A perfusion incubator liver chip for 3D cell culture with application on chronic hepatotoxicity testing

Liver chips have been developed to recapitulate in vivo physiological conditions to enhance hepatocyte functions for assessing acute responses to drugs. To develop liver chips that can assess repeated dosing chronic hepatotoxicity, we need to ensure that hepatocyte functions be maintained at constant values over two weeks in stable culture conditions of sterility, temperature, pH, fluidic-flow of culture media and drugs. We have designed a perfusion-incubator-liver-chip (PIC) for 3D cell culture, that assures a tangential flow of the media over the spheroids culture. Rat hepatocyte spheroids constrained between a cover glass and a porous-ultrathin Parylene C membrane experienced optimal mass transfer and limited shear stress from the flowing culture media; maintained cell viability over 24 days. Hepatocyte functions were significantly improved and maintained at constant values (urea, albumin synthesis, and CYP450 enzyme activities) for 14 days. The chip act as an incubator, having 5% CO2 pressure-driven culture-media flow, on-chip heater and active debubbler. It operates in a biosafety cabinet, thus minimizing risk of contamination. The chronic drug response to repeated dosing of Diclofenac and Acetaminophen evaluated in PIC were more sensitive than the static culture control.

In long-term microfluidic culture systems, dissociated gas from the culture media often results in bubble accumulation in the microfluidic channels. This leads to clogging and dead volumes. Bubble traps commonly used in microfluidic systems can prevent large bubbles from entering the culture chamber; however, they cannot prevent the formation of dissociated gas or tiny bubble accumulation inside the culture chambers. Thus, we have integrated an active debubbler in the PIC to remove bubbles from the chamber in real time.
A proper design of a chronic liver toxicity-testing chip should have the following features: • Organotypic 3D cellular architecture [11][12][13][14][15] ; • Good mass transfer, allowing diffusion of O 2 and nutrients from media to the spheroids and removing the metabolites and by-products from the proximity of the spheroids 1,16-18 ; • Maintenance of mechanical forces and limited shear stress to the cells 1,7,18 ; • Stable cell culture conditions: temperature, pH, sterility and optimized culture media composition 1,7,18 ; • Ease of handling cells and replenishing media 7 ; Here we report a method and device for long-term 3D hepatocyte culture, which maintains the cell viability over 3 weeks and functions over 2 weeks, and its application on chronic hepatotoxicity testing. An essential aspect is that the design of the device assures a tangential flow over the cell culture, removing the metabolites and by-products from the proximity of the cells and refreshing the cell environment. A constrain spheroids 3D rat hepatocytes cell culture model was used for the testing. In this model, ultrathin Parylene C porous membrane constrained the hepatocyte spheroids formed underneath; it protects the hepatocytes from the shear stress while its relatively large pore size (20 µm) assures a good mass transfer. The membrane also immobilises the spheroids and prevents them from colliding into each other, which frequently happens on non-adhesive surfaces 19 and ligand-modified films 20 . PIC with integrated temperature, pH and bubble control provides contamination-free and clogging-free environment to maintain rat hepatocytes viability and metabolic activities for 2-3 weeks. We applied the PIC for acute and chronic repeat dosing drug safety testing using two model-drugs diclofenac and acetaminophen (APAP). PIC addressed the various issues affecting the robustness and performance of liver chips for chronic drug testing.

Results
PIC Structure and Functionality. Device structure and microfluidic setup. The perfusion incubator liver chip (PIC) was designed combining a rigid, well defined and reusable glass/silicon structure with the advantages of PDMS assemblies (elasticity and gas permeable). The chip structure consists of three main elements illustrated in Fig. 1a: • A glass/silicon structure containing a 3D microfluidic circuit, the cell culture chamber, a bubble trap chamber and a heater (Fig. 1b). Nanoport microfluidic connectors were mounted on the glass/silicon die. The microfluidic circuits were engraved on both sides of silicon die while the glass die assures the sealing of the bottom microfluidics circuit. On the other side of the glass die a heater is printed. • A cell culture support (in our case a 100 µm-thick glass coverslip and a Parylene C membrane mounted on a Silicon ring). The constrained spheroid structure can be inserted and removed from the culture chamber. • A PDMS/glass sealing structure closes the microfluidic circuit. The PDMS structure also acts as a debubbler to absorb the air bubbles trapped when assembling the chip and generated during the culture. A cross-section of the chip illustrating the functions of the PDMS/glass die is presented in Fig. 1c.
The main materials of the PIC are silicon and glass. The main advantage is their biocompatibility 21,22 , while having a hydrophilic surface of the microfluidic elements, there is less absorption of proteins and drug molecules. This is an important aspect for drug screening applications, especially when working with lower drug concentration 23 . The silicon structure assures a uniform temperature of the media, due to its good thermal conductivity. Moreover, the device can be reused to generate reproducible experimental results; and can be sterilized by flowing ethanol. For testing purposes, the silicon/glass structure was fabricated with different depths of the cell culture chamber (0.5 mm, 1 mm, 2 mm and 3 mm). When the cell culture chamber is clamped with a soft deformable PDMS cover, the high rigidity of silicon enables water-tight sealing of the microwell.
