Phenotypic and functional analyses show stem cell-derived hepatocyte-like cells better mimic fetal rather than adult hepatocytes

Graphical abstract

members encourage hepatic differentiation [5,6,8,9,[11][12][13]; and hepatocyte growth factor (HGF), the synthetic glucocorticoid, dexamethasone (DEX), and oncostatin M (OSM) support increased maturity [3,4,6,[8][9][10][11][12][13][14]16]. However, fully mature hepatocytes have not been produced, which raises two unanswered questions: are cells aberrant because human liver development has not been followed with adequate specificity; or, if the lineage is correct, are HLCs actually 'stuck' in a fetal-like state? Assessment of the latter is problematic for two largely unaddressed reasons. HLC maturity is over-estimated if compared to sub-optimal adult hepatocytes. Thawed cells taken into culture are challenging to maintain [17]; in a well-controlled example, over 90% of the cytochrome P450 (CYP) 3A activity was lost in cryopreserved cells compared to freshly plated cells [18]. This illustrates the risk of over-interpreting the HLC phenotype if compared against dedifferentiated controls. Secondly, fresh fetal hepatocyte controls have been lacking when assessing HLC function. This risks misunderstanding as we have recently shown human fetal hepatocytes possess proteins, such as CYP3A4, commonly interpreted as adult markers [19].
To address these persisting questions about the differentiation and maturity of HLCs, we implemented a protocol with sufficient commonality to allow comparison with multiple previous reports. We analysed a wide range of human ESC lines, derived under different conditions alongside H9 cells, the most popular line for generating HLCs [3,7,9,10,12,14,18,20]. HLCs were assessed by proteome analysis and in a series of assays against fresh human fetal and adult hepatocytes. We also included cells differentiated by a second protocol in an extended array of new tests for differentiation status, devised by unbiased proteomics and principal components analysis that distinguish fetal from fresh adult and dedifferentiated adult hepatocyte phenotypes [19].

Human tissue and cells, and their culture
Human embryonic stem cell (ESC) lines were obtained with consent either directly from the derivation laboratory or the UK Stem Cell Bank. Cells were maintained on inactivated mouse embryonic fibroblast (MEF) cells [21]. The differentiation protocol ( Fig. 1) was commenced 3-4 days post passage onto fresh MEFs using Wnt3a (R&D Systems, UK) and Activin A (Peprotech, UK), diluted in RPMI media (Sigma-Aldrich, UK); followed by BMP2, OSM, FGF2, HGF (all R&D Systems) and DEX (Sigma-Aldrich, UK), diluted in Hepatocyte Culture Medium (HCM) (Lonza, UK). Information on the human fetal and adult hepatocyte controls can be found in the Supplementary Materials and methods. Human induced pluripotent stem cells (IPSCs) were developed and differentiated as previously reported [6,22].

Immunoblotting, immunofluorescence, cell sorting and cell proliferation and apoptosis studies
Immunoblotting and immunofluorescence were conducted as previously reported (Supplementary Table 1) [19,23]. Fluorescent activated cell sorting (FACS), cell proliferation and apoptosis are described in Supplementary Materials and methods.

Protein isolation and proteomic analysis
Protein isolation from whole cell extracts and labelling for isobaric tagging for relative and absolute quantification (iTRAQ) proteomics was described by Rowe et al. [19]. Quantitation of proteins was relative to a common reference preparation included in each run across different experiments. Protein identification and interrogation are described in Supplementary Materials and methods.

Phenotypic analysis
Gene expression analysis by RNA sequencing (RNA-seq) and quantitative PCR is described in Supplementary Materials and methods. Albumin and urea secretion into the media was measured using a human albumin ELISA kit and the Quanti-Chrom™ urea assay kit (both from Bethyl Laboratories). Comparisons with fetal and adult hepatocyte data used the unpaired two-tailed Student's t test. CYP3A activity was assessed in duplicate by incubation with P450-Glo™ CYP3A4 assay reagent (Luciferin-PFBE; Promega Ltd). For CYP analysis by mass spectrometry, cells were incubated with 1 mM testosterone or 1 mM dextromethorphan (Sigma, UK) in HCM. Conditioned medium was collected and diluted 1:1 in 0.5 lM phenacetin (Sigma) stop solution in methanol. CYP activity was calculated per min incubation. Alcohol dehydrogenase activity of cell lysates was assessed using a detection kit following the manufacturer's instructions (Abcam, UK). Results were standardized to the amount of protein measured by Bradford assay.

