Paracrine activation of hepatic stellate cells in platelet-derived growth factor C transgenic mice; evidence for stromal induction of hepatocellular carcinoma

Abstract

Cirrhosis is the primary risk factor for the development of hepatocellular carcinoma (HCC), yet the mechanisms by which cirrhosis predisposes to carcinogenesis are poorly understood. Using a mouse model that recapitulates many aspects of the pathophysiology of human liver disease, we explored the mechanisms by which changes in the liver microenvironment induce dysplasia and HCC. Hepatic expression of platelet-derived growth factor-c (PDGF-C) induces progressive fibrosis, chronic inflammation, neoangiogenesis, and sinusoidal congestion, as well as global changes in gene expression. Using reporter mice, immunofluorescence, immunohistochemistry and liver cell isolation, we demonstrate that receptors for PDGF-CC are localized on hepatic stellate cells (HSCs), which proliferate, and transform into myofibroblast-like cells that deposit extracellular matrix (ECM) and lead to production of growth factors and cytokines. We demonstrate induction of cytokines genes at two months, and stromal cell-derived hepatocyte growth factors that coincide with the onset of dysplasia at four months. Our results support a paracrine signaling model wherein hepatocyte-derived PDGF-C stimulates widespread HSC activation throughout the liver leading to chronic inflammation, liver injury, and architectural changes. These complex changes to the liver microenvironment precede the development of HCC. Further, increased PDGF-CC levels were observed in livers of patients with non-alcoholic fatty steatohepatitis (NASH) and correlate with the stage of disease, suggesting a role for this growth factor in chronic liver disease in humans. PDGF-C transgenic mice provide a unique model for the in vivo study of tumor-stromal interactions in the liver.

Introduction

Hepatocellular carcinoma (HCC) is the fifth most common type of cancer worldwide, and is one of the few types of cancer for which the death toll is rising¹. In the majority of cases, HCC arises in cirrhotic liver, resulting from viral infection, alcoholic or non-alcoholic fatty liver disease (NAFLD), genetic disorders, or other causes ². The prognosis of HCC is grim; the only curative treatments are liver resection for rare early cases and liver transplant for more advanced disease. Unfortunately, only a small percentage of HCC patients qualify for these operations, given the cirrhotic liver's poor regenerative response to resection and necessarily strict selection criteria for transplantation².

Repetitive cycles of cell injury, proliferation, and inflammation lead to fibrosis and eventual cirrhosis ^{2, 3}; the mechanisms by which cirrhosis predisposes to HCC are not well understood. There are several well-characterized animal models of liver fibrosis, including bile-duct ligation and repeated injections of carbon tetrachloride⁴, but these models do not spontaneously lead to HCC unless additional insults are employed, such as a choline-deficient diet ⁵ or carcinogens⁴. We developed a transgenic (Tg) mouse model wherein platelet-derived growth factor (PDGF-C) is ectopically expressed in hepatocytes, and found that these mice develop progressive liver fibrosis with a high incidence of HCC⁶. Since hepatic over-expression of this mesenchymal growth factor results in epithelial neoplasia, we hypothesize that PDGF-C promotes carcinogenesis by altering the liver microenvironment ^7-9.

PDGF-C is a member of the PDGF ligand family that includes four isoforms, PDGF-A, -B, -C, and -D^{10, 11}. The ligand dimers bind to either PDGF receptor (PDGFR) α or β subunits, resulting in hetero- or homo-dimerization and activation of intracellular signaling pathways. PDGFs are potent mitogens for mesenchymal cells, and function in wound healing and angiogenic processes ^{10, 11}. Despite the fact that there is overlap in receptor utilization, all four ligands are required during embryogenesis, indicating that each growth factor has a non-redundant function¹¹

PDGFs also drive fibrogenesis in a number of organs, including lung, vasculature, skin, and liver ¹⁰. In normal human liver, PDGF-A, –B and –C expression is relatively low, but is frequently elevated in cirrhotic liver and HCC^12-14. Similarly, expression levels of all four PDGFs increase in rats with experimental liver fibrosis¹⁵. Ectopic expression of Pdgf-a, -b or -c in mouse liver all lead to fibrosis ^{6, 16, 17}, although the mechanisms of fibrogenesis appear to differ, and for reasons that are unclear, spontaneous HCC only develops with PDGF-C over-expression. Several recent reports suggest a specific role for PDGF-C in paracrine signaling in tumor stroma and angiogenesis.^18-20

In this report using the PDGF-C Tg model, we explore the molecular mechanisms by which the growth factor, PDGF-C, induces HCC. We show that PDGFRα and Rβ expression is specifically induced in perisinusoidal cells, co-localizes with hepatic stellate cell (HSC) ‘markers’, and is enriched in isolated HSCs, suggesting paracrine stimulation by hepatic expression of PDGF-C. We provide evidence that PDGF-C may indirectly induce HCC through stromal cell activation and changes in the liver microenvironment. We perform microarrays using RNA from young animals to analyze the global changes in hepatic gene expression that may contribute to carcinogenesis. PDGF-C induces chronic inflammation, liver injury, and fibrosis, causing changes in liver architecture and a reduced lifespan in Tg mice. In human liver samples, PDGF-CC protein levels correlate with worsening stages of NASH, suggesting a role for PDGF-C signaling in human chronic liver disease. Together these results add to a growing body of evidence that PDGF-CC activates tumor stromal cells, which in turn drive carcinogenesis.

