Chemical reprogramming of mouse embryonic and adult fibroblast into endoderm lineage

We report here an approach to redirecting somatic cell fate under chemically defined conditions without transcription factors. We start by converting mouse embryonic fibroblasts to epithelial-like cells with chemicals and growth factors. Subsequent cell fate mapping reveals a robust induction of SOX17 in the resulting epithelial-like cells that can be further reprogrammed to endodermal progenitor cells. Interestingly, these cells can self-renew in vitro and further differentiate into albumin-producing hepatocytes that can rescue mice from acute liver injury. Our results demonstrate a rational approach to convert mouse embryonic fibroblasts to hepatocytes and suggest that this mechanism-driven approach may be generalized for other cells.

We report here an approach to redirecting somatic cell fate under chemically defined conditions without transcription factors. We start by converting mouse embryonic fibroblasts to epithelial-like cells with chemicals and growth factors. Subsequent cell fate mapping reveals a robust induction of SOX17 in the resulting epithelial-like cells that can be further reprogrammed to endodermal progenitor cells. Interestingly, these cells can self-renew in vitro and further differentiate into albumin-producing hepatocytes that can rescue mice from acute liver injury. Our results demonstrate a rational approach to convert mouse embryonic fibroblasts to hepatocytes and suggest that this mechanism-driven approach may be generalized for other cells.
Cell fate is determined during development to ensure proper formation of tissues and organs in an individual. Amid intrinsic and extrinsic signals, transcription factors play a primary role in controlling cell fate by specifying the expression of genes unique to a particular cell type in vivo and in vitro. This remarkable insight led to the development of iPSC 3 reprogramming almost 10 years ago (i.e. four transcription factors, Oct4, Sox2, Myc, and Klf4, capable of converting embryonic fibroblasts back to the pluripotent state found only in cells derived from blastocysts) (1). In addition, various transcription factors have been discovered to convert fibroblasts into neurons (2)(3)(4)(5), hepatocytes (6 -8), cardiomyocytes (9), and hematopoietic cells (10).
Despite the ease and remarkable advances achieved with transcription factor-based cell fate reprogramming, safety concerns associated with the insertion of retroviral vectors and the potential reactivation of exogenous transcription factors may hamper any attempt to deploy this technology therapeutically (11)(12)(13)(14)(15). Therefore, alternative approaches have been developed to ameliorate this problem. For example, an episomebased delivery system has been used to reprogram cells without the insertion of exogenous factors (16 -18). Recently, small chemicals have been used to replace transcription factors and convert somatic cells into ciPSCs (19 -21) and chemically induced neurons (22,23), suggesting a way to reprogram cell fate.
Unlike transcription factors, chemicals are mostly synthetic and designed originally to regulate biological activities through cellular targets, mostly receptors and enzymes. Although designed as potential therapeutics, only a limited number of them eventually became registered drugs. The remaining pool of chemicals has now become a vast resource for biomedical research. Recent success in utilizing chemicals to reprogram cell fate demonstrates the utility of chemicals in stem cell and cell fate research (21). Therefore, it is plausible that further research could be able to develop ways to reprogram cell fate as reliably and rationally as transcription factors, without the safety concerns.
The main challenges ahead can be categorized into two main areas. First, there is a knowledge gap between the biological activities regulated by chemicals and their relevance to cell fate decision. Second, chemicals are less specific than transcription factors and have off-target effects well-known in the pharmaceutical research as side effects. For example, vitamin C is a small molecule found in nature. It is best known as an antioxidant important for human health. Recently, we have shown that vitamin C can enhance somatic cell reprogramming by promoting histone and DNA demethylation through histone and DNA demethylases (24 -27). As such, the relationship between chemicals and cell fates can become a fertile ground for further investigations. New tools may be developed such that cell fate can be reprogrammed with chemicals with relative ease. In this report, we attempted to develop a rational approach to convert one cell type to another in a chemically defined and mechanistically understood manner.

