In vitro ballooned hepatocytes can be produced by primary human hepatocytes and hepatic stellate cell sheets

Despite the increasing prevalence of Nonalcoholic steatohepatitis (NASH) worldwide, there is no effective treatment available for this disease. “Ballooned hepatocyte” is a characteristic finding in NASH and is correlated with disease prognosis, but their mechanisms of action are poorly understood; furthermore, neither animal nor in vitro models of NASH have been able to adequately represent ballooned hepatocytes. Herein, we engineered cell sheets to develop a new in vitro model of ballooned hepatocytes. Primary human hepatocytes (PHH) and Hepatic stellate cells (HSC) were co-cultured to produce cell sheets, which were cultured in glucose and lipid containing medium, following which histological and functional analyses were performed. Histological findings showed hepatocyte ballooning, accumulation of fat droplets, abnormal cytokeratin arrangement, and the presence of Mallory–Denk bodies and abnormal organelles. These findings are similar to those of ballooned hepatocytes in human NASH. Functional analysis showed elevated levels of TGFβ-1, SHH, and p62, but not TNF-α, IL-8. Exposure of PHH/HSC sheets to a glucolipotoxicity environment induces ballooned hepatocyte without inflammation. Moreover, fibrosis is an important mechanism underlying ballooned hepatocytes and could be the basis for the development of a new in vitro NASH model with ballooned hepatocytes.

Toxicity assay. We prepared four different types of media (A-D) to analyze the effects of glucose and free fatty acids (FFA) on hepatocytes. Normal glucose/insulin medium was composed of 1 g/L glucose and 6.25 ng/ mL insulin (physiological concentration). High glucose/insulin medium was composed of 4.5 g/L glucose and 6.25 μg/mL insulin. FFA medium contained 300 μM oleic acid and 150 μM palmitic acid (Kohjin Bio, Saitama, Japan). Simple co-culture. To confirm that the cell sheets maintained hepatocyte function, PHH/HSC simple coculture was performed on the collagen dishes using glucolipotoxicity medium group as a control group. The seeding density was the same as that of cell sheets. Details regarding the co-culture are presented in the Supplementary Information.
Morphological analysis. PHH/HSC sheets were imaged on day 11 using digital microscopy (VHX-900, KEYENCE Ltd., Osaka, Japan) and optical coherence tomography (IVS-2000, KEYENCE Ltd.). The area of the PHH/HSC sheets was analyzed using ImageJ ver. 1.53 (NIH, Bethesda, MD, USA). The sheet thickness of was calculated from the cross-sectional area by randomly imaging three different areas in each sheet using optical coherence tomography.
Histological analysis. The PHH/HSC sheets were fixed with 4% paraformaldehyde (PFA), paraffin embedded, and cut into 5-μm thick slices. HE staining, E-cadherin/4′, 6-diamidino-2-phenylindole (DAPI) staining, Vimentin/DAPI staining, CK8/18 staining, alpha-smooth muscle actin (αSMA) staining, and p62 staining were performed on the fixed sections. Immunofluorescence staining was performed for E-Cadherin/DAPI and Vimentin/DAPI. Immunohistochemical staining using 3,3′-Diaminobenzidine-4HCI (DAB) was performed to detect CK8/18, αSMA, and p62. Formalin-fixed human liver (#HLFFPE-B, lot: HL170064) was purchased from Samsara Sciences (Solana Beach, CA, USA) and used as a positive control for p62 staining (Fig. S1). Details of the immunohistochemical analysis, antibodies, and dilutions are described in the Supplementary Information (Table S1). To evaluate hepatocyte ballooning, sections were stained with E-cadherin/DAPI and randomly imaged using a fluorescence microscope (BX51, Olympus, Tokyo, Japan). The cross-sectional area of hepatocytes with stained nuclei was measured using the ImageJ software. At least 15 hepatocytes were analyzed in each experiment under each condition. The average value was set at n = 1. The number of PHHs counted for each condition is described in the Supplementay Information (Table S2). To visualize accumulated lipids, oil red O staining was performed after fixing the sheets with 4% PFA. Briefly, oil red O powder (#154-02072, Wako) was dissolved in 60% isopropanol to prepare the staining solution. The staining solution was added onto PHH/HSC www.nature.com/scientificreports/ sheets and incubated for 20 min, following which the staining solution was washed off, and the sheets were observed under a phase-contrast microscope (ECLIPSE TS2, Nikon Ltd, Tokyo, Japan). To quantify the fat accumulation, oil red O stain was eluted from the sheets using 100% isopropanol and the absorbance was measured at 492 nm using a microplate reader. In addition, fat staining using the osmium method was performed for longterm storage. Oil red O staining and the osmium method are described in the Supplementary Information. For transmission electron microscopy (TEM), PHH/HSC sheets were fixed with 2% PFA and 2% glutaraldehyde in 0.1 M phosphate buffer. The samples were then washed and postfixed with 2% osmium tetroxide (0.1 M). Dehydration, infiltration, embedding, and polymerization were then performed. Samples were ultra-thin sectioned at 70 nm using an ultramicrotome (Ultracut UCT; Leica Vienna, Austria) and stained with 2% uranyl acetate followed by secondary staining with lead stain solution (#18-0875, Sigma-Aldrich). The samples were observed under TEM (JEM-1400Plus, JEOL Ltd., Tokyo, Japan).
ELISA and assay. The bioactive substances present in the culture supernatant were analyzed using the TGF-β1 ELISA kit (#KE00002, Protein Tech, Rosemont, IL, USA), IL-8 ELISA kit (#KE00006, Protein Tech), and Sonic hedgehog (SHH) ELISA kit (#ab100639, Abcam) according to the manufacturer's protocol. The samples were analyzed on day 3 or day 4 when ballooning was first observed (Fig. S2). On day 11, PHH/HSC sheets were homogenized, and protein lysates were prepared. Lysates were analyzed using the p62 ELISA kit (#ADI-900-212, ENZO New York, NY, USA), and protein carbonyl ELISA kit (#STA-310, CELL BIOLABS, San Diego, CA, USA) according to the manufacturer's protocol. The BCA protein assay kit (#T9300A, Takara Bio) was used to determine the protein concentration in the lysates. The Human Albumin ELISA kit (#E88-129, Bethyl Laboratories, Inc., Waltham, MA, USA) and Urea assay kit (#DIUR-100, BioAssay Systems, Hayward, CA, USA) were used to assess hepatocyte function. The samples were analyzed on days 0, 1, 5, and 10.

