Fate-Mapping Evidence That Hepatic Stellate Cells Are Epithelial Progenitors in Adult Mouse Livers



Liver injury activates quiescent hepatic stellate cells (Q-HSC) to proliferative myofibroblasts. Accumulation of myofibroblastic hepatic stellate cells (MF-HSC) sometimes causes cirrhosis and liver failure. However, MF-HSC also promote liver regeneration by producing growth factors for oval cells, bipotent progenitors of hepatocytes and cholangiocytes. Genes that are expressed by primary hepatic stellate cell (HSC) isolates overlap those expressed by oval cells, and hepatocytic and ductular cells emerge when HSC are cultured under certain conditions. We evaluated the hypothesis that HSC are a type of oval cell and, thus, capable of generating hepatocytes to regenerate injured livers. Because Q-HSC express glial fibrillary acidic protein (GFAP), we crossed mice in which GFAP promoter elements regulated Cre-recombinase with ROSA-loxP-stop-loxP-green fluorescent protein (GFP) mice to generate GFAP-Cre/GFP double-transgenic mice. These mice were fed methionine choline-deficient, ethionine-supplemented diets to activate and expand HSC and oval cell populations. GFP(+) progeny of GFAP-expressing precursors were characterized by immunohistochemistry. Basal expression of mesenchymal markers was negligible in GFAP(+)Q-HSC. When activated by liver injury or culture, HSC downregulated expression of GFAP but remained GFP(+); they became highly proliferative and began to coexpress markers of mesenchyme and oval cells. These transitional cells disappeared as GFP-expressing hepatocytes emerged, began to express albumin, and eventually repopulated large areas of the hepatic parenchyma. Ductular cells also expressed GFAP and GFP, but their proliferative activity did not increase in this model. These findings suggest that HSC are a type of oval cell that transitions through a mesenchymal phase before differentiating into hepatocytes during liver regeneration.

Disclosure of potential conflicts of interest is found at the end of this article.


Author contributions: L.Y. and Y.J.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing; A.O., R.P.W., S.C., and H.M.V.: collection and/or assembly of data; J.H. and G.D.A.: provision of study material or patients; A.M.D.: conception and design, financial support, data analysis and interpretation, manuscript writing, final approval of manuscript.

The role of hepatic stellate cells in liver repair is being reconsidered because of growing evidence that these cells express certain stem cell markers [1, [2]–3] and produce various trophic factors for liver epithelial cells [4, [5], [6], [7]–8]. It is generally acknowledged that hepatic stellate cell (HSC) populations in adult livers are heterogeneous [9, 10]. In healthy livers, quiescent hepatic stellate cells (Q-HSC) store vitamin A and other lipids [11, [12]–13]. They localize between sinusoidal endothelial cells and hepatocytes in the space of Disse. Chronic liver injury activates resident HSC to a proliferative, myofibroblastic (MF) phenotype. Myofibroblastic hepatic stellate cells (MF-HSC) synthesize α-smooth muscle actin (α-SMA) and are contractile. They also produce extracellular matrix molecules, such as type 1 collagen. Because portal hypertension, liver fibrosis, and sometimes cirrhosis ensue when populations of MF-HSC expand, the transition of Q-HSC to MF-HSC is thought to portend a bad outcome in liver injury [9, 14, 15].

Other evidence, however, suggests that HSC may support regeneration of damaged livers. For example, HSC have been localized in canals of Hering, the stem cell niche in adult livers [16, [17]–18]. In addition, they are known to produce morphogens, including hepatocyte growth factor, epimorphin, pleiotrophin, and Hedgehog ligands [4, [5], [6], [7]–8]. In embryos and neonates, stellate cells may be important for the development of the intrahepatic biliary tree [19]. The trophic paracrine interactions between HSC and cholangiocytes persist into adulthood, as evidenced by reports that conditioned medium from adult HSC promotes the growth of cholangiocyte cell lines [20]. Therefore, it seems likely that in addition to remodeling matrix, HSC play a larger role in adult liver repair.

The minority opinion even posits that HSC may be progenitors for liver epithelial cells. This argument is based on data showing that Q-HSC from adults express markers of all three embryonic germ layers, such as glial fibrillary acidic protein (GFAP; an ectodermal marker) [21, [22]–23], desmin (a mesodermal marker) [24], and Hes-1(an endodermal marker) [25], as well as various stem cell markers, including nestin, CD105, p75 neurotrophin receptors, c-kit, and CD133 [1, [2]–3, 26, 27]. Recent cell culture findings support the plasticity of HSC by showing that Q-HSC are capable of differentiating into cell types other than MF [1]. For example, both primary rat HSC [1] and rat HSC lines that were clonally derived from primary rat HSC [7] differentiate into myofibroblastic, hepatocytic, bile ductular, or endothelial-like cells depending on the in vitro conditions. The opposing view, however, argues that the apparent multipotency of HSC isolates results from the outgrowth of rare other cell types that contaminated the initial cell preparations. The same logic would confound interpretation of studies in which HSC isolates are transplanted into recipients to assess the role of HSC progeny in liver repair.

Fate-mapping approaches overcome the inherent limitations of working with isolated cells and thus are widely used to investigate cell fate during embryogenesis [28]. Therefore, we used this strategy to evaluate the fate of Q-HSC following liver injury. To mark Q-HSC and their progeny, we crossed transgenic mice in which GFAP promoter elements regulate Cre recombinase expression with other transgenic mice bearing floxed repressor alleles that control expression of green fluorescent protein (GFP) to generate double-transgenic GFAP-Cre/GFP mice. In the double-transgenic strain, progeny of GFAP-expressing Q-HSC are identified by expression of the marker gene (GFP), which persists even after GFAP gene expression has become downregulated. Therefore, it is possible to determine whether types of cells that do not normally express GFAP, such as hepatocytes and fibroblasts [21, [22]–23], are derived from Q-HSC. In humans, many types of chronic liver injury activate Q-HSC to myofibroblasts and induce expansion and differentiation of liver progenitors to replace dead mature hepatocytes [29, [30]–31]. Because chronic ingestion of methionine/choline-deficient supplemented with 0.15% ethionine (MCDE) diets are known to cause HSC activation and liver repopulation by progenitors in rodents [32, [33]–34], we fed GFAP-Cre/GFP mice and controls (GFP-floxed mice) MCDE diets to model the pathophysiology that occurs clinically.

