Differentiation of human‐induced pluripotent stem cell under flow conditions to mature hepatocytes for liver tissue engineering

Abstract Hepatic differentiation of human‐induced pluripotent stem cells (hiPSCs) under flow conditions in a 3D scaffold is expected to be a major step forward for construction of bioartificial livers. The aims of this study were to induce hepatic differentiation of hiPSCs under perfusion conditions and to perform functional comparisons with fresh human precision‐cut liver slices (hPCLS), an excellent benchmark for the human liver in vivo. The majority of the mRNA expression of CYP isoenzymes and transporters and the tested CYP activities, Phase II metabolism, and albumin, urea, and bile acid synthesis in the hiPSC‐derived cells reached values that overlap those of hPCLS, which indicates a higher degree of hepatic differentiation than observed until now. Differentiation under flow compared with static conditions had a strong inducing effect on Phase II metabolism and suppressed AFP expression but resulted in slightly lower activity of some of the Phase I metabolism enzymes. Gene expression data indicate that hiPSCs differentiated into both hepatic and biliary directions. In conclusion, the hiPSC differentiated under flow conditions towards hepatocytes express a wide spectrum of liver functions at levels comparable with hPCLS indicating excellent future perspectives for the development of a bioartificial liver system for toxicity testing or as liver support device for patients.

production of hepatocytes required for BAL development. Furthermore, the use of the patients' own hiPSCs may allow for personalized treatment and thereby avoiding immunological reactions. Although hiPSC-derived hepatocyte-like cells have been shown to have certain liver-specific phenotypic characteristics and exhibit many of the liverspecific functions (Chen et al., 2012;Gieseck et al., 2014;Si-Tayeb et al., 2010;Song et al., 2009), most of these functions are expressed at levels several magnitudes lower than in fresh liver tissue or freshly isolated human hepatocytes (Ulvestad et al., 2013), suggesting that improvements in the differentiation protocols are still warranted.
In most of these studies, the induced pluripotent stem cell-derived hepatocyte-like cells were obtained by maturation in 2D cultures, and the cells are loaded in a bioreactor only after maturation. Hepatic differentiation and maturation directly in the 3D bioreactor may offer great advantages such as overcoming the need to harvest the total amount of cells needed for the BAL from the 2D culture and loading in a 3D bioreactor. However, relatively few studies have investigated the hepatic differentiation of stem cells directly in a 3D perfusion bioreactor or BAL using embryonic stem cells (ESC; Fonsato et al., 2010;Miki, Ring, & Gerlach, 2011;Pekor, Gerlach, Nettleship, & Schmelzer, 2015;Schmelzer et al., 2010;Sivertsson, Synnergren, Jensen, Bjorquist, & Ingelman-Sundberg, 2013;Wang et al., 2012), and only three of them used hiPSCs (Freyer et al., 2016;Giobbe et al., 2015;Luni et al., 2016).
Flow of the medium was shown to have beneficial effects on hepatic differentiation of ESC and fetal liver cells and to improve liver functions of ESC-derived hepatocytes (Lu et al., 2015;Pekor et al., 2015;Schmelzer et al., 2010;Yu et al., 2014). Even simple recirculation of medium in a rotating bioreactor improved the function of the differentiated hepatocyte-like cells (Fonsato et al., 2010;Wang et al., 2012). The flow may not only physically influence the cells by creating flow forces, but it also may improve mass as well as gas transfer between the cells and the medium and promote the removal of waste products (Pekor et al., 2015;Yu et al., 2014). However, in another study, perfusion inhibited adipogenic differentiation of adipose-derived stem cells possibly by washing away autocrine or paracrine factors (Hemmingsen et al., 2013).
A limitation of many studies is the absence of a proper benchmark to evaluate whether the cells are fully differentiated with respect to the expression levels of liver-specific markers and liver functions of the generated hepatocytes. For example, many studies have not used a benchmark at all, whereas others have used PHH cultured in vitro for 2 days or more (Iwamuro et al., 2012;Song et al., 2009;Takayama, Inamura, Kawabata, Katayama, et al., 2012;Takayama, Inamura, Kawabata, Sugawara, et al., 2012). However, it is known that PHH cultured beyond 24-48 hr rapidly lose their phenotype and liver-specific functions, and using these cells as a benchmark results therefore in an overestimation of the maturation level of the stem cell-derived hepatocytes (Ulvestad et al., 2013). By contrast, fresh human precision-cut liver slices (hPCLS) contain hepatocytes in their natural 3D tissue-matrix configuration, in contact to the other liver cell types, and retain expression as well as activity of Phase I and Phase II metabolic enzymes at levels comparable with the in vivo situation (Elferink et al., 2011). Therefore, hPCLS can be considered the gold standard for assessing the maturation of stem cell-derived hepatocytes into fully differentiated hepatocytes.
Here, we differentiate hiPSC-derived definitive endoderm (DE) cells into hepatocytes in situ in a perfusion bioreactor system. Hepatic differentiation and functionality of hiPSC-derived hepatocytes were assessed using fresh hPCLS as benchmark for ex vivo liver, and 2D static cultures were used to compare differentiation efficacy in 2D static and 3D flow systems.

