Modelling non-alcoholic fatty liver disease in human hepatocyte-like cells

Non-alcoholic fatty liver disease (NAFLD) is the most common cause of liver disease in developed countries. An in vitro NAFLD model would permit mechanistic studies and enable high-throughput therapeutic screening. While hepatic cancer-derived cell lines are a convenient, renewable resource, their genomic, epigenomic and functional alterations mean their utility in NAFLD modelling is unclear. Additionally, the epigenetic mark 5-hydroxymethylcytosine (5hmC), a cell lineage identifier, is rapidly lost during cell culture, alongside expression of the Ten-eleven-translocation (TET) methylcytosine dioxygenase enzymes, restricting meaningful epigenetic analysis. Hepatocyte-like cells (HLCs) derived from human embryonic stem cells can provide a non-neoplastic, renewable model for liver research. Here, we have developed a model of NAFLD using HLCs exposed to lactate, pyruvate and octanoic acid (LPO) that bear all the hallmarks, including 5hmC profiles, of liver functionality. We exposed HLCs to LPO for 48 h to induce lipid accumulation. We characterized the transcriptome using RNA-seq, the metabolome using ultra-performance liquid chromatography-mass spectrometry and the epigenome using 5-hydroxymethylation DNA immunoprecipitation (hmeDIP) sequencing. LPO exposure induced an NAFLD phenotype in HLCs with transcriptional and metabolomic dysregulation consistent with those present in human NAFLD. HLCs maintain expression of the TET enzymes and have a liver-like epigenome. LPO exposure-induced 5hmC enrichment at lipid synthesis and transport genes. HLCs treated with LPO recapitulate the transcriptional and metabolic dysregulation seen in NAFLD and additionally retain TET expression and 5hmC. This in vitro model of NAFLD will be useful for future mechanistic and therapeutic studies. This article is part of the theme issue ‘Designer human tissue: coming to a lab near you’.


Introduction
Non-alcoholic fatty liver disease (NAFLD) now affects around 25 -33% of the population and up to 75% of obese individuals in developed countries [1,2]. NAFLD encompasses a spectrum of liver disease and while simple steatosis is considered relatively benign, it can progress to non-alcoholic steatohepatitis (NASH), fibrosis and cirrhosis, and approximately 25% of those with cirrhosis will develop hepatocellular carcinoma (HCC) [3,4]. Progression is highly variable between individuals, with only a minority of individuals reaching end-stage liver disease and/or developing HCC [5]. Such inter-individual variability, coupled with a lack of understanding of the underlying mechanisms, has limited the development of effective therapeutic interventions.
Studies in whole liver may be confounded by shifting cell populations, and therefore the possibility of recapitulation of multiple facets of human NAFLD in a single cell-type in vitro would be a key advance, permitting the dissection of disease processes in mechanistic studies and enabling high-throughput screening for new medicines. Exposure of cancer cell lines to saturated (typically palmitate) or unsaturated (typically oleate) long chain fatty acids results in steatosis accompanied by upregulation of cytokines, interruption of insulin signalling, increased reactive oxygen stress and apoptosis signalling [6][7][8][9]. However, while hepatic cancer-derived cell lines are a convenient, renewable resource, they have many genomic and functional alterations so that their utility in modelling diseases of overnutrition is unclear [10,11]. Primary hepatocytes isolated from human tissue have also been employed to model disease and predict drug toxicity, however, purification is technically challenging and isolated cells rapidly lose phenotype in cell culture [12]. Furthermore, the gene expression changes that occur with adaptation to culture resemble alterations present in liver disease and may additionally be influenced by the donor's hepatic phenotype and pharmacological history [13]. The renewable, pluripotent nature of embryonic stem cells (ESCs) presents an opportunity for human differentiated tissue to be reproducibly generated in vitro for the investigation of human disease processes. Exposure of ESCs to a refined regime of chemical and biological growth stimuli over a three to four week period allows robust, scalable generation of hepatocyte-like cells (HLCs) that exhibit a similar morphology, function and transcriptome to human hepatocytes [13,14]. As a consequence, HLCs have been used extensively in mechanistic studies of drug-induced liver injury, hepatitis viral replication, fetal xenobiotic exposure and the characterization of inherited disorders of lipid metabolism [15][16][17][18].
Recent studies in human liver biopsy specimens and in animal models suggest that epigenetic dysregulation could play a role in NAFLD pathogenesis and progression [19][20][21]. Our studies suggest that the cytosine modification 5-hydroxymethylcytosine (5hmC) may be useful as a biomarker of both normal and abnormal liver physiology [22,23]. However, 5hmC is rapidly lost during somatic cell adaptation to culture and in all tested hepatocyte cell lines [24,25], making it difficult to undertake meaningful epigenetic analyses (particularly 5hmC) in cultured cell line models. Here, we show that HLCs derived from human ESCs retain the ability to form 5hmC, providing a non-neoplastic and renewable source of human cells for liver research. Additionally, exposure of HLCs to a nutrient cocktail of lactate, pyruvate and octanoic acid produces a robust NAFLD-like phenotype in vitro. We believe this model will be of importance for further mechanistic studies of NAFLD on defined genetic backgrounds.