Materials and Methods

Animals, necropsy, and plasma analyses

Specific pathogen-free male C57BL/6 PDGF-C Tg mice or WT littermates were used. Mice that express histone 2B-bound nuclear green fluorescent protein (GFP) driven by the endogenous PDGFRα promoter (PDGFRα-GFP) ²¹ were purchased from JAX, and intercrossed with PDGF-C Tg mice. The Institutional Animal Care and Use Committee of the University of Washington, which is certified by the Association for Assessment and Accreditation of Laboratory Animal Care International, approved all experiments. At the time of necropsy, gross examination of liver, spleen, lungs, kidney, heart and intestine was performed. Blood was obtained by cardiac puncture, and analyses of circulating factors performed as described in Supplemental Methods.

Mouse histology, immunohistochemistry (IHC), and immunofluorescence (IF)

For ex-vivo imaging, livers from PDGFRα-GFP mice were imaged within 30 min of necropsy as previously described ²². Briefly, livers were incubated with 1μM Hoechst dye (DNA) and 200nM MitoTracker Deep Red (M22426 Life Technologies), which stained hepatocytes, and imaged in a chambered cover slip using a Zeiss 510 Meta confocal microscope with appropriate excitation and emission filters. Images were collected with an optical slice thickness of 1μm and processed using NIH ImageJ software. Detailed procedures for IHC, IF, and other histological analyses are described in Supplemental Methods.

Liver perfusion, hepatocyte and HSC isolation, and RNA analyses

Mouse livers were perfused with collagenase, and primary hepatocytes and HSCs isolated using Percoll²³ and OptiPrep ²⁴ gradients, respectively. RNA was extracted using TRIzol (Invitrogen) from whole digested liver (containing all liver cell types), primary hepatocytes cultured for 48hrs, and HSCs cultured for seven days and relative gene expression was determined as described²³. Further details of RT-PCR methods including a list of all primers are in Supplemental Methods.

Global gene expression analysis

Liver RNA was isolated from 8.8-week-old PDGF-C Tg and WT littermates using TRIzol (Invitrogen) and re-purified using Qiagen RNeasy columns. RNA was used to probe Affymetrix microarrays. Details of array procedures and data analysis are described in Supplemental Methods; the complete dataset can be accessed with the GEO submission number GSE38199 (http://www.ncbi.nlm.nih.gov/geo/).

Analysis of human liver samples

Explanted livers with cirrhosis of different etiologies, and 48 additional specimens from NASH patients were obtained from an archival repository at the University of Washington Medical Center. Methods for in situ hybridization (ISH) and IHC are described in Supplemental Methods.

Statistical Analysis

All data are expressed as mean ± SEM. Statistical analyses were performed using non-parametric Mann-Whitney U-tests or Krusal-Wallis analyses with multiple comparison Dunn's posttest using Prism software (GraphPad Software Inc.). The number of animals analyzed is shown in figure legends. Differences between groups were considered to be statistically significant as follows: (*) p < 0.05, (**) p <0.01, and (***) p<0.001.

Results

Localization of PDGFRs in normal and fibrotic mouse livers

In PDGF-C Tg mice, the human PDGF-C gene is expressed in hepatocytes under the control of the albumin promoter⁶. PDGF-CC signals through PDGFRαα and αβ dimers; thus its actions are therefore dependent on the expression of PDGFRα. To determine the liver cell type(s) that respond to PDGF-CC, we used several experimental approaches We first performed ex vivo imaging ²² of livers from two- month old PDGFRα-GFP mice²¹. Cells with GFP+ nuclei were found throughout the lobule adjacent to hepatocytes (Figure 1A, green arrow), suggesting that non-parenchymal cells (NPCs) express PDGFRα. We did not appreciate any GFP expression in hepatocytes (white arrow). We confirmed that GFP expression recapitulated endogenous PDGFRα protein expression by performing IF for PDGFRα on fixed liver tissue from PDGFRα-GFP mice (Figure 1B). The cellular localization of PDGFRα-GFP was the same as PDGFRα detected by IF staining, except that antibodies against PDGFRα showed membrane proximal cytoplasmic staining as expected for the endogenous receptor, while the PDGFRα-GFP remains in the nucleus as designed.