Statistical analyses.
All experiments were performed at least thrice. The results were expressed as mean ± standard error of the mean (SEM). Intergroup differences between glucose and FFA exposed groups were analyzed using two-way ANOVA with Tukey's post hoc test (for the comparison of more than two groups). All analyses were performed using the GraphPad Prism software (version 9.0, La Jolla, CA, USA). A p value < 0.05 was considered statistically significant.

Results
Effect of glucose and FFA on morphology of PHH/HSC sheets. Following PHH and HSC co-culture, strong aggregation occurred, and cells were detached from the UpCell culture dishes, leading to the formation of PHH/HSC sheets ( Fig. 2A). We tested the effects of glucose and FFA on the fabricated PHH/HSC sheets.
In the non-FFA-treated group, the cells were more loosely connected to each other than in the FFA-treated group (Fig. 2B), whereas the area of PHH/HSC sheets was smaller in the FFA-treated-group (Fig. 2C)  Effects of glucotoxicity, lipotoxicity, and glucolipotoxicity on BH. HE staining showed cell layering and enlargement of PHHs with thinning of cytoplasm (ballooning) in the glucotoxicity, lipotoxicity, and glucolipotoxicity groups on day 11 (Fig. 3A). Vimentin/DAPI staining revealed Vimentin expression in the HSCs in each medium condition and was especially strong in Glucolipotoxicity (Fig. S3). Quantification of the cross-sectional area of PHHs was performed by E-Cadherin/DAPI staining. Results showed that PHHs were more than twofold enlarged in the glucolipotoxicity group than in the normal group (Fig. 3A,B). There was a statistical difference in the FFA group (two-way ANOVA; FFA factor: F(1,2) = 22.63, p = 0.0415; glucose factor: F(1,2) = 4.263, p = 0.1750; FFA × glucose: F(1,2) = 0.6088, p = 0.5169). αSMA staining labelled all cells, including periportal hepatocytes, in day 0 samples. At day 11, in the non-FFA-treated group, polarity in staining was observed as αSMA was localized to the basement membrane side, whereas in the FFA-treated group, staining was observed in all cells, including the periportal hepatocytes (Fig. 3A). Oil red O staining showed fat accumulation in the FFA-treated group (Fig. 4A,B). There was a statistically significant difference between the glucose and FFA treatment groups (two-way ANOVA; FFA factor: F(1,5) = 116.7, p = 0.0001; glucose factor: F(1,5) = 6.868, p = 0.0471; FFA × glucose: F(1,5) = 0.3450, p = 0.5825). Osmium method evaluation and electron microscopy revealed that fat droplets were not observed in HSCs, only in PHH (Fig. 4C,D). These results indicate that FFA induce the ballooning of PHHs, and hyperglycemia and FFA induce fat accumulation. Moreover, αSMA is strongly expressed during sheet detachment and FFA resolves polarity in αSMA localization.   TEM results showed MDBs as a dark-colored granular structure with radial distribution in the glucotoxicity and lipotoxicity groups (Fig. 5C). In addition, enlarged mitochondria with crystalline structures were observed in the glucolipotoxicity group (Fig. 5D).