We found that liver sinusoids of healthy adult GFAP-Cre/GFP mice were lined by stellate-appearing cells that expressed GFAP, GFP, and Cre-recombinase. As expected, ingestion of MCDE diets provoked liver injury and fibrogenesis, and complete recovery from liver damage eventually occurred after the hepatotoxic diets were discontinued. Liver samples were obtained at different time points during this process, and the fate of GFAP(+) Q-HSC was investigated using immunohistochemistry. Changes in gene expression profiles of primary HSC were also evaluated by quantitative reverse transcription (QRT)-polymerase chain reaction (PCR) and Western blot analysis of samples obtained during culture-induced activation. Our results support the concept that Q-HSC are capable of functioning as multipotent progenitors and can give rise to hepatocytes in adult livers.

Materials and Methods

Animal Model of Liver Injury

GFAP-Cre mice [35] and two strains of reporter mice, ROSA-loxP-stop-loxP-GFP [36] and ROSA-loxP-stop-loxP-LacZ [37], were obtained from Jackson Laboratory (Bar Harbor, ME, http://www.jax.org). Mice expressing Cre recombinase under the control of GFAP regulatory elements (GFAP-Cre mice) were crossed with transgenic mice in which LoxP sites flanked a stop codon that repressed expression of reporter genes that encode either GFP (LoxP-GFP mice) or LacZ (LoxP-LacZ mice). In GFAP-Cre/LoxP-reporter double-transgenic mice (hereafter referred to as GFAP-Cre/GFP mice and GFAP-Cre/LacZ mice), GFAP promoter elements drive expression of Cre recombinase to generate active enzyme that then removes the stop codon that is flanked by LoxP sites, thereby permitting expression of the marker gene (GFP or LacZ). Hence, in any cell (that at any time) activates GFAP expression, Cre recombinase is generated, and this, in turn, results in an event (removal of the Lox P-flanked repressor element) that unleashes the ability to express the marker gene, GFP or LacZ. This strategy permits identification not only of cells that are actively expressing GFAP but also of GFAP-negative cells that are descendants (progeny) of GFAP-expressing precursors that “passed” the capability of expressing GFP or LacZ to their daughters.

Preliminary studies demonstrated that LoxP-GFP and LoxP-LacZ mice had normal-appearing livers and that HSC from these mice displayed typical culture-related induction of myofibroblastic genes. Therefore, since reporter gene expression is silenced in LoxP-GFP and LoxP-LacZ mice, we used these mice as controls in all subsequent experiments. In contrast, reporter gene expression is evident in GFAP-expressing cells and their progeny in GFAP-Cre/-Cre/GFP or GFAP-Cre/LacZ double-transgenic mice.

All mice were housed in a facility with a 12-hour light/dark cycle and allowed free access to food and water. Adult mice were used between 3 and 4 months of age. To induce oxidative liver injury, inhibit replication of mature hepatocytes, and activate HSC and liver progenitor populations, GFAP-Cre/GFP mice (n = 15) were fed an MCDE diet [38, [39], [40], [41]–42]. Surviving mice were sacrificed after being fed MCDE diets for either 1 week (n = 4) or 3 weeks (n = 4). Another group of GFAP-Cre/GFP mice that were fed MCDE for 3 weeks (n = 4) was switched back to normal diet for another 3 weeks to allow liver recovery and then sacrificed. Chow-fed GFAP-Cre/LacZ mice (n = 4) were also sacrificed to obtain tissue for localization of β-galactosidase activity. Animal care and surgical procedures were approved by the Duke University Medical Center Institutional Animal Care and Use Committee as set forth in the Guide for the Care and Use of Laboratory Animals published by the NIH.

Cell Isolation and Culture

Hepatic stellate cells (HSC) were isolated from mice as previously described [43]. Briefly, after in situ perfusion of the liver with pronase (Boehringer Mannheim, Indianapolis, IN, http://www.boehringer.com) followed by collagenase (Crescent Chemical, Hauppauge, NY, http://www.crescentchemical.com), dispersed cell suspensions were layered on a discontinuous density gradient of 8.2% and 15.6% Accudenz (Accurate Chemical, Westbury, NY, http://www.accuratechemical.com). The resulting upper layer consisted of >95% stellate cells, as defined by morphology and vitamin A autofluorescence. The viability of all cells was verified by phase contrast microscopy, as well as the ability to exclude propidium iodide. The viability of all cell cultures used for study was >95%. Isolated stellate cells were seeded at a density of 3 × 102 cells per mm2 with Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 100 units/ml streptomycin, and 100 units/ml penicillin. Rat cholangiocytes were isolated as described as before [44] and were kindly provided by Dr. Gianfranco Alpini (Texas A&M University).