Material and methods (a) Embryonic stem cell derived hepatocyte differentiation and LPO treatment
Female H9 ESCs were differentiated into HLCs as previously described [14]. At day 20 of differentiation, cells were exposed to sodium L-lactate (L), sodium pyruvate (P) and octanoic acid (O) (all Sigma, Gillingham, UK) at low (L : P : O 10 mM :

(c) Metabolome profiling
Metabolome profiling on cell media and HLCs was undertaken using ultra-performance liquid chromatography-mass spectrometry (UPLC-MS) using a Thermo Scientific Ultimate 3000 UPLC system coupled to an electrospray Q Exactive Focus mass spectrometer. A hydrophilic interaction liquid chromatography (HILIC) assay to investigate water-soluble metabolites and a C 18 reversed phase method to investigate lipid metabolites were applied. Univariate and multivariate data analyses were performed in METABOANALYST 3.0 [26], including principal components analysis (PCA), Mann Whitney U-tests or Kruskal-Wallis tests to identify metabolites demonstrating a statistically significant change in relative concentrations between two or three biological classes. Fold changes were calculated by division of the mean peak response for one biological class by the mean peak response of the second biological class. For detailed methods, see the electronic supplementary material.

(d) 5hmC profiling
For slot blotting, serial dilutions of extracted DNA were blotted onto Hybond Nþ nitrocellulose membrane (Amersham, Buckinghamshire, UK) using a slot blot manifold (HSI, GE Healthcare, Buckinghamshire UK), washed in 2xSSC buffer and dried overnight. Membrane was then probed with 1 : 10 000 anti 5hmC antibody (Active Motif, cat:39769) for 1 h at room temperature followed by goat anti-rabbit IRDye 800CW (1 : 10 000) for 1 h at room temperature and subsequent imaging on the Li-cor Odyssey infrared imaging system (both Li-cor, Cambridge, UK). DNA loading was confirmed by staining with 0.02% methylene blue in 0.3 M sodium acetate ( pH 5.2) and destaining with ddH 2 O. rstb.royalsocietypublishing.org Phil. Trans. R. Soc. B 373: 20170362 5hmC DNA immunoprecipitation (hmeDIP) was performed and libraries sequenced on an Ion Torrent semiconductor sequencer using the Ion PI TM Hi-Q TM Sequencing Kit and an Ion PI TM Chip Kit v3 (Thermo Fisher Scientific, Paisley, UK) to a depth of approximately 30 million reads (see the electronic supplementary material). Raw sequencing data were quality controlled, filtered and aligned using Ion Torrent suite software (Life Technologies, Paisley, UK) and then normalized to total reads in R using bespoke scripts. Relative 5hmC levels per 150 bp window were determined using the 'sliding windows' function on the Galaxy server. Genomic annotation data for human (hg19) analyses were downloaded from the University of California Santa Cruz Genome Bioinformatics Resource. Further details on hmeDIP bioinformatic processing can be found in Thomson et al. [27]. Raw and processed data files are available for download from the GEO repository accession number GSE109139.

