FGF19 and CCND1 are both overexpressed in HCCs harboring the 11q13.3 amplicon

The 11q13.3 amplicon containing CCND1 is one of the most frequent amplification events in human tumors and is well characterized; thus, it was surprising to find another tumor-promoting gene (FGF19) in the same region. FGF19 lies within 45 kb of CCND1 and the two genes are invariably co-amplified in the samples we analyzed, leading to an increase in expression of both genes (Figure 2A). FGF4 and FGF3 are also frequently co-amplified with CCND1, though they are further away than FGF19 (120 and 155 kb, respectively). However, CCND1 and FGF19 are often amplified in the absence of co-amplification with these two other FGF genes (Figure 2A). Furthermore, we found that although amplification results in significant increases in CCND1 and FGF19 expression in HCC, amplification of FGF4 and FGF3 does not (Figure 2A-B). This lack of correlation of amplification and overexpression for these latter two FGF genes was previously noted in breast cancer (Fantl et al., 1990).

Open in a separate window

Figure 2

Epicenter mapping and expression of genes in the 11q13.3 amplicon in HCC and the difference in the effect of amplification on FGF19 and CCND1 expression between breast and liver tumors

(A) Individual boundaries and the region of common overlap for the 14 11q13.3 amplicons, along with the underlying RefSeq genes in the depicted 1.5 Mb region, are displayed. The genes are color-coded (see inserted scale) to indicate the degree of correlation between DNA copy number and gene expression. Correlation coefficients between DNA copy number and expression are for FGF3 (r= −0.20, p=0.36) and FGF4 (r= 0.17, p=0.45), statistically insignificant in HCC. (B) Scatter plots with associated correlation coefficients showing the relationship in HCC samples (both tumors and cell lines) between DNA copy number and expression for CCND1 (left) and FGF19 (right). (C) As in (B) but with breast cancer cell line samples. See also Figure S3.

Curiously, the only report in the literature regarding the effect of amplification of FGF19 on its expression is with oral cancer, where it has been reported to not be overexpressed (Huang et al., 2006), similar to FGF4 and FGF3. Because FGF19 was discovered after most analyses of the 11q13.3 amplicon in breast cancer were conducted, there are no reports in the literature regarding the effect of amplification on its expression in this tumor type. We have found that, similar to oral cancer, FGF19 is not overexpressed when amplified in breast cancer (Figure 2C and Figure S3), nor does FGF19 appear to be overexpressed when amplified in lung cancer or melanoma (Figure S3). Thus, amplification does not invariably cause overexpression of FGF19; rather, overexpression appears restricted to a specific tissue type. We have previously reported this same phenomenon for the amplified gene TTF1, which is overexpressed when amplified in lung adenocarcinomas but not in lung squamous carcinomas (Kendall et al., 2007).

ORAOV1, which is located between FGF19 and CCND1 (Figure 2A), is overexpressed in all amplified tumors that have been tested, including in our HCC dataset. However, our screening showed that it does not promote tumorigenicity in p53^−/−;Myc hepatoblasts, nor does it show any cooperativity with FGF19 or CCND1 (data not shown).

Effects of FGF19 and CCND1 overexpression on hepatocellular tumorigenicity

We wanted to confirm that both FGF19 and CCND1 could induce tumorigenicity using an orthotopic transplantation assay. When hepatocytes overexpressing either FGF19 or CCND1 were transplanted into the liver of mice, tumors developed within 8 weeks (Figure 3A). Microscopic examination of the resultant in situ liver tumors classified them as aggressive solid hepatocellular carcinomas. The tumors were composed of a population of undifferentiated cells growing as a sheet without any histological evidence for gland formation or any other structure. The cells were large with a more basophilic-staining cytoplasm compared to normal liver and resembled human HCC. We established that the tumors arose from the transfected hepatoblasts because the carcinoma cells were positive for the GFP marker. In addition, cellular proliferative status was examined by immunohistochemical staining for PCNA. The tumors formed by either FGF19 or CCND1-expressing hepatoblasts were clearly positive for PCNA as well, indicating that the tumors induced by these genes were significantly proliferative. Similar morphological and molecular changes were observed in orthotopic tumors induced by MET, POFUT1, or HCK (Figure S4).

