cardiac models to measure electrical and mechanical deficits from off-target toxicity. For this study, the systems were used to investigate the efficacy, metabolism, and off-target organ effects of the anticancer small molecules imatinib and diclofenac (system 1) and tamoxifen (system 2) coadministered with the MDR1 inhibitor verapamil.

Imatinib is a first generation BCR-ABL tyrosine kinase inhibitor used to treat chronic myelogenous leukemia (10, 11) with high efficacy and little to no toxicity in other organ systems and has resulted in about 83% event-free survival and 93% freedom from progression to accelerated or blast phase in patients at 6 years after treatment (12). The nonsteroidal anti-inflammatory drug diclofenac inhibits tumor cell proliferation potentially through increasing cellular reactive oxygen species by inhibition of superoxide dismutase (13). It has also been shown to cause some liver toxicity (14).

Tamoxifen, discovered in the 1960s, has been used as an effective treatment for estrogen receptor–positive (ER+) breast cancer (15) and is a prodrug compound with little affinity for the ER. Therefore, tamoxifen needs to be metabolized in the liver into the high-affinity metabolites 4-hydroxytamoxifen and N-desmethyl-4-hydroxytamoxifen (endoxifen) by the cytochrome P450 (CYP) isoforms 2D6 and 3A4 for maximum efficacy (16). 4-Hydroxytamoxifen and endoxifen competitively bind to ERs, creating a complex that inhibits DNA synthesis and induces cell cycle arrest in the G0 and G1 phases (1618). At doses associated with clinical breast cancer treatment, tamoxifen shows few adverse cardiac effects and has indicated potential as an adjuvant therapy for reduction of cholesterol and high-density lipoprotein/low-density lipoprotein (19). However, when dosed at the high concentrations (~10 μM in plasma) necessary to treat MDR+ cancers, tamoxifen adversely affects Ca2+ and K+ handling in cardiac cells and causes the functional deficit QT elongation and torsade de pointes syndrome (2022). Verapamil, a voltage-gated calcium channel blocker used to treat high blood pressure and cardiac arrhythmias, also functions as an MDR1 transport inhibitor when administered in the micromolar concentration range (23). Verapamil is also a known inhibitor of the human ether-a-go-go (hERG) (Kv 11.1) channel (24). Coadministration of verapamil with anticancer therapeutics such as paclitaxel improved the PK leading to greater drug efficacy in patients with breast cancer (25). Therefore, verapamil increases the efficacy of anticancer compounds in multidrug-resistant tumors but could cause cardiotoxicity at higher doses (24).

The array of cell types chosen highlights the adaptable nature of these in vitro systems while simultaneously displaying its accuracy in predicting target efficacy and off-target functional toxicity of standard anticancer therapeutics. Here, we show that this system can be used to measure the effects of both anticancer parent drugs and metabolites on multiple cancer subtypes in response to in vitro liver metabolism in a recirculating serum-free medium. The integration of two-dimensional (2D) cultures with bioMEMS devices created hybrid 3D systems that focus on the recreation of function rather than anatomy.

RESULTS

System design, fabrication, assembly, and simulation

The BoaC microfluidic device was designed using Autodesk Inventor and with experimentally supported computational fluid dynamics (CFD) modeling. The outer materials were composed of polymethyl methacrylate that was laser etched to define the location of chips and access ports for functional readouts and medium removal and replacement. These outer supporting materials were sealed by two poly(dimethylsiloxane) (PDMS) membranes that were laser etched to hold the bioMEMS chips in place and to define the microfluidic channels. Cardiac cells were plated and cultured separately on the electrical and mechanical devices to allow for maturation in their optimal medium (table S1) before being assembled into the device. Liver cells were plated on collagen-coated glass coverslips in a medium that maintained full activity (table S2) before assembly. Cancer cells were thawed and expanded in T75 flasks before being embedded in hydrogels or plated on extracellular matrix (ECM)–coated glass coverslips. The individual components were assembled within the device (Fig. 1) using a custom serum-free medium formulation (table S3), and cells maintained viability for a minimum of 14 days (fig. S1). The orientation and location of the organ components were designed such that drugs would initially pass over the liver to mimic first pass metabolism before moving on to the cancer and cardiac tissues, where efficacy and off-target toxicity were monitored. The liver was scaled to the total volume of the system to allow metabolites to be generated at physiologically relevant concentrations.

Fig. 1 Five-chamber reconfigurable multi-organ system.

(A) Photograph of the multi-organ system filled with green-colored dye for visualization. Scale bar, 2 cm. (B) Exploded view schematic representation of the platform assembly and design used in the system 2 study of tamoxifen. Chamber 1 houses hepatocytes on coverslips. Chambers 2 and 4 are cardiac cantilevers and microelectrode arrays (MEAs), respectively. Chambers 3 and 5 are for cancer cells SW-962 and MCF-7. Drugs were applied to medium access port A and initially passed over the liver to mimic aspects of first pass metabolism.

The information gained from modeling each configuration using the rocking speed and amplitude, individual tissue chamber depth, and other geometrical parameters was used to determine physiologically relevant shear stresses, fluid residence times, and flow rates (7, 26, 27). At full tilt under the designated 1°, 1 oscillation per minute rocking parameters, the maximum flow rate and sheer stress within the system were 227 μl/min and 0.055 dyne/cm2, respectively. Because of the customizable nature of these systems, they can easily be assembled with and without the incorporation of liver tissue to examine the effects of metabolism of the test compounds on the function of other tissues as in system 2.