Figure 1. PDGFRα is expressed in hepatic stellate cells.

(A) An ex-vivo image of fresh, unfixed liver from mice expressing nuclear GFP expressed by the endogenous PDGFRα promoter¹⁶ illustrates that PDGFRα is localized to non-parenchymal cells (NPCs). Nuclei of cells expressing GFP are green (green arrow). MitoTracker Deep Red was used to visualize hepatocytes (white arrow), while nuclei are visualized with Hoechst dye (blue, 400x). (B) Nuclear GFP and PDGFRα protein colocalize in the same cells. IF for PDGFRα in fixed liver tissue from PDGFRα-GFP reporter mice is shown with GFP seen as green nuclei, and PDGFRα protein (red) localized to the cell membrane of the same cells (600x). (C) IF on fixed fibrotic PDGF-C Tg liver shows PDGFRα–driven nuclear GFP co-localizes with PDGFRβ, and both receptors are expressed in cells that also express desmin and αSMA. Detection of PDGFRβ (blue), desmin or αSMA (red), immuno-epitopes with GFP are shown separately, with merged images in the last panel of each row (600x). (D) HSCs isolated from WT mice express Pdgfrα and Pdgfrβ mRNA. Total RNA isolated from whole liver, primary hepatocyte, and HSC fractions underwent RT-PCR for Pdgfrα and Pdgfrβ, along with markers for hepatocytes (albumin) and activated HSCs (Crbp1 and desmin). Data are normalized to Gapdh mRNA levels, and error bars represent the range of values from two cell cultures from different mice.

We next intercrossed PDGFRα GFP reporter mice with PDGF-C Tg mice to assess PDGFRα localization in fibrotic liver. Liver sections from PDGFRα-GFP;PDGF-C Tg mice were fixed and underwent IF staining with antibodies against α-smooth muscle actin (αSMA) and desmin, two HSC markers. As shown in Figure 1C, αSMA and desmin co-localized with PDGFRα-GFP, indicating that they are expressed in the same cells. IF staining with anti-PDGFRβ indicated that this receptor also co-localizes with PDGFRα-GFP, and HSC markers (Figure 1C).

We next isolated and separated cells from wild type livers and measured the relative levels of Pdgfr mRNA in whole liver, hepatocyte, and HSC enriched fractions (Figure 1D). Pdgfrα and Pdgfrβ expression were low in collagenase-digested liver and cultured primary hepatocytes, but high in cultured HSCs. Albumin mRNA was highest in whole liver and hepatocyte fractions, while desmin and cellular retinol binding protein 1 (Crbp-1) were extremely low in these cell fractions, but elevated in HSCs (Figure 1C), consistent with gene expression changes that occur when HSCs are activated in vitro and in vivo²⁵. Taken together these data indicate that PDGFRα and PDGFRβ are predominantly expressed in HSCs.

To determine the pattern of PDGFR expression in mice with worsening fibrosis, we performed IHC on liver sections from 4-month-old PDGF-C Tg and WT mice. In Tg mice, perisinusoidal staining for PDGFRα (Figure 2A, left panels) and PDGFRβ (right panels) can be appreciated, while WT mice had little expression of either receptor (data not shown). We reasoned that if PDGFRs are primarily expressed in activated HSCs, then HCCs should have relatively less expression than surrounding fibrotic liver. Total RNA was isolated from HCCs and surrounding livers of 12 month-old PDGF-C Tg mice, and expression of Pdgfrα and Pdgfrβ was compared to WT littermates (Figure 2B). Levels of both Pdgfrα and Pdgfrβ mRNA in HCCs were lower than in paired fibrotic livers, with Pdgfrα reaching statistical significance. Since primary liver tumors are comprised of heterogeneous cell types, these results are consistent with the hypothesis that a sub-population of cells within HCCs expresses PDGFRs, perhaps infiltrating stromal cells. If, on the other hand, transformed hepatocytes in these tumors aberrantly express PDGFRs, it is not widespread. Together with our previous observations that primary hepatocyte cultures and hepatocyte cell lines do not proliferate in response to PDGFs in vitro⁶, these data suggest that PDGF-CC acts directly on HSC, rather than hepatocytes, to stimulate receptor-mediated signaling pathways.

Figure 2. PDGFRα and PDGFRβ expression in the stroma of PDGF-C Tg livers.

(A) PDGFRα and PDGFRβ protein is localized to perisinusoidal areas in 4 month-old Pdgf-c Tg mice. Protein was detected in fixed tissue by IHC with antisera specific for each receptor (brown staining). Magnification: upper panels (100x) and lower panels (400x). (B) Pdgfr RNA levels were lower in HCCs than adjacent fibrotic liver in Tg mice. RNA was extracted from tumors (n= 6) and adjacent fibrotic liver from 12 month-old PDGF-C Tg mice (n=5), with livers from 12-month old WT mice as controls (n=3). Relative mRNA levels of Pdgfrα and Pdgfrβ from WT livers (clear bars) and Tg tumor (speckled bars) and non-tumor liver (black bars) are shown. p < 0.05 (*).