mRNA Quantification by Real-Time Reverse Transcription-PCR

mRNAs were quantified by real-time reverse transcription (RT)-PCR per the manufacturer's specifications (Mastercycler Real-Time PCR; Eppendorf AG, Hamburg, Germany, http://www.eppendorf.com). The sequences of primers for mouse 18S, collagen I α1, α-SMA, GFAP, peroxisome proliferator-activated receptor γ (PPARγ), aquaporin-1, NCAM, MPK, cytokeratin 7 (CK7), enhanced green fluorescent protein (EGFP), CK19, and α-fetoprotein (AFP) were as follows: 18S: sense, 5′-TTGACGGAAGGGCACCACCAG-3′; antisense, 5′-GCACCACCACCCACGGAATCG-3; collagen I α1: sense, 5′-GAGCGGAGAGTACTGGATCG-3′; antisense, 5′-GCTTCTTTTCCTPPTGGGGTTC-3′; α-SMA: sense, 5′-AAACAGGAATACGACGAAG-3′; antisense, 5′-CAGGAATGATTTGGAAAGGA-3′; GFAP: sense, 5′-GCTTCCTGGAACAGCAAAAC-3 ′; antisense, 5′-ATCTTGGAGCTTCTGCCTCA-3′; PPARγ: sense, 5′-CACAATGCCATCAGGTTTGG-3′; antisense, 5′-GCTGGTCGATATCACTGGAGATC-3′; Aquaporin-1: sense, 5′-GCTGGTCCAGGACAACGTG-3′; antisense, 5′-CCGCAGCCAGTGTAGTCAAT-3′; NCAM: sense, 5′-GACGTCCGGTTCATAGTCCT-3′; antisense, 5′-GGCAGTGGCATTCACGA-3′; MPK: sense, 5′-GCGTGTAGTGCCTGTACCTT-3′; antisense, 5′-GTAGGGCCCTGAATAATAGCTG-3′; CK7: sense, 5′-TAGAGTCCAGCATCGCAGAG-3′; antisense, 5′-CACAGGTCCCFATTC CGTC-3′; EGFP: sense, 5′-ACGTAAACGGCCACAAGTTC-3′; antisense, 5′-AAGTCGTGCTGCTTCATGTG-3′; CK19: sense, 5′-GTGAAGATCCGCGACTGGT-3′; antisense, 5′-AGGCGAGCATTGTCAATCTG-3′; AFP: sense, 5′-GGTCGCTGGATCTCTAGGCT-3′; antisense, 5′-GCGGAAAGTCTCTCGGTCT-3′.

Total RNA was extracted from cells or whole livers using TRIzol (Invitrogen, Carlsbad, CA, http://www.invitrogen.com). One microgram of RNA was reverse-transcribed using random primer and Superscript RNase H-reverse transcriptase (Invitrogen). Samples were incubated at 20°C for 10 minutes and 42°C for 30 minutes; reverse transcriptase was inactivated by heating at 99°C for 5 minutes and cooling at 5°C for 5 minutes. Amplification reactions were performed using a SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA, http://www.appliedbiosystems.com). Five microliters of diluted cDNA samples (1:5 dilution) were used for quantitative two-step PCR (a 10-minute step at 95°C, followed by 50 cycles of 15 seconds at 95°C and 1 minute at 65°C) in the presence of 400 nM specific forward and reverse primers, 5 mM MgCl2, 50 mM KCl, 10 mM Tris buffer (pH 8.3), 200 μM dATP, dCTP, dGTP, and 400 μM dUTP, and 1.25 U of AmpliTaq Gold DNA polymerase (PerkinElmer Life and Analytical Sciences, Boston, http://www.perkinelmer.com; Applied Biosystems). Each sample was analyzed in triplicate. Target gene levels in treated cells or tissues are presented as a ratio to levels detected in corresponding control cells or tissues, according to the ΔΔCT method.

Western Blot Analysis

Protein lysates were also prepared from the aforementioned HSC cultures and analyzed by immunoblot analysis as described [8]. Blots were incubated with primary antisera to either GFAP (1:1,000; Santa Cruz Biotechnology Inc., Santa Cruz, CA, http://www.scbt.com); a mesenchymal marker, α-sma (1:1,000; Sigma-Aldrich, St. Louis, http://www.sigmaaldrich.com); or an epithelial marker, CK7 (1:1,000; Santa Cruz Biotechnology). The expression of β-actin (Sigma-Aldrich), a constitutively expressed housekeeping protein, was used as a loading control.


Specimens fixed in formalin and embedded in paraffin were cut into 4-μm sections, dewaxed, hydrated, and subsequently incubated for 10 minutes in 3% hydrogen peroxide to block endogenous peroxidase. Antigen retrieval was performed by heating in 10 mM sodium citrate buffer (pH 6.0) for 10 minutes or incubation with 0.25% pepsin for 10 minutes. Sections were blocked in Dako protein block (X9090; Dako, Glostrup, Denmark, http://www.dako.com) for 30 minutes and incubated with primary antibodies, GFP (ab6556; 1:2,500; Abcam, Cambridge, MA, http://www.abcam.com), GFAP (M0334; 1:4,000; Dako), cre (69053-3; 1:10,000; Novagen, La Jolla, CA, http://www.emdbioscience.com), α-smooth muscle actin (M0851; 1:1,000; Dako), Ki-67 (ab15580; 1:1,000; Abcam), and AE1/3 (18-0132; 1:500; Invitrogen) at room temperature 2 hours. Other sections were also incubated at room temperature for 2 hours in nonimmune sera. Polymer-horseradish peroxidase (HRP) anti-rabbit (K4003; Dako) or anti-mouse (K4001; Dako) was used as secondary antibody. DAB reagent was used in the detection procedure. Omitting primary antibodies from the reactions eliminated staining, demonstrating staining specificity.

