Since the initial description of β-catenin mutations in HCCs in 1998 , Wnt signaling became a center of interest for these tumors. A large set of Wnt ligands and a large array of receptors are implicated in different cell processes by initiating canonical, but also noncanonical Wnt signals . The antagonism between canonical and noncanonical Wnt pathways has also been reported [37, 38, 46]. Thus, both β-catenin and Wnt signaling are involved in highly complex cellular events of which only some are mediated by canonical Wnt pathway. This complexity is also observed during liver development and as well as in adult liver homeostatic events. Canonical Wnt signaling contributes to liver growth and regeneration, but also to liver "zonation" by controlling some liver-specific metabolic programs. (). In addition, it contributes to the activation of liver stem or progenitor cells, as well as HCC-initiating cells [48–51].
Mutational activation of canonical Wnt signaling is not a frequent event in HCC, in contrast to hepatoblastoma displaying very high rates . Mutations of β-catenin were restricted to a group of HCCs associated with low p53 mutation rate, negative HBV status and chromosomal stability, as stated earlier. These mutations were also associated with lower histological grade and better patient survival. Unexpectedly, β-catenin mutations are rare in more advanced and poorly differentiated HCCs [20, 23]. Therefore, although considered to play an active role in HCC malignancy, the activation of canonical Wnt signaling may not be necessary for, or even repressed in advanced HCCs. Thus, β-catenin mutation and constitutive activation of canonical Wnt signaling may be differentiation-dependent events with mechanistic implications in HCC initiation and progression. We attempted to address this issue by using HCC-derived cell lines.
We first classified 11 HCC cell lines into "well-differentiated" and "poorly differentiated" subtypes using hepatocyte lineage, epithelial and mesenchymal cell markers, and in vitro migration assays. Well-differentiated HCC cell lines shared many features with hepatocytes such as expression of HNF-1α, HNF-4α, and E-cadherin, and epithelial morphology. Poorly differentiated cell lines were usually deficient in the expression of hepatocyte lineage and epithelial markers, but they expressed different mesenchymal markers strongly. These two types of HCC cell lines were also distinguished from each other by their in vitro behaviors. Poorly differentiated cell lines were usually more motile and more invasive than well-differentiated cell lines (Table 1). A global expression profiling study classified HCC cell lines in Group I and Group II . Our well-differentiated and poorly differentiated cell line subtypes showed perfect correlation with Group I and Group II, respectively. Well-differentiated Group I was characterized by the activation of oncofetal promoters leading to increased expression of AFP and IGF-II, whereas poorly differentiated Group II was characterized by overexpression of genes involved in metastasis and invasion. Our well-differentiated and poorly differentiated subtypes were also in perfect correlation with respectively epithelial and mesenchymal HCC cell line types that have been identified very recently . Mesenchymal cancer cells are considered as the products of EMT that is believed to be a key mechanism for the acquisition of invasive and metastatic capabilities by tumor cells . Higher motility of poorly differentiated HCC cell lines reported here is in line with this concept. Thus, our two classes of cell lines share many similarities with well-differentiated and poorly differentiated HCC tumors. We used this model to compare the status of Wnt pathway according to HCC differentiation status.
A comprehensive analysis of Wnt signaling components in liver or hepatocytes is lacking. However, the expression of Wnt ligands and Frizzled receptors in mouse hepatocytes has been published . Mouse hepatocytes expressed canonical receptors Fzd7 and Fzd9, as well as noncanonical Fzd2, Fzd3, Fzd4 and Fzd6. We observed highly similar pattern of expression in HCC cell lines with the exception of Fzd9 that showed weak expression. In addition, we detected increased expression of canonical Fzd1 and Fzd5 in most HCC cell lines. The expression frequency of these receptors was not associated with HCC cell differentiation status. Mouse hepatocytes expressed canonical Wnt1 and Wnt2, and noncanonical Wnt4, Wnt5a, Wnt5b and Wnt11. All or most HCC cell lines have lost the expression of canonical Wnt1 and Wnt2, but they displayed increased expression of canonical Wnt3 and Wnt10b ligands. Another canonical ligand, Wnt8b was expressed selectively in well-differentiated cell lines. In contrast, noncanonical Wnt4, Wnt5a and Wnt5b ligands were expressed in the majority of poorly differentiated cell lines, but not in most of well-differentiated cell lines. In addition, most HCC cell lines (poorly differentiated cell lines in particular) also displayed increased expression of noncanonical Wnt7b. Among Wnt ligands and Frizzled receptors that we found to be expressed or upregulated in HCC cell lines, Wnt3, Wnt4, Wnt5a, Fzd3, Fzd6 and Fzd7 have been previously reported to be overexpressed also in primary HCC tumors [55–57]. Overexpression of Wnt10b was also reported in HCC cels . Increased levels of Wnt5a transcripts were detected in chronic hepatitis, cirrhosis and HCC . A C-terminally mutated HBV X protein was shown to upregulate Wnt5a expression in HCC cells . Thus, Wnt5a upregulation observed in clinical samples might be related to HBV at least in HBV-related liver diseases. Based on our observations that associate noncanonical Wnt ligand expression to poorly differentiated HCC cell lines, it will be interesting to test the predictive value of noncanonical Wnt expression for HCC prognosis.