PDGF-C over-expression affects global gene expression in the liver

We used microarray analysis of gene expression in young mice to identify PDGF-C-driven pre-neoplastic changes in fibrotic liver. We found statistically significant changes in expression of several thousand genes. In our first analysis, we parsed these genes into two groups: 2,079 genes whose expression was significantly increased and 1,596 genes whose expression was significantly decreased in PDGF-C Tg vs. WT mice. Each group of genes was subjected to pathway analyses; examples of these analyses are shown in Figure 3A, with up-regulated genes shown on the right (black bars), and down-regulated genes shown on the left (gray bars). Pathways that were up-regulated, such as ubiquitin-mediated proteolysis, glycolysis, and purine metabolism, are all consistent with extensive tissue remodeling. Pathways that are down-regulated in PDGF-C Tg livers are those involved in normal liver function, such as lipid metabolism and hormone synthesis. Interestingly, several pathways associated with known tumor suppressors, e.g. p53 and mTOR, were also decreased in Tg mice.

Figure 3. Biological pathways affected in pre-neoplastic PDGF-C Tg fibrotic liver.

(A) DAVID analysis of Affymetrix array gene expression profiles from fibrotic livers from pre-neoplastic 8.8 week old PDGF-C Tg (n=7) and WT (n=9) mice as described in Supplemental Methods. Pathways with an EASE score of < 0.1 were selected from BioCarta, Panther and KEGG analyses and are shown as horizontal bars. Pathway analysis was performed on two separate gene sets: increased (black bars) or decreased (grey bars) in PDGF-C Tg mice relative to WT animals. The x-axis shows the number of genes in each pathway that changed significantly in PDGF-C Tg mice relative to WT mice. (B) A heat map is shown for the relative expression of genes in the chemokine biosynthetic processes pathway (Gene Ontology ID: GO42033) identified from microarray data using Gene Sorting Analysis (GSA). Red indicates increased expression, and green, decreased. WT mice are shown as grey number 1-9, and PDGF-C Tg mice are shown as black numbers 1-7.

We next subjected the selected gene lists to gene ontology for biological function (GOfatBP) analysis; the top 37 gene sets that were increased (Supplemental Table 1) and decreased (Supplemental Table 2) in PDGF-C Tg mice are presented. Biological functions that are increased include cell cycle, adhesion, and chromatin organization, while components of many metabolic pathways are decreased, suggesting that normal hepatic function is inhibited in fibrotic liver.

The large sample size used in our array allowed us to detect small changes in gene expression in the context of whole liver RNA, including those in minority stromal cell populations. Decreases in components of the retinol metabolism pathway and an increase in transforming growth factor β (TGFβ) signaling may reflect widespread activation of HSCs early in this model (Figure 3A, Supplemental Tables 1 and 2). In addition, PDGF-C induced changes in genes that are associated with HSC activation in vitro and in vivo²⁵, including Col1a1, TIMP-1, Plexin C1, Col4a1, and Pdgfrα, (Supplemental Figure 1A); we confirmed these latter microarray data with qPCR analyses (Supplemental Figure 1B).

Finally, we conducted Gene Sorting Analysis (GSA; Broad Institute) wherein all 30,000 genes on the arrays were analyzed to look for small, concordant changes within biologically relevant groups to identify altered cellular mechanisms. We found that several categories related to inflammation were significantly altered (p< 0.001), including cytokine production (GO01816), regulation of cytokine production (GO01817), and chemokine biosynthetic processes (GO42033); expression of specific genes is shown as a heat map in Figure 3B. We again corroborated these results with qPCR. Tumor necrosis factor (Tnfα) was significantly increased in PDGF-C Tg mice, while neutrophil elastase (Ela-2) and interleukin 1β (IL-1β) were increased but not significantly (Supplemental Figure 2). GSA analysis also revealed significant increases in pathways regulating protein kinase activity (GO45860), PDGFR signaling (GO48008), and wound healing (GO35313). The multitude of expression changes seen in pre-neoplastic livers suggest that PDGF-C overexpression induced dramatic changes in the liver that accompany fibrosis, and are consistent with alterations in the microenvironment prior to development of tumors. These results are also consistent with PDGF-C inducing an inflammatory response that causes cell injury and death, perhaps, in part by PDGF-C's ability to induce chemokines, such as CCL2 (also known as monocyte chemoattractant protein-1, MCP-1) ²⁶.