| Scaffolds fabrication and perfusion cell differentiation culture
PDMS was chosen as scaffold material instead of hydrogels because of its good biocompatibility and structural stability and possibility to scale it up enabling production of litre-sized scaffolds (Mohanty et al., 2016). Random porous scaffolds (Figure 1e) were fabricated from PDMS by using salt leaching techniques similar to that described previously (Yuen, Su, Goral, & Fink, 2011). Hexagonal combined structured or porous scaffolds ( Figure 1e) were made using a sacrificial mould with hexagonal pattern fabricated by 3D printing using commercially available water dissolvable polyvinyl alcohol (MakerBot, USA) and packed with salt crystals as described before. The scaffolds were treated with oxygen plasma (125 W, 13.5 MHz, 50 sccm, and 40 millitorr) to render their surfaces hydrophilic and sterilized by autoclaving. They were coated with Hepatocyte Coating (from Cellartis Hepatocyte Differentiation Kit, Cat. No. Y30050) by centrifugation at 300 g for 5 min and then left overnight at 37°C. The scaffolds were subsequently washed with phosphate buffered saline centrifugation at 300 g for 5 min and then left in a medium at 37°C for 2 hr prior to being used to experiments.
A self-sustained perfusion system with 16 parallel reactors was constructed ( Figure S1) holding PDMS scaffolds. The scaffold bioreactor array, glass vials, caps, and poly(tetrafluoroethylene) tubing were sterilized by autoclaving before assembling in a laminar flow bench. A 0.5 M NaOH was flushed throughout the system to ensure a sterile fluidic path. The system was subsequently flushed with sterile water and then with culture medium. Coated scaffolds were placed in cylindrical holes in a custom built tray. A 2.5 × 10 6 freshly thawed DE cells in 30 μl of Hepatocyte Thawing and Seeding Medium was pipetted into each scaffold, and cells were allowed to adhere for 3 hr at 37°C under 95% air/5% CO 2 . The seeding tray was inverted as well as placed vertically in four different positions to allow the cells to distribute throughout the scaffolds during 3 hr. The scaffolds were then placed in the 4 × 4 bioreactor array of the fluidic platform, and media were perfused through the scaffolds at flow rates of either 1 or 5 μl/min. The entire system was incubated at 37°C under 95%air/5% CO 2 .
Cells were cultured and differentiated for 25 days.

| Human liver tissue
Human liver material was obtained from liver tissue of 10 individual patients, remaining as surgical waste after reduced liver transplantation patients, from liver tissue donated after cardiac death but not suitable for transplantation due to the age, or from patients undergoing hepatectomy for the removal of carcinoma. This study was approved by the Medical Ethical Committee of the University Medical Centre Groningen, according to Dutch legislation and the Code of Conduct for dealing responsibly with human tissue in the context of health research (http://www.federa.org), refraining the need of written consent for "fur-

| Phase II metabolism
For Phase II metabolism studies, hPCLS and cells in perfused systems or in static condition were exposed to 100 μM of 7-hydroxycoumarin (7-HC; Sigma-Aldrich, St.Louis, MO, USA) for 1-3 hr. Medium was collected at outlet tubes or from the incubation medium and stored at −20°C until further analysis, using 7-HC, 7-HC-glucuronide, and 7-HC-sulfate as standards. The metabolite production was normalized per milligram protein and per hour.

| Gene expression
Total cellular RNA from cells or PCLS was purified by using the RNeasy Micro kit (Qiagen, 74004) or using Maxwell 16 LEV simplyRNA Tissue Kit (Promega, USA), respectively. The RNA was converted to cDNA using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, 4374966), and quantitative real-time PCR was conducted using the TaqMan Gene Expression Assays (Applied Biosystems 4331182). More details are provided in the Supporting Information. which is lower than the corresponding 2D static cultures (usually 80,000-100,000 stem cell-derived hepatocytes/cm 2 ; data not shown).