System 1: Bone marrow proliferation system

The system was configured for use with two types of bone marrow–derived cells (Kasumi-1 myeloblasts and MEG-01 megakaryocytes) for the two cytostatic cancer drugs imatinib and diclofenac on their proliferation. On the basis of the configuration of the system (Fig. 1), the liver component was placed in chamber 1, and MEG-01 and Kasumi cells were placed in compartments 4 and 5, respectively. The plating densities for both the Kasumi-1 and MEG-01 cells were optimized to provide a linear proliferation rate over a 7- or 14-day time course for ideal measurements of an acute application (7 days) or for chronic or repeat application (14 days) for future studies (Fig. 2). Imatinib, diclofenac, or carrier control dosing solutions were placed in the portal chamber closest to the liver and allowed to incubate for 24 hours under recirculation. After 24 hours, 30% of the medium was removed every 24 hours for 3 days for replenishment of nutrients and to mimic elimination of the drug compound. The cell density was measured preapplication and before every feeding for 72 hours. The cytostatic effect of imatinib and diclofenac was evident for both MEG-01 and Kasumi-1 cells (Fig. 3), with control-treated cells in systems proliferating unabated, whereas imatinib- or diclofenac-treated cells exhibited arrested proliferation. The bone marrow–derived cell-based configuration of the multi-organ system effectively demonstrated the cytostatic effects on cancer cells observed clinically (2830).

Fig. 2 Kasumi-1 and megakaryocyte (MEG-01) proliferation over 14 days.

(A) Representative phase contrast images of Kasumi-1 and megakaryocytes growing inside the multi-organ system chamber on days in vitro (DIV) 1, 7, and 14. (B) Quantification of proliferation, represented as both a linear fold change and a log-transformed change in cell number over time. Error bars are represented as SD of at least six replicates across a minimum of three biological replicates.

Fig. 3 Effect of imatinib and diclofenac on proliferative capability of megakaryocyte and Kasumi-1 cells.

(A) MEG-01 and (B) Kasumi-1 cells in bone marrow/liver system under continuous flow. Drug addition was performed on day 4, and effects on proliferation were evaluated at DIV 7 (day 3 after drug addition). Error bars are represented as SEM. *P < 0.02 [one-way analysis of variance (ANOVA) followed by Dunnett’s test (α = 0.05)] and represents a minimum of n = 4 biological replicates.

The enzymatic activity and viability of the liver compartment were measured after system dosage with imatinib or diclofenac as an indication of specific off-target effects. Imatinib was chosen because it effectively halts proliferation of myeloid leukemia–derived cells without apparent clinical off-target toxicity effects. There was no effect on liver CYP 1A1, 3A4, or 2C9 activity or any corresponding effect on liver viability (Figs. 3 and 4), which mimics the effects seen in published clinical and preclinical data for imatinib (12, 28, 31, 32).

Fig. 4 Liver functionality and viability in bone marrow/liver system.

End point measurements of CYP isoforms 3A4 (A), 1A1 (B), and 2C9 (C). (D) Relative viability after 3 days of drug exposure. Liver viability was assessed after drug exposure to determine drug effects on liver function and survival at DIV 7. Error bars are represented as SEM. CYP values are double-normalized to control values and to cell number. *P < 0.003 and **P < 0.0002 [one-way ANOVA followed by Dunnett’s test (α = 0.05)] and represent a minimum of n = 3 biological replicates.

Diclofenac has been shown to have antiproliferative effects and concentration-dependent toxicity on liver, including a dose-dependent effect on liver viability (13, 14). When applied in high concentrations, diclofenac produces protein adducts that decrease the activity of the liver (13, 14). The protein adduct formation and resultant toxicity are a direct result of intermediate metabolite formation in the CYP breakdown of diclofenac (33, 34). Diclofenac reduced liver viability by 30% in the bone marrow systems (Fig. 4). The observed resultant toxicity due to diclofenac conversion provides evidence of liver metabolism in the system and adduct production. In the system, diclofenac caused about a 2.5-fold induction in both CYP 3A4 and 2C9 activities while having no effect on CYP 1A1 activity (Fig. 4, A to C).

System 2: Multidrug-resistant cancer model

The efficacy of drug combination therapy on both an MDR+ cancer and an MDR cancer was examined in a system configured with these two cancer lines, primary human hepatocytes, and functional cardiac electrical and mechanical bioMEMS devices. When treating the system with tamoxifen, the MDR MCF-7 breast cancer cells underwent a viability reduction of 20 ± 3% when the liver was present to metabolize tamoxifen to its active metabolites (Fig. 5A). This loss of MCF-7 viability was in contrast to the minimal cell death (viability of 98 ± 5%) when the liver was not present (Fig. 5A) but similar to the loss of viability when only adding an equal concentration of the active metabolite 4-hydroxytamoxifen (29 ± 3%). Together, these data indicated that the liver compartment effectively metabolized the relatively inactive tamoxifen to its active metabolites and that the MCF-7 breast cancer cells responded to the tamoxifen only after its metabolism. The activity of CYP 2D6, which is primarily responsible for tamoxifen metabolism, is difficult to measure because it is heavily outweighed by the coactivation of CYP 3A4 in the enzyme-linked immunosorbent assay kits used for measurement; however, the data from the system with and without the liver component indicate that CYP activity and metabolite formation occurred.