For immunofluorescent staining, cells were fixed, permeabilized, and processed for immunostaining with primary antibody GFAP (1:1,000; Calbiochem, San Diego, http://www.emdbiosciences.com). Alexa Fluor 488 (Molecular Probes, Eugene, OR, http://probes.invitrogen.com) was used as secondary antibodies. For double immunohistochemical staining, frozen liver sections were used. Samples were fixed and permeabilized, saturated, and processed for immunostaining with primary antibody GFP (ab6556) and albumin (MAB1455; 1:100; R&D Systems Inc., Minneapolis, http://www.rndsystems.com). Alexa Fluor 568 and Alexa Fluor 488 (Molecular Probes) were used as secondary antibodies. 4,6-Diamidino-2-phenylindole (DAPI) counterstaining was used to demonstrate nuclei. To quantify Ki-67 and GFP staining, seven portal tract (PT) and central vein (CV) areas were randomly selected per section at ×40 for each mouse. PT selected for analysis contained a portal vein that ranged from 130 to 180 μm. The average number of Ki-67 or GFP-positive cells was obtained by dividing the total number of positive cells by the total number of cells around the PT or CV.

Anonymized liver sections were also examined from three patients without chronic liver disease who had liver resections for colorectal metastases. Tissues were obtained from the Duke University School of Medicine Tissue Bank Shared Resource and studied in accordance with NIH and institutional guidelines for human subject research.

Double Immunostaining

To determine whether expression of certain proteins colocalized in cells, formalin-fixed, paraffin-embedded mouse liver sections were double-immunostained for GFP and α-SMA, α-SMA and the S-phase marker Ki-67, and GFP and the liver epithelial progenitor marker AE1/AE3. Polymer-HRP anti-rabbit (K4003; Dako) and the MACH3 mouse AP polymer kit (MP530; Biocare Medical, Concord, CA, http://www.biocare.net) were used as secondary antibodies. In each double-immunostaining experiment, GFP or Ki-67 was identified by DAB (Dako) to generate a brown color, and each of the other markers was identified by the Ferangi Blue chromogen kit (FB812S; Biocare Medical) that generated a blue color. Deidentified, double-stained sections were examined by two independent observers who counted the numbers of single- and double-positive cells in seven fields per section per mouse under ×40 magnification. Interobserver variability was negligible. Mean ± SEM results are presented as number of double(+) cells per field. To determine whether GFAP expression colocalized with CK7, a marker of immature biliary epithelial cells, frozen human liver sections were costained to demonstrate GFAP (M0334; 1:4,000; Dako) and CK7 (M7018; 1:750; Dako). Alexa Fluor 568 and Alexa Fluor 488 (Molecular Probes) were used as secondary antibodies. DAPI counterstaining was used to demonstrate nuclei. Bile ductules composed of double-positive cells were identified and photographed under ×63 magnification.

Analysis of LacZ Expression

For qualitative analysis of lacZ expression, fixed tissues were stained using the LacZ Detection Kit for Tissues (Invivogen, San Diego, http://www.invivogen.com) according to the manufacturer's instructions.

Statistical Analysis

Results are expressed as mean ± SEM. Significance was established using Student's t test and analysis of variance when appropriate. Differences were considered significant when p < .05.


Liver sinusoids of healthy adult control mice (Fig. 1A) and GFAP-Cre/GFP mice (Fig. 1D) were lined by stellate-appearing cells that expressed GFAP. In GFAP-Cre/GFP mice, expression of GFP and Cre-recombinase was similarly localized (Fig. 1E, 1F). Control mice did not exhibit expression of GFP or Cre-recombinase (Fig. 1B, 1C). The staining characteristics of Q-HSC in GFAP-Cre/GFP mice suggested that these mice would be useful for tracking the progeny of Q-HSC. Unexpectedly, however, expression of GFAP (Fig. 2A), Cre-recombinase (Fig. 2B), and GFP (Fig. 2C) was also demonstrated in bile duct cells and ductular-appearing cells in peri-portal canals of Hering in GFAP-Cre/GFP mice. A number of experiments were done to determine whether ductular cell expression of these markers was artifactual. First, additional GFAP-Cre recombinase mice were crossed with mice harboring floxed-LacZ alleles to generate double-transgenic GFAP-Cre/LacZ mice. β-Galactosidase staining confirmed that intrahepatic and extrahepatic bile ducts of these GFAP-Cre/LacZ mice were derived from GFAP(+) cells (Fig. 2D, 2E). The biliary tree in control mice exhibited no β-galactosidase activity (Fig. 2F), confirming the specificity of this approach for detecting LacZ expression. Second, immunohistochemistry was used to demonstrate ductular cell staining for GFAP in nondiseased liver sections from healthy control mice (Fig. 2G) and patients who were undergoing resection of metastatic colorectal cancers (Fig. 2H). In patients, frozen liver sections were also costained for both GFAP and CK7, a marker of bipotent liver epithelial progenitors and immature biliary epithelial cells. Bile ductules composed of GFAP/CK7 double-positive epithelial cells were demonstrated by immunofluorescence microscopy (supplemental online Fig. 1). Finally, expression of GFAP was demonstrated at the RNA level in freshly isolated primary cholangiocytes and HSC, but not hepatocytes, from healthy adult rats (Fig. 2I). Thus, contrary to current dogma [21, [22]–23], Q-HSC are not the only type of cell that expresses GFAP in adult livers. In several species, ductular cells also express this marker. Hence, Q-HSC, ductular cells, and their progeny are all specifically marked by GFP in GFAP-Cre/GFP mice.

Figure Figure 1..

Glial fibrillary acidic protein (GFAP), Cre-recombinase, and green fluorescent protein (GFP) expression in hepatic stellate cells (HSC) of control (Ctr) and GFAP-Cre/reporter mice. Immunohistochemistry for GFAP (A, D), Cre-recombinase (B, E), and GFP (C, F) was performed in representative Ctr (A–C) and GFAP-Cre/GFP (transgenic [TG]) (D–F) mice. GFAP was expressed by hepatic stellate cells (HSC) in a representative Ctr mouse (A) and a representative GFAP-Cre/GFP (TG) mouse (D). HSC expressed Cre recombinase (E) and GFP (F) in TG mice but not in Ctr mice (B, C). Magnification, ×100.