To minimize the risk of microbial contamination, the PIC was kept inside a biosafety cabinet at all times. To achieve this, we integrated a heater on chip. The heater was connected in series with a temperature controller and a thermocouple. Using a second thermocouple, the temperature of the cell culture media in the cell culture chamber and at the outlet was calibrated. The measured temperature is continuously adjusted to the set point of the temperature controller. The integration of the heater on the chip avoids the use of the incubator. Pressurized CO 2 was used to drive media perfusion through the chip (Supplementary 1). CO 2 dissolves in water forming carbonic acid, which keeps pH of the media constant.
We integrated bubble trap and active debubbler on chip to remove bubbles accumulated in the culture chamber. We avoided the use of external bubble traps to reduce the complexity of the system (more fluidic connectors) and surface area in contact with the cell culture media. Cells exposed to air bubbles are subjected to higher shear stress due to the stretching force at liquid-air interface. Large bubbles can cause cell death. As shown in (Fig. 1c), the debubbler consists of a 70 µm-thick PDMS membrane (gas permeable) bonded to a PDMS molded chamber with pillars that support the membrane. The PDMS structure is connected to external vacuum (through a pressure controller). Gas bubbles trapped in the microwell can diffuse through the PDMS membrane due to negative pressure in the vacuum chamber while culture media remains inside the culture chamber. Time-lapse images show that the vacuum eliminated the bubbles inside the microwell in 30 min (Supplementary 2).
ScIeNtIfIc RepoRTs | 7: 14528 | DOI:10.1038/s41598-017-13848-5 Fabrication of microfluidic glass/silicon structure. The glass/silicon structure was fabricated in four versions with varying depth of the cell culture chamber (0.5 mm, 1 mm, 2 mm and 3 mm). We will describe the fabrication process for the device with the depth of the chamber of 1mm. A 4" silicon wafer, 1mm-thick, with the crystallographic orientation <100> was used for the fabrication of the microfluidic circuits. Main steps of the fabrication process are illustrated in Fig. 2. A similar process is described in 23 .
The silicon wafer was cleaned in piranha (H 2 SO 4 /H 2 O 2 in ratio 2/1) in a quartz tank at 120 °C for 20 min, rinsed in DI water and spun dried. A 2 µm-thick thermal SiO 2 layer (wet process) was grown on the Si surface in a furnace (Tystar, USA) at 1050 °C for 11 hours (Fig. 2a). The pattern with the microfluidic circuit was transferred on the SiO 2 layer (bottom side of the wafer) using a 2 µm-thick photoresist mask (AZ7217-Clariant) and a classical RIE etching process in CHF 3 /O 2 using and (SPTS reactor) - Fig. 2b. After the RIE process, the photoresist mask was removed in an ultrasonic tank with NMP at 70 °C for 30 min, rinsed in DI water and spun dry. The opposite microfluidic circuit was aligned and imprinted in a similar manner on the other side of the Si wafer (Fig. 2c). A 10 µm-thick photoresist mask, having the pattern of the etch-through features (inlet outlet holes, via-holes, cell culture chamber and bubble-trap chamber) was aligned and applied on the bottom of the wafer. Through this mask, trenches were etched in Silicon (400 µm in depth) using a classical Bosch process (on an Alcatel 101SE ICP Deep RIE) (Fig. 2d). The photoresist mask was striped using O 2 plasma process in the same equipment, the SiO 2 mask was then revealed. Another anisotropic etching of silicon was performed through the SiO 2 mask (100 µm deep). It generated the microfluidic circuit on the bottom of the Si wafer (Fig. 2e). A Teflon layer was uniformly deposited in the Deep RIE system using C 4 F 8 chemistry for protection of the microfluidic structure during deep RIE etching from the opposite side of the wafer. In the next step (Fig. 2f), the process described in Fig. 2d (400 µm deep trenches using Bosch process through a photoresist mask) was repeated, this time from the topside of the wafer. After removing the photoresist mask in O 2 plasma and temporary bonding onto a dummy Si wafer with wax 24 , the microfluidic circuit from the top side of the wafer was defined using the same Bosch process till etch-through holes were completed (Fig. 2g). After wax removing in NMP ultrasonic tank and rinsing with DI water, the remaining SiO 2 masks (top and bottom side of the wafer) were removed in BOE (Buffer Oxide Etch) solution, rinsed in DI water and cleaned in piranha. A 200nm-thick dry SiO 2 layer was grown on the surface of Si wafer in a furnace. This SiO 2 layer assures a hydrophobic surface of the microfluidic structure. The silicon wafer is then anodically bonded on a 4", 500 µm-thick Pyrex glass wafer (Corning 7740) (Fig. 2h). The heater of the chip consists of a Cr/Au metal deposition (40 nm/1 µm)(CHA e-beam evaporator). It was defined using a classical photolithographic process (AZ7217) and Cr/Au wet etching solutions (Fig. 2i). A passivation layer SiO 2 /SiC (PECVD) was deposited on the heater and patterned using a photolithographic process and RIE process (Fig. 2j). This layer assures electrical isolation and chemical protection of the heater. Finally, the wafer was diced. Nanoport microfluidic connectors (Upchurch) (Fig. 2k) and an electrical connector were mounted on the chip (Fig. 2l). PIC System optimization. Flow simulation. Figure 3 shows the profiles of flow velocity and O 2 concentration at different depths in the cell culture chamber. Although the velocity is high near the inlet and outlet, it becomes relatively low at the membrane plane due to the large cross-sectional area of the cell culture chamber. This may help to lower the shear stress on the cells under the porous membrane. The O 2 concentration decreases along the flow direction due to the O 2 consumption by the cells at the bottom.