| Statistical analysis
However, as determined by visual inspection, the cells adhered typically in clusters, and therefore, the local cell densities were probably higher.
Differentiation of hiPSC-derived DE cells loaded in the scaffolds was performed at a flow rate of 1 μl/min (exchange rate every 50 min) and 5 μl/min (exchange rate every 10 min). It was calculated that in both flow regimens, shear forces were very low (1.1e-4 dyn/ cm 2 or less; Figures S7 and S8). Furthermore, the exchange rate of once per 10 min was shown to be compatible with differentiation into hepatocytes in 2D flow cultures ( Figure S2) and has previously shown to support differentiation of adipose-derived stem cells into adipocytes using conditioned medium (Hemmingsen et al., 2013).

| Comparison of gene expression of hepatocyte markers of human iPSC-derived hepatocytes and hPCLS
We investigated the hepatic differentiation of hiPSC-derived DE cells in the 3D scaffold at two different flow rates, 1 and 5 μl/min, and in two different scaffold designs ( Figure 1e) and compared the results with cells differentiated under static 2D conditions in standard PS well plates or in well plates coated with a PDMS layer. hPCLS were used as a benchmark for cells differentiated in the scaffold under perfusion. hPCLS were prepared from 10 individual livers of human donors, aged 20-73 years (60% female) as described (de Graaf et al., 2010). Due to the limited amount of liver material, not every test was performed on all donor livers. The morphological appearance and ATP content of the hPCLS after a preincubation of 1 hr to restore ATP levels (0 hr) and after 24 hr of incubation indicated that the slices were viable. ATP levels were 9.7 ± 1.3 and 8.24 ± 0.76 pmol/μg protein at 0 and 24 hr, respectively (mean ± standard error of the mean). Morphology showed intact liver tissue at 0 and after 24 hr of incubation ( Figure 1d).
The gene expression of the liver-specific genes of the cells differentiated under the conditions outlined above and the hPCLS is depicted in Figure 2. A summary based on classification of gene expression into broader groups is presented in Figure 3. When  most of the CYP genes, the expression of the epithelial biliary cell markers (CK7 and BGP), and the drug transporter ABCB1 (multidrug resistance protein, P-gp) in the cells were in the range of that seen in hPCLS. Large differences in gene expression were observed for the genes CAR, ALB, and BSEP. These genes were clearly expressed lower in differentiated cells compared with hPCLS. These three genes were, however, expressed higher in differentiated cells than in the DE cells (Figures 2 and 3). Furthermore, the differentiated cells showed higher expression of AFP and HNF4a than the hPCLS. The differentiated cells on PS therefore had a mixed phenotype, where some genes suggest a partly to fully maturated phenotype (HNF4a, CYP3A4, 3A5, 3A7, and 2B6 and P-gp), whereas others suggest a less mature phe- Flow modulated the gene expression of differentiated cells only to a small extent, with 5 μl/min performing slightly better than 1 μl/min for CYP 3A4 and 2B6. Flow also modulated the ALB and AFP expression; the ALB expression was suppressed by flow compared with the corresponding static cultures. The AFP expression was lowest in cells exposed to the 5 μl/min perfusion compared with static and perfusion with 1 μl/min but did not result in the very low levels observed in hPCLS (Figures 2 and 3).

| Liver functions
Phase I and Phase II metabolic activities, albumin, urea, and total bile acid production were analysed in hPCLS and the hiPSC-derived hepatocytes for each of the different conditions described above.

| Phase II metabolism
Differentiated cells exhibited high uridine UDP-glucuronyltransferase and sulfotransferase activities when exposed to 7-HC ( Figure 5). Both Phase II activities were higher in hiPSC-derived hepatocytes at a flow of 5 μl/min than at 1 μl/min. Although the activities in cells cultured under static conditions were similar to those in hPCLS static cultures, the 7-HC-glucuronide production by cells cultured at 5 μl/min flow in both hexagonal and random scaffolds was on average twofold higher than in liver slices at flow conditions. In addition, the sulfation rate of 7-HC resulting in 7-HC-sulfate was 30-40-fold higher in cells differentiated under flow condition compared with differentiation under static condition and the hPCLS.

| Albumin production
Albumin production by the hiPSC-derived hepatocytes was overlapping with the lower range of that of hPCLS ( Figure 6). No difference was observed between the two types of scaffolds at both flow rates or between static and perfusion cultures.