Figure Figure 2..

Glial fibrillary acidic protein (GFAP), Cre-recombinase, and green fluorescent protein (GFP) expression in Chol of GFAP-Cre/reporter mice. GFAP (A), Cre-recombinase (B), and GFP (C) were expressed by bile ductular cells in portal triads of a representative GFAP-Cre/GFP (transgenic) mouse. (D): Eosin-counterstained liver section from a representative GFAP-Cre/LacZ mouse showing LacZ(+) bile ductular cells. Extrahepatic bile ducts, cystic duct, and gallbladder exhibited β-galactosidase activity in a representative GFAP-Cre/LacZ mouse (E) but not in a control mouse (F). Bile ductular cells in healthy adult mice (G) and a typical sample of nondiseased human liver (H) also expressed GFAP. Primary rat Chol and HSC expressed GFAP mRNA, but primary Hep did not (I). Magnification, ×63. Abbreviations: Chol, cholangiocytes; Hep, hepatocytes; HSC, hepatic stellate cells.

To evaluate the effects of liver injury on the GFAP(+) cell populations, GFAP-Cre/GFP mice were fed MCDE diets. Some mice were sacrificed 1 or 3 weeks later. Others were withdrawn from the MCDE diets at the 3-week time point and placed back on normal chow diets for an additional 3 weeks before being sacrificed. Ingestion of MCDE diets provoked liver injury and fibrogenesis, as evidenced by increased serum aminotransferase levels (Fig. 3A), hyperbilirubinemia (Fig. 3B), upregulation of matrix gene expression (Fig. 3C), and loss of liver mass (Fig. 3D). However, complete recovery from liver damage eventually occurred after the hepatotoxic diets were discontinued (Fig. 3A–3D). Liver sections were obtained from mice that had been fed the diets for either 1 or 3 weeks and from mice that had been fed MCDE diets for 3 weeks and then returned to normal chow. Immunohistochemistry (IHC) was done to track changes in the populations of cells that expressed GFAP and Cre-recombinase (supplemental online Fig. 2). Results in the groups that had been fed MCDE diets were also compared with findings in healthy transgenic mice before exposure to MCDE diets (Figs. 1D–1F, 2A–2C). Bile ductular cells expressed both GFAP and Cre-recombinase before (Fig. 2A–2C), during (supplemental online Fig. 2A–2D), and after (supplemental online Fig. 2E, 2F) MCDE diet exposure, whereas GFAP-positive (+ve) sinusoidal cells were noted only in mice with healthy livers (Fig. 2). Mature-appearing hepatocytes were not noted to be GFAP+ve in any of the groups at any of the time points that were evaluated.

Figure Figure 3..

Effect of MCDE diet on serum liver enzymes levels, liver Col I α1 mRNA, and liver mass. Serum ALT (A) and TB (B) levels were measured in CON mice fed MCDE diets for 3 W (MCDE 3W) and mice fed MCDE diets for 3 W and then returned to normal chow for 3 W before sacrifice (MCDE restored) (n = 4 mice per group per time point). Col I α1 mRNA expression was evaluated in liver samples from the same mice using real-time polymerase chain reaction (C). (D): Liver weight in the CON and both MCDE diet groups. *, p < .05 compared with CON mice; n = 4 mice per group per time point. Abbreviations: ALT, alanine aminotransferase; Col 1 α1, collagen I α1; CON, controls; MCDE, methionine/choline-deficient supplemented with 0.15% ethionine; TB, total bilirubin; W, weeks.

Changes in populations of cells that were derived from GFAP+ve cells (i.e., GFP-expressing cells) were assessed in the same animals (Fig. 4). After 1 week of exposure to the hepatotoxic diets, the number of GFP+ve cells in portal tracts was similar to that of healthy livers; however, large numbers of GFP+ve cells had accumulated in hepatic sinusoids (Fig. 4A, 4B). Accumulation of GFP+ve, fibroblastic-appearing sinusoidal cells was particularly prominent in perivenular and midzonal areas (acinar zones 3 and 2, respectively) (Fig. 4A). Occasional hepatocytic-appearing cells in these areas also expressed GFP (Fig. 4B) at this time point. After 3 weeks of MCDE diet treatment, GFP staining had become localized closer to portal tracts (Fig. 4C). Hepatocytic cells were the predominant GFP+ve cell type in these areas (Fig. 4D). Mice that had been withdrawn from MCDE diets and allowed to recover for a 3-week period retained GFP expression in hepatocytes (Fig. 4F), and this was also most intense peri-portally (Fig. 4E). Thus, before and early after the onset of liver injury, GFP was expressed by sinusoidal and bile ductular cells. However, with time, livers accumulated hepatocytic cells that expressed GFP. Such cells were apparent first in perivenular and midzonal areas (zones 3 and 2, respectively) but later accumulated predominately around portal tracts (i.e., in zone 1). During the 6-week time period of these experiments, the net numbers of GFP+ve cells (i.e., bile duct cells, sinusoidal cells, and hepatocytes) increased significantly (Fig. 4G), with hepatocytic cells accounting for the largest numbers of GFP+ve cells in mice that had recovered from MCDE diet-induced liver damage.

Figure Figure 4..