Chemical induction of epithelial-like cells (ELCs) from mouse embryonic fibroblast (MEFs)
One of the earlier insights we gained in analyzing the reprogramming of MEFs into iPSCs by the Yamanaka factors is the realization that the starting fibroblasts undergo a mesenchymalepithelial transition (MET) process to become epithelial cells. This is accomplished by the suppression of the fibroblastic characteristics reinforced mostly through the TGF␤ signaling pathway with Oct4, Sox2, and Myc and then the activation of E-cad and other epithelial features by Klf4 (28). Inspired by this mechanistic insight, we wished to formulate a chemical recipe that could reliably convert fibroblasts into epithelial cells. To this end, we have developed a mixture called F 2 BRFCYT that can convert MEFs into ELCs in a chemically defined medium (Fig. 1A). The starting rationale is to first inhibit TGF␤ signaling with RepSox (R) and then activate epithelial characteristics with BMP4 (B) based on our studies (28,29). In addition to RB, we also included Fgf2, CHIR, and Y27632 used previously in our reprogramming medium (30). Furthermore, through compound screening, we also determined that Forskolin (F) and TTNPB (T), two ingredients important for ciPSC generation (21), are very helpful in promoting the MET process. Specifically, we observed cell morphological changes typical of MET when MEFs were treated with F 2 BRFCYT for 4 and 8 days (Fig.  1B). Accordingly, mesenchymal genes, such as Cdh2, Cd44, and Vim, were inhibited (Fig. 1C, bottom), whereas epithelial genes, such as Cdh1, Epcam, and Ocluddin, were up-regulated (Fig.  1C, top), confirming at the molecular level a MET process. We then wished to determine the cellular state for the emerging ELCs and therefore performed RNA-seq to map their molecular signatures. Interestingly, the ELCs appear to have acquired endoderm characteristics as demonstrated by robust induction of Sox17, Gata4, and Gata6 at the mRNA level (Fig. 1D), apparently without appreciable induction of pluripotency genes (supplemental Fig. S1). These ELCs showed positive staining for SOX17 in the nuclei (Fig. 1E), further confirming that they have transitioned into the endoderm lineage.

Chemically induced endoderm progenitor cells (ciEPCs) from MEF-derived ELCs
Given the fact that chemicals are not designed to mediate cell fate changes as specifically as transcription factors have been evolved to do, we were surprised by the acquisition of endodermal markers, such as Sox17 and Gata4/6, in the ELCs (Fig. 1E). To understand and improve this induction process further, we systematically optimized the induction process (supplemental Fig. S2A). We first performed drop-out experiments to assess the role of each component in mediating the induction of individual markers, such as Sox17, Foxa2, Hnf4a, and Cdh1. It became apparent that RepSox plays a critical role for the induction of Sox17, Foxa2, Hnf4a, and Cdh1 (supplemental Fig. S2B). As for Sox17, TTNPB and BMP4 are also critical (supplemental Fig. S2B). On the other hand, all except for RepSox appear to be dispensable for Foxa2, Hnf4a, and Cdh1 (supplemental Fig.  S2B). Furthermore, small chemicals, such as TTNPB, appear to be detrimental for the induction of Cdh1, Foxa2, and Hnf4a (supplemental Fig. S2B). In addition to these markers, we also