Fig. 5 Tamoxifen treatment effects on cancer cell viability in the presence and absence of liver tissue.

(A) Normalized viability of MCF-7 cells treated with 4-hydroxytamoxifen or tamoxifen with or without verapamil in the presence or absence of liver tissue. (B) Normalized viability of SW-962 cancer cells, which express Pgp, treated with 4-hydroxytamoxifen or tamoxifen with or without verapamil in the presence or absence of liver tissue. Error bars are represented as SEM. *P < 0.02 [one-way ANOVA followed by Tukey’s test (α = 0.05)] and represents a minimum of n = 4 biological replicates.

In contrast to the MDR MCF-7 cells, the MDR+ SW-962 vulva cancer cells did not lose viability when the system was treated with either tamoxifen (with or without the liver present) or 4-hydroxytamoxifen. This lack of response to tamoxifen or its metabolites is consistent with the multidrug resistance of the SW-962, mitigated through MDR1 expressed by these cells, demonstrating the ineffectiveness of this drug alone on multidrug-resistant cancer cells.

The system was further tested with a combinatorial therapy of tamoxifen, in the presence of the liver for active metabolism, and verapamil, a first generation MDR1 inhibitor. The combinatorial treatment elicited a 40 ± 2% decrease in SW-962 viability, a significant decrease from the control (P < 0.002) (Fig. 5B), without producing an additional decrease in MCF-7 viability (Fig. 5A). These data demonstrated the effectiveness of the system to respond to combinatorial therapy, with the MDR+ SW-962 only responding to the anticancer drug tamoxifen when the Pgp pump responsible for inducing multidrug resistance was blocked by the inhibitor verapamil. In addition, verapamil alone did not cause toxicity in either SW-962 or MCF-7 cells in monoculture in concentrations up to two orders of magnitude higher than that used in the system (fig. S2).

Off-target organ effects of tamoxifen with or without verapamil on cardiac function were evaluated using functional cardiac models integrated on electrical and mechanical bioMEMS chips, as well as liver function and viability. Tamoxifen without the addition of a liver component had little to no effect on MCF-7 viability. The addition of the liver allowed for the metabolism of tamoxifen into its metabolites, which are about 100 times more effective, that produced a 20% reduction in viability after a 24-hour incubation. This 20% reduction was similar to the 29% drop that occurred if the metabolite alone was added. The addition of verapamil had no additive effect on MCF-7 viability because this line does not express MDR1. Cardiac functional readouts were performed before and after drug addition. Representative readouts of custom MEA (cMEA) recordings are shown in fig. S3. Each cMEA was monitored over a period of 20 s to determine the spontaneous beat frequency and conduction velocity (CV) of the cardiomyocytes. Cardiac force output was evaluated by the change in position of signal on the detector resulting from deflection of the cantilevers in response to cardiac contraction (fig. S3). A protocol similar to those previously published was used to convert this signal to a force output (35, 36). Pre- and postdrug application recordings were compared to determine the effect of drug addition on mechanical and electrical output of the heart.

The 24-hour tamoxifen treatment reduced cardiac CV by 60%, with the cardiac cells recovering to near the baseline CV after an additional 48-hour recovery period (Fig. 6A). With a concomitant addition of verapamil, the combinatorial treatment resulted in further reduction of CV to an almost nonexistent state and had a substantial effect on calcium and sodium handling such that the CV did not recover after an additional 48-hour washout period. A similar pattern was observed with cardiac force production, with force output of the cells dropping by more than 60% with tamoxifen addition alone and to near zero with concurrent verapamil addition (Fig. 6B). With tamoxifen alone, the force recovered after the 48-hour washout period, similar to the recovery of the CV. With the combinatorial drug treatment of both tamoxifen and verapamil, the force did not recover.

Fig. 6 Effects of tamoxifen on cardiac function.

Normalized (A) CV, (B) cardiac force, (C) beat frequency, and (D) cell viability of cardiac cells treated with tamoxifen with or without verapamil. Error bars are represented as SEM. *P < 0.02 [one-way (A) or two-way (B, C, and D) ANOVA followed by Tukey’s test (α = 0.05)] and represents a minimum of n = 3 biological replicates.

Cardiac beat frequency was much less affected by drug addition than either force or CV, exhibiting a reduction from baseline of 40% with tamoxifen alone and 60% with the addition of verapamil and tamoxifen. Under both single and combinatorial therapies, the beat frequency recovered after 48 hours (Fig. 6C). Whereas single and combinatorial drug treatment affected functional measurements of the cardiomyocytes, indicating off-target effects on cardiac function, no significant (P > 0.5) loss of cardiomyocyte viability was observed (Fig. 6D). Verapamil had an additive adverse effect on the sodium and potassium handling already affected by tamoxifen and compounded the cardiac effects seen with tamoxifen addition alone but were ameliorated by a 48-hour washout period and returned to pre-addition levels (Fig. 6, A to C). The 48-hour washout period proved to be an insufficient length of time to allow the cardiac cells to recover from the combinatorial drug therapy (A to C). However, these data indicate that not only were the functional effects of the drugs on cardiomyocytes not due to loss in viability but also significant changes in electrical and force function can be observed in our system before any observable loss in viability and were a more sensitive measure than beat frequency.