(e) Statistics
Prism GraphPad software (GraphPad Software Inc.) was used for analysis of qPCR and cell culture variables. Data were routinely analysed for outliers and normality of distribution. Non-parametric data or data with significant outliers were log 10 transformed. If data remained non-parametric, a non-parametric test was used as indicated. Data point and bar graphs are shown as mean + s.e.m. pluripotency markers, including: POU class 5 homeobox 1 (POU5F1) and Nanog homeobox (NANOG) with a concurrent increase in the hepatocyte-specific markers ALB and hepatocyte nuclear factor 4 alpha (HNF4a) (electronic supplementary material, figure S1b). At day 21, CYP1A2 and CYP3A4 activity and secreted human albumin were evident and consistent with previous reports [28,29] (electronic supplementary material, figure S1c,d).

Results
HLC LPO exposure induced a dose-and time-dependent increase in lipid vacuole generation (figure 1b,c). LPO exposure increased mitochondrial stress in a dose-dependent manner at both 48 and 96 h of exposure as determined by an increase in mitochondrial inner membrane potential allowing uptake of MitoTracker Deep Red dye (https://www.nature. com/protocolexchange/protocols/3673#/equipment figure 1d). While LPO treatment did not affect cell functionality as determined by CYP3A assay (figure 1e), high-dose LPO over a longer duration increased apoptosis (figure 1f ). Thus, to avoid studying secondary effects associated with cell death, we opted for the lowest dose of LPO treatment for 48 h in all further analyses. qPCR analysis showed that LPO exposure consistently stimulated the expression of multiple putatively causal genes in NAFLD pathological processes, including: fatty acid synthesis (ACACA, LXR, SREB1c), gluconeogenesis (PCK, G6PD) and lipid vesicular transport proteins Perilipin 1 and 2 (PLIN1, PLIN2) (figure 1g).   figure S4a). Breakdown of free fatty acids in liver involves b-oxidation. Initial transport of unmodified long-chain fatty acids through the mitochondrial membrane requires the addition of carnitine by carnitine palmitoyl transferase 1 (CPT1), generating acyl carnitine species that are shuttled into mitochondria by carnitine -acylcarnitine translocase. Consistent with previous data [31], LPO-induced steatosis was associated with activation of the b-oxidation transcriptional pathway (electronic supplementary material, figure S4b), with upregulation of long-chain-fatty acid-CoA ligase 1 (ACSL1), CPT1A, acyl coenzyme A dehydrogenase (ACADM), very long-chain-specific acyl-CoA dehydrogenase (ACADVL), and acetyl CoA acetyltransferase (ACAT1) (electronic supplementary material, figure S4b). Defective b-oxidation in NAFLD results in the accumulation of acyl carnitines and diacylglycerols [31], and in support of this we observed substantial dysregulation of carnitine and acyl carnitine metabolites and glycerol species in cells and media with LPO ( figure 3a). Further, we observed marked alterations in fatty acids, oxidized fatty acids and acyl glycine species in LPO-treated cells and media (figure 3b). Defective oxidative phosphorylation is also feature of NAFLD [31] and we identified altered expression of transcripts involved in all five of the mitochondrial complexes in LPO-treated HLCs (table 1).
When b-oxidation is saturated/dysfunctional, surplus fatty acids can be degraded through microsomal v-oxidation, to produce hydroxylated fatty acids. Crucially, this comparatively inefficient process increases cellular oxidative stress, a key pathological mechanism in NAFLD progression. LPO treatment was associated with activation of v-oxidation, with upregulation of cytochrome P450 4A11 (CYP4a11) (table 1) and accumulation of hydroxylated fatty acids in cells and media ( figure 3c). An alternative method of fatty acid clearance involves the synthesis and sequestration of triglycerides within lipid vesicles, the primary histological finding in hepatic steatosis. Our findings of an LPO-induced increase in glycerol-3-phosphate within cell media with induction of phospholipid phosphatase 2 (PLPP2) and suppression of the lipolytic enzyme carboxyl ester lipase (CEL) support an increase in triglyceride synthesis in LPO-treated HLCs (table 1). In addition to this, upregulation of the lipid vesicle transport proteins PLIN1, PLIN2 (figure 1g) and CIDEC and the structural lipoprotein APOA4 (figure 2b) are reflected in the increase in lipid vacuoles with LPO exposure.
(d) Other aspects of lactate, pyruvate and octanoic acidinduced transcriptional and metabolic dysregulation are consistent with non-alcoholic fatty liver disease LPO treatment was associated with increased expression of cytosolic phosphoenolpyruvate carboxykinase (PCK1 ;  table 1; electronic supplementary material, table S2), the rate-limiting enzyme in gluconeogenesis and an important regulator of TCA cycle activity. An increase in gluconeogenesis in LPO-exposed HLCs was supported by increased phosphoenolpyruvate and glucose in cell media (table 2). PCK1 may be an important therapeutic target because knocking down PCK1 can prevent oxidative stress and inflammation in mice on a high-fat diet [33]. Examination of the KEGG pathway for NAFLD (hsa04932) demonstrated transcriptional dysregulation of mediators of insulin resistance (IRS2, P1K3R1, AKT1), steatohepatitis ( JUN, ERN1) and neutrophil infiltration and inflammation (CXCL8) (table 1).