Changes in expression of hepatocyte growth and survival factors

We hypothesized that aberrant hepatocyte growth in PDGF-C Tg livers may be stimulated by factors produced by stromal cells based, on the lack of hepatocyte PDGFR expression in our model. We determined whether hepatocyte growth factor (HGF) and heparin-binding EGF-like growth factor (HB-EGF), which are expressed and secreted by stromal cells^{27, 28}, are induced in PDGF-C Tg mice. Hb-egf mRNA was significantly increased in the livers of PDGF-C Tg at six months of age compared to WT littermates (Figure 4A), while Hgf levels were increased at both 4 and 6 months of age (Figure 4B). Conversely, no significant increase in the hepatocyte autocrine EGFR ligand transforming growth factor-α (Tgfα) was seen (Figure 4C). In a separate experiment, we prepared RNA from enriched hepatocyte and NPC fractions isolated from collagenase-perfused WT livers, and found that Hgf RNA was enriched in the NPC fraction compared to whole liver or hepatocytes, while Hb-egf RNA was present in all cell fractions (data not shown). These results confirm that HGF is stromal derived as was previously published²⁷. We also examined the survival factors insulin-like growth factor 1 and 2 (Igf-1 and Igf-2) as IGF-2 specifically is frequently increased in patients with HCC²⁹. Igf-2 mRNA expression was significantly increased in PDGF-C Tg mice (Figure 4D); while there was no change in Igf1 levels (data not shown).

Figure 4. Elevated expression of hepatocyte growth factors and an increase in liver injury and inflammation in PDGF-C Tg mice.

(A-D) Relative levels of growth factor expression were determined using total RNA from 4 month old (4M) and six month old (6M) mice (n= 3 to 7) by RT-PCR for (A) Hb-egf (B) Hgf (C) Tgfα (D) Igf2. Relative expression levels were normalized to 18s rRNA levels. Data are presented as fold change relative to 4 month-old WT mice. Bars represent WT (clear), and PDGF-C Tg (black) mice, p < 0.05 (*), p < 0.01 (**). (E) Inflammation increases in PDGF-C Tg mice with age. Semi-quantitative assessment of liver inflammation from mice at the indicated ages was determined as described in Supplemental Methods; n=7 per age &genotype; p< 0.05 (*), p < 0.01 (**) as determined by using Kruskal-Wallis analysis with Dunn's multiple comparison test for PDGF-C Tg animals at different ages. (F) Serum levels of ALT, AST, and AP from 12 month old WT (n=4, clear bars) and PDGF-C Tg (n=7, black bars) mice.

Chronic liver inflammation and injury in older PDGF-C Tg mice

In addition to progressive fibrosis, we found evidence for chronic inflammation and liver injury in PDGF-C Tg mice. Focal inflammation was seen by histology (Supplemental Figure 3 A-D), which became more widespread as the mice aged. Using a semi-quantitative scoring system ³⁰, nine month-old PDGF-C Tg mice had significantly more inflammatory foci than six month-old mice (Figure 4E). We previously reported that serum levels of alanine aminotransferase (ALT), aspartate aminotransferase (AST), and alkaline phosphatase (AP) were not significantly elevated in young PDGF-C Tg mice ⁶, but when we re-assessed these markers of liver injury at 12 months of age, we found a consistent significant increase in Tg mice (Figure 4F). In addition, circulating IL-6 was detected in a subset of 12 month-old PDGF-C Tg mice, but not in WT littermates, also suggesting low-level liver injury (Supplemental Figure 3E).

Evidence of apoptosis in multiple cell types is apparent in 1.5-month-old PDGF-C Tg livers as detected by IHC for activated caspase 3 (Supplemental Figure 3F), and the presence of councilman bodies in eight-month-old mice (Supplemental Figure 3B, black arrow), suggesting a low level of apoptosis or other forms of cell death in this model. Together, these data suggest that liver inflammation, and injury and cell death are ongoing in PDGF-C Tg mice, similar to humans with chronic hepatitis who are predisposed to HCC.

A plausible hypothesis is thus that HCC arises due to mutations caused by oxidative DNA damage that accumulates as hepatocytes proliferate in an inflammatory microenvironment². We used the random mutation capture (RMC) assay³¹ to measure the frequency of point mutations in nine-month-old PDGF-C Tg mice, however we did not observe an increase in the mutation frequency compared to WT mice [mutation frequencies: 7.6 +/− 1.6 × 10⁻⁷ (n=8, PDGF-C Tg); 5.0 +/− 1.3 × 10⁻⁷ (n=8, WT) mutation/base pair]. The similar mutation frequencies detected by RMC analysis indicate that PDGF-C-induced inflammation, injury, and fibrosis per se did not increase point mutations, but we cannot rule out changes in gross chromosomal rearrangements or large deletions/insertions using this method.