High-performance liquid chromatography analysis and PK modeling of drug concentration in systems

Simulation protocols were used to model drug concentrations in the system by taking into account mixing, adsorption, and metabolic conversion to predict the concentrations (Fig. 7A). High-performance liquid chromatography with mass spectrometry (HPLC-MS) analysis of the medium in the system, specifically with respect to recirculation due to rocking, was used to determine the PK area under the curve (AUC) for compound concentration, by removing a small fraction of the total medium in the system to compare to the simulation results. The concentration of the drug at time points from 10 min to 24 hours was determined to obtain the exposure over time profile, and an integration of this profile provided the AUC (Fig. 7B). Using the modeling data, the concentration of added drug was adjusted to account for losses due to adsorption onto housing materials to produce desired final concentrations that were verified with HPLC-MS. Tamoxifen is a relatively hydrophobic compound that adsorbs readily onto the PDMS material and requires a bolus application of 100 μM to reach a maximum systemic medium concentration, analogous to a systemic plasma concentration, of 7.3 μM after 30 min (Fig. 7B).

Fig. 7 CFD modeling and HPLC drug analysis in a multi-organ BoaC system.

(A) CFD modeling was performed to determine PD parameters, incorporating cyclic rocking flow, drug metabolism by liver cells, and complexing of drug and albumin. (B) A time-dependent HPLC analysis was performed on sample aliquots drawn from the system to generate an AUC tamoxifen concentration profile within the system 2 configuration.

DISCUSSION

We have established a reconfigurable multi-organ system to investigate anticancer drug efficacy and off-target effects in two different cancer-derived models. The system uses serum-free medium in microfluidic channels to mimic blood circulation and functional bioMEMS devices for monitoring drug effects. CFD modeling enabled prediction of an in vitro PK-like AUC of drugs circulating in the system. These data informed system design to mimic drug retention times and sheer stresses in a physiologically relevant manner. By coupling this geometry with additional CFD simulations, the system can be reconfigured for cell types that can be grown on a coverslip, embedded in a hydrogel, or incorporated onto microfabricated bioMEMS devices that are primarily planar, provided that the cell type is viable under these conditions. With additional modeling for specific devices, nonplanar bioMEMS devices could be incorporated. Currently, the field of in vitro tissue engineering has two guiding philosophies: One argues that anatomy is important for in vitro systems, whereas the other suggests that proper function, whether achieved in a 2D or 3D system, is the main goal. The successful fabrication of our hybrid 3D systems was guided by two principles aligned with the “function over form” philosophy: (i) to engineer an organ mimic in the simplest form that maintains the functions of interest and only adds complexity as necessary and (ii) to combine surface chemistry and physiologically relevant ECM adhesion cues with serum-free medium to support the maintenance of cell function and differentiation that has proven effective in the maturation of induced pluripotent stem cell (iPSC)–derived tissues or the maintenance of function in primary cells (8).

Initially, the system was configured to study the effect of two well-characterized compounds, imatinib and diclofenac, on two bone marrow tumor cell lines and human primary hepatocytes. A 24-hour imatinib treatment produced cytostatic effects on Kasumi-1 and MEG-01 cells with minimal effect on liver functionality as expected from preclinical and clinical data (12, 28, 31, 32). Diclofenac, with known metabolite effects on liver function, was used to study efficacy and off-target toxicity effects in the system. Diclofenac treatment (1 mM) caused a cytostatic effect on Kasumi-1 and MEG-01 cells and also resulted in a 37 ± 7% decrease in liver viability. These data are consistent with established PK/PD data from human studies (14). We observed an induction of CYP 3A4 and CYP 2C9 activities in diclofenac-treated systems compared to controls. Although diclofenac is known to be a substrate of both 3A4 and 2C9, there is no published literature supporting or contradicting this observation. Further investigation is beyond the scope of this manuscript; however, this unexpected finding should be explored in future studies.

To demonstrate the system’s versatility, we reconfigured it to study the efficacy of a combinatorial drug therapy on multidrug-resistant (MDR+) and non–multidrug-resistant (MDR) cancer cell lines. This BoaC design incorporated cardiomyocytes grown on MEAs and cantilevers as a sentinel tissue to allow for simultaneous efficacy studies of tamoxifen and its more effective metabolite and off-target studies of verapamil and tamoxifen. All tissues remained viable and functional in control systems for the 7-day time course of these experiments and have shown long-term viability and function up to 28 days (8). Previously published in vitro cardiac systems are primarily limited to measuring a drug’s effects on beat frequency. Recently, electrical impedance of cardiac cell layers has also been measured to monitor contraction; however, impedance of the cell layer is not a measurement of force and instead monitors indirect changes in morphology (3741). Unlike our current multi-organ system, previously published studies used systems, whether frequency or impedance based, that cannot analyze the multifaceted effects a drug may have on the complex arrangement of the heart (6, 3741).

We speculate that the recovery in beat frequency, whereas CV and force failed to recover, under combinatorial addition of tamoxifen and verapamil was due to the complexity by which these actions occur. For CV and force to be generated and measured, a syncytium of cells is necessary, requiring ion channels within individual cells and electrical signal propagation through gap junctions between cells for these parameters to be measured. Beat frequency, however, does not require coordination among cells and can be generated to a measurable degree by a single cardiomyocyte. Clinically, standard low-dose tamoxifen therapies show no cardiac toxicity (42). However, high-dose tamoxifen therapies (7 to 10 μM in plasma), used for more resistant cancers such as MDR+ and gliomas, have shown adverse effects on Ca2+ and K+ handling leading to QT elongation and torsade de pointes syndrome (2022). Effects on K+ and Ca2+ handling not only affect polarization and repolarization of the cell, altering electrical activity, but also will affect the excitation-contraction coupling and contractile machinery. Similarly, verapamil has been shown to affect cardiac output and beat frequency via calcium channel blockade (43, 44).

