Fibroblasts Derive from Hepatocytes in Liver Fibrosis via Epithelial to Mesenchymal Transition*

Activated fibroblasts are key contributors to the fibrotic extracellular matrix accumulation during liver fibrosis. The origin of such fibroblasts is still debated, although several studies point to stellate cells as the principal source. The role of adult hepatocytes as contributors to the accumulation of fibroblasts in the fibrotic liver is yet undetermined. Here, we provide evidence that the pro-fibrotic growth factor, TGF-β1, induces adult mouse hepatocytes to undergo phenotypic and functional changes typical of epithelial to mesenchymal transition (EMT). We perform lineage-tracing experiments using AlbCre. R26RstoplacZ double transgenic mice to demonstrate that hepatocytes which undergo EMT contribute substantially to the population of FSP1-positive fibroblasts in CCL4-induced liver fibrosis. Furthermore, we demonstrate that bone morphogenic protein-7 (BMP7), a member of the TGFβ superfamily, which is known to antagonize TGFβ signaling, significantly inhibits progression of liver fibrosis in these mice. BMP7 treatment abolishes EMT-derived fibroblasts, suggesting that the therapeutic effect of BMP7 was at least partially due to the inhibition of EMT. These results provide direct evidence for the functional involvement of adult hepatocytes in the accumulation of activated fibroblasts in the fibrotic liver. Furthermore, our findings suggest that EMT is a promising therapeutic target for the attenuation of liver fibrosis.

the liver and its functional loss, which is caused by a progressive fibrosis, occurs due to a variety of insults, such as viral hepatitis, metabolic or autoimmune diseases, toxic injury, or congenital abnormalities (1)(2)(3). Fibrosis in the liver, as with fibrosis in other organs such as lung, kidney, heart, or skin, is described as excessive deposition of extracellular matrix (ECM), 6 which leads to the destruction of organ structure and impairment of organ function. Activated fibroblasts are key mediators of organ fibrosis. In the fibrotic liver, such activated fibroblasts are considered to derive via activation and proliferation of resident stellate cells and periportal fibroblasts (1, 2, 4 -9). Because specific treatments to halt progressive fibrosis of the liver are not yet available, further understanding of the underlying pathogenic mechanisms involved in the accumulation of scar-forming fibroblasts is required.
Evidence is evolving that fibroblasts are more heterogeneous than traditionally thought (10) and recent studies reveal additional mechanisms such as recruitment of bone marrow-derived cells and conversion of resident epithelia as contributors to the accumulation of activated fibroblasts in fibrotic tissues (10,11). In this study, we attempted to gain further insights into the role of adult hepatocytes, the parenchymal epithelial cells of the liver, in the accumulation of fibroblasts in the inflamed liver. A direct contribution of hepatocytes to liver fibrosis is considered to be relatively minor, even though hepatocytes perform the majority of liver associated functions and constitute more than 80% of the liver cellular mass (1,12). Therefore, we tested the hypothesis that hepatocytes could directly contribute to the accumulation of activated fibroblasts during liver fibrosis by actively engaging in cellular transition.
Epithelial to mesenchymal transition (EMT) is defined as a process in which epithelial cells loose their phenotypic characteristics and acquire typical features of mesenchymal cells such as fibroblasts (13,14) (15). Plasticity resulting in cells shifting between epithelial and mesenchymal phenotypes is essential during embryonic development, as it permits anchored epithelial cells to re-orient their phenotypic characteristics in a developing organism (16). In adults, EMT involving resident epithelia is speculated to occur in response to injury, which protects them from cell death by serving as an additional source of fibro-blasts, which are essential for the repair of injured tissue (11,17,18). During organ fibrosis, such as in the kidney or lung, conversion of epithelium into fibroblasts is considered detrimental, as it leads to the disruption of polarized epithelial layers and an increase in the fibrotic scar formation (17). In the adult liver, however, evidence for EMT involving hepatocytes is lacking. Recent studies suggest that EMT is a promising therapeutic target for inhibition or even reversal of fibrosis in the kidney (19,20). In the fetal liver, hepatocytes are detected in various intermediate stages of EMT (defined as cells which co-express mesenchymal and also hepatocyte markers), further suggesting that at least during fetal development, hepatocytes can contribute to the emergence of cells with mesenchymal features (21,22).
In the present study, we demonstrate that in addition to ␣SMA-positive myofibroblasts a second population of fibroblast-specific protein-1 (FSP1) positive fibroblasts contribute to the progression of liver fibrosis. We demonstrate for the first time that under the influence of TGF-␤1 up to 45% of FSP1positive fibroblasts derive from hepatocytes via EMT. We further demonstrate that administration of BMP7, a growth factor which is known to inhibit TGF-␤1 action, inhibits progression of liver fibrosis and that such protection of the liver is partially mediated by prevention of EMT.