The decrease along the depth direction is less significant. This may be attributed to the shallowness of the cell culture chamber (H = 0.5 mm) in which the O 2 transfer from the top to bottom is relatively fast. Figure 4a and b show the averaged O 2 concentration and shear stress at the membrane plane. The simulation was performed for different depth of the bioreactor (0.5, 1, 2 and 3 mm). With the increasing flow rate, more O 2 is carried into the system by the culture medium while the consumption rate is not significantly affected. As a result, the O 2 concentration at the membrane level also increases, although the change is less significant when the flow rate is beyond 0.1 mL/h. On the other hand, for an increased depth of the cell culture chamber a lower O 2 concentration (at the same flow rate) was achieved, which is not surprising due to the corresponding higher resistance for O 2 transfer. Moreover, the shear stress on the membrane can increase almost linearly with the increasing flow rate, as shown in Fig. 4b. It can be observed from the slope of the lines that a shallow depth corresponds to a high shear stress. For example, at the same flow rate, the shear stress for H = 0.5 mm is about 38 times higher than that for H = 3 mm. A typical enigma in designing a cell culture chamber is to achieve a compromise between the requirement for mass transfer and the protection of cells from shear stress. A shallow chamber operating at high flow rates may provide better nutrient supply and waste removal than a deeper one at low flow rates. However, the former usually generates higher shear stress on the cells than the latter, as shown by the simulation results. In an environment with either insufficient nutrients and O 2 or too much shear stress, the cells cannot exhibit proper metabolism or even become unviable. With the help from the porous membrane, used in this study to shield cells from most of the shear stress, it is possible to design a shallow chamber (e.g. H = 0.5 mm) and operate it at a relatively high flow rate (e.g. Q = 0.1 mL/h), in an effort to provide sufficient O 2 to the cells for their metabolic activities. We will further demonstrate that the cell viability can be affected by the shear stress at high flow rates (Fig. 5a).
The O 2 equilibrium is affected by not only the supply and transfer but also the consumption rate. Figure 4c shows the O 2 level at the membrane plane with different number of cells cultured in the chamber. Obviously, a larger number of cells correspond to higher O 2 consumption, resulting in a lower oxygen level in the region with cells. Optimization of the flowing conditions. Flow rate was optimized with flow simulation and verified by selecting a flow rate at which mass transfer is maximized without too much shear stress damage to the cells to ensure the best performance of the on-chip culture. Hepatocyte spheroids were cultured under different flow rates: 0.02 mL/h, 0.06 mL/h, 0.1 mL/h, 0.2 mL/h and 0.4 mL/hr. MTS assay and urea assay is performed after 3 days of experiment to evaluate hepatocyte viability and function for different flow rates. To compare with static cultures, MTS activity of spheroids in PIC was measured in 48 well-plate, after transferring the coverslips from the PIC back to the well-plate. The results are presented in Fig. 5a (hepatocytes viability) and Fig. 5b (urea production). The spheroid viability in the chip and collagen sandwich were benchmarked against the cell viability of freshly isolated hepatocytes seeded in the 48 well-plate. Due to cell loss in the process of spheroid formation and transfer of coverslip, the viability presented are around 30-40% of control. The working range is between 0.06 and 0.2 mL/h, having an optimum flow rate at 0.1 mL/h. Cell culture chamber depth affects cell viability and function. The influence of chamber depth was studied. Theoretically, a tangential flow over the surface of the membrane helps the removal of the metabolites and by-products and refresh the media around the cells. Moreover, the concentration of the O 2 and nutrients decrease with the depth of the cell culture chamber (Fig. 6a), so we expect an optimal hepatocyte function and viability can be identified at optimal depth. We experimented with different depth of the cell culture chamber: 0.5 mm, 1 mm, 2 mm and 3 mm in order to appreciate the relevance of this parameter. The experiments are presented in detail in Supplementary 3 the cells cultured in chamber with 0.5 mm depth showed the highest secretion level.  (using a modified PDMS cover) inside the cell culture well as well as in the outlet microfluidic connector. There was no difference between the indication of the temperature controller and the measurements, showing good thermal transfer of the design system.