Initial characterization of HLCs
Transcripts for albumin and alpha1-antitrypsin (AAT, officially designated SERPINA1) were barely identified in DE-like cells but were readily detected in early HLCs following stage 3A (Supplementary Fig. 2A and B). We have previously shown that the transcription factors GATA4 and SOX17 become restricted from the early human embryonic liver compared to the adjacent foregut [23]. Between DE-like cells and early HLCs, GATA4, and SOX17 expression declined by approximately 75% and >90% respectively (data not shown). The transcription factor HNF4a encoded by HNF4A, is a master regulator of the hepatocyte phenotype [24][25][26]. During development there is a switch from an upstream immature P2 promoter to a downstream 'liver' P1 promoter, generating alternative first exons of the HNF4A gene [27,28]. This was mirrored in our cultures: in DE-like cells, HNF4A was expressed from the 'immature' P2 promoter; however, in HLCs the downstream first exon was preferentially transcribed from the P1 'liver' promoter ( Supplementary Fig. 2C, red boxes).
Final HLC morphology mimicked that of freshly plated human adult hepatocytes (Fig. 2D). Across the five different ESC lines, albumin was present in >75% of differentiated H9, SHEF1 and HUES7 cells (Fig. 2E). At least half of the H9 and HUES7 albumin-positive HLCs also contained AAT (Fig. 2E). All five cell lines showed a progressive increase in albumin secretion starting from day 11 (Fig. 2F). At the end of differentiation, levels were at least comparable to those from freshly plated fetal hepatocytes, which were approximately 8-fold lower than those from freshly plated adult hepatocytes. HLCs also showed urea secretion (mean ± S.E.: 2.87 ± 0.18 lg/ml/mg protein/day) comparable to fresh fetal hepatocytes (2.71 ± 0.09 lg/ml/mg protein/day), but approximately 18-fold lower than freshly plated adult cells (50.6 ± 6.11 lg/ml/mg protein/day).
We wanted to assess whether HLCs mimic periportal or pericentral hepatocytes. Glutamine synthase, a pericentral marker, was readily detected by immunocytochemistry in contrast to carbamoyl-phosphate synthase, a periportal protein ( Supplementary  Fig. 3A). This correlated to their transcript profiles (Supplementary Fig. 3B) and was true for a range of other genes, differentially expressed between pericentral and periportal hepatocytes [29]. All the pericentral genes except UDP-glucuronosyltransferase 1A (UGT1A) were expressed at increased or equivalent levels to  those in DE-like cells. In contrast, only phosphoenolpyruvate carboxykinase 2 (PCK2) of the periportal markers was expressed in HLCs, but at levels lower than in DE-like cells.
While these data were encouraging of liver-specific differentiation, it is an assumption based on limited user-selected proteins and assays and ignores potential similarity to cell types from other organs. We performed unbiased proteomic assessment of whole cell extracts from undifferentiated H9 cells and their HLCs at the end of stage 3. 61 proteins showed significant upregulation (>2-fold) including known liver markers, such as AAT/SERPINA1 and phase 1 (e.g. aldehyde dehydrogenases) and phase 2 enzymes (e.g., nicotinamide N-methyltransferase [NNMT] and glutathione S-transferase [GST] Mu3) (Supplementary Table 2). Cytokeratin (KRT) 8 and KRT18, both upregulated, function together as heteropolymers to protect hepatocytes from mechanical and non-mechanical stress [30]. KRT7, apparent in <10% of cells by immunocytochemistry (data not shown), may reflect slight permissiveness in our protocol to cholangiocyte differentiation but has also been reported as a hepatocyte progenitor cell marker [31,32]. Some upregulated proteins were not recognized as hepatocyte markers, such as KRT5 and KRT6A. Therefore to assess the broad proteome phenotype, we compared the upregulated protein dataset against data from 30 other human organs and tissues in the EBI Gene Expression Atlas (Fig. 3). By heatmap, the upregulated HLC proteome most closely resembled the liver, followed by another anterior derivative of foregut endoderm, the thyroid. Other foregut endoderm derivatives (stomach, small intestine and pancreas) showed recognizable similarity, in contrast to a marked divergence of HLCs from mesodermal and ectodermal derivatives such as bone marrow, skeletal muscle and brain.