EXPERIMENTAL PROCEDURES
Materials-Recombinant human TGF-␤1 and antibodies specific to mouse albumin were purchased from R&D Systems (Minneapolis, MN). Matrigel TM , mouse type I collagen, and mouse type IV collagen were obtained from BD Biosciences (Bedford, MA). E-cadherin antibodies were purchased from Sigma-Aldrich. Alexa Fluor 488 and 568-labeled secondary antibodies were purchased from Molecular Probes (West Grove, PA). FSP1 antibodies are a gift from Dr. Eric G. Neilson.
Mice-C57BL/6 mice and CD1 mice were purchased from Charles River Laboratories (Cambridge, MA). Alb1Cre and R26RstoplacZ transgenes were described previously (23,24). In AlbCre and R26RstoplacZ double transgenic mice, the stop codon of the lacZ gene, which is ubiquitously expressed under the control of the Rosa promoter, is deleted upon expression of Cre recombinase. Because in AlbCre mice, the Cre recombinase is expressed in hepatocytes, the lacZ gene is irreversibly expressed in all cells of hepatocyte origin in livers of AlbCre; R26RstoplacZ double transgenic mice. To obtain Alb1Cre; R26stoplacZ mice matings were set up between AlbCre-positive and homozygous R26RstoplacZ mice. All mice were housed under standard conditions in the animal facility of the BIDMC. All experiments were conducted with the ethical approval of the institutional Animal Care and Use Committee of the Beth Israel Deaconess Medical Center.
Administration of rhBMP7-rhBMP7 (300 g/kg) was administered intraperitoneally to mice every other day, starting 1 day prior (prevention) or 2 weeks after the first injection of CCl 4 (intervention) for the entire duration of the study (6 weeks). rhBMP7 was a research gift from Curis, Inc.
Isolation of Primary Mouse Hepatocytes-We used a new method (modified) to isolate primary mouse hepatocytes with insignificant contamination with stellate cells (25). Livers of C57BL/6 mice were perfused through the portal vein with HBSS buffer, which contained 10 units/ml heparin. The livers were removed from the mice and were perfused with HBSS buffer, containing 0.0025% type I collagenase and 0.075% CaCl 2 for 30 min at 25°C. Then they were triturated and digested for an additional 30 min at 25°C in HBSS, containing type I collagenase. The liver suspension was filtered through an 80 micron sieve and the collected solution was sedimented for 20 min. To avoid contamination of hepatocytes with stellate cells, we used an additional purification step. The cell pellet was resuspended in 10 ml of Dulbecco's modified Eagle's medium and an equal volume of Nycodenz (18% w/v) was added. The suspension was then centrifuged for 30 min at 1,200 ϫ g. The interface containing stellate cells was discarded, and the sedimented hepatocytes were washed three times in Ham F12 medium containing 10% fetal bovine serum. The hepatocytes were then added onto a Matrigel-coated dish in F12 medium, containing 10% fetal bovine serum and 2 g/ml insulin. The purity of the hepatocytes was screened by morphology of viable cells and confirmed by albumin staining. Primary hepatocyte cultures utilized in the present study were 97% pure as determined by morphology and immunocytochemistry. To exclude potential contamination by significant numbers of stellate cells, the cells were doublestained for the expression of albumin and FSP1. About 97% of all cells were just positive for albumin expression.
Induction of EMT in Vitro-Primary mouse hepatocytes (1 ϫ 10 6 /dish) were cultured for 48 h in F12 medium containing 10% fetal bovine serum and 2 g/ml insulin until they had adhered. To induce EMT, the media was replaced with F12 media supplemented with 0.5% fetal bovine serum and 2 g/ml insulin containing TGF-␤1 for 48 h.
Migration Assay-Migration assays using a Boyden chamber were performed as previously described (26). The bottom wells of chamber were filled with F12 medium containing supplements according to the specific experimental protocol. 20,000 freshly isolated hepatocytes were added to each well in the upper chamber and the Boyden chamber was incubated at 37°C for 6 h or 48 h, to allow possible migration of cells into the lower chamber. Membranes were stained with Hema3 stain according to the manufacturer's recommendations (Biochemical Sciences, Inc., Swedesboro, NJ). Cells that migrated through the membrane were counted using a counting grid, which was fitted into an eyepiece of a phase contrast microscope. All experiments were repeated at least three times.