Absorption of drugs to the PIC. The accuracy of drug testing experiments performed in microfluidic devices depends on achieving a consistent and reproducible concentration of drug molecules in the solution. However, in microfluidic devices with PDMS components, hydrophobic molecules tend to be absorbed to the walls of microfluidic devices, which reduces their concentration and affect the accuracy of drug testing results. We quantified the absorption of drugs to the tubing and microfluidic chip, the results are presented in Supplementary 5. The surface area of the PDMS cover (~450 mm 2 ) exposed to the media in the PIC was much smaller than the surface area in the tubing (~2300 mm 2 ). Majority of the drug absorption was found to be attributed to the tubing. Initially, after 30 minutes of perfusion, 74% of the APAP can be retrieved from the outlet, only 52% of the diclofenac can be retrieved. After 48 hours of perfusion, drug molecules absorbed to the surface starts to saturate, 98% of APAP and 95% of diclofenac can be retrieved. Diclofenac absorbs more easily to PDMS surface than APAP because of its hydrophobicity and higher log P value. To avoid cross-contamination of the drugs due to absorption, the PDMS cover of the chip and the tubing were not reused in the experiment.
Comparison with a laminar flow bioreactor. To benchmark the performance of the PIC against other perfusion culture systems, we compared the functions of cells in a bioreactor reported previously 26 and the PIC (in both cases we use the same constrained spheroids cell culture model). The functions are compared with the cells cultured in 3 other control configurations. Monolayer setting refers to seeding hepatocyte directly to cell culture plates. Spheroids on AHG has the same substrate modification method and cell-culture configuration as PIC but cultured in multi-well plates. In this setting, hepatocytes are seeded on a low-adhesion substrate (glass coverslips modified with poly(ethylene glycol) (PEG) and galactose ligand 1-O-(60-aminohexyl)-D-galactopyranoside (AHG, MW 279)) to promote spontaneous spheroid formation 20 . We previously demonstrated that hepatocyte spheroids cultured on AHG-modified substrate exhibited improved polarity and liver-specific functions 27 . Collagen sandwich configuration is the industry standards that has often been recommended in FDA guidelines for in vitro hepatotoxicity testing 28 .
We measured the urea synthesis and albumin secretion levels in both perfusion culture systems over 8 days (Fig. 6). The cells cultured in PIC and the bioreactor produced similar levels of urea while both higher than static culture as expected. The albumin production in PIC is significantly higher than the laminar flow bioreactor. These results indicate that tangential flow that characterize the PIC structure is essential in maintaining cell functions.
We used experiment success rate to gauge the robustness of PIC compared with bioreactor. All the 12 experiments conducted with PIC and collagen sandwich were successful without contamination. 9 out of 12 experiments with spheroids cultured on AHG were successful, with 3 experiments having extensive cell detachments after 7 days of culture. Only 3 out of 12 experiments were successful for laminar flow bioreactor, mostly because of bacterial contamination issue during the culture. All the experiments were conducted by the same operator.

Maintenance of cell function in PIC.
Long-term cell viability on chip. The PIC system for hepatocyte constrained spheroid culture has been tested for long-term cell culture for 24 days. Optical images taken during this period of the constrained spheroids are presented in Fig. 7a. At day 24 a live/dead staining was performed. The images presented in Fig. 7b demonstrates the viability of the cultured cells can be maintained for 24 days.
Maintenance of hepatocyte differentiated function. To investigate whether the PIC can support chronic hepatotoxicity testing of drugs, we studied whether the hepatocyte functions can be maintained over 14 days. The results of urea secretion and albumin synthesis are presented in Fig. 8a and b respectively.
Urea synthesis in spheroids on AHG increased compared to collagen sandwich culture (Fig. 8a); the increase was more significant than the increase in albumin secretion (Fig. 8b). The urea produced by hepatocytes cultured in collagen sandwich configuration is 87. Hepatocytes cultured in PIC showed significant increase in albumin secretion compared to collagen sandwich throughout 14 days of culture (Fig. 8b)  We quantified the gene expression levels of CYP1A2, CYP2B1/2 and CYP3A2 in PIC and compared with monolayer and collagen sandwich culture. These CYPs are rat homologs of human CYP1A2, CYP2B6, and CYP3A4 respectively. They have important roles in drug metabolism. CYP1A2 is the second most abundant CYPs in human body and is inducible by many of the carcinogens and aromatic hydrocarbons. CYP2B6, although only account for 5% of total CYP component, is responsible for the metabolism of more than 25% of drugs in the market. CYP3A4 is the most abundant in human liver and responsible for metabolism of two-thirds of all marketed drugs.
The activities of CYP1A2, CYP2B1/2 and CYP3A2 enzymes were measured using their respective specific substrates: phenacetine, bupropion and midazolam (Fig. 9a). The quantity of their respective metabolite acetaminophen, OH-bupropion and 1′-OH-midazolam were measured with LC-MS. We found that hepatocytes cultured in spheroid model shows enhanced CYP1A2 activity; the amount of acetaminophen (metabolite of phenacetine metabolized by CYP1A2) produced by collagen sandwich cultured hepatocytes was measured at 49.2ng/10 6 cells/90 min, while hepatocytes in PIC produced 195.1ng/10 6 cells/90 min on Day 8.