In our system, tamoxifen treatment alone caused a decrease in the viability of ER+ and MDR MCF-7 cancer cells while having no significant effect on MDR+ SW-962 cells. However, cardiomyocytes exhibited a significant decline in the functional readouts of CV, force output, and beat frequency after 24-hour treatment but no changes in viability. These data are in support of preclinical and clinical findings for PK/PD of tamoxifen at these concentrations (42). The addition of verapamil to the tamoxifen-containing system facilitated a synergistic effect resulting in cytotoxicity to the MDR+ SW-962 cells but exacerbated the cardiac off-target effects. On the basis of the drugs’ mechanisms of action and the unaltered cardiomyocyte viability, we expected the off-target cardiac effects to be transient. After a 48-hour recovery period that included two 30% medium changes to mimic drug excretion, cardiomyocytes in the systems treated with tamoxifen alone exhibited recovery in all parameters measured. Although recovery of cardiac physiology from the combinatorial tamoxifen and verapamil therapy was expected, only incomplete recovery for beat frequency was observed. Previous studies have indicated that verapamil alters gap junction function; therefore, the functional syncytium necessary for sustained force generation and CV may be affected, which could explain the observed effects (43). Consequently, beat frequency, a parameter not dependent on gap junctions, would be expected to recover more quickly. Both of these observations were apparent from the data collected in these studies.

As shown here, the ability to culture cell systems representing organs of interest in the presence of a liver module allowed for direct efficacy and toxicity analysis of a parent drug compound and its liver metabolites due to the pumpless system enabling a low-volume platform. This design is superior to current multi-organ systems for multiplexed and repeat drug dosage regimes, and the metabolites generated can be monitored in real-time using functional readouts for acute administrations but also allows the observation of functional recovery. We believe that these functional readouts are more comparable to clinical observations than general viability or biomarker analysis. Furthermore, this human cell–based BoaC design retains the ability to analyze standard toxicity and biomarker assays associated with typical in vitro efficacy and toxicity studies.

Indirect determination of liver metabolism was performed by testing drugs requiring liver metabolism in systems with and without liver tissue. For example, in the system configured to test the MDR cancer line, treatment with tamoxifen had minimal effect on cancer viability in the absence of the liver cells. However, in the presence of liver, tamoxifen had similar effects to systems dosed directly with the active metabolite 4-hydroxytamoxifen. The rapid reconfigurable nature of our system allows for the determination of functional drug metabolism, if present, without the need for complex analysis systems. However, this indirect measure of metabolite generation would be limited to detoxification of parent compound or generation of toxic metabolites. The size of other tissues relative to the liver could be scaled differently if a dynamic interplay between organs becomes necessary as would be the case for a diabetes model.

Although the drug dosing parameters presented here are acute, a chronic, low-dose treatment regimen may better mimic clinical anticancer therapies. The ability of these systems to assess cardiac function noninvasively and monitor biomarkers over time provides an opportunity to run long-term studies. For example, a longer recovery period may have resulted in contractile strength and CV measurements returning to baseline in our system. If, as is the case in most in vitro assays, we had restricted analysis to viability and beat frequency alterations, then the range of drug effects would be unclear, leading to an incomplete understanding of the drug efficacy and toxicity profile.

Overall, the two configurations designed to test anticancer therapies using functional readouts of efficacy and toxicity generated data similar to previous in vivo data (2830, 45). Application of this type of system in pharmaceutical testing may increase the accuracy of preclinical models and reduce the use of animals by enabling the study of repeat dose effects in drug discovery by providing the ability to establish the therapeutic window for compound evaluation in human cells. In addition, the ability to culture patient-derived cells, including cancer, in the system enables direct utilization of this multi-organ system for personalized medicine applications.

MATERIALS AND METHODS

Study design

In each experimental study, as many experimental conditions as needed were tested simultaneously to block for sample group variation, and at least three separate sample groups were tested for each condition. Sample size was determined from initial variance measurements and using a type I error rate (α) of 0.05 and power of 0.8 in detecting differences in means of at least 30%. Drug compounds were applied to two separate configurations of the system to evaluate the system’s ability to recapitulate known drug effects. All compounds in each system configuration were added with identical concentrations of dosing vehicle, and the control was dosed with the equivalent vehicle dosage to enable comparison among compounds and to the control. Systems were selected at random for each treatment, and the run order for experimental testing was also randomized. Analysis was performed by a blinded operator, via assignment of a nonidentifiable batch and system number to data labels for later data aggregation. Sampling replicates were used, e.g., images from different locations of a coverslip or isolated cantilevers on a bioMEMS chip, to reduce the effect of sampling variation on the measurement of each experimental replicate. Except in end point measurements, a repeated measures analysis was performed to block for small variations among systems before treatment, and normalization to the controls was performed to block for experiment-to-experiment variation.

System configurations

Two configurations of the system were developed to demonstrate both reconfigurability and the ability of the system to recapitulate several different in vivo responses.