Immunocytochemistry-EMT in primary mouse hepatocytes was visualized by immunofluorescent double staining with antibodies to albumin, FSP1, ␤-galactosidase and E-cadherin as described with minor modifications (27). Cells were fixed with acetone at 4°C and incubated for 1 h with the primary antibodies. For E-cadherin/FspI double staining experiments, the cells were fixed for 20 min in ice-cold acetone, and the slides were incubated with the primary antibodies at 4°C overnight. Cells were then washed again and incubated for 45 min at room temperature with Alexa Fluor 488-and Alexa Fluor 568-conjugated secondary antibodies. The nuclei were counterstained using TOPRO3, and sections were covered using Vectashield mounting media (Vector, Burlingame, CA). Staining was analyzed using a confocal microscope. CCl 4 -induced Liver Fibrosis-Liver fibrosis was induced by CCl 4 injection in mice as previously described (28). Briefly, C57BL/6 mice were injected with CCl 4 (250 l of 20% v/v CCl 4 ) intraperitoneally twice a week. Mice were sacrificed after 4 weeks and their livers were harvested. The tissue was fixed in 10% formalin for histopathology or snap-frozen in Tissue-Tek TM for immunohistochemistry. We started treatment with 300 g/kg rhBMP7, which was determined as the optimum dose in several animal studies the day before the first CCl 4 injection (prevention study) or 2 weeks after the first CCl 4 injection (intervention trial) as previously reported (19,29). RhBMP7 was administered by intraperitoneal injections every other day until week six (n ϭ 6 mice per group), when the mice were sacrificed and livers and serum were obtained.
Tissue Preparation and Histological Assessment of Liver Fibrosis-The livers were removed, washed in phosphate-buffered saline, and cut in cross-sections of about 3 mm. Of each liver one piece of tissue was fixed in 4% paraformaldehyde, and embedded in paraffin for morphometric analysis. The other piece was embedded in OCT and snap-frozen in liquid nitrogen to perform immunohistochemistry. Paraffin embedded tissue was cut into 5-m thick cross-sections, and sections were stained with hematoxylin and eosin, and Masson's trichrome. To quantify liver fibrosis, three independent Masson's trichrome-stained sections were analyzed from each mouse, using a counting grid. The percent area of liver fibrosis was calculated as previously described (30).
Immunofluorescence Staining-We performed immunofluorescence staining as previously described (29). Frozen tissue was cut into 5-m thick cross-sections which were fixed in 100% acetone at Ϫ20°C for 10 min. We incubated the sections with primary antibodies FSP-1 (31), LacZ (Alpha Diagnostics), Collagen III (Southern Biotechnology) or ␣-smooth muscle actin (Sigma) at 4°C overnight. After the sections were washed with Tris-buffered saline, they were subsequently stained with secondary antibodies. The nuclei were counterstained with DAPI (Vectashield) for fluorescence microscopy or with TOPRO-3 (Molecular Probes) for confocal microscopy. FSP-1 antibody was a gift from E. G. Neilson (Vanderbilt University, Nashville, TN).
Staining Analysis-Relative stained area was assessed using NIH ImageJ software as described in our previous publications (32). For quantification of FSP1-positive and FSP1/␤-gal double-positive cells a total of 1000 cells were evaluated for each group.
Immunohistochemistry-Immunohistochemistry of 10% formaldehyde-fixed, paraffin-embedded liver sections was performed as described previously with minor modifications (33). Briefly, sections were de-paraffinized and microwaved in Antigen Unmasking Solution (Vector Laboratories, Burlingame, CA) according to the manufacturer's recommendations.
Immunohistochemical staining was performed using primary antibodies to Alk2 (R&D systems), Alk3 (R&D Systems) and Alk-6 (Santa Cruz Biotechnology) and the Vector Alkalinephosphatase immunoassay kit (Vector Laboratories). Positive staining was visualized with the Vector Alkaline Phosphatase Substrate Kit I, and sections were counterstained with methylene-green.
Albumin ELISA-Serum albumin was determined using an ELISA kit (Alpha Diagnostics, San Antonio, TX) according to the manufacturer's recommendations. Briefly, serum samples were diluted 1:70,000 in sample diluent. The precoated ELISA plate was incubated with 20 l of the diluted samples for 60 min at room temperature, followed by vigorous washing and incubation with a secondary antibody-horseradish peroxidase conjugate solution. After washing horseradish peroxidase-substrate solution was added, and color development was assessed by measurement of the absorbance at 450 nm using an ELISA reader. The serum albumin concentration was calculated by plotting the net absorbance values against appropriate albumin standard concentrations. The normal value for serum albumin in C57BL/6 mice, assessed by testing 25 different adult C57BL/6 mice sera samples, was 20.3 mg/ml.