Open in a separate window

Figure 3

FGF19 and CCND1 cooperate to promote liver carcinoma formation

(A) Images of mouse livers and liver sections taken eight weeks following transplantation of p53^−/−;Myc hepatoblasts expressing either empty vector, CCND1, or FGF19. The five panel columns are, from left to right, intact livers; fluorescent imaging of intact liver for GFP-positive transplanted cells; hematoxylin and eosin staining of liver tissue sections showing the border between normal liver and carcinoma (arrows); immunohistochemical detection of GFP; and immunohistochemical detection of PCNA. The last three are from the same tissue block. Size bars = 100μm. (B) Kaplan-Meier plot showing the percentage of mouse survival at various times post-transplantation. The livers of mice were transplanted with p53^−/−;Myc hepatoblasts infected with either control vectors, FGF19 alone, CCND1 alone, or both genes in combination. (C) Subcutaneous growth of p53^−/−;Myc hepatoblasts infected with either control vector pMSCVpuro, control vector pMSCVhygro, FGF19 alone, CCND1 alone, or FGF19 with CCND1 (n=10 injections, asterisks indicate that the indicated tumor group is significantly different than controls, error bars denote ± SD, *p<0.05, **p<0.0005). Tumor volumes were determined on 28 (red columns), 35 (green columns), and 42 (blue columns) days after injection. See also Figure S4.

Next we explored for possible cooperative effects when both FGF19 and CCND1 genes were co-expressed in murine hepatoblasts. For the transplantation assays, we measured the survival of mice transplanted with p53−/−;Myc hepatoblasts ectopically expressing the two genes alone and also in combination. None of the control mice transplanted with hepatoblasts transfected with empty vectors died within the 100-day observation period, but 100% of the FGF19 alone, CCND1 alone, and CCND1 plus FGF19 groups did eventually succumb to tumors (Figure 3B). According to the log rank test for significance, the survival of the FGF19 or CCND1 alone and CCND1 plus FGF19 groups was significantly less than control (p<0.05). There was an increase in morbidity when comparing the CCND1 plus FGF19 group to the FGF19 or CCND1 alone groups (Figure 3B). This difference was clearly significant when compared to the FGF19 group alone (p=0.003), but the difference compared with the CCND1 alone group did not pass the p=0.05 cutoff normally used for significance, although the p value obtained indicates only a 15% chance that the null hypothesis was correct (p=0.15). In the subcutaneous tumorigenicity assay, which uses tumor volume as a readout therefore providing a greater range of quantitative values than survival, the combination of both CCND1 and FGF19 was very clearly significantly greater than either gene alone (p<0.0005, Figure 3C). Taken together, these results suggest that the combination of overexpressing CCND1 and FGF19 is more tumorigenic than when either single gene is overexpressed.

FGF19 requires β–catenin to mediate cyclin D1 protein levels

Since many growth factors are known to regulate cyclin D1 protein production, we wanted to determine whether FGF19 levels in turn regulated cyclin D1 levels in human HCC cells. Towards this end, we tested and validated two shRNAs targeting FGF19 and two shRNAs targeting CCND1 that were each effective at reducing target protein levels (Figure S5). We found that RNAi-mediated silencing of FGF19 in the HCC cell line Huh-7, which harbors the 11q13.3 amplicon and overexpresses both FGF19 and CCND1 (Figure S5), caused what appeared to be complete suppression of FGF19 protein as well as almost complete elimination of cyclin D1 protein (Figure 4A). Furthermore, we found that silencing the expression of either FGF19 or CCND1 significantly inhibited clonogenic growth of Huh-7 cells (Figure 4B). To control for off-target effects of the shRNAs, we performed RNAi rescue experiments. We found that the addition of recombinant FGF19 protein to the culture medium restored high levels of cyclin D1 protein and rescued the growth-defect caused by shRNA knock-down of FGF19, but that it could not do either to cells with the shRNA knock-down of CCND1 (Figure 4B and Figure S5). On the other hand, overexpression of an RNAi-insensitive CCND1 construct restored high levels of cyclin D1 protein and completely rescued the growth defects of both FGF19 and CCND1 shRNA knockdowns (Figure 4B and Figure S5). These results show that the shRNA effects were not off-target, and they also establish a clear hierarchy of oncogene dependence, in that FGF19 functions upstream of cyclin D1 in human HCC cells.