Vascular changes and splenomegaly in PDGF-C Tg mice

Hepatic expression of PDGF-C induces extensive vascular changes, as evidenced by increased CD34 staining in PDGF-C Tg livers and the development of venous malformations^{6, 32}. Given the juxtaposition of HSCs and sinusoidal endothelial cells (SECs) and the activation of HSCs in PDGF-C Tg mice, we used scanning electron microscopy to determine whether defenestration of SECs (i.e. capillarization) was evident. We discovered congested sinusoids with extensive extracellular matrix deposition (ECM) and an apparent lack of fenestrae, suggesting that capillarization had occurred (Supplemental Figure 4A-B). Images of a sinusoid from a WT mouse liver are shown for comparison (Supplemental Figure 4 C-D.) These dramatic changes in sinusoidal architecture suggest that PDGF-C induces capillarization, which is associated with hepatocellular neoplastic transformation in other systems³³.

All PDGF-C Tg mice have splenomegaly (Supplemental Figure 5), and many have grossly dilated mesenteric veins (data not shown). These anatomical changes may result from portal hypertension, as seen in patients with cirrhosis, resulting from sinusoidal congestion³⁴. It is also possible that splenomegaly is a direct consequence of circulating PDGF-CC⁶, but since this is the only significant extra hepatic phenotype in these mice, we suggest that it is caused by portal hypertension, similar to another mouse model³⁵.

As early as 10 months of age, gross venous malformations are visible on the surface of PDGF-C Tg livers⁶, indicative of neoangiogenesis. An increase in circulating levels of vascular endothelial growth factor (VEGF) is seen in 12 month old PDGF-C Tg mice (Supplemental Figure 6), and we previously showed that CD34-postive cells are proliferating in Tg mice³². Taken together these data indicate that liver sinusoidal endothelial cells are altered as a consequence of PDGF-C expression and/or fibrosis³³, and support the notion that neoangiogenesis may contribute to carcinogenesis in this model.

Decreased lifespan and tumor incidence in PDGF-C Tg mice

PDGF-C Tg mice have a decreased life span, with a median survival of 16 months compared to 24 months for WT littermates (Figure 5). The cause of death in PDGF-C Tg mice is usually hemoperitoneum, with extensive blood and fluid found in the peritoneal cavity during necropsy. All Tg mice had liver tumors at the time of death, and since hemoperitoneum is a common cause of death in patients with HCC³⁶, it appears that premature death in these mice is related to their chronic liver disease and HCC.

Figure 5. PDGF-C Tg mice die earlier than WT littermates.

Median survival for PDGF-C Tg mice is 16 months, compared to 24 months for WT mice (p <0.0001) as determined by Kaplan-Meier analysis for WT (n= 10, clear circles) and PDGF-C Tg (n= 13, black squares) mice. Note that some WT mice were euthanized at 26 months of age, so the median survival of WT mice may be under-estimated.

To delineate of the precise progression from dysplasia to HCC in PDGF-C Tg mice, we determined the incidence of dysplastic foci and HCCs with age. Dysplastic foci were observed in 44% of the livers at 4-5 months, 68% at 8 to 9 months old, and 100% at 12 months. No HCCs were detected at 4 months, while 26% had macroscopic tumors by 9 months, and 85% by 12 months⁶. None of the WT littermates had dysplasia or HCCs at any age.

PDGF-C expression in human cirrhosis and NASH

To determine whether expression of PDGF-C is elevated in human liver disease, we analyzed PDGFC mRNA in patient samples by in situ hybridization (Figure 6A-I). PDGFC mRNA was detected in cells within cirrhotic bands from patients with hepatitis C, alcoholic, and NASH- induced cirrhosis (Figure 6A, D and G). These cells also expressed ACTA2 (αSMA) mRNA (Figure 6B, E, and H). ISH with negative control probes are shown for the same patients (Figure 6C, F, I). We further explored the relationship between PDGF-C expression and severity of liver disease in biopsies from a cohort of 48 patients with NASH. We detected cleaved, activated PDGF-CC in all NASH livers by IHC (examples shown in Figure 6K-M), but not in normal liver (Figure 6J). We then used a semi-quantitative scoring system to determine whether PDGF-CC levels increased with increasing severity of NASH and/or pattern of αSMA expression³⁷. Spearman's correlation analysis revealed a positive correlation between PDGF-CC protein level and more severe disease stage, though there was no correlation between αSMA and PDGF-CC (Supplemental Table 3). These results suggest PDGF-C is present in human liver disease of different etiologies, and that it specifically correlates with disease severity in NASH. These data are consistent with published reports of elevated PDGF levels in human cirrhotic liver and HCC ^{12-14, 38, 39} and confirm the translational validity of the PDGF-C Tg mouse model.

Figure 6. PDGF-C expression in human liver disease (A-I) Detection of PDGFC RNA by ISH.