Statistical Analysis-Results are expressed as means Ϯ S.E. Multiple comparisons were performed by one-way analysis of variance. p values lower than 0.05 were considered significant.

Accumulation of FSP1-positive Fibroblasts in Liver Fibrosis-
The key mediators of liver fibrosis are activated fibroblasts, which are traditionally detected by means of ␣-smooth muscle actin (␣SMA) expression (1,34). However, recent studies in the setting of fibrosis in kidney, solid tumors, lung, and skin reveal that the heterogeneity of activated fibroblasts is greater than commonly believed (35). FSP1-positive fibroblasts (characterized by their expression of fibroblast specific protein 1, a member of the S100 family of calcium-binding proteins), is one class of fibroblasts present in many tissues (17,36). To gain insights into the contribution of FSP1-positive fibroblasts to liver fibrosis, we established a disease progression curve by sacrificing cohorts of mice with CCl 4 -induced liver fibrosis at various time points after initial injection of CCl 4 (2 weeks, 4 weeks, 6 weeks, n ϭ 6 in each group) (Fig. 1). In control livers, few FSP1-positive fibroblasts are noticed, predominantly surrounding central veins and within the periportal region (Fig. 1A). Immunofluorescence double-labeling experiments using antibodies specific to FSP1 and Collagen III revealed that FSP1-positive fibroblasts accumulate during fibrosis progression, predominantly in areas of collagen III deposition (Fig. 1, A-C). Experiments in which liver sections were double-labeled with antibodies to ␣SMA and FSP1 reveal that these two markers rarely co-localize ( Fig.  1, D-I). In control livers, ␣SMA expression is restricted to vascular smooth muscle cells, distinct from the FSP1-positive cells (Fig. 1, D-G). During early stages of liver fibrosis (after 2 weeks of CCl 4 administration), both ␣SMA-positive cells and FSP1positive cells increase in numbers, but do not overlap (Fig. 1, E  and H). During advanced stages of fibrosis (after 6 weeks of CCl 4 administration), both ␣SMA-positive cells and FSP1-positive cells substantially accumulate in the areas of fibrotic septae ( Fig. 1, F and I). However, only ϳ10% of FSP1-positive cells are also positive for ␣SMA (Fig. 1I).
TGF-␤1 Induces EMT in Primary Mouse Hepatocytes-FSP1-positive fibroblasts have been well-characterized in fibrogenesis in kidney, lung, and solid tumors. As opposed to ␣SMA fibroblasts, which predominantly derive via activation of resi-dent stellate cells and fibroblasts or via shedding from vascular smooth muscle cells, FSP1-positive fibroblasts have been shown to derive by a significant extend from resident epithelia via epithelial-mesenchymal transition. We hypothesized that hepatocytes could contribute to the pool of FSP1-positive fibroblasts in the setting of liver fibrosis. TGF-␤1 is considered to be the principal mediator of liver fibrosis (37,38), and it is the most potent inducer of EMT in different organ systems (39,40). Therefore, we tested the capacity of TGF-␤1 to induce EMT in primary cultures of adult mouse hepatocytes (41).
In the absence of growth factors, 97% of cells isolated from adult mouse livers were albumin-positive hepatocytes (see methods). FSP1 staining and ␣SMA staining was negligible (less than 3% of the cells) ( Fig. 2A-C). After exposure of these primary mouse hepatocytes to 3 ng/ml of human TGF-␤1 for 48 h (optimum stimulus for the induction of EMT in several different epithelial systems), the albumin-positive cells also uniformly express FSP1 (Fig. 2, D-F). The negligible number of albumin negative cells (Ͻ 3% of the cells) did not increase. After 96 h of TGF-␤1 incubation, albumin staining was lost and the cells were uniformly spindle-shaped and positive for FSP1, typical of an EMT-derived fibroblast (Fig. 2, G-K). Gain of FSP1 expression upon exposure to TGF-␤1 also coincided with decreased staining for the epithelial marker E-cadherin (Fig. 2, L-Q) The findings suggest that after 48 h of TGF-␤1 incubation an intermediate phenotype (positive staining for both the hepatocyte marker and fibroblast marker) is observed, whereas after 96 h a more complete EMT occurs.