(e) Hepatocyte-like cells maintain 5hmC and TET enzyme expression in culture
Global epigenetic DNA modification levels and patterns are frequently altered in cultured cells, particularly 5hmC, which is vastly reduced genome-wide [24]. To test if our system maintained a normal 5hmC landscape, we carried out an antibody-based quantitative analysis of 5hmC in HLCs and compared these to ESCs and HepG2 cells. HLCs retain 5hmC levels at equivalent levels to mouse liver, which is much higher than the minimal level observed in HepG2 cells (figure 4a). Additionally, although mammalian cell culture is associated with loss of Ten-eleven-translocation (TET) enzyme expression [24], the expression of all three TET isoforms is maintained in HLCs (figure 4b). TET1 is highly expressed in ESCs and plays a critical role in the maintenance of pluripotency [34] and consistent with this, TET1 mRNA expression is downregulated at the later stages of the hepatic differentiation protocol. TET2 is the dominant TET isoform expressed in liver and expression was appropriately increased during differentiation [35]. In order to validate that HLCs have a liver-like hydroxymethylome and have transitioned from an ESC state, we performed hmeDIPsequencing (hmeDIP-seq) on HLCs and compared profiles with published datasets for 5hmC in human liver and human ESCs [36,37]. We additionally performed hmeDIPseq on human kidney to assess tissue-specificity. Accordingly, we observed high levels of 5hmC across key liver genes, e.g. albumin (Alb) exclusively in HLCs and human liver datasets, while low levels of 5hmC were observed over ESC and kidney 5hmC-enriched loci, e.g. the HoxA cluster (electronic supplementary material, figure S5a,b). In both human ESCs and liver, high levels of 5hmC are present at transcription start site (TSS), distal promoter and proximal gene body regions [38,39] and bioinformatic analysis of 5hmC datasets in HLCs showed a similar 5hmC profile (figure 4c). As in previous studies, 5hmC patterns followed transcriptional states, with highly transcribed genes containing more gene body 5hmC than lowly transcribed genes [40] (figure 4c). Thus, 5hmC profiles are consistent with a liver-like epigenome, supporting the utility of this cell line for the study of human metabolic liver disease. We also profiled 5hmC in LPO-treated and control HLCs using hmeDIPseq. Sliding window analysis showed no global change in 5hmC in LPO-exposed HLCs (figure 4d), although we observed some 5hmC enrichment within bodies of induced genes involved in lipid synthesis and transport (electronic supplementary material, figure S6). Such changes allowed for stratification by treatment (figure 4e) and highlight the utility of combined transcriptomic and epigenetic profiling to investigate molecular mechanisms.