Open in a separate window

Figure 4

FGF19 and CCND1 functionally interact through β-catenin signaling

(A) FGF19 and cyclin D1 protein expression in Huh-7 (11q13.3-amplified) cells following stable transfection with one shRNA targeting luciferase (control) and two independent shRNAs targeting FGF19 (19K4 and 19K5). (B) Quantification of clonogenicity of Huh-7 cells infected with shRNAs against FGF19 (19K4) and CCND1 (D1K2) are shown relative to results obtained with a shRNA against luciferase (control). Recombinant FGF19 protein was added to the medium, or a shRNA-insensitive CCND1 expression construct was transfected into the cells, where indicated (error bars denote ± SD, *p<0.005). (C) Active β-catenin levels in Huh-7 cells expressing the two shRNAs targeting FGF19, as revealed by immunoblotting with an antibody specific for the activated form of β-catenin, relative to total β–catenin levels. (D) TCF reporter activity, relative to constitutively-expressing renilla luciferase activity, in Huh-7 cells infected with shRNAs against FGF19 (19K4) or β-catenin (BcatK1), compared to a shRNA against luciferase (control) (error bars denote ± SD, *p<0.05). (E) Quantification of clonogenicity of Huh-7 cells infected with shRNAs against CTNNB1 (BcatK1 and BcatK4) are shown relative to a non-targeting shRNA (control) (error bars denote ± SD, *p<0.05). (F) Time-course effects of adding FGF19 to the medium of SNU423 cells (with a single copy of 11q13.3) on active β-catenin levels, as well as the effects of added FGF19 on cyclin D1 protein levels, as detected by immunoblotting. (G) TCF reporter activity, relative to constitutively-expressing renilla luciferase activity, in SNU423 cells treated for 24 hours with recombinant FGF19, compared to non-treated cells (error bars denote ± SD, *p<0.05). (H) Quantification of clonogenicity in SNU423 cells infected with an effective shRNA against CTNNB1 (BcatK1), compared to cells infected with a non-targeting shRNA (error bars denote ± SD, p=0.169). (I) SNU423 cells were treated with FGF19, EGF or FGF2. β–catenin and MAPK activity were determined by immunoblotting after 15 minutes, while cyclin D1 protein was detected after 24 hours exposure. (J) SNU423 cells infected with either a non-targeting shRNA (control) or an effective shRNA targeting β–catenin (BcatK1) were treated with FGF19, EGF or FGF2 for 24 hours and then cyclin D1 protein levels were detected by immunoblotting. See also Figure S5.

We aimed to identify a potential mechanism through which FGF19 is regulating cyclin D1 levels. Recently, one of us (D.F.) showed that in colon cancer cell lines, expression of FGF19 activates β-catenin signaling whereas its inhibition reduces β-catenin signaling (Pai et al., 2008). β-catenin has been proposed to activate CCND1 transcription (Tetsu and McCormick, 1999) although it is not always an immediate transcriptional target (Sansom et al., 2005) and it can also influence cyclin D1 levels by stabilization of CCND1 mRNA (Briata et al., 2003). This led us to predict that FGF19 may be signaling through β-catenin to regulate cyclin D1 protein levels. To test this, we determined if the knockdown of FGF19 would have an effect on β–catenin activation. Levels of activated β-catenin protein were analyzed by immunoblotting using an antibody directed against NH₂-terminally dephosphorylated β-catenin. We found that both shRNAs targeting FGF19 caused a clear reduction of β-catenin activation (Figure 4C). We wanted to determine if this held true if we used a dual-luciferase TCF reporter as a readout for β-catenin activity. We found that in the 11q13.3-amplified Huh-7 cell line, TCF reporter activity was reduced by 25% when FGF19 was knocked down, and that a shRNA targeting β–catenin (CTNNB1) reduced activity by 55% (Figure 4D).

We then used shRNAs targeting CTNNB1 to determine if reducing β–catenin would have an effect on cell growth of the 11q13.3-amplified HCC cell line Huh-7. Two shRNAs targeting CTNNB1 significantly reduced the clonogenic growth potential of Huh-7, as compared to cells expressing a non-targeting shRNA (Figure 4E & Figure S5). This effect was not seen in the non-amplified HCC cell line SNU423 (Figure 4H). This further supports our model that β-catenin signaling is critical in FGF19-amplified HCC cell lines.