We next confirmed that the observed phenotypic change was indeed reflective of an EMT (loss of epithelial properties combined with gain of functional fibroblast properties) as opposed to hepatocyte de-differentiation (loss of hepatocyte properties). Migration is a hallmark feature of activated fibroblasts, which enables them to fulfill their physiological function during wound repair by quickly appearing at the site of injury (42), and like other cells of epithelial lineage, hepatocytes are immobile and tightly integrated into the epithelial cell layer (43). In order to provide further evidence that TGF-␤1 induces a gain of fibroblast-like function involving hepatocytes, we evaluated migratory capacity of primary mouse hepatocytes. We used a two-compartment Boyden chamber, which utilized a polycar- Nuclei were visualized using TOPRO-3 staining. The pictures display representative photomicrographs, which were acquired by confocal microscopy at an original magnification of ϫ60. Whereas FSP1 staining was absent in control hepatocytes (A-C), hepatocytes were positive for both albumin and FSP1 after 48 h (F). After 96 h, albumin staining was neglible (G-I). J and K, light microscopy. While primary mouse hepatocytes displayed a homogenous round-shaped appearance (J), stimulation with TGF-␤1 induced an acquisition of a fibroblast-like phenotype (K). L-Q, E-cadherin/FSP1 double staining. Normal hepatocytes express E-cadherin (red). Incubation of primary hepatocytes with TGF-␤1 induces FSP1 expression (green) associated with decreased E-cadherin staining.  bonate membrane coated with type IV collagen and type I collagen (both up-regulated during fibrosis) to study migration. The migration rate was assessed after 6 h and after 48 h of stimulation with TGF-␤1 (44). After stimulation with TGF-␤1 for 6-h migration of cells into the bottom chamber was already increased as compared with the untreated control. (2.2-fold increase compared with control) (Fig. 3, A-C). The total number of cells that had migrated after 6 h (an average of 144 cells out of a total 20,000 seeded cells) was rather small at this time point. Incubation with TGF-␤1 for 48 h (allowing hepatocytes enough to undergo EMT) induced significantly increased migration of primary hepatocytes into the bottom compartment (an average of 2460 cells out of 20,000 originally seeded hepatocytes) (Fig. 3, D-F). In summary, these results suggest that EMT of primary hepatocytes is associated with a significantly enhanced motility.
Evidence for EMT Involving Adult Hepatocytes in a Mouse Model for Liver Fibrosis-Our studies suggest that hepatocytes could acquire a fibroblast-like phenotype in the in vitro experiments. Next, we evaluated the presence of hepatocyte-derived fibroblasts in mouse models of liver fibrosis. We first evaluated for possible EMT involving hepatocytes in the mouse model of CCl 4 -induced liver fibrosis. In this model, TGF-␤1-dependent liver fibrosis begins 2 weeks post-CCl 4 injection, eventually leading to severe liver fibrosis after 6 weeks (28). Among the accumulated FSP1-positive fibroblasts, up to 60% (after 2 weeks of CCl 4 injection) co-labeled with albumin (Fig. 4, A-G), consistent with the intermediate phenotype of EMT observed in our in vitro studies. In the advanced stages of fibrosis (after 6 weeks of CCl 4 injection), 17% of all cells were FSP1 positive and up to 28% of these FSP1 positive fibroblasts were positive for albumin, indicating their hepatocyte origin. While these studies demonstrate the presence of hepatocytes, which reflect a possible intermediate transitional stage of EMT (positive for both the fibroblast marker FSP1 and the hepatocyte marker albumin), these studies could not demonstrate a complete conversion of hepatocytes to fibroblasts, because the methodology design was insufficient to distinguish fibroblasts which potentially derive de novo from resident fibroblasts via proliferation, from those which are derived via EMT loss of albumin expression.
Genetic Lineage Tag Evidence for EMT Involving Hepatocytes during Liver Fibrosis-To provide further evidence for EMT involving hepatocytes during liver fibrosis, we next performed lineage analysis of fibroblasts recruited during fibrosis, using double transgenic mice which express Cre-recombinase under control of the albumin promoter (AlbCreϩ) and which also harbor an allele in which a STOP cassette flanked by two loxP sites is inserted between the R26Rosa promoter and the LacZ gene (R26RstoplacZ) (Fig. 5A) (24,45). The albumin gene is expressed by mature hepatocytes. In AlbCreϩ mice, the Cre recombinase is expressed under the Alb promoter and thus present in mature hepatocytes (45). In the R26RstoplacZ indicator strain, the LacZ reporter gene is activated after Cre-mediated excision of a LoxP stop cassette (24). Therefore, in the AlbCre; R26RstoplacZ double transgenic mice the LacZ gene remains activated in hepatocytes and in all cells with a previous hepatocyte origin, despite possible phenotype changes.