Chemical induction of mesenchymal-to-epithelial transition
analyzed their impact on cell proliferation and determined that Fgf2 and, to a lesser degree, TTNPB affect cell proliferation as expected (supplemental Fig. S2C) (30).
Based on the role of each component in mediating marker induction and also cell proliferation, we decided to further optimize the induction process toward the endodermal lineage. We first performed dose-response experiments for BMP4 (0 -50 ng/ml), CHIR99021 (0 -20 M), RepSox (0 -20 M), TTNPB (0 -5 M), FGF2 (0 -50 ng/ml), and FSK (0 -50 M) and analyzed the induction of Cdh1, Sox17, Foxa2, and Hnf4a. As shown in supplemental Fig. S2D, it is apparent that each component must be titrated carefully. For example, CHIR appears to work optimally at 10 M. We then performed time-course experiments as shown in supplemental Fig. S2E and demonstrated that TTNPB is time-sensitive for Sox17 and Hnf4a induction and detrimental to Cdh1 and Foxa2, yet CHIR is time-sensitive for Sox17 and Foxa2 and detrimental to Cdh1 and Hnf4a. Based on these results and insights, we developed an optimized schedule as shown in Fig. 2A. We reduced the concentration of CHIR based on its inhibitory effect on Cdh1 for the first 4 days and removed TTNPB after day 4 to allow the induction of Hnf4a, Cdh1, and Foxa2 (supplemental Fig. S2B). To facilitate the induction of endoderm markers, we also included activin A based on its known inductive activity. Based on this improved protocol, we can consistently induce the formation of ELCs and the further reprogramming of these cells morphologically (Fig. 2B). To quantify the reprogramming efficiency, we stained the cells with SOX17 antibody in site ( Fig. 2C and supplemental Fig. S3A) and counted the Sox17 ϩ cell clusters. More than 150 Sox17 ϩ cell clusters were formed per 1.5 ϫ  10 4 cells at day 24, thus with an induction efficiency of Ͼ1% (Fig. 2D). Furthermore, to analysis the optimized protocol in dynamics, we show by qRT-PCR that Sox17 is induced rapidly, followed by Foxa2, Hnf4a, and Cdh1 (Fig. 2E), and also a robust induction of Gata6 early and Epcam and Gata4 relatively late (supplemental Fig. S3B). Western blot analysis confirmed the early induction of SOX17 protein ahead of CDH1 (Fig. 2F). We further showed by flow cytometry that more than 50% of the chemically induced cells in the dishes were Sox17 ϩ /Epcam ϩ at day 24, indicating a robust endoderm cell induction (Fig. 2G).

Characterization of ciEPCs
To see whether ciEPCs could self-renew, we derived stable lines and characterized them in depth. As shown in Fig. 3A, these ciEPCs can be passaged without any appreciable differentiation based on cell morphology, as passages 2 and 13 are essentially the same. We then cultured and passaged them continuously for 70 days and can demonstrate that they proliferated by ϳ10 14 times to ϳ10 20 cells from a starting cell population of 500,000 cells, whereas the MEFs can grow for 5-10 days and cannot be maintained for more than 30 days (Fig. 3B). We then performed RNA-seq for cells collected between passages 1 and 22 and demonstrated that they maintain very stable transcriptomes (Fig. 3C). Consistently, we could detect SOX17, FOXA2, HNF4A, and CDH1 with immunostaining for two independently derived ciEPC lines ( Fig. 3D and supplemental Fig.   S4A). We also performed qRT-PCR for MEFs and ciEPCs at different passages to show that they were quite stable in expressing the key endoderm markers, such as Sox17, Foxa2, Hnf4a, and the epithelial marker Cdh1 (Fig. 3E).
To evaluate the potential risk of tumorigenesis, we injected the ciEPCs into an SCID mouse and found no teratoma formed in the ciEPC groups during a 6 -10-week period, whereas all of the ESC/iPSC controls could form teratoma (supplemental Fig.  S4, B and C). We then tested the karyotypes of ciEPCs at different passages and found that the ciEPCs have normal karyotypes when cultured for more then 30 passages (supplemental Fig.  S4D). Taken together, these data indicated that the ciEPCs could maintain in vitro stability.

The ciEPCs have differentiation potential restricted to endoderm
To evaluate the differentiation potential of the ciEPCs, we first compared the RNA-seq data sets from ciEPCs generated in different experiments with MEFs, liver, intestines, and colon. The data show that the ciEPCs are positive for additional endoderm markers Hind1b and Tead4, in addition to Sox17, Foxa2, Gata4/6, and Sox7, while lacking ectoderm or mesoderm markers (Fig. 4A). Interestingly, the ciEPCs also express Krt8, Krt18, and Krt19 strongly, markers known for liver/foregut. We then induced ciEPCs and mouse ESCs with conditions established for promoting mESCs differentiation toward neuroectoderm  (31) or mesoderm. We showed that the neuroectoderm markers Nestin and Sox1 (Fig. 4B) or mesoderm markers T and Mixl1 (Fig. 4C) were up-regulated significantly in the ES group, but not in the ciEPC group.