On day 14, the amount of acetaminophen measured in collagen sandwich and PIC were 26.9ng/10 6 cells/90 min and 156.4ng/10 6 cells/90 min respectively.
The CYP2B1/2 activity of hepatocytes was quantified by measuring the amount of OH-bupropion metabolized from bupropion (Fig. 9b). On day 8, the amount of OH-bupropion produced by collagen sandwich was 17.78 ng/ 10 6 cells/90 min, lower than spheroids on AHG, which produced 32.49 ng/10 6 cells/90 min of OH-bupropion. By comparison, Hepatocytes in PIC produced significantly higher OH-bupropion: 67.32 ng/10 6 cells/90 min on day 8. On day 14, the amount of OH-bupropion produced by collagen sandwich and spheroids on AHG were 12.38 ng/10 6 cells/90 min and 22.54 ng/10 6 cells/90 min respectively. The amount of OH-bupropion produced by PIC cultured hepatocytes was slightly higher at 76.56 ng/10 6 cells/90 min but this is not statistically significant.
To quantify the activity of CYP3A2, we measured the amount of 1′OH-midazolam metabolized by CYP3A2 from midazolam (Fig. 9c). On day 8, the amount of 1′-OH-midazolam produced by collagen sandwich and spheroids on AHG were 57.2ng/10 6 cells/90 min and 140.6 ng/10 6 cells/90 min respectively. The amount of 1′-OH midazolam produced by spheroids in PIC was higher at 167.2ng/10 6 cells/90 min. On day 14, collagen sandwich Application of PIC-cultured hepatocytes in drug safety testing. We studied the hepatocyte spheroids under static and perfusion (PIC) conditions for two weeks for the evaluation of chronic drug effects. Acute toxicity and chronic toxicity of two drugs: Diclofenac and Acetaminophen (APAP) were tested and compared. Diclofenac causes rare but significant cases of serious hepatotoxicity including liver necrosis, jaundice, fulminant hepatitis with and without jaundice, and liver failure 29 . Diclofenac toxicity is difficult to be evaluated due to the problematic estimation of the dose response. Its hepatotoxic effects are not easily reproducible in current animal models, indicating idiosyncratic toxicity of diclofenac 29 . Therefore, the liver function should be monitored on long-term therapy with diclofenac since increased AST/ALT levels are observed in clinical treatment 30 . In vivo, delayed diclofenac induced hepatotoxicity usually occurs 29 . Previous studies showed that in vitro, only high concentrations of diclofenac (>200 µM) induce acute toxicity 30,31 in primary cultures of human hepatocytes. Thus diclofenac repeatedly dosed over 2 weeks in our in vitro PIC was tested. For APAP as a most frequently used pain killer in the world, there is large body of literature characterizing its safety in vitro and in vivo [32][33][34] . Both acute and chronic overdosing of APAP can cause liver toxicity and liver failure 32 .
Acute toxicity (24 h) of diclofenac is presented in Fig. 10a. In this experiment, we observed a significant toxic effect at 500 µM and 1 mM concentration. At the highest concentration (1 mM), the cell viability significantly decreased to 10%. There is no relevant difference between PIC and collagen sandwich culture for diclofenac acute toxicity. The IC50 was calculated to be 495.61 µM and 493.66 µM respectively. For APAP acute toxicity, we observed that cells cultured in PIC were more sensitive to APAP treatment than the cells cultured in collagen sandwich (Fig. 10c). The IC50 of APAP was 22.51 mM and 38.57 mM for PIC and collagen sandwich. Chronic drug toxicity of diclofenac and APAP were evaluated by measuring the viability of hepatocyte spheroids perfused with different drug concentrations for 14 days. For chronic hepatotoxicity, static cultured hepatocytes (Fig. 10b), the lowest cell viability was found at 100 µM diclofenac. In the perfusion culture, spheroids were exposed to repeated dosing of diclofenac every 2 days. The perfusion system is more sensitive for testing chronic drug response of diclofenac: the IC 50 is 50.86 µM for PIC and 121.53 µM for the collagen sandwich control. At 100 µM diclofenac chronic dosage, cell viability is reduced to less than 40% in PIC after 14 days of culture, whereas 100 µM diclofenac dosing doesn't show a significant acute toxicity. Similarly, cells cultured in PIC is more sensitive to chronic APAP toxicity (Fig. 10d): at 1 mM concentration, cell viability was reduced to 21% in PIC, whereas cells cultured in the collagen sandwich control had 32% viability compared to the DMSO control. The IC 50 is 2.391 mM for PIC and 3.059 mM for the collagen sandwich control.