GFP expression in the livers of transgenic mice after injury. Glial fibrillary acidic protein-Cre/GFP mice (n = 15) were fed MCDE diets, and survivors were sacrificed after 1 W (n = 4) or 3 W (n = 4). Other mice that survived 3 W of MCDE treatment (n = 4) were sacrificed after being switched back to normal diets for 3 W. GFP expression was assessed by immunohistochemistry. Shown are representative results in mice that were sacrificed at the end of 1 W of MCDE diet treatment (A, B), 3 W of MCDE treatment (C, D), or 3 W after stopping a 3-W course of MDCE diets (E, F). Numbers of GFP-positive cells visualized at ×40 were counted in different parts of the liver lobule (i.e., near CV and PT) (G). *, p < .05; **, p < .005. Original magnifications, ×10 (A, C, E), ×63 (B, D, F). Arrows and arrowheads indicate portal triads and CV, respectively. Abbreviations: CON, controls; CV, central veins; GFP, green fluorescent protein; MCDE, methionine/choline-deficient supplemented with 0.15% ethionine; PT, portal tracts; W, weeks.

To clarify the origins of the GFP+ve hepatocytic cells that accumulated during the regeneration of injured livers, expression of Ki-67, an S-phase marker, was assessed before, during, and after liver injury (Fig. 5A). Hepatocytes in healthy livers rarely expressed Ki-67 and did not upregulate this proliferation marker during or after liver injury. Rare bile ductular cells in portal tracts were Ki-67+ve at baseline, but numbers of Ki-67+ve bile ductular cells did not change much during or after liver injury. In contrast, the numbers of sinusoidal cells expressing Ki-67 increased more than 10-fold in perivenular areas during liver injury and then returned to baseline with recovery. Ki-67+ve fibroblastic-appearing cells were easily demonstrated in liver sinusoids after 1 week of MCDE diet treatment (Fig. 5B). Most of these cells coexpressed α-SMA, a marker of myofibroblastic HSC (Fig. 5C). The number of α-SMA/Ki-67 double-positive cells was more than 10-fold higher in the livers of mice that had received 1 week of MCDE diet treatment than in healthy mice but quickly declined to basal levels (Fig. 5D). Thus, liver injury induced a transient wave of proliferative activity in α-SMA+ve sinusoidal cells, and this preceded the accumulation of GFP-expressing hepatocytic cells. Together, these findings suggest that the GFP+ve hepatocytic cells may have been derived from the fibroblastic (i.e., α-SMA+ve) cells.

Figure Figure 5..

Ki-67 expression in the livers of transgenic mice. Glial fibrillary acidic protein-Cre/green fluorescent protein mice (n = 15) were fed MCDE diets, and survivors were sacrificed as described above (n = 4 mice per group per time point). Immunohistochemistry for Ki-67 or double immunohistochemistry for Ki-67/αSMA was performed. Numbers of Ki-67-positive cells visualized at ×63 (high power) were counted in different parts of the liver lobule (i.e., near CV and PT) and in different cell types (i.e., Hep, BD, and Sinu). Results are graphed as numbers of positive cells per field in (A). *, p < .05. Shown are Ki-67 expression (B) and Ki-67 and αSMA coexpression (C) in mice that were sacrificed at the end of 1 W of MCDE treatment. Numbers of DP cells for Ki-67and αSMA in different areas of the liver lobule (CV or PT) were counted at different time points, and results are graphed in (D). *, p < .05; **, p < .005. Magnification, ×100 (A, C). Abbreviations: BD, bile ductular cells; CON, controls; CV, central veins; DP, double-positive; Hep, hepatocytes; MCDE, methionine/choline-deficient supplemented with 0.15% ethionine; PT, portal tracts; Sinu, perisinusoidal lining cells; αSMA, α-smooth muscle actin; W, weeks.

IHC was done to track α-SMA expression before, during, and after MCDE diet-induced liver injury (supplemental online Fig. 3). Before exposure to MCDE diets, control GFAP-Cre/GFP mice exhibited only rare α-SMA+ve cells in portal tracts. Liver sinusoids (which contained GFAP+ve stellate cells; Fig. 1) lacked α-SMA-expressing cells (supplemental online Fig. 3A, 3B). After 1 week of treatment with MCDE diets, however, livers contained large numbers of α-SMA+ve cells, and these were localized predominately in perivenular and mid-zonal sinusoids. Occasional hepatocytic cells in zone 3 also expressed this marker (supplemental online Fig. 3C). Because GFAP-expressing Q-HSC are an important source of the GFAP-negative/α-SMA+ve fibroblasts that accumulate in injured livers [21, [22]–23], livers were double-immunostained to determine whether these markers colocalized in the cells of GFAP-Cre/GFP mice (Fig. 6A, 6B). As expected, numerous sinusoidal cells that coexpressed GFP and α-SMA were noted in hepatic sinusoids of mice that had been fed MCDE diets for 1 week. Scattered GFP/α-SMA double+ve hepatocytic cells were also evident. The latter were particularly prominent near clusters of double+ve sinusoidal cells and were also observed immediately adjacent to terminal hepatic venules. At this time point, these areas also harbored rare hepatocytic cells that expressed GFP, but not α-SMA, although the scattered GFP/α-SMA double(+) hepatocytic cells were much less prevalent than double+ve sinusoidal or hepatocytic cells. Over the course of the study, the numbers and lobular distribution of the GFP/α-SMA double+ve sinusoidal cells (Fig. 6C) corresponded to those observed for α-SMA-expressing cells in general (Fig. 5), peaking after 1 week of liver injury and then quickly declining to basal levels. Thus, MCDE diet-induced liver injury transiently expanded populations of GFP/α-SMA double+ve sinusoidal cells. This process was accompanied by the appearance of occasional GFP/α-SMA double+ve hepatocytic cells, and both events occurred weeks before large numbers of hepatocytic GFP-positive/α-SMA-negative hepatocytic cells accumulated.

Figure Figure 6..