Chemical induction of mesenchymal-to-epithelial transition
EPCs could be derived from ES cells by a stepwise differentiation protocol (32-34). We then compared the gene expression profile of the ESC-derived EPCs (ESC-EPCs) with the protocol published previously by Sakano et al. (35), with our ciEPCs. We showed that ESC-EPCs were quite similar to ciEPCs in gene expression (Fig. 4D). However, the ESC-EPCs have residual pluripotent property, as indicated by Pou5f1, and also primitive stick or mesoderm markers, such as Cer1, Fgf8, and Mixl1, which is consistent with the recent report by Cheng et al. (36) ( Fig. 4E). Compared with ESC-EPCs, the ciEPCs have a higher level of expression for definitive endoderm markers, such as Sox17, Foxa2, and Hnf4a (Fig. 4E). We then further confirmed the RNA-seq data by qPCR and observed the same result (Fig.  4F). These data suggest that the ciEPCs are restricted to the endoderm lineage and closely resemble endoderm progenitor cells.

Differentiation of ciEPCs toward hepatocytes (ciHeps)
We then attempted to direct the ciEPCs toward hepatic lineage. To this end, we devised a two-step process consisting of a specification phase between 7 and 10 days and a maturation phase of 10 days (Fig. 5A). At the end of the specification phase,

Chemical induction of mesenchymal-to-epithelial transition
we obtained a quite homogeneous cell population with flat cell morphology (Fig. 5A, left). After a maturation period of 10 days, these cells give rise to cells of typical hepatocytic morphology (Fig. 5A, right). We then performed RNA-seq to show that the resulting ciHeps have acquired expression profiles similar to primary hepatocytes (supplemental Fig. S5A). We then checked the expression of hepatocyte markers by qPCR. Specifically, ciHeps express higher levels of Afp and Cyp3a13 than primary hepatocytes and comparable levels of Ck8, Ck18, and Cyp2b10 but lower levels of Alb, Cyp3a11, and Cyp1a1 (Fig. 5B). Compared with starting MEFs, ciHeps express ALB, AFP, HNF4A, CK18, CK8, CDH1, and ZO-1, as demonstrated by immune staining (Fig. 5C and supplemental Fig. S5B). Given the fact that albumin is a hallmark for hepatocytes, we further characterized ciHeps for their expression of albumin and demonstrated that albumin can be detected by both Western blotting (Fig. 5D) and ELISA (Fig. 5E). We further detected the expression of ALB by FACS and found that 66.7% of the ciHep cells were ALB-positive (supplemental Fig. S5C). We further analyzed the liverspecific function of ciHeps and showed that they are capable of storing glycogen as well as taking up LDL and indocyanine green (ICG) (Fig. 5F).
To test their function in vivo, we injected ciHeps into concanavalin A (ConA)-treated mice and showed that they can rescue mice from liver failure-related death (Fig. 5, G and H). As shown in supplemental Fig. S5 (D and E), ciHeps are as effective as primary hepatocytes in reducing the serum levels of alanine aminotransferase and aspartate aminotransferase in mice.