Discussion
In vitro liver cell models based on conventional cell culture platforms such as 2D monolayer and collagen sandwich can support hepatocyte function for a week. 3D cell culture configurations can enhance functions to higher level with their own limitations such as mass transfer and heterogeneous environments etc. Bell et al. proposed a 3D primary human hepatocyte model for in vitro study of liver functions and diseases and drug response 35 Perfusion culture models have been developed to regulate the soluble microenvironment and can sustain hepatocyte function in a confined space for a longer period, such as suspension spheroids perfusion-stirred bioreactor 36 , encapsulated spheroids bioreactor 37 and laminar flow spheroids bioreactor 27 . Various microfluidic chips have been developed to provide physiological shear stress and liquid to cell ratio within the microfluidic compartments for a few days [38][39][40][41] . Carraro et al. 42 proposed a perfusion hepatocyte cell culture having a "vascular-like" microfluidic structure in hepatocyte culture seeded on a nanoporous membrane. Vernetti et al. 43 proposed a 3D human liver model organized in four-cell sandwich structure (hepatocytes, endothelial cells, stellate cells and Kupffer cells). Lee et al. 44 fabricated a PDMS device which mimics the liver anatomy. The device features an artificial liver sinusoid with an artificial barrier layer mimicking endothelial barrier layer. We have also integrated cell lines and primary cells into 3D microfluidic channels to evaluate the cellular response to drug exposure 4,26,27,45-47 . However, these devices have not been utilized in chronic drug safety testing applications due to the deterioration of hepatocyte functions and contamination problems associated with repeating dosing of drugs to the cells. A number of chips have been developed for long-term culture [48][49][50][51][52] . Wagner et al. 53 combined traditional PDMS microfluidic device with transwell inserts and created a multi-organ chip capable of supporting long-term cell culture. Maschmeyer and colleagues 54 developed a PDMS multi-chamber chip for long-term co-culture of 4 tissue types. Recently, Kang et al. 49 presented a long-term transwell microfluidic device using PDMS microchannels and polyethyleneterephthalate (PET) membranes to mimic the liver sinusoid structure. These devices have the advantage of adjusting fluid flow and controllable liquid to sample ratio, enabling prolonged cell culture. However, these PDMS devices all rely on traditional incubator and involve complex fluidic connections to operate. The drawback of these devices is that when they undergo frequent transfer of devices from incubator to microscope for observation, changing media and adding drug doses, the risk of contamination and bubble generation-once the chip is reinserted in the microfluidic setup-becomes higher. Our PIC is such a device that integrates incubator functions such as temperature and pH control, and with active debubbler for 3D long-term culture and systematically ensure contamination-free long-term cell culture for repeatedly dosed drug testing. This PIC system doesn't need to be inserted in an incubator system, it works independently in a biosafety cabinet. It enabled long-term maintenance of 3D hepatocyte spheroid culture. We tested cell viability for 24 days, while the cell function was verified for 2 weeks. The integration of all the elements such as temperature controller and active debubbler allowed constrained spheroid model to be cultured in a stable microenvironment for weeks. Heaters have been used in cell culture platforms adapted to robotic systems and real-time imaging systems [55][56][57][58] . We applied this concept to our long-term cell-culture chip and implemented a thermocouple and a temperature controller to achieve precise maintenance of temperature during prolonged culture period. We have created a closed microenvironment for cell culture and manipulation inside a sterile biosafety cabinet, eliminating the need of an incubator. Moreover, the tangential flow allowed refreshing of media, removing of by-products and protection of the spheroids cell culture from shear stress.
Microfluidic cell-culture systems typically need to be pre-conditioned and sterilized before cell seeding 59 . Bubbles can form at fluidic connections or generated from the dissociated gas in the media. A great deal of care is needed during the operation of microfluidic chips to ensure bubble-free condition in the cell culture. In the long-term cell-culture chip, the chance of bubble accumulation in the microfluidic channels is even higher, amplifying the effects of bubbles to perfusion flow and cell function. Bubble traps [60][61][62] are usually used to prevent bubbles from entering cell culture area of the microfluidic chips. When the bubble trap is filled with bubbles over weeks, additional bubbles can still enter cell culture areas. Therefore, we integrated a bubble trap and an active-debubbler to trap and remove the bubbles generated during long-term culture.
A long-term cell-culture system must also maintain a microenvironment with favourable mass transport and limited shear stress. In this study, we used an established hepatocyte culture model: the constrained spheroid model, due to its superior ability to maintain cell function and minimize cell loss 27 . Here, we further optimized the spheroid seeding and perfusion culture process. We pre-formed the spheroids on glass coverslips and constrained them with parylene C ultra-thin porous membrane in 48-well plates. After spheroid stabilization, they are transferred to the PIC chip to take advantage of the perfusion flow. We demonstrated improved and well-maintained activity of hepatocytes' CYP1A2, CYP2B1/2 and CYP3A2 enzymes. We validated the use of PIC for acute and chronic drug toxicity studies. We found that for APAP, the acute toxicity response is more sensitive in the PIC compared to static culture. Similar APAP toxicity testing results 63 were reported in perfusion bioreactor: perfusion of APAP resulted in a shift in dose response such that 20 mM of APAP in perfusion culture lead to significant acute toxicity compared with 40 mM in static culture. The reason is that, the metabolic activity is elevated in perfusion culture microenvironment. Hepatocytes can metabolize APAP faster and produce higher quantity of its reactive intermediate NAPQI, which causes free-radical damage of cellular structures 64 . We further validated the use of our PIC chip for chronic drug testing by evaluating hepatotoxicity after 2 weeks' exposure to diclofenac and APAP and obtained a more sensitive chronic toxicity response in the perfusion-cultured spheroid model compared with the static cultured collagen sandwich control. Chronic toxicity of diclofenac is more sensitive in the PIC compared with static culture. Diclofenac chronic toxicity is largely attributed to one of its phase II metabolite: diclofenac-acylglucuronide due to covalent binding 65 . Phase II metabolism requires longer time for metabolite production than primary metabolites. Therefore at 100 mM concentration, diclofenac doesn't show acute toxicity but results in chronic toxicity.