| Bile acid secretion and urea synthesis
Bile acid secretion by the hiPSC-derived hepatocytes was at the same level of 25-30 pmol·hr −1 ·mg −1 protein in cells differentiated under static conditions for 22 or 24 days as in hPCLS ( Figure S9). We could not detect bile acids in the samples of the outflow medium obtained of the scaffolds. This was probably because the excreted compounds were highly diluted by the perfusion flow rate. The observed total bile acid secretion of 25-30 pmol·hr −1 ·mg −1 protein by differentiated cells or hPCLS would result in a concentration of about 20-100 pmol/ml at a flow rate of 1 and 5 μl/min, respectively, in the BAL, which is below the detection limit of the bile acid assay.
On average, the urea production by hiPSC-derived hepatocytes was below or in the lower range of that of hPCLS (0.06-7.6 μg·hr −1 ·mg −1 protein for hiPSC-derived hepatocytes and 1.6-11.9 μg·hr −1 ·mg −1 protein for hPCLS; Figure S10). Cells differentiated at 5 μl/ min flow and under static conditions tended to show higher urea

| DISCUSSION
We have obtained highly differentiated hepatocytes from hiPSCs under flow conditions in a BAL. To assess their differentiation status, we compared the expression and function of the cells in this BAL model with fresh human liver slices that have in vivo-like activities (de Graaf et al., 2010; and found as yet unprecedented liver functions in the differentiated cells. Moreover, we found that DE cells can be successfully differentiated into hepatocyte-like cells in a 3D scaffold in a bioreactor under flow conditions, to a similar or only slightly better differentiation grade than under static 2D conditions, especially with respect to Phase II sulfation activity and a lower AFP expression, which can make the production of a BAL easier and more effective in the future.