Hepatocyte function in the PHH/HSC sheets. Albumin secretion and urea synthesis were immedi-
ately deactivated in the simple co-culture, while they were maintained in cell sheets. The activity of CYP1A2 on day 11 was maintained compared to day 1 even in the presence of fatty acids. (Fig. S4).

Discussion
The presence of a BH is an important finding in the histological diagnosis of human NASH. BHs are considered a special form of cellular degeneration that is caused by the accumulation of fatty acids and glycogen, abnormal protein accumulation due to unfolded protein response dysfunction, and autophagy suppression. However, the mechanism underlying BH formation is not well understood 5,14-17 . Histological features of BHs include hepatocyte ballooning (1.5-to twofold increase in size compared to normal hepatocytes), accumulation of fat droplets, abnormal arrangement of CK, abnormalities in organelles such as those in the mitochondria and endoplasmic reticulum, and the presence of MDBs 5,[18][19][20] . In this study, we found that the exposure of PHH/HSC sheets to a glucolipotoxicity environment resulted in the production of BHs with features similar to those observed in human NASH. Functional analysis showed the involvement of fibrosis factors such as SHH and TGF-β1 but not the inflammatory cytokines such as TNFα and IL-8. Interestingly, BH was produced without the involvement of Kupffer cell (KC) or inflammatory cytokines. These results suggest that fibrosis is the major factor whereas www.nature.com/scientificreports/ inflammation is a supporting factor in BH formation. This observation is supported by the pathological correlation between fibrosis and BH formation 21,22 , and both correlate with clinical outcomes 10 . Although fibrosis is the most important factor in BH formation, it is not well understood why exposure of PHH/HSC sheets to a glucolipotoxicity environment results in BH formation. In a previous study, it was shown that ballooning did not occur in co-culture sheets with 3T3-J2 cells 12 , which have been used as a feeder for maintaining PHH function 23 , but ballooning did occur in co-culture with NHDF, a fibroblast cell line 12 . These results suggest that fibrosis, hyperglycemia, and insulin resistance are important for ballooning. One of the factors that could produce BH was the change of co-culture cells to HSCs. HSCs are non-parenchymal cells of the liver and have a variety of functions 24 One pathway of HSCs is activated by SHH, which functions as a protection against apoptosis when the liver is damaged, and are transformed into myofibroblasts 25 , to produce extracellular matrix (ECM) by increasing desmin and αSMA 24 . After ECM production, activated HSCs regress to inactive forms or undergo apoptosis 26 . However, HSCs continue to be activated under chronic damage such as fatty acid and oxidative stress and produce excessive ECM while further producing fibrosis-inducing factors such as TGF-β, causing liver fibrosis 24 . It has also been reported that hyperglycemia and insulin resistance indirectly promote fibrosis 27 . The action of HSCs as fibroblasts, chronic stimulation by fatty acids, and indirect effects of hyperglycemia and insulin resistance may have been factors involved in BH formation. Since ballooning does not occur in simple co-cultures 28,29 , it is assumed that there are some factors that lead to the production of BH in PHH/HSC sheets. We hypothesized that cells in the sheet will become three-dimensional (3D). This is supported by our observation that cells in the glucotoxicity, lipotoxicity, and glucolipotoxicity groups were layered compared to the normal group. Moreover, the 3D PHH/HSC sheets better represent the complex human liver due to enhanced cell-tocell interaction 11,30,31 . Recently, the sinusoidal pressure hypothesis has been proposed, and a correlation between pressure and fibrosis has been shown 32 . We suspect that aggregation of PHH/HSC sheets during detachment exerts a strong force on cells ( Fig. 2A), which in turn triggers fibrosis and HSC activation. In support of this, we observed that αSMA staining on day 0 of sheet fabrication showed strong activation of HSCs (Fig. 3A). During PHH/HSC sheet fabrication in the absence of FFA, αSMA was mainly localized at the basal side, while in the presence of FFA, αSMA showed no polarity and was localized to all regions including the periphery of hepatocytes www.nature.com/scientificreports/ (Fig. 3A). Thus, during sheet fabrication, stress due to external forces and glucolipotoxicity environment may have contributed to BHs formation. MDBs are not specific to NASH but are also observed in various other diseases such as alcoholic steatohepatitis, liver cell carcinoma, and Wilson's disease 33 . MDBs are thought to be formed as a result of the complex interaction between several factors, such as fibrosis, inflammation, suppression of autophagy, and cellular stress. All these factors lead to an imbalance in protein homeostasis within the cell and accumulation of abnormal intracellular proteins such as ubiquitin and p62 in addition to CK [33][34][35] . However, the detailed underlying mechanisms remain