The PIC is currently a low throughput chip that addressed the contamination issues associated with long-term culture of cells in 3D to preserve high levels of cell function for drug testing applications. In the near future, such in vitro model could be used to study the complex spatiotemporal cues that govern long-term changes in cell fates. The PIC can be further developed into high-throughput format, to involve multiple cell types in co-culture, and to support industry-scale drug screening applications.
In conclusion, we presented a method and a system for robust long-term hepatocytes constrained spheroid culture for chronic drug testing. The PIC integrates a heater and temperature controller to control temperature, CO 2 pressure-driven culture-media flow to control pH, and active debubbler on chip, in a biosafety cabinet, to minimize the risk of contamination. The tangential flow over the parylene membrane (that stabilize the spheroids and limit the shear stress) assures a controlled microenvironment. The PIC maintained the cell functions for 2-3 weeks, supporting repeated dosing chronic drug testing. This PIC is a robust long-term 3D perfusion culture platform that can be used also for culturing different cell types for other sub-acute and chronic drug testing applications.

Methods
Flow simulation. As a powerful tool based on the transport theories of momentum, mass and energy, computational fluid dynamics (CFD) simulation yields the profiles of shear force and O 2 concentration in this study. The geometry and mesh of the flow domain were constructed using GAMBIT 2.3.16, and the created mesh was introduced into FLUENT 6.2.16 for calculation. The flow was considered as laminar, steady and three-dimensional, with the operating temperature of 37 °C. The inlet and outlet of the bioreactor were modelled as "velocity-inlet" and "outflow", respectively, while the walls were considered as no-slip. In addition to the Navier-Stokes equations, a species transport equation was solved to calculate the oxygen concentration in the bioreactor. The concentration of dissolved O 2 at the inlet was set as 6.08e-3 kg/m 3 according to its solubility in plasma (α) at 37 °C. The zone occupied by the hepatocytes under the membrane was modelled as a "source term" for O 2 , from which the dissolved oxygen was consumed at the Oxygen Uptake Rate (OUR) of the hepatocytes 66 .
where c is the local O 2 concentration; V m is the maximum rat hepatocytes O 2 consumption rate; and K m is the half-saturation constant for hepatocytes. The parameters were taken from Rat hepatocytes isolation and spheroid pre-culture. We used a constrained spheroid model established previously 27 as the hepatocyte cell culture model. Hepatocytes aggregated in spheroids are able to retain polarity and form functional bile canaliculi 67 . Functional parameters such as albumin and urea production, CYP450 activities are also improved compared with collagen monolayer 19,68 and collagen sandwich culture 69,70 .
To obtain hepatocyte spheroids, we isolated primary rat hepatocytes from male Wistar rats of 250-300 g in weight using a two-step collagenase perfusion method 37 . Animals were handled according to the Institutional Animal Care and Use Committee (IACUC) protocol. Isolation protocol was reviewed and approved by the Biological Resource Center (BRC) Institutional Animal Care and Use Committee. Freshly isolated hepatocytes were seeded onto collagen-coated 48-well plate or PEG-galactose modified glass coverslips at 1 × 10 5 cells/cm 2 to pre-form the spheroids 27 . The glass coverslips were modified according to the protocol 27 . The cells were cultured in William's E medium supplemented with 1 mg/mL BSA, 10 ng/mL EGF, 0.5 μg/mL insulin, 5 nM dexamethasone, 50 ng/mL linoleic acid, 100 units/mL penicillin and 100 μg/mL streptomycin; and incubated with 37 °C, 5% CO 2 , 95% humidity. Cell seeding glass coverslips 10mm in diameter were purchased from Paul Marienfeld GmbH & Co.KG (Lauda-Königshofen, Germany). Silane-PEG-COOH, MW 5000 was purchased from Nanocs Inc. (New York, USA). 1-O-(6′-aminohexyl)-D-galactopyranoside (AHG, M.W. 279), the galactose ligand was synthesized in house as reported previously and verified by NMR spectrum 20,71 . Wafers used for chip fabrication were purchased from Bonda Technology (Singapore). Other chemicals were purchased from Sigma-Aldrich (Singapore) unless otherwise stated. Parylene C membranes used for sandwich constrained spheroid culture were fabricated and surface modified 27 . After 1 day in static culture, the coverslips with attached spheroids were manually moved to PIC using sharp point forceps and covered with parylene C membrane. The PIC was sealed by clamping the PDMS-glass cover on top of the culture chamber. To evaluate the robustness of the PIC, we conducted cell culture experiments in cell culture plates, in a laminar flow bioreactor previously developed 26 and in the PIC in parallel. Six independent 14-day cell culture experiments were performed in duplicates for each setup.