Coexpression of GFP with αSMA, AE1/3, or albumin in the livers of transgenic mice after liver injury. Glial fibrillary acidic protein-Cre/GFP mice (n = 15) were fed MCDE diets, and survivors were sacrificed as described above (n = 4 mice per group per time point). Double immunohistochemistry for GFP and αSMA (A, B) or AE1/3 (D) in the 1-W MCDE diet group. Numbers of DP cells for GFP and αSMA in different areas were counted at different time points, and results are graphed in (C). *, p < .05; **, p < .005. Shown are costaining for GFP (E) and albumin (F) and a merged photomicrograph demonstrating coexpression of these markers (G) in a representative mouse that was switched from MCDE diet to normal chow for 3 W. Original magnifications, ×100 (A, B, D), ×63 (E–G). Abbreviations: CON, controls; DP, double-positive; GFP, green fluorescent protein; MCDE, methionine/choline-deficient supplemented with 0.15% ethionine; αSMA, α-smooth muscle actin; W, weeks.

In the aggregate, the IHC data support the concept that the accumulation of GFP+ve hepatocytic cells that occurred as MCDE diet-injured livers regenerated resulted from the differentiation of GFP/α-SMA double+ve sinusoidal cells. To investigate this possibility further, liver sections obtained before, during, and after exposure to hepatotoxic diets were stained to demonstrate liver progenitor cell cytokeratins (supplemental online Fig. 4). Double immunostaining was also done to determine whether the GFP+ve hepatocytic cells that appeared early after the onset of liver injury coexpressed AE1/AE3-reactive cytokeratins that are known to exist in subpopulations of hepatic epithelial progenitors [45]. Occasional hepatocytic cells in zones 2 and 3 coexpressed GFP and hepatic progenitor cytokeratins after 1 week of treatment with MCDE diets (Fig. 6D). Hence, early after the onset of liver injury in GFAP-Cre/GFP mice, GFP+ve hepatocytic cells that coexpressed α-sma and epithelial progenitor cell cytokeratins emerged. This event was transient and preceded the accumulation of large numbers of GFP(+) hepatocytic cells that lacked expression of either of the other markers. Because mature hepatocytes normally do not express either α-SMA or AE1/AE3-reactive cytokeratins, IHC was done to determine whether the GFP(+) hepatocytic cells expressed albumin protein, a marker of mature hepatocytes. Livers were examined when the animals had completely recovered from their diet-induced liver injury (Fig. 3), at a time when GFP(+) hepatocytic cells made up almost one-third of the hepatic parenchyma (Fig. 4). Virtually all of the GFP(+) hepatocytic cells coexpressed albumin at this time point (Fig. 6E–6G), confirming that restoration of liver mass (Fig. 3) resulted from reconstitution of the liver by functioning mature hepatocytes.

Because the findings in MCDE-treated mice suggested that liver injury provoked Q-HSC to transition through a proliferative, myofibroblastic phase before differentiating into mature hepatocytes, we examined primary HSC for markers of hepatocyte progenitors and evidence of epithelial-mesenchymal transitions. QRT-PCR analysis of RNA obtained from freshly isolated HSC revealed expression of classic markers for Q-HSC (GFAP and PPARγ; Fig. 7A, 7B), as well as biliary epithelial cells (CK19 and aquaporin-1; Fig. 7C, 7D), immature hepatocytes (AFP; Fig. 7H), and more primitive epithelial progenitors (mpk, NCAM, and CK7; Fig. 7E–7G). Thus, the gene expression profile of our Q-HSC suggested that these isolates were enriched with putative hepatocyte and cholangiocyte progenitors. During standard culture conditions, these cells downregulated their expression of GFAP and PPARγ (Fig. 7A, 7B) and became myofibroblastic, upregulating their expression of α-SMA and collagen I α1 (Fig. 7I, 7J). The latter results are consistent with published data about gene expression changes that typically occur as wild-type Q-HSC transition to MF-HSC [9, 46]. During HSC activation to a myofibroblastic phenotype, expression of S100A4 (also called fibroblast-specific protein) (Fig. 7K), a marker of epithelial-derived fibroblasts [47], increased, and expression of both biliary epithelial markers (Fig. 7C, 7D) and markers of immature hepatocytes (Fig. 7G, 7H) fell. Expression of mpk and NCAM (Fig. 7E, 7F), markers of immature liver progenitors, increased concomitantly. Western blot analysis verified that changes in mRNA expression of representative stellate cell, mesenchymal, and epithelial markers were accompanied by similar changes in protein content (Fig. 7M). Throughout this process, primary HSC from GFAP-Cre/GFP mice retained GFP expression at both the mRNA and protein levels (Fig. 7L, 7N, 7O), demonstrating that the activated HSC were progeny of Q-HSC that initially expressed GFAP.

Figure Figure 7..

GFP expression in hepatic stellate cells (HSC) from GFAP-cre/GFP mice. Analysis of gene expression by freshly isolated and 5-day culture-activated primary HSC from adult mice. Real-time reverse transcription-polymerase chain reaction was used to evaluate GFAP (A), PPARγ (B), AQP (C), CK19 (D), MPK (E), NCAM (F), CK7 (G), AFP (H), α-SMA (I), collagen I α1 (J), S100A4 (K), and GFP (L) mRNA levels. (M): Western blot analysis for CK7, αSMA, and GFAP. Cultures of HSC were stained with anti-GFP antibody on day 0 (N) and day 5 (O). Original magnification, ×63. *, p < .05 vs. day 0. Abbreviations: AFP, α-fetoprotein; AQP, aquaporin-1; CK, cytokeratin; Col 1 α1, collagen I α1; GFAP, glial fibrillary acidic protein; GFP, green fluorescent protein; PPARγ, peroxisome proliferator-activated receptor γ; αSMA, α-smooth muscle actin.