Another important finding of this study is the differential activity of canonical Wnt signaling in different HCC subtypes. Well-differentiated cell lines displayed active canonical Wnt signaling at variable degrees. In addition to strong signaling activity associated to β-catenin and Axin1 mutations in two well differentiated cell lines, we also observed autocrine canonical Wnt signaling in two other well differentiated cell lines, as reported for some other cancer cell lines . The functional significance of autocrine canonical Wnt signaling in these cell lines is not known.
However, small molecule antagonists of Tcf4/beta-catenin complex were shown to inhibit TCF reporter activity and down-regulate the endogenous Tcf4/β-catenin target genes c-Myc, cyclin D1, and survivin in Huh7 cells . This observation strongly suggests that the autocrine canonical Wnt signaling is functional in well-differentiated HCC cell lines. Canonical Wnt signaling has been linked to both stem cell and cancer cell self-renewal in other cancer types. It was proposed that some adult cancers derive from stem/progenitor cells and that canonical Wnt signaling in stem and progenitor cells can be subverted in cancer cells to allow malignant proliferation . Indeed, well-differentiated-HCC cell lines identified here such as HepG2, Huh7 and PLC/PRF/5 have been reported to harbor HCC stem cells 
Our third noteworthy observation was the lack of detectable canonical Wnt signaling activity in six out of seven poorly differentiated cell lines. Even a poorly differentiated cell line with a deleterious Axin1 mutation (SNU475) lacked detectable signaling activity. Thus, most probably, the canonical Wnt signaling was not only inactive, but also repressed in poorly differentiated HCC cell lines. In confirmation of this expectation, transient or Tet-regulated expression of mutant β-catenin failed to generate significant canonical Wnt signaling activity in two different poorly differentiated cell lines. Furthermore, we linked this weak activity to poor nuclear accumulation of β-catenin protein in SNU449.cl8 cell line. Thus, unlike well-differentiated cell lines, poorly differentiated HCC cells displayed strong resistance to canonical Wnt signal activation.
The mechanisms of resistance to canonical Wnt signal activation in poorly differentiated HCC cells are presently unknown. We provide here one potential mechanism. Wnt5a has been previously implicated in canonical Wnt signaling as an antagonist and regulator of β-catenin levels in other cell types [48, 49]. Using both ectopic and endogenous mutant β-catenin expression systems in two different cell lines, we demonstrated that co-transfections with Wnt5a-expressing plasmid can significantly inhibit canonical Wnt signaling in HCC cells. The mechanism of Wnt5a antagonism on canonical Wnt signaling in HCC cells is not known. In breast cancer cells, the loss of Wnt5a signaling resulted in stabilization of nuclear beta-catenin and expression of Wnt/beta-catenin target genes . However, both ectopically and endogenously expressed mutant β-catenins used in our experiments were N-terminally truncated devoid of their Ser/Thr phosphorylation motifs. Thus, Wnt5a appears to inhibit canonical Wnt signaling in HCC cells, downstream to β-catenin, independent of its glycogen synthase 3-β- and bTrCO-dependent degradation. Wnt5a has been shown to inhibit canonical Wnt signaling either by bTrCP-independent proteasomal degradation , or, by downregulating β-catenin-induced reporter gene expression without influencing β-catenin levels  in kidney epithelial cells. Wnt5a may use similar mechanisms in Huh7 and HepG2 cells. Further studies with downregulation of noncanonical Wnt ligands in poorly differentiated HCC cell lines may help to better define the implications of such ligands in liver cancer biology.