System 1—Bone marrow/liver hepatocyte. This system included two bone marrow lines, Kasumi-1 myeloblasts (CRL-2724, the American Type Culture Collection (ATCC)] and MEG-01 megakaryocytes (CRL-2021, ATCC), along with primary human hepatocytes to measure the effects of drugs on bone marrow proliferation and on liver with organ-organ interactions between the liver and bone marrow. Primary human hepatocytes were cultured on coverslips derivatized with an amine-containing silane, and Kasumi-1 and MEG-01 cells were suspended in alginate-based hydrogels in separate chambers. Each system was run for a total of 7 days, with drug addition occurring at day 4 to allow for the monitoring of cell proliferation over several days. The proliferation of each of the bone marrow cell types was quantified via phase contrast microscopy and image analysis. Measured values for liver included viability and CYP enzymatic activity.

System 2—Cancer/cardiomyocyte/hepatocyte. The second system contained both an MDR+ vulva carcinoma–derived cell line (SW-962) and an MDR breast cancer cell line (MCF-7), a liver compartment, and two separate bioMEMS devices to measure cardiac electrical and mechanical function. Cardiac mechanical function was evaluated by incorporating the cardiomyocytes onto custom arrays of microscale cantilevers and calculating force and frequency dynamics from laser-based measurements of cantilever bending resulting from cardiomyocyte contractions (44). Cardiac electrical function was measured via a MEA amplifier system by incorporating and chemically patterning cardiomyocytes onto cMEAs to produce a defined conduction path along a series of surface-embedded microelectrodes (44, 46). Electrically stimulated cardiomyocyte activity was generated inside the system with housing-embedded electrodes for cantilevers and via stimulation through the MEA chip for electrical measurements. In this configuration, the efficacy of drug combinations with respect to multidrug resistance was evaluated on a single device with drug metabolism from the liver compartment, and simultaneously, off-target cardiac effects were measured using the two primary determinants of heart function: electrical and mechanical function. In this configuration, the systems ran for a total of 7 days, with drug dosage applied as a single dose at day 4 and effects measured on days 5 and 7.

System fabrication

A multichamber flow system (housing) was designed to allow for the incorporation of multiple organ tissues in a pumpless, recirculating system with integrated measurements for cardiac electrical and mechanical function by modifying a design from a system containing four organs (7). The design of the flow chambers and microfluidic device was driven by CFD modeling to establish shear stresses near the cell layers within physiological ranges. Housings were fabricated from 6-mm-thick clear cast acrylic sheets (McMaster-Carr), and gaskets were made from 0.5-mm-thick PDMS elastomer sheet material (Grace Bio-Labs). Housings and gaskets were laser-cut on a Universal Laser Systems VersaLASER PLS 75-W laser cutter, with additional postprocessing performed for counterboring screw holes and inserting brass screw inserts (93465A107, McMaster-Carr) and sealed with stainless steel screws. A VWR Signature rocking platform shaker was used to produce a defined rocking profile of 1 oscillation per minute at an amplitude of 1°. This rocking action produces recirculating flow in the system, with medium flowing between the two reservoirs through the organ chambers in a 37°C incubator with 100% relative humidity and 5% CO2.

cMEA and cantilever preparation

Cantilever array chips for force measurements were fabricated following protocols previously described (36, 44, 4749). Briefly, cantilever chips containing 4-μm-thick, 100-μm-wide, and 737-μm-long cantilevers were fabricated from silicon-on-insulator wafers with a 4-μm device layer and 1-μm buried oxide. The cantilever devices were created in the device layer, and a window underneath the cantilevers was produced using standard photolithographic patterning techniques and deep reactive ion etching. cMEA chips were designed with ten 200-μm-diameter recording/stimulating electrodes and one 2000-μm-diameter ground electrode and fabricated with standard microfabrication procedures. The wires and electrodes were composed of electron beam–evaporated 10-nm titanium and 50-nm platinum and deposited on a fused silica substrate and patterned via a liftoff process. The titanium/platinum wires were insulated with a three-layer stack of 150-nm layers of silicon oxide, silicon nitride, and silicon oxide produced via plasma-enhanced chemical vapor deposition and etched using RIE. After fabrication, the surfaces of the cantilever chips and cMEA chips were surface-modified using silane chemistry, as previously described (44, 46). The surfaces of the cantilever chips were modified with 3-(trimethoxysilyl)propyl)diethylenetriamine silane (DETA) in toluene, followed by adsorption of fibronectin. The cMEA chips were surface-patterned with a combination of poly(ethylene glycol) (PEG)–containing silane and fibronectin. In this method, the surfaces were modified with 2-[methoxypoly(ethyleneoxy)propyl]trimethoxysilane in distilled toluene to produce a cytophobic surface. This layer was patterned using a 193-nm ArF excimer laser (Lambda Physik) through a quartz photomask to remove areas of the PEG chemistry and allowing the deposition of fibronectin, as described in the cell culture procedures.

Cell culture

Cardiomyocyte culture. The patterned cMEA chips and cantilevers were coated with human plasma fibronectin (Millipore) diluted in 1× phosphate-buffered saline (PBS) (Thermo Fisher Scientific) to concentrations of 10 μg/ml for cMEAs and 50 μg/ml for cantilevers and then incubated at 37°C for 30 min. MEAs were then washed once, and cantilevers were washed three times with 1× PBS. iPSC-derived cardiomyocytes (Cellular Dynamics Inc.) were thawed and seeded directly onto the surfaces at 50,000 cells per MEA and 500,000 cells per cantilever in our cardiac medium (table S1) (7).