Collagen sandwich culture. We used both the hepatocyte monolayer and the collagen sandwich as standard controls for hepatotoxicity testing of drugs, in order to compare with the results achieved from the PIC. The bottom collagen-coating substrate was prepared by adding 40 µL neutralized collagen type I solution (Invitrogen, Palo Alto, USA) onto the 10mm glass coverslip before incubation at 37 °C overnight for gelation. Hepatocytes were seeded on the collagen-coated coverslip at 1 × 10 5 cells/cm 2 density; they were incubated for 1 h for full attachment before media replenishment and then cultured for 24 h. The culture medium was removed; subsequently, 40 µL of un-gelled collagen was overlaid on top of the cells. Gelation of the collagen overlay was allowed to occur at 37 °C for 3 h before fresh medium was replenished. The collagen sandwich culture was placed in the incubator with daily change of culture media.
Urea and albumin synthesis measurement. 1 mL of culture media were collected for urea and albumin synthesis measurements each day. Urea content in the culture media was measured using Urea Nitrogen Kit (Stanbio Laboratory, USA). Albumin concentration was measured using Rat Albumin ELISA Quantification Kit (Bethyl Laboratories Inc., USA). The absolute urea and albumin amounts were calculated based on media volume collected and was normalized against the cell number by the end of culture.

Diclofenac and APAP concentration measurement.
To determine the amount of diclofenac and APAP absorbed to the PIC and tubing, concentration of diclofenac 72 and APAP 73 were measured by colorimetric method using a Tecan infinite m200 plate reader (Tecan, Switzerland).
Cell viability staining. The viability of hepatocytes was visualized by a dual staining method. 2 fluorescent nuclear dyes: Propidium iodide (PI) (Molecular Probes, USA) and Calcein AM (Molecular Probes, USA) were used to stain the necrotic and viable cell population. The staining was done by moving the glass coverslip from the PIC to a microscope slide. The cells were stained by incubating in 100 μl of 25 mg/mL PI (30 minutes) and 1 μM Calcein AM (30 minutes); the samples were washed 3 times with 1X PBS before imaging. All stained samples were imaged with a confocal microscope at 488 nm and 543 nm excitation.
Liquid chromatography-mass spectrometry (LC-MS) measurement. For CYPs specific activity analysis, hepatocytes were cultured for 8 or 14days before adding CYP-specific substrates diluted in Krebs-Henseleit buffer (KHB) and incubated for 1.5 hours. Hepatocytes in perfusion cultured were removed from the bioreactor and transferred to multi-well plates for this experiment. The samples were collected and stored at −80 °C until LC-MS measurement. To conduct the LC-MS measurement, 50 µL of 100ng/mL internal standards were added to the samples and the mixture was dried using Techne ® Sample Concentrator (Techne, UK). The dried residues were reconstituted using 100 µL of methanol containing 0.1% formic acid. The supernatant was then analysed using LC-MS system (LC: 1100 series, Agilent, US; MS: LCQ Deca XP Max, Thermo Finnigan, US) with 100 × 3.0 mm onyx-monolithic C18 column (Phenomenax, USA) as reported previously 74 . Acute toxicity. The hepatocyte spheroids were tested for acute toxicity. 1 day after culturing in PIC, The cells were exposed to 4 concentrations of diclofenac (10-1000 μM) and APAP (10-40 mM) for 24 hours under perfusion condition. DMSO vehicle controls were established by using medium supplemented with 1% DMSO without drugs. Viability was tested after removing the glass slides containing cells from PIC to a 48-well-plate. The viability was assessed with MTS assay (Promega, USA).
Chronic toxicity. For the assessment of chronic toxicity of diclofenac and APAP, the hepatocytes were exposed to four concentrations of diclofenac (1-100 μM) and APAP (0.1-10 mM). The selection of these concentrations range corresponds to the therapeutic serum concentration (C max ) of diclofenac (6.4 μM) 75 and APAP (1 mM) 76 . Vehicle control (0.1% DMSO) was used. Serum free William's E medium was used during drug treatment. Repeated doses were given every 48 h for static culture upon medium change. The viability was assessed with MTS assay (Promega, USA) upon 14 days after seeding of the hepatocytes. Statistical analysis. At least 3 experimental replicates were used for each data point. In each experiment, 2 chips were used to generate duplicated data. Statistical comparisons were performed using 2 way ANOVA. Results are expressed as means ± standard error of the means (sem) of 3 independent experiments.