Both PDGF-C and ACTA2 (αSMA) mRNAs (red signal) are concentrated in the cirrhotic bands in biopsies from patients with hepatitis C- (A-C), NASH- (D-F) and alcohol-induced (G-I) cirrhosis. Hepatocytes did not express either marker. The left panels show localization of the PDGF-C probe (red signal; A, D, G); the center panels show localization of the ACTA2 probe (red signal; B, E, H) and the right panels are images with a negative control probe (C, F, I). All images were taken at 20x magnification. (J-M) Detection of cleaved, active PDGF-CC protein by IHC. PDGF-CC staining correlates with stage of NASH. Images are: histologically normal liver (J; 100x), stage 1 NASH (K; 200x), stage 3 NASH (L; 200x), and stage 3 NASH (M; 400x). Black arrows indicate location of PDGF-CC staining.

Discussion

The importance of the tumor microenvironment in the development of cancer has recently become more fully appreciated. Stromal cell sub-populations undergo dynamic changes, creating a microenvironment that leads to genetic and epigenetic changes in tumor cells⁹. In a study by Hosihida et al, gene expression signatures from liver specimens adjacent to HCCs were more predictive of recurrence than the signatures of HCCs themselves⁴⁰. These results illustrate the importance of the liver microenvironment, and suggest that changes throughout the liver may be permissive for carcinogenesis, consistent with the observation that cirrhosis is the primary risk factor for HCC. PDGF-C Tg mice represent a relevant pre-clinical model to investigate microenvironmental factors that predispose epithelial cells to transformation. We demonstrate that this model recapitulates many of the pathophysiological features of human chronic liver disease, including progressive fibrosis⁶, chronic inflammation and injury (Figure 4), sinusoidal congestion, and splenomegaly consistent with portal hypertension (Supplemental Figures 4 and 5). PDGF-C Tg mice have a shorter life span than WT littermates (Figure 5), likely due to hemoperitoneum, an often fatal consequence of HCC rupture in humans ³⁶.

To explore how PDGF-C overexpression results in HCC, we identified the liver cells that express PDGFRα and thus respond directly to PDGF-CC. A key finding of this study is elevated expression of PDGFRα in activated HSCs and/or myofibroblasts (Figure 1). In the PDGF-C Tg model, the human transgene is expressed and secreted by hepatocytes and directly activates HSCs, resulting in induction of collagen, ECM deposition, and cell proliferation. Extensive activation of HSCs throughout the liver is accompanied by increased production of cytokines and inflammatory infiltrates that precede elevation of transaminases, dysplasia, and HCCs (Figures 3B and 4E). The precise nature of the mechanisms by which PDGF-C induces inflammation in this model is under investigation. Similar to kidney fibrosis, PDGF-C may induce cytokine production in activated hepatic stellate cells^{26, 41}. Once cell death is present, macrophages, stellate cells and infiltrating monocytes will phagocytose cell debris eliciting more pro-inflammatory and fibrotic cytokines^{42, 43, 46}. In addition, PDGF-C induces proliferation of SECs³³ and thus remodeling of liver sinusoids (Supplemental Figure 4), indicating that significant architectural changes and NPC proliferation precede the onset of dysplasia in this model.

Stromal cell-derived HGF and HB-EGF^{27, 28} are potent hepatocyte mitogens and are frequently elevated in HCCs ^{44, 45}. Elevated hepatic expression of Hgf and Hb-egf in PDGF-C Tg mice at the age when dysplastic foci are present (Figure 4A-B) suggests that these growth factors may drive tumorigenesis. Our array data revealed that expression of two other EGFR ligands, Egf and Epiregulin, were also significantly increased in PDGF-C Tg mice (data not shown). We also measured the expression of c-Met, the receptor for HGF using RNA extracted from tumors and adjacent liver dissected from 12 month old PDGF-C Tg mice. We observed a high level of c-Met expression in PDGF-C Tg HCCs as well as surrounding liver compared to WT tissue (data not shown). As c-Met is expressed on hepatocytes, similar to EGF receptors, and both respond to ligands derived in the stroma that are elevated in this model, these growth promoting pathways⁴⁵ may contribute to abnormal hepatocyte growth in PDGF-C Tg mice.

We suggest that the presence of these factors, along with alterations in the ECM, chronic inflammation, and capillarization, together lead to aberrant hepatocyte growth and dysplasia. PDGF-CC also likely contributes to neoangiogenesis in this model; likewise PDGF-C has been shown to function in tumor angiogenesis²⁰. Thus PDGF-C overexpression induces a myriad of changes in the liver microenvironment, which ultimately lead to the development of HCC.

Normal liver has very little hepatocyte turnover, but PDGF-C over-expression creates an environment that promotes hepatocyte growth, as evidenced by the hepatomegaly and baseline hepatocyte proliferation in Tg mice⁶. In addition to growth promoting signals, carcinogenesis requires mutational events within hepatocytes to allow abnormal growth in the face of homeostatic mechanisms that preserve organ architecture. When we measured point mutation frequencies in WT and PDGF-C Tg littermates, surprisingly we found no difference. We recently demonstrated that hepatic mutation frequency increases with age in C57BL/6 mice, such that the frequency is 50-fold higher in nine-month old compared to five week-old mice ³¹. We thus speculate that the age-induced increase in mutation frequency, coupled with chronic inflammation and fibrosis induced by PDGF-C, may promote abnormal hepatocyte growth and thus cancer.