Our findings support the concept that HSC are capable of differentiating into hepatocytes. Published evidence had suggested this possibility, but definitive proof has been difficult to acquire in experimental animals [16, 18, 48, 49]. More direct evidence for this concept was provided by studies that cultured a subpopulation of HSC that had been enriched for cells expressing the multipotent progenitor marker CD133. Depending on the culture conditions, such cells were shown to generate various cell types, including myofibroblasts, cholangiocytes, and hepatocytes [1]. However, because primary HSC isolates were the source of the CD133(+) cells, it was impossible to exclude the possibility that differential outgrowth of rare cell types that contaminated the original preparation might have accounted for the findings. Earlier analyses of clonal lines that were derived from HSC that had been isolated from a single adult CCl4-treated rat [50] argue against cell contamination as an explanation for HSC heterogeneity. Different HSC clones from that rat exhibited variable coexpression of liver epithelial and mesenchymal markers. Moreover, clones that predominately expressed mesenchymal genes when cultured in media with a high serum content acquired the epithelial-predominant phenotype of other AFP/CK19-coexpressing clones when cultured in serum-depleted medium [7]. Although compelling, however, the findings in cell lines might merely have reflected changes that were acquired during the cloning process.

The fate-mapping approach used in the present study overcomes some of the inherent limitations of the earlier strategies that were used to characterize HSC. In rats, mice, and humans, Q-HSC express GFAP and downregulate this marker as they become myofibroblastic [21, [22]–23, 51]. By using GFAP promoter elements to regulate expression of Cre-recombinase in transgenic mice carrying floxed repressor elements that controlled expression of GFP or LacZ alleles, we were able to track the fate of Q-HSC during liver injury and regeneration. We confirmed that basal expression of various mesenchymal markers is negligible in Q-HSC. During both injury-related activation in mice and “spontaneous” activation that occurs during culture on plastic dishes, HSC become highly proliferative and begin to coexpress markers of mesenchyme and progenitors. In GFAP-Cre/GFP mice, these transitional cells disappeared as GFP-expressing hepatocytes emerged, and eventually the hepatocytic cells repopulated large areas of the hepatic parenchyma. Freshly isolated primary HSC expressed several epithelial genes and acquired a mesenchymal-type phenotype when we cultured them under standard conditions that are known to encourage myofibroblastic transformation and growth [11, [12]–13]. Another group has already demonstrated that primary HSC can differentiate into hepatocytes when they are cultured with appropriate growth factors [1]. Hence, our new findings complement and extend earlier in vitro and in vivo evidence that HSC populations contain hepatocyte progenitors.

HSC also appear to be related to ductular cells. Although intermediate filament expression often varies over the course of cellular differentiation, qualitative differences in intermediate filaments are often used to differentiate cell types [52]. Others have reported that both HSC and ductular cells express synemin [48], and the present study demonstrates that they also share expression of GFAP, which is another type of intermediate filament [53]. The similarities in intermediate filament subtypes that exist in HSC and ductular cells suggest that HSC and ductular cells may have a common lineage. This concept is supported by our observation that freshly isolated primary HSC also express several other biliary epithelial markers, such as CK19, aquaporin-1, and CK7. Ductular cells have long been implicated as hepatocyte progenitors [54]. In livers injured by MCDE diets, however, ductular cell proliferation did not increase. Therefore, this cell population did not expand to replace hepatocytes and recover the liver mass that was lost during MCDE diet exposure. Yin et al. reported similar findings in allyl alcohol-treated rats, another model in which liver regeneration is accomplished by oval cells [18]. In healthy livers, oval cells are thought to reside along canals of Hering, the most proximal extensions of the biliary tree [55]. These cells are considered to be the adult equivalent of fetal hepatoblasts, that is, bipotent hepatic progenitors that can differentiate into either hepatocytes or cholangiocytes [56, 57]. Thus, the heterogeneity of oval cell populations is thought to reflect the admixture of primitive oval cells and their progeny that are at various stages of differentiation [58]. Although our studies do not resolve whether or not Q-HSC are the progeny or precursors of ductular cells, these cells are clearly capable of generating hepatocytes. Therefore, HSC may belong to the oval cell family.

Some may consider the concept that HSC are a type of oval cell to be heretical. However, in addition to the literature that has already been discussed, there is actually a considerable body of published data about fetal livers that also supports this hypothesis. Liver epithelial cells and some HSC are both thought to originate from endoderm [59, [60]–61]. Fetal livers contain large numbers of bipotent hepatoblasts (oval cells), as well as cells that coexpress liver epithelial and HSC markers (e.g., cytokeratins 7/8, desmin, and α-sma) [60]. Hence, the simplest hypothesis to explain all of our observations is that HSC are a type of oval cell, and in certain circumstances, these cells transition through a mesenchymal phase before differentiating into mature liver epithelial cells, including hepatocytes. This logic is appealing because it does not preclude the possibility that the cells may terminally differentiate into fibroblasts in other microenvironments. It also allows for the possibility that some epithelial progeny of HSC may be capable of undergoing epithelial-mesenchymal transition to reconstitute more fibroblastic populations, as was recently demonstrated in some hepatocytic cells that were cultured under profibrogenic conditions [62] and in bile ductular cells in patients with primary biliary cirrhosis [63]. Viewed from this perspective, HSC appear positioned to dictate the ultimate outcome of liver injury, and efforts to differentiate mechanisms that promote their maturation into epithelial, as opposed to fibroblastic, cells have important clinical implications.

Disclosure of Potential Conflicts of Interest

The authors indicate no potential conflicts of interest.


This work was supported in part by National Institute on Alcohol Abuse and Alcoholism Grant 5R01-AA0 10154 (to A.M.E.D.) and by a Department of Veterans Affairs (VA) Research Scholar Award and VA Merit Award (to G.D.A.). L.Y. and Y.J. contributed equally to this work.