Discussion
NAFLD is strongly associated with obesity, insulin resistance (IR), type 2 diabetes (T2DM) and cardiovascular disease [1] and the increasing prevalence of these disorders is a substantial public health burden. Indeed, NAFLD is an early predictor of, and important determinant for, the development of T2DM and the metabolic syndrome [41]. While the use of rodent models of NAFLD to explore mechanisms and test therapies is common, lipid metabolism in mice and humans differs so that the applicability of mouse models to human  Human HLCs possess hepatocyte-like morphology, gene expression and function [14]. They can be derived from renewable cell populations at scale, are robust, and are comparable to cryopreserved human hepatocytes in terms of predicting human compound toxicity and model metabolic differences [29]. Exposure of cells to fatty acids has been employed to model NAFLD in cell culture systems with various reports of effects on steatosis, insulin signalling and activation of apoptosis pathways [6,7,9]. The one previous report of the use of HLCs to model NAFLD used oleic acid-induced steatosis and reported perturbation of multiple metabolic pathways, including activation of the proliferatoractivated receptor (PPAR) pathway [42]. However, the aetiology of NAFLD may also include the response to nutritional excess, hyperinsulinaemia, oxidative stress, adipokine fluctuations, cytokine signalling and the influence of gut-derived factors [43]. Therefore, more representative models are required. The high-energy substrates lactate and pyruvate have been used in combination with the medium chain fatty acid octanoate and ammonia (LPON) to mimic energy excess in human hepatoblastoma cell lines [44,45]. This cocktail generates hepatic steatosis, impaired mitochondrial function, increased reactive oxygen species and perturbs energy metabolites and the cellular proteome, without compromising cellular viability [43,44]. Here we show that HLC exposure to LPO induces an increase in cellular steatosis and stimulates the expression of multiple genes associated with NAFLD, demonstrating the importance of HLCs as a cell-based model of human NAFLD. We also identified transcriptional and metabolic perturbations consistent with mitochondrial dysfunction, a key feature of NAFLD, with multiple alterations in TCA cycle metabolism, b-oxidation and oxidative phosphorylation.
Recent studies using both genome-wide and candidate gene analysis in human liver biopsy specimens and in animal models have identified alterations in DNA methylation in NAFLD, suggesting that epigenetic dysregulation may play a role in its pathogenesis and progression [19 -21]. DNA methylation (5-methylcytosine, 5mC) is important in the regulation of gene expression and plays a key role in transcriptional silencing [46]. By contrast, the cytosine modification 5hmC is enriched over the bodies of expressed genes as well as at enhancer elements and some promoter regions [47 -49]. While 5hmC appears to function as part of an active DNA demethylation pathway, catalysed by the TET enzymes [50], it may also act as a functional DNA methylation mark. We have suggested that 5hmC patterns may be useful as an identifier of cell/tissue type and a marker of cell state and may be a useful tool to identify novel therapeutics and assess drug response [22,51]. However, 5hmC profiles in whole liver from animal models or human biopsy specimens may be confounded by changes in cell populations as a consequence of the disease process, because 5hmC patterns are highly cell-and tissue-specific. While the use of single cell-types overcomes this problem, there are particular problems with studying 5hmC in vitro: 5hmC is rapidly lost during somatic cell adaptation to culture, including in all tested hepatocyte cell lines [24]. Additionally, established HCC cell lines are unsuitable because resident 5mC/5hmC changes are a feature of neoplasia that would confound analyses in these models following interventions to induce steatosis [25,52]. By contrast, we demonstrate that HLCs maintain TET expression, have 5hmC profiles consistent with a liver-like epigenome, show 5hmC enrichment within bodies of induced genes and therefore represent a transformative resource with which to study epigenetic changes in human liver disease.
In conclusion, our study shows that HLCs treated with LPO recapitulate the transcriptional and metabolic dysregulation seen in NAFLD and additionally retain expression of the TET enzymes and 5hmC. This in vitro model of 'NAFLD in a dish' using a single cell-type will be useful for future mechanistic and therapeutic studies.