Conversely, we added FGF19 to the culture medium of a HCC cell line (SNU423) that has the normal copy number of both FGF19 and CCND1 and that does not express detectable levels of FGF19 protein by immunoblotting (Figure S6). These cells were incubated in medium supplemented with FGF19 and β-catenin activity was analyzed both by immunoblotting for NH₂-terminally dephosphorylated β-catenin and by measuring TCF reporter activity. As determined by immunoblotting, the addition of FGF19 induced activation of β-catenin within 10 minutes, and baseline levels returned within 24 hours (Figure 4F). Correspondingly, TCF reporter activity was increased 2.1-fold in the presence of recombinant FGF19 (Figure 4G). We found that cyclin D1 protein levels were subsequently elevated after 24 hours of exposure to exogenously added FGF19 (Figure 4F), supporting a mechanism by which FGF19 induces elevated cyclin D1 through β–catenin signaling.

Our proposed mechanism for how FGF19 induces higher levels of cyclin D1 protein differs from how other mitogens have been shown to increase cyclin D1 protein in fibroblasts, a mechanism that requires MAP kinase activation (Lavoie et al., 1996). To test if this was also true in human HCC cells, we treated serum-starved SNU423 cells with either FGF19, FGF2 (basic FGF) or EGF for 15 minutes and then analyzed β-catenin and MAPK1/2 activation. We also treated the cells for 24 hours to measure cyclin D1 protein levels. We found that all three growth factors were able to induce elevation of cyclin D1 protein, however, only FGF19 activated β-catenin while only EGF and FGF2 activated MAPK1/2 (Figure 4I). We also determined, using an effective shRNA against CTNNB1, that β-catenin function was selectively required by FGF19 to induce cyclin D1 but that this was not true for EGF or FGF2 (Figure 4J). We conclude that there are two distinct pathways in HCC cells through which mitogens induce elevation of cyclin D1 protein, the well-established pathway involving RAS/RAF/MAPK signaling, and the β-catenin pathway.

CCND1 and FGF19 oncogene dependency in human HCC cell lines

We wanted to test whether amplification of CCND1 and FGF19 in human HCC cell lines led to dependence on their continued expression, and if so whether such oncogene dependence would hold true in HCC cell lines that were not amplified for 11q13.3. Towards this end we used the previously described shRNAs targeting FGF19 and CCND1 to test oncogene dependence in a panel of six HCC cell lines: three harboring amplification of 11q13.3, and three that were single-copy for this locus. We introduced these shRNAs into each of the six cell lines and tested their effects on growth using a clonal growth assay. Strikingly, the clonogenic growth potential of each of the three CCND1/FGF19-amplified cell lines was significantly reduced by silencing of either FGF19 or CCND1, whereas none of the CCND1/FGF19 single-copy cell lines were significantly affected (Figure 5A-B). These results establish a clear link between genotype (CCND1/FGF19 copy number status) and oncogene dependence.

Open in a separate window

Figure 5

CCND1 and FGF19 oncogene dependency in human HCC cell lines

(A) Clonogenicity assay of Huh-7 cells (11q13.3-amplified) and SNU182 cells (single-copy for 11q13.3) infected with shRNAs against luciferase (control), FGF19 (19K4 and 19K5), and CCND1 (D1K2 and D1K4). (B) Quantification of clonogenicity in six cell lines (three with 11q13.3 amplification and three without) infected with shRNAs against FGF19 (19K4 and 19K5) and CCND1 (D1K2 and D1K4) relative to a shRNA against luciferase (control). For each of the six cell lines, the results are displayed in this order (from left to right): cells infected with control shRNA (blue column), 19K4 and 19K5 shRNAs (green columns), and D1K2 and D1K4 shRNAs (yellow columns) (error bars denote ± SD, *p<0.001). (C) Subcutaneous tumor growth in nude mice of Huh-7 cells infected with indicated shRNAs (n=12 injections, error bars denote ± SD, *p<0.005). (D) As in (C) but with JHH-7 cells (n=10 injections, error bars denote ± SD, *p<0.01, **p<0.0001). (E) Subcutaneous growth of established tumors from Huh-7 cells treated either with PBS, control antibody, or anti-FGF19 antibody (1A6). Treatment was on the days marked with red asterisks. Dashed lines indicate that mice were terminated before the end of the study (n=20 injections, error bars denote ± SEM, **p<0.05). (F) Growth inhibition of HCC cell lines grown in vitro with anti-FGF19 antibody (1A6) relative to the indicated 11q13.3 amplification status. Error bars denote ± SEM. The bracket above the four amplified cell lines indicates that by Student s t-test the average growth inhibition by the anti-FGF19 antibody was significantly greater than that of the non-amplified control group. See also Figure S6.