Chemical induction of mesenchymal-to-epithelial transition Chemical reprogramming of fibroblasts into EPCs and hepatocytes
To ultimately confirm that the ciEPCs were induced from fibroblast, we used a lineage tracing system to track the origin of the ciEPCs. As shown in Fig. 6A, transgene mice that express Cre recombinase under the control of fibroblast-specific protein 1 (Fsp1 or S100A4) promoter were crossed with the R26R tdTomato mice. Because Fsp1 is specifically expressed in the fibroblast, the offspring of the mice (Fsp1-Cre:R26R tdTomato ) would express tdTomato in the fibroblast. The Fsp1-Cre: R26R tdTomato MEFs, which were isolated from Fsp1-Cre: R26R tdTomato mice and purified by FACS sorting, were used for EPC induction (Fig. 6B). The tdTomato-MEFs underwent a similar morphology change and could form EPC-like colonies as wild-type MEFs when induced toward EPCs by chemicals at day 24 (Fig. 6B). The tdTomato-ciEPC could passage continuously (Fig. 6C) and express key endoderm markers, such as SOX17, FOXA2, and epithelial marker CDH1 (Fig. 6D). We then differentiated the tdTomato-ciEPCs into hepatic lineage following the protocol described in MEF-derived ciEPCs. Consistent with the MEF-ciHep, the tdTomato-ciHeps widely express the hepatocytes markers, such as Afp, Alb, CK8, CK18, Cyp3a11, Cyp1a1, Cyp3a13, and Cyp2b10 (Fig. 6E). The expression of ALB, AFP, HNF4A, and CK8 were further confirmed by immune staining (Fig. 6F). These data definitely demonstrate that fibroblasts could be induced into EPCs by chemicals.
We also showed that mouse neonatal dermal fibroblasts (MNFs), defined adult fibroblast cell, could be induced to EPCs and hepatocytes in the same chemical mixture. As shown in supplemental Fig. S6A, MNF cells had a morphologic change

Chemical induction of mesenchymal-to-epithelial transition
similar to that of MEFs when induced by chemicals. The induction of SOX17 expression was confirmed by immune staining at day 24 (supplemental Fig. S6B). The derived MNF-ciEPCs cell lines could be passaged stably (supplemental Fig. S6C) and express a high level of endoderm markers Sox17, Foxa2, Hnf4a, Gata4, and Gata6 and the epithelial markers Cdh1 and Epcam but no appreciable level of mesenchyme cell marker Chd2 or liver makers Alb and Afp (supplemental Fig. S6D). The expression of SOX17 and FOXA2 was further confirmed by immune staining (supplemental Fig. S6E). We then differentiated the MNF-ciEPCs into hepatic lineage following the protocol described in MEF-derived ciEPCs. RNA-seq data showed that the MNF-ciHeps were very similar to the MEF-ciHeps in whole-gene expression profiles (supplemental Fig. S6F). The expression of hepatocyte markers ALB, AFP, HNF4A, CDH1, CK8, and CK18 in MNF-ciEPCs were confirmed by immune straining (supplemental Fig. S6G). Functionally, the MNF-ciEPCs were proven to be capable of storing glycogen, taking up LDL and ICG (supplemental Fig. S6H). We further detected the expression of albumin by FACS and found that 64.3% of the MNF-ciHep cells were ALB-positive (supplemental Fig. S6I).