HLCs have a metabolic profile comparable to human fetal rather than adult hepatocytes To gain broad developmental insights iTRAQ was performed on whole cell extracts rather than enriched microsomes. Consequently, this restricted identification of phase 1 enzymes, especially CYPs, similar to our previous study of human fetal and adult hepatocytes [19]. We used quantitative RT-PCR to analyse the expression of phase 1 enzymes, including CYPs (Fig. 4A), alcohol dehydrogenases (Fig. 4B), flavin-containing monooxygenases (Fig. 4C), aldehyde dehydrogenases (Fig. 4D), esterases (Fig. 4F), and other enzymes (Fig. 4E). The expression of 51/63 enzymes (81%) was significantly increased in H9 HLCs and 30 (48%) enzymes were significantly increased in HUES7 HLCs compared to their undifferentiated counterparts. CYP3A family members, CYP1B1, and DPYD were clearly increased in HLCs (Fig. 4A) with more modest increases in a number of alcohol dehydrogenases and flavin-containing monooxygenases ( Fig. 4B and C). The increase in the aldehyde dehydrogenase gene, ALDH1A1, in H9   SPR  IDH2  ABHD14B  SERPINA1  ALDH1A1  ANXA4  KRT18  ALDH5A1  CRYAA  CTSA  GLB1  FGB  KRT8  MYD88  ANXA13  TPP1  CLIC6  FGG  HSPB1  RAD23A  TGM2  CRYBB2  S100A14  FGA  GNS  ATP7B  HEXB  DPP4  TGFBI  ARHGDIA  CMPK1  DCN  GSTM3  GSTP1  LUM  AFP  MIF  DDAH2  NNMT  COL6A3  CTSB  GGH  RBP1  MME  COL1A2  NUCB2  ISG15  GALNS  MXRA7  KRT7  COL2A1  LAMP2  FN1  TCEAL3  DFNA5  MARCKS  CKB  FBXO2  KRT6A  KRT5      HLCs approached levels in fetal hepatocytes, while ALDH1A2 levels surpassed those of both adult and fetal cells (Fig. 4D). However, more generally, HLC transcript levels were markedly reduced compared to those in fresh human adult hepatocytes. In contrast, expression of 33 of the 63 phase 1 enzymes in H9 HLCs (52%) was statistically either greater or no different than in fresh human fetal hepatocytes, indicating a major overlap between the HLC and fetal phase 1 metabolic phenotype. CYP3A activity, mostly via CYP3A4 in adult liver, and CYP2D6 are two major mechanisms for drug metabolism that require CYP oxidoreductase (CYPOR) to donate electrons during catalysis. To help further gauge maturity of HLCs, we measured protein levels by immunoblotting and metabolic activity by mass spectrometry. Immunoreactivity for CYP3A (current antibodies fail to distinguish the different CYP3A isoforms) in HLCs was approximately 10% of the levels in fresh adult hepatocytes (Fig. 5A). For CYP2D6, two closely positioned bands were visible, the upper of which corresponded to the size of the purified protein (Fig. 5A). CYPOR was robustly detected. CYP2D6 metabolizes dextromethorphan to dextrorphan. This activity was detected in both H9 and HUES7 HLCs, but not in ESCs or HepG2 cells. Conversion by HUES7 HLCs was similar to that by fetal hepatocytes but was 527-fold less than detected in fresh adult hepatocytes (1313-fold lower than adult cells for H9 HLCs) (Fig. 5B). CYP3A-mediated metabolism of testosterone to 6b-hydroxytestosterone by HLCs was at least 100-fold greater than by ESCs or HepG2 cells and at least comparable to fresh fetal hepatocytes (Fig. 5C). Nevertheless, CYP3A metabolism of testosterone was 47-fold and 66-fold higher in fresh adult hepatocytes than in HUES7 and H9 HLCs, respectively. Mass spectrometry provides a 'gold standard' for metabolic assay. In contrast, our experience of measuring CYP3A4 activity by commercially available luciferase assay (PFBE reagent, Promega) was unhelpful. Although HLCs from all five ESC lines matched fresh fetal hepatocytes, fetal hepatocytes misleadingly showed greater activity than their adult counterparts ( Supplementary Fig. 4).