To investigate the contribution of EMT involving hepatocytes in liver fibrosis, CCl 4 was administered to AlbCre; R26RstoplacZ adult mice. In the normal livers of AlbCre; R26RstoplacZ mice, LacZ expression can be detected in ϳ70% of hepatocytes, whereas LacZ staining was completely absent in livers of wt; R26RstoplacZ control mice (Fig. 5B). These results confirm previous studies, demonstrating a high specificity (no LacZ staining in cells other than hepatocytes, absence of staining in livers of AlbCre-negative control mice) albeit limited sensitivity (ϳ70%) of this system (45). Administration of CCl 4 over a period of 6 weeks results in liver fibrosis similar to that observed in the wt C57BL6 mice (Fig. 5C). LacZ staining of fibrotic livers reveals several lacZ positive cells within fibrotic zones (Fig. 5D). Double-labeling of liver tissues with FSP1 antibodies (red fluorescence staining) and ␤-gal antibodies (green fluorescent staining) reveals that in fibrotic livers (after 6 weeks of CCL 4 ) 18.4% (ϩ4.24%) of all cells were FSP1-positive and 8.2% (ϩ1.13%). This means that up to 45% of FSP1-positive fibroblasts are also positive for ␤-gal, indicating that about half of FSP1-positive fibroblasts in liver fibrosis exhibit a hepatocyte imprint (Fig. 5D). To further verify that FSP1/␤-gal doublepositive cells were indeed representative of hepatocyte-derived fibroblasts, we isolated primary hepatocytes from AlbCre; R26RstoplacZ mice and incubated them with TGF-␤1. Hepatocytes derived from AlbCre; R26RstoplacZ mice labeled positive for ␤-gal while they were negative for FSP1 in absence of TGF-␤1. When TGF-␤1 was added for 48 h, cells co-expressed FSP1 and ␤-gal (Fig. 5E), further supporting our analysis using anti-albumin antibodies in the previous experiments (Fig. 2,  D-F). However, even after 96 h of TGF-␤1 incubation, when albumin expression is lost in our earlier experiments (Fig. 2, G-I), the AlbCre; R26RstoplacZ hepatocytes remain positive FIGURE 5. Lineage tracing of hepatocytes using AlbCre; R26RstoplacZ mice. A, schematic illustration. Crossing of AlbCre;wt mice with R26RstoplacZ mice generates AlbCre ϩ ; R26RstoplacZ mice. Cre recombinase removes the STOP cassette from between the Rosa promoter and the LacZ gene in albumin-positive cells. In these cells the LacZ gene is constitutively expressed under control of the R26 Rosa promoter, even if albumin expression is lost. B, ␤-galactosidase staining. ␤-galactosidase staining was performed on livers from AlbCre Ϫ ; R26RstoplacZ mice (left picture) and from AlbCre ϩ ; R26RstoplacZ mice. When this method is used, the presence of ␤-galactosidase (product of the LacZ gene) is indicated by a blue precipitate. The absence of blue staining in wt; R26RstoplacZ mice indicates high specificity of the system. Approximately 70% of hepatocytes stain blue in AlbCre; R26RstoplacZ livers. C, liver fibrosis in AlbCre; R26RstoplacZ mice. AlbCre; R26RstoplacZ mice received bi-weekly injections of CCl 4 for 6 weeks. MTS staining revealed liver fibrosis after 6 weeks. Beta galactosidase staining in fibrotic livers from AlbCre; R26RstoplacZ mice revealed the presence of single ␤-gal-positive cells within fibrotic lesions. The right panel shows a higher magnification (ϫ63) of the boxed region in the left picture. The dotted line in the right picture indicates the outer margin of the fibrotic area, the arrows indicate single ␤-gal-positive cells within the fibrotic lesion. D, ␤-gal/FSP1 double labeling in vivo. We performed immunofluorescence double labeling using antibodies specific to ␤-gal (green) and FSP1 (red). In fibrotic livers FSP1-positive cells were present that were also positive for ␤-gal, revealing their hepatocyte origin. The graph summarizes the average percentage of FSP1 positive fibroblasts and FSP1-␤-galactosidase double-positive cells (hepatocyte-derived fibroblasts) per ϫ40 visual field. E, ␤-gal/FSP1 double labeling in vitro. Primary hepatocytes were isolated from AlbCre; R26RstoplacZ mice and subjected to TGF-␤1 for 96 h. Acquisition of a spindle shaped morphology was associated with de novo expression of FSP1. ␤-gal expression remained stable, demonstrating the hepatocyte origin of these cells. The photomicrographs were acquired using a confocal microscope at a magnification of ϫ60.
for ␤-gal (Fig. 5E), demonstrating the utility of the system to detect without ambiguity the contribution of hepatocytes in the generation of fibroblasts during liver fibrosis.