hiPSC-derived cells differentiated under flow in a 3D bioreactor
resulted in a BAL model with overlapping Phase I metabolism (except for CYP2B6) and similar or higher Phase II metabolism, compared with fresh human liver slices. Urea production was present in the hiPSC but was below or in the lower range of the hPCLS. However, the capacity of urea production in the BAL from externally added ammonia, in order to mimic ammonia from extrahepatic tissues, was not assessed because no ammonia was added to the medium. Further studies with exposure to extracellular ammonia are needed to show the ability of the cells to detoxify ammonia, which is important for patients with liver diseases where high concentrations of neurotoxic ammonia are detected. Bile acids were produced by hiPSC-derived hepatocytes at the same rate as in fresh tissue slices. The gene expression of P-gp in hiPSC-derived hepatocytes was shown to be higher than in hPCLS, which is remarkable as a 10-20-fold lower expression was found in differentiated hiPSC compared with human hepatocytes by Lu et al. (2015). However, the gene expression of BSEP in the differentiated cells was lower than in the hPCLS. The hiPSC-derived cells also expressed both CK-7 and BGP, similar to the hPCLS, indicating that in addition to the hiPSC-derived hepatocytes, also some biliary epithelial cells are present. This bipotent differentiation potential of iPSCderived hepatic progenitor cells was also found previously (Freyer et al., 2016;Yanagida, Ito, Chikada, Nakauchi, & Kamiya, 2013). The albumin secretion of stem cell-derived hepatocytes achieved here is similar to hiPSC-derived hepatocytes found by Gieseck et al. (2014) and 10-100-fold higher than in human ESC-derived hepatocytes (Miki et al., 2011) and 3-40-fold higher than mouse iPSC-derived hepatocytes (Iwamuro et al., 2012) but lower than in fresh tissue slices.
Although the mRNA expression was high for HNF4a indicating hepatic differentiation and low for CYP3A7, which is a fetal enzyme with low expression in the adult liver, the relatively high expression of the fetal protein AFP indicates that maturation of the cells is not fully complete.
This has also been observed by others (Chen et al., 2012;Freyer et al., 2016;Gieseck et al., 2014;Iwamuro et al., 2012;Kim et al., 2015;Sasaki et al., 2013). It still needs to be addressed how relevant this is for the functioning of the BAL in patients who need liver support or for toxicity testing. Taken together, these results show that hiPSC differentiated under static conditions as well as under flow in a scaffold have liver functions close to those in fresh human liver tissue.
The significant improvements with respect to liver functions of the differentiated cells presented here compared with other studies could be due to better differentiation protocols, whereas the difference between the cells differentiated under 2D static and 3D perfusion conditions could be ascribed to better nutrient delivery and waste removal.
Because PHH functions decrease rapidly and drastically (10-1000-fold after 48 hr culture) during in vitro culture (Ulvestad et al., 2013), the use of these PHH as standard tends to overestimate the metabolic function of hiPSC-derived hepatocytes. Therefore, the comparison of those data with our study is difficult. Moreover, comparison of the metabolic activity data between different studies is further hampered of those studies with our data shows that the CYP3A, CYP2C9, and CYP1A activities of hiPSC-derived hepatocytes in our study were 10 to several hundred folds higher. In the study of Freyer et al., no CYP2C9 activity at all was detected in the derived hepatocytes. We were the first to measure Phase II metabolism in hiPSC and found that glucuronidation was comparable with PCLS, but sulfation was remarkably higher after differentiation under flow, which requires further studies.
The gene expression of CYP-enzymes and their activity varied notably between donor livers. This variation is very well known in the human population, which is among others a result of polymorphisms and induction by environmental and physiological factors. For example, CYP1A2, CYP2D6, CYP2C9, CYP2C19, CYP2B6, and CYP3A4 are known to be polymorphic and highly inducible enzymes in human (Zhou, Liu, & Chowbay, 2009). With this in mind, it is noteworthy that the gene expression and enzyme activities in the hiPSCderived cells overlapped in most cases with those in the hPCLS, with a few discrepancies noted below. For example, the CYP3A5 gene was higher expressed in hiPSC-derived hepatocytes compared with hPCLS, whereas CYP3A4 gene expression only reached up to the lower range of human livers (Figure 2). The fact that CYP3A4 and 3A5 have strongly overlapping substrate specificities (Emoto & Iwasaki, 2006) may explain why the total CYP3A metabolism of midazolam was similar in hiPSC-derived hepatocytes and hPCLS ( Figure 4). Although the gene expression of CYP2B6 in the differentiated cells was in the lower range of hPCLS, possibly due to low CAR expression, the activity of this enzyme was at least 10 times lower in hiPSC-derived hepatocytes than in hPCLS indicating a post-transcriptional regulation. Future research will be focused to improve the as yet under-expressed CAR-mediated pathway.
We found a limited influence of the flow rate on the hepatic differentiation of hiPSC in the BAL. A flow of 5 μl/min resulted in a slightly better hepatocyte differentiation and maturation than the 1 μl/min flow. This might be explained by the better nutrient and oxygen supply and removal of waste metabolites at the higher flow rate. Also, the type of scaffold had no obvious impact on the differentiation.
In conclusion, most of the drug metabolism enzyme activities of the developed hiPSC-based BAL were in the same order of magnitude as in the fresh human tissue, which is an important achievement in liver tissue engineering and is thus promising for future applications in drug metabolism and toxicity testing. A limitation of this study is that besides hepatocytes and biliary epithelial cells, which were present in the developed BAL according to gene expression profiling, no non-parenchymal cells are present yet. Inclusion of non-parenchymal liver cells is essential in the development of a fully functional BAL and will be a future step in this field. Also, for future toxicity tests, it is necessary to add these other liver cell types, as it is well known that these non-parenchymal cells play an important role in drug-       Figure S6). The surface area of a salt particle is 0.35*0.35*6 = 0.735 mm 2 .
1556*0.735 = 11.39 cm 2 . However because the salt crystals need to touch each other in order to form a network, some of the area is lost. In the idealized situation, each cube loses about 1.5 sides in surface area as it shares that area with other salt crystals. Therefore the area is estimated to be 11.39*9/12 = 8.5 cm 2 In Estimate 2 we used the measured porosity (determined to be 65%) of a random scaffold as input parameter and calculated the number of salt molecules that could fit into pores with a total volume of 0.65 multiplied by the scaffold volume: r 2 *pi*h*0.65 = 3*3*pi*5*0.65 = 92 mm 3 . Volume of salt particle is 0.35*0.35*0.35mm 3 = 0.042875 mm 3 . Number of salt particles in scaffold are 92/0.042875 = 2145 particles. Number of particles multiplied by the surface of each salt cube (6*0.35*0.35mm 2 = 0.735 mm 2 ) = total surface volume: 1577 mm 2 = 15.8 cm 2 . However, just as in the case above, some of the sides of the cubes are shared between each sugar cube. Using the estimate above, it is suggested that the surface area is 15.8 cm 2 *9/12 = 11.85 cm 2 . In this case, the surface area is estimated to be about 12 cm 2 , which is close to Estimate 1 of 11.39 cm 2 . For simplicity we estimate the surface of a scaffold to 10 cm 2 (1000 mm 2 ).