Primary human hepatocyte culture. Primary human hepatocytes were sourced from Massachusetts General Hospital (MGH lot Hw54) from patient biopsies and thawed and cultured in specific medium (table S2) on rat tail collagen type I (60 μg/ml)–coated surfaces according to previously published protocols (7).

Cancer and bone marrow culture. MEG-01 (CRL-2021, ATCC) megakaryocytes and Kasumi-1 (CRL-2724, ATCC) myeloblasts were initially cultured in T75 flasks according to described vendor protocols. When cells had reached optimum density (300,000 to 500,000 cells/ml), aliquots of each cell type were harvested and pelleted at 150g for 5 min. The pellets were then resuspended in Hibernate E without calcium (HibE − Ca2+) (BrainBits) with alginate (5 mg/ml; Sigma-Aldrich) and 25% 10 mM CaCl2 solution and triturated gently. The suspension (50 μl) was quickly pipetted into the housing system with Kasumi-1 cells in chamber 5 and MEG-01 in chamber 4 (Fig. 1). After the cells had been plated, 10 μl of 30 mM calcium chloride solution was gently dripped over the gel to further facilitate hardening of the alginate hydrogel. The alginate hydrogels were allowed to harden for 30 min before medium was added, and the systems were assembled.

MCF-7 breast cancer–derived cells (HTB-22, ATCC) and SW-962 vulva carcinoma–derived cells were initially cultured according to the manufacturer protocols, and aliquots were frozen at passage 1. For MCF-7 cells, sterile 15-mm round glass coverslips were coated with collagen solution [collagen type 1 (60 ng/ml) dissolved in a 0.12% acetic acid in 1× PBS solution] at room temperature for 2 hours and then washed once with sterile water. SW-962 cells were plated directly onto DETA-derivatized 15-mm round coverslips. Cell aliquots were transferred from liquid nitrogen and snap-thawed in a 37°C water bath diluted 1:10 in growth medium and pelleted at 150g for 5 min. Each cell pellet was resuspended in culture medium and plated on their respective surfaces at a density of 250 cells/mm2 for SW-962 and 450 cells/mm2 for MCF-7. Seeding densities were determined for optimal confluency levels based on length of time in culture and doubling rate of cells. The cultures were allowed to proliferate 3 days before being assembled into the housings.

Drug application

Imatinib mesylate (SML1027, Sigma-Aldrich) was dissolved in purified distilled water to a concentration of 125 μM. Diclofenac sodium (D6899, Sigma-Aldrich) was dissolved in purified distilled water to a concentration of 100 mM. Tamoxifen citrate (0999, Tocris Bioscience) and 4-hydroxytamoxifen (3412, Tocris Bioscience) were dissolved in ethanol to final concentrations of 100 mM. Verapamil HCl (V4629, Sigma-Aldrich) was dissolved in purified distilled water to a concentration of 300 μM. Each drug was administered dissolved in medium at 1:1000 dilution of the final overall volume of 2 ml within the housing system and added to the reservoir closest to the liver chamber (chamber 1; Fig. 1).

Noninvasive MEA and cantilever recordings

Cardiac contractions were measured using a cantilever deflection system that has been described previously (7, 48, 49). In summary, longitudinal deflection of the cantilever due to contractions results in changes in position of the reflected beam on the lateral effect sensor. The voltage output from the position sensor was converted directly to force using a modified form of Stoney’s equation, as published previously (7, 36, 47, 50).

cMEAs were fabricated to simultaneously stimulate and record electrical activity of cardiomyocytes, allowing CV to be recorded at different points (14-mm total path length). Printed circuit boards and flexible elastomeric connectors (1 mm wide by 18.2 mm long by 9 mm tall; ZEBRA connectors, Fujipoly) were incorporated into each system to create an interface between the MEA and a commercially available 60-electrode amplifier (MEA1060, Multichannel Systems). A stimulus generator (STG 1002, Multichannel Systems) was used to stimulate the cells (800-mV rectangular pulse, 0.5 to 3.0 Hz in 0.25-Hz increments). The multichannel systems software suite was used to control both the amplifier and stimulator and was used to record action potentials (APs).

Drug additions were performed on day 4 of the experiment, immediately after baseline recordings of spontaneous and stimulated cell activity were taken. Cell activity was recorded again at 24 and 72 hours after drug exposure. CV and spontaneous beat frequency were extracted from the data using Clampfit (Axon Instruments). Spontaneous beat frequency was obtained from spontaneous recordings of cardiac activity 20 s in length. Stimulated conduction was initiated by stimulation using 40-ms pulses at 1 Hz for 15 s. For each recording, the frequency was calculated as the number of measured APs divided by the length of time between the first and last APs during the recording interval. The number of APs used in the calculation was decreased by 1 to account for the distances between beatsBF(Hz)=(no.of peaks in interval1)/time between first and last peaks (s)CV was measured by determining the length of time for an AP to travel from an electrically stimulated electrode along the patterned cardiomyocyte path, following the method described by Stancescu et al. (44). From the recording of stimulated conduction, the CV was determined by averaging several measurements of the length of time for the AP to travel along the defined path between two defined electrodes and dividing the physical distance between those two electrodes by that average propagation time.CV(m/s)=distance between electrodes (m)/average signal propagation time (s)