Prevention of EMT by Recombinant Human Bone Morphogenesis Protein-7 (rhBMP7) Inhibits Progression of Liver Fibrosis-
BMP7 is a member of the TGF-␤ superfamily, involved in morphogenesis associated with a variety of organs. In the liver, BMP7, in concert with other BMPs such as BMP-4, regulates proliferation and differentiation of the liver bud, eventually leading to complete hepatocyte differentiation (46). Functional BMP7 receptors Alk3 and BMPR-II are expressed on adult hepatocytes, and because our data suggest that EMT involving hepatocytes is TGF-␤1-dependent, we hypothesized that administration of recombinant human BMP7 (rhBMP7) could potentially inhibit progression of liver fibrosis by preventing EMT involving hepatocytes.
To test our hypothesis, we first subjected primary mouse hepatocytes to either TGF-␤1, BMP7, or both TGF-␤1 and BMP7 together and evaluated the impact on EMT by albumin and FSP1 co-labeling. Whereas TGF-␤1 induced loss of albumin expression associated with newly acquired FSP1 expression (indicative of EMT), incubation with BMP7 (100 ng/ml) alone had no visible effect on hepatocyte phenotype (Fig. 6, A-F). However, BMP7 blocked TGF-␤1 induced EMT when primary mouse hepatocytes were incubated with both TGF-␤1 and BMP7 (Fig. 6, A-F).
To evaluate the therapeutic effect of rhBMP7 on hepatocytes in vivo, we used the CCl 4 -induced liver fibrosis model in mice. We initiated treatment with rhBMP7 before induction of disease (prevention group) and also 2 weeks after the initial CCl 4 injection, when fibrotic lesions were established (intervention group). All animals were sacrificed after 6 weeks of disease and liver histology and serum albumin were compared between the BMP7 administered mice and the control mice. After 6 weeks, fibrotic lesions were substantially less in both the groups that received rhBMP7, as compared with the control group (Fig. 6, G-N). Improved histology was associated with improved liver function as determined by serum albumin levels (Fig. 6J). We next performed albumin/FSP1 double-labeling experiments to establish whether BMP7 inhibited EMT involving hepatocytes in vivo. Similar to our results obtained using primary mouse hepatocytes, cells that labeled for both albumin and FSP1 (indicative of EMT) were significantly diminished in rhBMP7 administered mice (Fig. 6, O-R).
We further analyzed liver from control mice and from mice that received CCl4 for 6 weeks to gain insights into the receptors that mediate BMP7-action in hepatocytes. BMP7 is known to bind to the BMP type II receptor, which then forms complexes with specific type I serine-threonine kinase receptors (Alk receptors). BMP7 binds specifically to Alk-3 and Alk-6 and also to the activin receptor Alk-2 (47,48). Immunohistochemistry using specific antibodies to Alk-2, Alk-3, and Alk-6 revealed that in control livers mainly Alk-3 is present on hepatocytes (Fig. 7, A-C). Interestingly, in fibrotic livers we also detected Alk-2 and Alk-6 expression on the hepatocytes (Fig. 7, D-F), suggesting that each of these receptors is potentially involved in mediating BMP7 action in the fibrotic liver.

DISCUSSION
Activated fibroblasts are the principal mediators of organ fibrosis. In the liver, several studies have unequivocally demonstrated that resident adult stellate cells can respond to many different forms of insult, including TGF-␤1, by converting or contributing to activated fibroblasts seen in the injured liver (7,49,50). In addition, other cell types of the fibroblast lineage, such as portal fibroblasts or vascular myofibroblasts are also being recognized in this setting for their fibrogenic potential FIGURE 6. Inhibition of EMT by rhBMP7 is associated with amelioration of liver fibrosis. A-F, albumin-FSP1 double labeling of primary hepatoctyes. Primary mouse hepatocytes were subjected to either TGF-␤1, BMP7, or combination of TGF-␤1 and BMP7 or to control media without growth factors. After 96 h, cells were stained using antibodies specific for FSP1 (green) and albumin (red). TGF-␤1 induced FSP1 expression (D), whereas BMP7 alone had no effect on either albumin of FSP1 expression (E). In combination with TGF-␤1, BMP7 prevented loss of albumin expression and FSP1 expression was not detected (F). G-N, prevention of liver fibrosis by BMP7. Livers from untreated control mice with CCl 4 -induced liver disease, from mice that received rhBMP7 starting from day 0 of disease (prevention) and from mice that received treatment after 14 days of disease (intervention), were evaluated by MTS staining (G-I). The pictures in the upper row display representative MTS-stained liver sections of each group. Treatment with rhBMP7 resulted in substantial amelioration of fibrotic lesions (original magnification ϫ20). The graph summarizes serum albumin measurements for each group (J). The pictures in the bottom row display representative photomicrographs of liver sections that were