Bespoke tests to distinguish hepatocyte maturity show that HLCs are fetal-like
We have previously shown that CYP3A4 protein, detected by iTRAQ, is relatively ineffective at determining hepatocyte maturity or whether cells have dedifferentiated [19]. In contrast, principal components analysis provided new protein combinations and simple assays not requiring mass spectrometry to distinguish human adult hepatocytes from their fetal counterparts: AFP, GSTp and heat shock protein (HSP) 47 with negligible alcohol dehydrogenase (ADH) activity or CYP2A6 discriminates fetal cells; conversely, abundant CYP2A6 and ADH activity are hallmarks of adult cells [19]. AFP, GSTp, and HSP47 were readily detected in our HLCs (Fig. 6Aa and B). CYP2A6, another marker of perivenous cells, was weakly detected in HLCs, which also showed slightly more ADH activity than fetal cells (Fig. 6C); however, both CYP2A6 levels and ADH activity were much higher in adult hepatocytes. Although our protocol has marked similarity to that used widely by others [3][4][5][6][7][8][9][10][11][12][13][14][15][16], we wanted to extend these discriminatory tests further by adding immunocytochemistry and FACS analysis of HLCs differentiated from IPSCs by another very well-established protocol in a different lab [6]. We also examined whether our markers could distinguish dedifferentiated adult hepatocytes from fetal-like immaturity. Although we were unable to analyse ADH by immunocytochemistry, CYP2A6, AFP, GSTp and HSP47 all gave the expected distribution in triplicated samples of freshly plated human adult and fetal hepatocytes (Fig. 7). In contrast, minimal staining was apparent for any of the markers in dedifferentiated adult hepatocytes. HLCs from both sources produced fetal-like staining patterns with few cells stained for CYP2A6. The corresponding profiles were evident by FACS (Fig. 8). Taken together, these data consistently indicate a fetal phenotype for HLCs.
Uncertainty on HLC maturity has been discussed by others [16,20]; functional comparison with fresh fetal hepatocytes has been lacking [6]. We included undifferentiated stem cells and both fresh first trimester fetal and adult hepatocytes and consistently demonstrated major similarities with the fetal rather than the adult cell type. Our data were consistent with a previous report that included cryopreserved human fetal hepatocytes from 25 weeks of gestation [34]. Moreover, by studying fold increments over undifferentiated stem cells our data can be integrated with those from others. Our fetal-like HLC albumin secretion showed at least a 200-fold increase from the parent ESC line (Fig. 2F). This increment matches or surpasses that reported by others [3,9,11,18,22,35,36] but was still lower on average than secretion from freshly plated adult cells. However, lower increments in albumin secretion from ESCs to HLCs have been reported to match adult hepatocytes following cryopreservation [35,36]. These data imply the ease with which adult control cells can dedifferentiate and that in fact HLCs, differentiated through the use of soluble factors, mimic human fetal cells in keeping with the common AFP detection [3,[5][6][7][9][10][11][12]14,16,22,37]. A similar fetal-like conclusion can be drawn from urea secretion. Others have observed relatively low urea secretion in HLCs compared to adult cells [9,11,36], which may also reflect a pericentral rather than a periportal phenotype [29]. H9 HLCs matched fetal hepatocytes for the expression of half of the CYPs tested but were inferior for others (e.g. CYP2C8, CYP3A, CYP4, and CYP7 family members, and CYP8B1). This included transcripts for what have been previously considered 'adult' CYPs [6], warning that detection alone does not reliably indicate maturity. Phase 1 enzyme expression was commonly massively higher for fresh adult cells (up to 62,000-fold over ESCs for CYP2E1). This was not apparent in a recent protocol [38], using cryopreserved adult hepatocyte controls that had transcript levels only approximately 20-fold higher than undifferentiated stem cells for CYP2B6 (445-fold here), 10-fold for FMO3 (8178-fold here), and 20-fold for CYP2A13 (664-fold here). The human fetal liver contains both CYP3A4 and CYP2D6 proteins prior to midgestation [19]. Here we showed fetal-like HLC function by mass spectrometry, despite HLCs possessing between 10-and 60-fold lower transcript levels than fetal hepatocytes for CYP3A4, CYP3A5, and CYP3A7 (Fig. 6). Our experience with a commonly used commercial 'CYP3A4' luciferase assay was misleading. Luciferase CYP3A activity from HLCs was equivalent to data from others [14] and to fetal hepatocytes; but fetal hepatocytes showed more luciferase activity than adult hepatocytes, which is incompatible with the approximate 100-fold superiority of adult cells by mass spectrometry. Instead, we extended our previous discovery of a signature of proteins, capable of distinguishing fetal from adult hepatocytes [19]. Here, by adding immunocytochemistry and FACS, the combination of CYP2A6, AFP, GSTp and HSP47 accurately distinguished human fresh adult and fetal and dedifferentiated adult hepatocytes. HLCs, generated by our main protocol and another established one [6], possessed fetal discriminators rather than an adult or a dedifferentiated phenotype.
In summary, our HLC differentiation, and by inference that of others, mimics human liver development. However, by a wide range of analyses, these HLCs share pronounced similarities with human fetal rather than adult hepatocytes. This is important as it implies currently elusive soluble factors require discovery for the transition of HLCs in vitro into a truly mature adult phenotype.

Financial support
This work was funded by the Stem Cells for Safer Medicine Consortium (grants to NAH and CEG), the Engineering and Physical Sciences Research Council (to NAH), and a Medical Research Council (MRC) Centre grant. NAH is a Wellcome Trust Senior Fellow (funded by WT088566 and WT097820). SW is a Biotechnology and Biological Sciences Research Council (BBSRC) PhD student.