Discussion
In the last 10 years, the iPSC field has generated key insights into cell fate decisions. One such insight is the realization that cell fate can be reprogrammed if the tissue-specific or lineagespecific transcription factors have been identified or characterized through developmental biology or genetics. However, these factors must be delivered through efficient methods, such as retroviral transduction, and integrated into the genomes of recipient cells, thus raising serious safety issues that may prevent their ultimate use in regenerative medicine (37). With high-throughput screening and chemical biology, the role of transcription factors may be gradually replaced by chemicals (small chemicals and growth factors). To this end, several groups have succeeded in converting somatic cells to iPSCs and neural stem cells/neurons by small molecules (19,(21)(22)(23)38). In this report, we show that a similar approach can be used to convert fibroblasts into self-renewing ciEPCs and then eventually functional ciHeps. Furthermore, we were also able to differentiate the ciEPCs toward cells of the pancreatic lineage that express markers such as Pdx1, Nkx2.2, Neurod1, and Hnf6 by qRT-PCR (supplemental Fig. S7, A and B) and PDX1 and HNF6 by immune staining (supplemental Fig. S7C). On the whole, we established a feasible method to generate the main cell types of endoderm by chemicals and growth factors (Fig. 6G). This approach may have several implications.
First, our study suggest that cell fate can be manipulated rationally based on the molecular signatures of the starting cells and the induction conditions defined by chemicals and growth factors. Whereas the initial morphological transformation of fibroblasts into epithelial-like cells (i.e. MET) provided the first informative guide, detailed transcriptomic signatures obtained by RNA-seq provided the most critical parameters to design and optimize the induction process. By recognizing that the cells start to assume endodermal characteristics, we then initiated subsequent work to direct them toward endoderm progenitors and hepato-cytes. Similar approaches should be considered for the development of progenitors of the mesoderm or ectoderm lineages.
Second, we were surprised to find that TTNPB, an agonist of the retinoic acid receptor, could activate endoderm marker Sox17 robustly but impair the activation of the other two endoderm markers, Foxa2 and Hnf4a, indicating a complex regulating pattern in chemical-triggered cell fate transition. Given this, the mechanism by which one compound exhibits a different, even contrasting, effect on genes belonging to the same lineage requires further investigation. To obtain a designed cell type with special expression of a group of biomarkers together, we rationally optimized the time window for treatment at the beginning to guarantee the expression of Sox17, and then we substituted TTNPB with activin A in the mixture to allow the successful activation of two other endoderm marker genes. This method could be a general way to guide the cell fate transition by chemicals.
Third, in this study, to rule out the possible contamination of progenitor cells and non-fibroblast cells in the embryo, we used MEFs, MNFs, and fibroblast-specific protein (FSP) MEFs for chemical induction, respectively. All of the three cell types showed a similar response to the chemicals, indicating the universality of our protocol for cell fate transition. In a future study, we will extend this protocol for the transition of other special fibroblasts, such as those in monkeys and humans. Collectively, this discovery has promised application in future drug screening and regeneration medicine.
Last, in our study, we have identified a critical factor in the induction process, SOX17. SOX17 has been implicated in many cell fate decisions, including in endoderm progenitors and other cells (39 -43). The activation of Sox17 by chemicals offered a new platform to understand the mechanism of endoderm development and will offer new insights into cell lineage transition. In future studies, we will continue to analyze the induction process for SOX17 and use this as a marker to further improve the process in terms of timing as well as simplicity.

MEF cell isolation and culture
MEFs were derived from 13.5-day postcoitum mouse embryos from a cross of male Oct4-GFP transgenic allele-carrying mice (CBA/CaJ X C57BL/6J) to 129Sv/Jae female mice and maintained in DMEM supplemented with 10% FBS, Glu-taMAX, and NEAA.

Primary hepatocyte isolation and culture
Primary hepatocytes were isolated with the standard twostep collagenase perfusion method. Briefly, the mouse liver was preperfused through the portal vein with calcium-free buffer (0.5 mM EGTA, Hanks' balanced salt solution without Ca 2ϩ and Mg 2ϩ ) and then perfused with collagenase (0.1 mg/ml collagenase type IV (Sigma), Hanks' balanced salt solution with Ca 2ϩ and Mg 2ϩ ). After perfusion, the livers were excised and pelleted into small pieces in DMEM (Hyclone, high glucose) with an additional 9 mg/ml glucose. Then the pellets were washed and centrifuged at low speed (1,500 rpm, 10 min) 2-3 times. The purified primary hepatocytes were strained with 0.04% trypan blue to evaluate the activity, which should exceed 80%, and cultured in HCM Chemical induction of mesenchymal-to-epithelial transition medium (Lonza, CC-3198) in Matrigel-coated dishes for 4 days before harvesting for RNA extraction or other analysis.

Induction and expansion of endoderm progenitor cells (ciEPCs)
Induced ELCs were continually cultured in a modified medium, which mixed the HCM medium (Lonza, CC-3198) with the induction medium described above (days 4 -14), for another 10 days. Then cells were harvested with 0.25% trypsin and passaged in Matrigel-coated dishes at 5,000 -6,000 cell/ cm 2 in the same medium.