unclear. Autophagy, a mechanism through which homeostasis is maintained by the degradation of unwanted cytoplasmic components and abnormal proteins, is thought to be important for MDB formation 36 . In general, damaged organelles and abnormal proteins are modified with p62 and degraded. Once autophagy is inhibited, degradation cannot be performed, and p62 accumulates in the cell, which leads to MDB formation 37 . Histological and functional analyses in our study showed that Ballooning, fibrosis, and fat accumulation are mainly induced by FFA exposure. The presence of MDBs suggest the involvement of hyperglycemia and hyperinsulinemia because these bodies appeared in the glucotoxicity and glucolipotoxicity groups. Interestingly, MDBs appeared in the glucotoxicity group, even though autophagy was only mildly suppressed as evidenced by a slight increase in p62 expression. It might be possible that mechanisms involved in autophagy suppression are different in PHHs and PHH/HSC sheets. It has been reported that autophagy is suppressed in hepatocytes with TGF-β1 levels were significantly higher in lipotoxicity and glucolipotoxicity groups compared to control and glucotoxicity groups. n = 6, 3 independent experiments, **p < 0.01, *p < 0.05. (B) Evaluation of SHH levels in culture supernatants by ELISA. SHH levels were significantly higher in lipotoxicity and glucolipotoxicity groups compared to control and glucotoxicity groups. n = 7, 3 independent experiments, **p < 0.01, *p < 0.05. SHH: Sonic hedgehog. (C) Evaluation of IL-8 levels in culture supernatant by ELISA. IL-8 levels were significantly higher in the lipotoxicity and glucolipotoxicity groups compared to control and glucotoxicity groups. n = 6, 3 independent experiments, **p < 0.01 *p < 0.05. (D) Evaluation of protein carbonyl levels in PHH/HSC sheet lysates by ELISA. Protein carbonyl levels were not significantly different between groups. n = 4. (E) Evaluation of TNFα-IP8 mRNA expression in PHH/HSC sheets by RT-qPCR. GAPDH was used as normalizer. No significant difference was observed between groups. n = 3, 3 independent experiments.  38 , which is consistent with the results of this study. We investigated the presence of MDBs in PHHs from three different donors and found that MDBs were present in PHHs from only one donor (HU8317), even though all donors showed BH phenotypes such as ballooning and fatty degeneration. Clinically, MDBs have been reported only in 10%-70% of patients with NASH 33,39 . Reportedly, MDBs are particularly difficult to observe in pediatric NASH 40 and in cases of acute liver injuries such as acute hepatitis and drug poisoning 33 . This suggests that aging-related cumulative oxidative injury or genetic variables may play a contributing role. The background of the PHH lot used in this experiment was reviewed (Table S3), but no clear factor causing donor differences was found. On the other hand, in vitro, p62 aggregation by NF-κB activation 41 and dysfunction of the ubiquitin-proteasome system are relevant to MDB 33 . Therefore, there may be differences in the expression of ubiquitin and NF-κB among lots. Although various molecules are involved in NASH development and progression, the key factors are thought to be metabolic abnormalities due to hepatocyte steatosis, progression of fibrosis due to HSC activation, and inflammatory stimulation by KCs 2 . This study has some limitations. Our PHH/HSC model Our PHH/HSC model only partially represented NASH; i.e., it only demonstrated BHs. It could partially reproduce hepatocyte steatosis and fibrosis but could not reproduce inflammation. Since no inflammation was observed in our in vitro model, oxidative stress and endoplasmic reticulum stress could not be detected as well. Further, reportedly, KCs and inflammatory cytokines can cause fibrosis and cellular damage 42,43 . Therefore, a tri-culture model with KCs is essential to better reproduce human NASH, and further research is required to evaluate cell behavior in a tri-culture model. Our model could produce BHs and fibrosis in approximately 2 weeks; therefore, it could also be used as a fibrosis model with BHs 13 . Furthermore, patients with MDBs have been reported to have a worse prognosis, and thus, the presence of MDBs can be a clinically important indicator 44 . Hence, the use of healthy hepatocytes from the donors in this model and screening for the appearance of MDBs may be applied to predict the prognosis of the donor after future development of NASH.
In summary, in vitro BH can be produced by exposing PHH/HSC sheets to a glucolipotoxicity environment. BH formation in this model is not due to inflammatory cytokines production but due to transformation of cells to a 3D structure and fibrosis. Our model will further help in understanding the mechanisms underlying BH formation. In the future, this model could be further developed by incorporating inflammatory cells in the PHH/ HSC sheets and creating a new type of in vitro NASH model with BHs.

Data availability
Data supporting the findings of this study are available in the article and Supplementary Information files, or from the corresponding author upon request.