To complement our analyses of murine fibrosis and HCC, we explored endogenous PDGF-CC expression in human cirrhosis. Previous studies demonstrated that PDGFs and PDGFRs are induced in cirrhosis compared to normal liver^{12, 13}. We found that PDGF-C mRNA was present in cirrhosis from differing etiologies (Figure 6A), and PDGF-CC protein levels correlate with disease severity in NASH (Figure 6B and Supplemental Table 3). It is interesting to note that in steatotic liver, cleaved, active PDGF-CC appears to associate with NPCs rather than hepatocytes. Further experiments are needed to determine the nature of PDGF-CC binding in human liver diseases. These observations, together with previous reports that platelets and Kupffer cells secrete PDGFs and TGFβ in inflamed liver⁴⁶ suggest that PDGF-C may be an important driver of the development of HCC in patients with NASH.

Evidence for induction of HCC through activation of stromal cells

In PDGF-C Tg mice, PDGFRα is expressed in HSCs/myofibroblasts (Figures 1 and 2), perisinusoidal cells that also express αSMA, desmin and PDGFRβ. In addition, PDGFR expression was lower in tumor than in surrounding tissue, at least at the RNA level (Figure 2B). The most parsimonious interpretation of these data is that PDGF-C's effects on hepatocyte growth and transformation are paracrine or indirect. We suggest that PDGF-C is critical to the development of tumor-promoting stroma, and facilitates paracrine signaling between hepatocytes and various stromal cell populations^7,8. Precedent for a mesenchymal growth factor promoting epithelial cancer by paracrine mechanisms can be found in other models. For example, PDGF-C has angiogenic and paracrine functions that have been attributed to cancer associated fibroblasts or tumor cells themselves^18-20. Although PDGFR alleles are not commonly mutated in HCCs⁴⁴, changes in their expression have been investigated in chronic liver disease and HCC. Bedossa's group found that PDGFRA is elevated in cirrhosis but decreased in HCC, albeit with large sample-to-sample variation³⁹. Additionally, Llovet and co-workers observed PDGFRA expression was decreased in early HCCs, compared to dysplastic lesions ³⁸. In contrast, Stock et al. observed an increase in PDGFR mRNA and protein in HCCs¹⁴. Together with data from hepatoma cell lines ^{14, 15, 47, 48}, these reports indicate that aberrant PDGFR expression occurs in epithelial liver tumor cells. We did not find evidence for PDGFRα protein in hepatocytes or widespread RNA expression in HCC tumors in PDGF-C Tg mice, but it is possible that oval cell/progenitor cells emerge, or a small sub-clonal population of transformed hepatocytes acquires mutations that lead to aberrant expression of PDGFRα under selective microenvironmental pressure. Critical future experiments to test our proposed paracrine model will require the generation of PDGF-C Tg mice with HSC-, oval-, and hepatocyte-specific deletion of PDGFRα. For example, it would be intriguing to intercross PDGF-C Tg mice with a recently described strain of mice with hepatocyte-specific deletion of PDGFRα ⁴⁹ and determine whether loss of PDGFRα in hepatocytes would affect the incidence of HCC.

Utility of the PDGF-C Tg mouse as a stromal model for cancer

Development of new therapeutic approaches that prevent or reverse liver inflammation and fibrosis is critical to the prevention of primary and recurrent HCC, and relies on preclinical models that recapitulate human disease. Currently, immunocompetent in vivo models wherein stromal-tumor interaction can be investigated are limited. Novel therapies that target the stroma are thus often tested in xenograft or allograft models in immunodeficient or ‘humanized” mice⁵⁰. Here we show that ectopic expression of PDGF-CC induces chronic hepatitis and liver fibrosis, stimulating long-term injury and HCC, as is seen in humans. To this end, we treated PDGF-C Tg mice with the tyrosine kinase inhibitor imatinib, which targets PDGFRs, and found that it blocked proliferation of HSCs and SECs but not hepatocytes, and decreased markers of angiogenesis³². In another study, treatment of PDGF-C Tg mice with the acyclic retinoid, peretinoin, decreased PDGF signaling, fibrosis, and the incidence of HCC⁴⁸. PDGF-C Tg mice are thus of great utility in testing the ability of novel drugs to reverse fibrosis, block angiogenesis, or prevent cancer, as well as to elucidate their mechanisms of action in vivo. Moreover, PDGF-C Tg mice provide a new means of exploring the interaction between stroma and tumor cells, and advance our understanding of how chronic inflammation leads to cancer in the liver and other organs.