The role of Wnt5a in cancer is complex. It may play tumor-promoting or tumor-suppressing functions depending on cellular context. Wnt5a has been described as a tumor promoter in melanoma, gastric, pancreas, prostate cancer, but as a tumor suppressor in HCC, neuroblastoma, leukemia, colon, and thyroid cancers . The inability of Wnt5a to transform cells or signal through the canonical β-catenin pathway pointed that it cannot promote tumorigenesis by upregulation of canonical Wnt signaling, unlike canonical Wnt ligands . Our results suggest that Wnt5a, upregulated in poorly differentiated highly motile mesenchymal-like HCC cells may play a role in tumor progression by inducing EMT. Upregulation of Wnt5a expression during EMT has been reported . Furthermore, the Wnt5A/Protein kinase C pathway was shown to mediate motility in melanoma cells via of an EMT . Similarly, CUTL1-upregulated Wnt5a significantly enhanced migration, proliferation and invasiveness in pancreas cancer cellls . These effects were accompanied by a marked modulation of marker genes associated with EMT. Wnt5a may promote EMT in HCC cells by a similar mechanism.
The expression status of Wnt5a in HCC is not well known. To our knowledge, only one report addressed this issue . Compared to normal tissue, Wnt5a mRNA expression was strongly induced in HCC, as well as in chronic hepatitis and cirrhosis. However, immunostaining of Wnt5a protein showed a bell-shaped pattern: low to undetectable levels were present in normal tissue and in tumor samples, whereas strong immunostaining was seen in chronic hepatitis, cirrhosis and dysplastic liver cells. The reasons of the discrepancy between transcript and protein expression in HCC tissues are not known presently. However, it appears that peritumoral liver tissues express high levels of Wnt5a protein that could trigger noncanonical Wnt signaling in adjacent tumor cells. It will be important to further investigate the role of Wnt5a in HCC tumor progression.
Taken together, our studies demonstrate that canonical Wnt activity is active in well-differentiated, but repressed in poorly differentiated HCC cell lines. This correlates with in vivo tumor studies indicating that β-catenin mutations are prevelant in well-differentiated, but not in poorly differentiated tumors. In addition, we showed that poorly differentiated cell lines express noncanonical Wnt ligands such as Wnt5a acting as an antagonist of canonical Wnt signaling. Thus, it appears that HCC cells may activate or repress their canonical Wnt signaling, using autocrine/paracrine systems based on selective use of canonical and noncanonical Wnt ligands.
We hypothesize that the active canonical Wnt signaling observed in well-differentiated HCC cells contributes to tumor initiation, but not necessarily to tumor progression. Instead, noncanonical Wnt signaling may be used by poorly differentiated HCC tumors to promote cell motility and invasion. Selective use of canonical and noncanonical Wnt signaling at different stages may be a key mechanism involved in hepatocellular carcinogenesis. Recent studies showed that canonical Wnt signaling contributes to the self-renewal and expansion of HCC-initiating cells with stem/progenitor cell features [50, 70]. However, the lack of HCC development in β-catenin transgenic mice strongly suggests that canonical Wnt signaling activation has limited tumorigenic potential in liver tissue. Indeed, recent studies showed that canonical Wnt signaling plays a major role in the specification of mature hepatocytes for perivenous-specific gene expression ([71, 72]. Such a hepatocyte differentiation function of canonical Wnt signaling may not be compatible with cellular dedifferentiation that goes along with HCC development. Therefore, alternative pathways such as Wnt5a-mediated noncanonical Wnt signaling may be necessary for sustained growth and progression of HCC tumors. Melanoma may serve as a demonstrated model to our hypothesis. Similar to HCC, canonical Wnt signaling activation is an early event and nuclear β-catenin accumulation is associated with better patient survival in melanoma. Nuclear β-catenin is lost in more aggressive melanomas that express Wnt5a that promotes EMT, cell motility and metastasis. Chien et al.  have recently demonstrated that canonical Wnt signaling induces growth inhibition and differentiation in melanoma cells, whereas Wnt5a can antagonize some of these effects. These findings clearly establish a dual function of Wnt signaling in melanoma. In light of these recent developments, our findings call for further investigations on respective roles of canonical and noncanonical Wnt signaling in HCC.