We wanted to determine if the selective inhibition of the 11q13.3-amplified tumor cells by shRNAs targeting either FGF19 or CCND1 was reflected in correspondingly different levels of expression of the gene products in untreated cells. We found that FGF19 protein could be detected in all three 11q13.3-amplified HCC cell lines but none of the non-amplified cell lines (Figure S6), an expected result based on the notion that gene amplification drives increased mRNA and protein expression. However, cyclin D1 protein levels did not vary significantly between the two groups (Figure S6), which could be explained by the high levels of mitogens found in the cell culture conditions keeping cyclin D1 levels high regardless of amplification status. However, this result indicates that the selective dependence of the 11q13.3-amplified cells for cyclin D1 is not due to addiction to higher levels of cyclin D1 protein but rather to selective dependence on cyclin D1 downstream effector functions.

We then tested whether elevated expression of these two genes in the CCND1/FGF19-amplified cell line Huh-7 played a significant role in vivo. The shRNAs silencing either FGF19 or CCND1 significantly slowed the growth of Huh-7 cells transplanted subcutaneously into nude mice (p<0.005, Figure 5C). Significant inhibition of tumor growth by shRNAs targeting FGF19 or CCND1 was also observed with CCND1/FGF19-amplified JHH-7 cells (p<0.0001, Figure 5D). These results establish key tumor maintenance functions for both FGF19 and CCND1 specifically in hepatocellular carcinomas harboring the 11q13.3 amplicon, but not in those without.

Despite the fact that most cancers contain several oncogenetic alterations affecting multiple genes, oncogene dependence has almost always been evaluated for only a single oncogene, although it has been shown that in some circumstances inhibition of multiple altered oncogenes can be beneficial (Podsypanina et al., 2008). This led us to test whether shRNA-mediated silencing of both driver genes in the 11q13.3 amplicon would be more effective than silencing of FGF19 or CCND1 alone. To test this, we co-expressed effective shRNAs against each gene in the 11q13.3-amplified JHH-7 line. By measuring clonogenic growth, we found that the dual shRNA knockdown was no more effective in suppressing growth than shRNA knockdown of either gene alone (Figure S6). Nor did the dual shRNA knockdown show more effective inhibition of tumor development (Figure S6). We believe this result supports our model that the tumor-promoting effects of FGF19 are mediated by its ability to increase cyclin D1 protein levels, and that because single shRNAs targeting either FGF19 or CCND1 can effectively lower cyclin D1 protein levels (Figure 4A and Figure S5), additional lowering of cyclin D1 protein levels has no growth-inhibitory effect. We do not believe however that this negative result should be extrapolated to other situations where driver genes may operate in different pathways.

Oncogenomic selection of genes/cDNAs from focal amplicons and construction of the amplicon-focused and randomly selected cDNA libraries

We selected high-level amplicons (segmented value ≥ 1.5) that were ≤ 20 Mb in size and applied an algorithm similar to the MCR (Minimal Common Region) method to determine the region of common overlap (Tonon et al., 2005). This analysis resulted in 29 commonly amplified regions, ranging in size from 230 kb to 10 Mb, with a total of 812 RefSeq genes (Table S1). Starting with genes from the smallest amplicon, we obtained from Open Biosystems, a distributor of plasmids from the Mammalian Gene Collection (MGC) (Gerhard et al., 2004), all available (as of June 2007) human or murine cDNA expression plasmids in the pCMV-SPORT6 vector until we had reached our target screen size of 150 cDNAs.

We thank L. Bianco, J. Marchica, T. Wang, A. Bakleh and Q. Liu for technical support, J. Duffy for his help preparing figures, and A. Ashkenazi for discussions. This work was supported by the Hope Funds for Cancer Research (E.T.S.) and NIH grants CA124648 (S.P.), CA105388 (S.P., S.W.L.) and CA076905 (R.S.F.). S.W.L. is a Howard Hughes Investigator. R.S.F. is a recipient of an NIH-LRP award. E.T.S., S.P. and S.W.L. are members of the CSHL Cancer Target Discovery and Development Center (CTD², supported by CA148532) and part of the NCI CTD² Network.