Cell viability, density, and enzymatic assays

Multiple images of each bone marrow chamber were taken using a 20× objective using a Zeiss Axioscope phase contrast microscope. Cells were monitored at day 0 of drug addition and at DIVs 5, 6, and 7. The density of megakaryocytes and Kasumi-1 cells were analyzed via image processing of phase contrast images of each of the chambers. The imaging processing was developed for and performed in ImageJ (51) and incorporated a contrast enhancement and manually directed brightness thresholding operation, followed by an area determination occupied by the cells. A series of five focal plane images were collected and analyzed for each data point, and the cell density among the five images was averaged. Cell viability was assessed in one of two ways, depending on the plating format. Cells plated on coverslips (MCF-7, SW-962, and hepatocytes) were assessed for viability after disassembly via standard 3-(4,5-dimethylthiazol-z-yl)-2,5-diphenyltetrazolium bromide (MTT) assays. Briefly, MTT powder was dissolved in growth medium to a final concentration of 5 mg/ml. Cells were incubated in 500 μl of the solution for 90 min at 37°C and 5% CO2. The medium was removed, and resultant crystals were dissolved in a lysis buffer consisting of 10% SDS with 0.5% acetic acid in dimethyl sulfoxide. The solution (100 μl) was placed into a 96-well plate and absorbance read at 570 nm using a BioTek Synergy HT plate reader (BioTek). The viability of cardiac cells plated on the MEA was assessed using an Alamar blue solution (Thermo Fisher Scientific). Cardiac surfaces were incubated in 500 μl of a 10% solution (v/v) of Alamar blue in growth medium at 37°C and 5% CO2 for 4 hours. The solution (100 μl) was placed into a 96-well plate and read at fluorescence excitation wavelength of 570 nm and emission at 590 nm using the BioTek Synergy HT plate reader.

High-performance liquid chromatography with mass spectrometry

Medium samples collected from the systems were spiked with an amiodarone internal standard (25 nM final measured concentration) and combined with acetonitrile at a 1:3 ratio to extract proteins. The mixtures were then centrifuged at 6600 rpm for 10 min, and supernatants were collected and diluted 1:10 in the initial mobile phase. A gradient elution method was run more than 5.5 min beginning at a ratio of 60:40 and ending at 5:95 of 0.1% formic acid in water: 0.1% formic acid in acetonitrile through a 4.6-mm internal diameter × 100 mm, 3.5-μm Agilent ZORBAX C18 column installed in a 1260 Infinity Agilent LC system with a 6490 triple quadrupole MS detector (Agilent). The monitored transitions were 372 ➔ 72 with a 28-eV collision energy for tamoxifen, 388 ➔ 72 with a 24-eV collision energy for 4-hydroxytamoxifen, and 646 ➔ 58 with a 48-eV collision energy for amiodarone. Calibration curves of 4-hydroxytamoxifen and tamoxifen were prepared using standards ranging from 0.1 to 100 nM. Regression lines were linear with a 1/X weighting applied.

Computational simulation of fluid dynamics

CFD was performed on a 3D computer-aided design model of the multi-organ housing system with commercially available CFD-ACE+ (ESI Group). A previously developed (7) transient model for gravity-driven flow on the rocking system was applied to the CFD model to simulate oscillatory flow. The CFD model was used to determine and optimize resulting shear stress profiles in each of the organ chambers, with shear stresses below 0.05 dynes/cm2 (within acceptable physiologic ranges) (Fig. 7). This model was extended to include reactions for the metabolic conversion of compounds to conversion products and complexing of compounds with albumin.

Statistical analysis

To account for any system-to-system variation in initial conditions (slight differences in number of cells seeded into the cancer-tissue chambers), measured parameters were analyzed using repeated measures analysis, normalizing data to the initial value for the specific system. Measurements were compared statistically using Fisher’s least significant difference test (α = 0.05) following one-way ANOVA (α = 0.05) for end point measurements or two-way ANOVA (α = 0.05) for time course measurements across two or more conditions. Measured values are presented as means with error bars representing SEM. Primary data are reported in data file S1.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/11/497/eaav1386/DC1

Fig. S1. Overview of four-organ system with cancer, cardiac, and liver and phase contrast microscopy of tissues at 14 days.

Fig. S2. Escalating dose of verapamil effect on MCF-7 and SW-962 viability.

Fig. S3. In situ functional cardiac and liver readouts after 14 days.

Table S1. Formulation of human iPSC-derived cardiomyocyte culture medium.

Table S2. Formulation of primary human liver culture medium.

Table S3. Formulation of multi-organ system common medium.

Data file S1. Primary data.

REFERENCES AND NOTES

Funding: We would like to acknowledge support from Roche Pharmaceuticals as well as the NIH, SBIR grant number R44TR001326-02, and Hesperos Internal Development funds. Author contributions: C.W.M., J.J.H., T.S., A.B.R., and F.S. designed the experiments; C.W.M., D.E., T.S., L.R.B., J.W.R., C.M., M.S., and Y.W. performed the experiments; C.W.M., C.J.L., T.S., and L.R.B. analyzed the data; C.W.M., C.J.L., F.S., A.B.R., C.F., M.L.S., and J.J.H. interpreted the results; C.W.M. prepared the figures; C.W.M. and C.J.L. drafted the manuscript; J.J.H., J.W.R., F.S., A.B.R., and M.L.S. made revisions to the manuscript. Competing interests: C.W.M., D.E., T.S., L.R.B., J.W.R., and M.L.S. are or were employed at Hesperos Inc., a for-profit company seeking to market services for human-on-a-chip platforms. F.S., A.B.R., and C.F. are employees of Roche Pharmaceuticals, which is a major pharmaceutical company. All other authors declare that they have no competing interests. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article