Immunofluorescence
Cells growing on a confocal dish (NEST catalog no. 801002) or coverslips were washed three times with PBS and then fixed with 4% PFA for 30 min and subsequently penetrated and blocked with 0.1% Triton X-100 and 3% BSA for 30 min at room temperature. Then the cells were incubated with primary antibody for 2 h. After three washes in PBS and 1 h of incubation in second antibodies, cells were then incubated in DAPI for 2 min. Then the coverslips were mounted on slides for observation on the confocal microscope (Zeiss, 710 NLO). The following antibodies were used in this project: rat anti-E-cad (

Immunoblotting
Cells were collected and lysed in lysis buffer supplemented with protease inhibitor mixture (Roche Applied Science) on ice for 15 min, and then cells were boiled at 100°C for 10 min. After centrifugation, the cell supernatants were subjected to SDS-PAGE and detected with corresponding primary antibody and second antibodies. The following antibodies were used in the project: anti-E-cad (1:1,000; CST3195), anti-c-Sox17 (1:200; R&D Systems, AF1924), and anti-GAPDH (1:5,000; Bioworld, AP2063).

qRT-PCR and RNA-seq
Total RNAs were prepared with TRIzol. For quantitative PCR, cDNAs were synthesized with ReverTra Ace (Toyobo) and oligo(dT) (Takara) and then analyzed by qPCR with Premix Ex Taq (Takara

FACS analyses
For intracellular staining of albumin, 10 6 cells were harvested and fixed with 4% PFA for 15 min and then permeabilized in Perm/

Periodic acid-Schiff (PAS) stain, DiI-ac-LDL, and ICG uptake assays
Cells were stained by PAS (Sigma) and DiI-ac-LDL (Invitrogen) following the manufacturer's instructions. For the ICG (Sigma) uptake assay, ICG was dissolved in double-distilled H 2 O and added to culture medium to a final concentration of 1 mg/ml. After cells were cultured at 37°C for 1 h, they were washed with PBS three times, and then uptake of ICG was examined by microscopy.

ALB ELISA
To determine ALB secretion, MEFs, ciEPCs, ciHeps, and primary hepatocytes were cultured in HCM medium (Lonza, CC-3198). Culture supernatant was collected 24 h after medium change. The amount of ALB in the supernatant was determined by the mouse albumin ELISA kit (Bethyl Laboratory) according to the manufacturer's instructions.

Transplantation of ciHep to ConA-induced acute liver failure mice
C57B16/J mice were injected with ConA through the tail vein at a dose of 35 mg/kg body weight. 4 h later, MEF cells, ciHep cells, or primary hepatocytes were injected into the acute liver failure mice through the tail vein. Mice in the control group were dead at 24 h after the ConA injection. Blood samples were collected from the orbital vein of the surviving mice. Liver samples were collected from the surviving animals after transplantation. All of the animal experiments were performed with the approval of and according to the guidelines of the Animal Care and Use Committee of the Guangzhou Institutes of Biomedicine and Health.

Statistics and reproducibility
Data are presented as mean Ϯ S.D., as indicated in the figure legends. For unpaired two-tailed Student's t tests, the p values were calculated with Prism version 6 software. A p value Ͻ 0.05 was considered statistically significant (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). No statistical method was used to predetermine sample size. The experiments were not randomized. The investigators were not blinded to allocation during the experiment and outcome assessment.
Author contributions-J. L. designed and performed the experiments and analyzed the data. S. C. performed the cell experiments with S. Y. S. C. and Y. C. performed primary hepatocyte isolation experiments. X. W., Y. Q., and H. L. performed RNA-seq experiment and analyzed the data. C. Z., Y. Liu, L. W., L. G., and J. K. identified the ciEPC and ciHep cell lines. D. L. and C. S. performed the cell transplantation experiment. Y. Li supervised the hepatocyte transplantation experiment, X. S. supervised the hepatocyte differentiation experiments, and J. C. and D. P. supervised the whole study. D. P. conceived the whole study, wrote the manuscript, and approved the final version.