labeled with antibodies to type III collagen (green) and TOPRO-3 nuclear counterstain (blue). K-M, the bar graph summarizes average type III collagen-stained area in each group (N). O-R, FSP1/albumin double staining. After 6 weeks of CCl 4 -induced disease, livers from untreated control mice and from mice that received treatment with rhBMP7 for 6 weeks were stained with antibodies to FSP1 (red) and albumin (green) and the percentage of FSP1-positive fibroblasts and FSP1-positive/albumin double-positive cells (hepatocytes undergoing EMT) from total liver cells was assessed using a confocal microscope at an original magnification of ϫ100. In normal control livers and in BMP7-treated livers FSP1/albumin double-positive cells were absent. BMP7 treatment significantly decreased the number of FSP1-positive cells. The graphs summarize the quantification of six different livers, the pictures display representative staining for each group. *, p Ͻ 0.01; **, p Ͻ 0.005; ***, p Ͻ 0.001. FIGURE 7. Immunolocalization of BMP receptors in the liver. Liver sections obtained from control mice (A-C) and mice that received CCL4 for 6 weeks (D-F) were stained with antibodies specific for ALk-2 (A and D), Alk-3 (B and E), and Alk-6 (C and F). Positive staining is indicated by red precipitate, nuclei are stained green. The pictures display representative photomicrographs of each group at an original magnification of ϫ40. In control livers, Alk-2 staining is predominantly located on endothelial cells (A, arrows), whereas Alk-2-positive hepatocytes are detected in fibrotic livers (D, arrows). In control livers robust Alk-3 staining is present, predominantly nuclei stain positive (B, arrows). In fibrotic livers, hepatocytes, and fibroblasts are Alk-3-positive (E, arrows). In fibrotic livers, few hepatocytes were also Alk-6 positive (F, arrows). (7). In this regard, a contribution of damaged, injured or TGF-␤1-influenced adult hepatocytes to this process is undetermined. Here we provide evidence that hepatocyte-derived fibroblasts are an additional and significant lineage of mesenchymal cells that contribute to progression of liver fibrosis.
In the present study we demonstrate that a substantial population (up to 45%) of FSP1-positive fibroblasts derives from hepatocytes via EMT. These results are similar to studies in the kidney, which report that about 40% of all fibroblasts are derived via EMT (17). While the importance of fibroblasts in the progression of liver fibrosis is undisputed, this study demonstrates that fibroblasts can increase in their number via different mechanisms. The relative individual contribution of stellate cell-derived fibroblasts, resident liver fibroblasts and hepatocyte-derived fibroblasts to the overall activated fibroblast population might be crucial for the progression of liver fibrosis. Whereas our studies demonstrate that hepatocytes contribute to the accumulation of fibroblasts by undergoing EMT, more extensive studies are required to determine if all hepatocytes, or just a select subpopulation, have the capacity to undergo EMT.
We demonstrate for the first time that administration of rhBMP7 effectively inhibits progression of fibrosis in the liver. The most striking effect of rhBMP7 administration was the almost complete absence of hepatocytes undergoing EMT. This confirms our working hypothesis that BMP7 is an inhibitor of TGF-␤1-induced EMT. Additionally, strong correlation for the absence of EMT coupled with the inhibition of liver fibrosis suggests that EMT involving hepatocytes contributes significantly to the pathogenesis of liver fibrosis. We note that in rhBMP7-administered mice, the overall liver structure was improved associated with lesser number of inflammatory cells, decreased number of fibroblasts and less steatosis when compared with the livers from control mice. Additional studies are required to delineate underlying molecular mechanisms of the total therapeutic effect of BMP7, including its anti-inflammatory effect. BMP7 is not expressed in the liver, but the functional receptors for BMP7 are present on hepatocytes. We demonstrate that each of the three known BMP type I receptors are expressed on hepatocytes in diseased livers, and we previously demonstrated that in the setting of partial keratectomy neutralization of endogenous circulating BMP7 slows liver regeneration, suggesting that physiological levels of endogenous BMP7 have a putative role for hepatocytes in the setting of acute liver injury (51).
In summary, we demonstrate that hepatocytes can contribute to the accumulation of activated fibroblasts via EMT. EMT involving hepatocytes plays a functional role in the progression of fibrosis in two different mouse models for chronic liver disease. Inhibition of EMT by rhBMP7 leads to attenuation of liver fibrosis and improvement of liver function. Therefore, rhBMP7 may have a value as a therapeutic agent to control liver fibrosis.