Oncogenicity of the Developmental Transcription Factor Sox9

SOX9 [sex-determining region Y (SRY)-box 9 protein] is a member of the SOX family of transcription factors. These are developmental regulators that possess high mobility group (HMG) box DNA-binding domains as well as other conserved features, such as transactivation domains (1). However, their action on target genes is context dependent, relying on other transcription factors with which they may directly interact for specificity (1). SOX proteins coordinate disparate functions, such as maintaining stem cell properties, lineage restriction, and terminal differentiation, through precise temporal and spatial expression patterns that differ between particular cell types and tissues (2). In particular, SOX9 plays a pivotal role in a number of developmental processes and its levels need to be strictly controlled for normal embryogenesis. In humans, SOX9 heterozygous mutations result in campomelic dysplasia, a syndrome characterized by severe skeletal malformations, defects in the central nervous system and several other organs, frequent XY female sex reversal, and perinatal lethality (3). In mice, Sox9 homozygous null mutant embryos die around embryonic day 12 (E12) due to heart defects, whereas heterozygotes die around birth with phenotypes similar to human patients with campomelic dysplasia, although without sex reversal (4). In addition, duplications of the Sox9 gene or its deliberate misexpression have been linked with XX male sex reversal and fibrosis-related disorders (5, 6) and show that dysregulation of the gene can cause disease. Further analyses have shown that SOX9 is crucial for Sertoli cell differentiation, chondrogenesis, neural crest development, and differentiation of some of its derivatives, as well as, for the development of specific cell types and lineages within the central nervous system, pancreas, prostate intestine, skin, pituitary, heart, kidney, and sensory systems (7).

Colorectal cancer is the third leading cause of cancer-related deaths worldwide and its incidence is steadily increasing (8). Colorectal carcinomas derive from the intestinal epithelium, which is constantly self-renewing and requires a high rate of cell division to maintain epithelial homeostasis (9). It is becoming increasingly clear that mutations in developmentally regulated genes can cause the initiation and progression of these cancers. Particularly, genes constituting the Wnt signaling pathway are crucial for the maintenance of undifferentiated progenitors in the crypts and for the maintenance of the postmitotic Paneth cells. Wnt activating mutations are often found in colorectal cancers, frequently targeting the tumor suppressors APC or Axin2 or the oncogene β-catenin (10). Members of other important developmental programs, such as Notch and bone morphogenesis protein (BMP), are also deregulated in colorectal cancers (11).

Despite the importance of SOX9 in multiple developmental programs, its role in cancer remains unclear. On the one hand, SOX9 is often overexpressed in cancers of the skin, prostate, lung, and brain (12–15). On the other hand, SOX9 may behave as a tumor suppressor, at least in some melanomas (16). This ambiguity also concerns colorectal cancers where SOX9 has been found to be overexpressed (17), but at the same time, it has been reported to reduce the tumorigenicity of HT-29Cl.16E colorectal cancer cells (18). A number of observations suggest that Sox9 could be a critical player in the homeostasis of colorectal epithelium. Sox9 is localized in proliferating crypt cells, promotes stem/progenitor cell proliferation, and is required for Paneth cell differentiation in the intestinal epithelium during development and in adult stages (19–21). Also, Sox9 is a downstream effector and a regulator of the Wnt pathway in the regulation of intestinal epithelium homeostasis (22). Beyond the intestinal epithelium, Sox9 is also a target of BMP, Notch, or Hedgehog in different processes such as chondrogenesis or regulation of stem cells from several tissues (7). SOX9 is activated in bladder injury repair and constitutively upregulated in bladder cancer (23). Given that activated stem cells are required for injury repair and that chronic injury increases the risk of forming cancer, SOX9 might contribute to carcinogenesis through effects on stem cells. Furthermore, Sox9 promotes survival and provides competence for epithelial-to-mesenchymal transition in the neural crest (24) and contributes to cancer cell invasion and migration in urothelial carcinomas (23). In this work, we directly test the role of SOX9 in cancer and reveal new cellular and molecular mechanisms that reinforce and explain its involvement in tumor biology, and particularly in colorectal cancer.

Mice were housed in the pathogen-free barrier area of the MRC National Institute for Medical Research (London, UK) and of the Spanish National Cancer Research Center (CNIO), Madrid, Spain. The pCAGGS-ERT2iCre line was generated as follows: the chicken β-actin enhancer/cytomegalovirus (CMV) promoter fragment of the plasmid pCAGGS was released with SalI/XhoI digestion and inserted into the SalI site upstream of the PBS.ERT2iCrepA plasmid, generating the expression plasmid CAGGSET. The plasmid was sequenced using an on-site sequencing service. The transgene fragment was released with SalI/NotI digestion and was purified for pronuclear injection into fertilized eggs of B10/CBAF1 mice. Sox9^Flox/flox and Z/Sox9 mice were maintained on a mixed C57BL6/129 genetic background. Isolation, culture, and assays with mouse embryo fibroblasts (MEF) were carried out as previously described (25). For detailed methods, see Supplementary Data.

Human cancer cell lines were obtained from the American Type Culture Collection and cultured under recommended conditions in Dulbecco's Modified Eagle's Media supplemented with 10% FBS at 37°C and 5% CO₂. The cell lines were tested for mycoplasma contamination, and the levels of expression of oncogenes and tumor suppressors for the various lines correlated with the American Type Culture Collection profiles of these cells. For SOX9 or BMI1 knockdown by short hairpin RNA (shRNA), cells were transfected using Lipofectamine (Invitrogen), according to instructions of the reagent supplier. Cells were cultivated in the presence of puromycin for 3 weeks. Cells were transfected using 2 different SOX9 shRNAs (OriGene; sh1 or sh75 and sh2 or sh73) and shRNA specific for BMI1. A nonspecific shRNA (pRS) was used as a control.

Total RNA was extracted with TRIzol (Life Technologies). Reverse transcription was carried out using random priming and Superscript Reverse Transcriptase (Life Technologies), according to the manufacturer's guidelines. Quantitative real-time PCR was carried out using Absolute SYBR Green mix (Thermo Scientific) in an ABI PRISM 7500 thermocycler (Applied Biosystems). Variations in input RNA were corrected by subtracting the number of PCR cycles obtained for β-actin. For information on primers, see Supplementary Data.

Immunoblots were done following standard procedures. Typically, equal amounts of protein (30 μg) were separated on 15% SDS-PAGE and blotted onto nitrocellulose membranes (BioRad). For detection of human and mouse SOX9, we used AB5535 (Chemicom) and sc-20095 (Santa Cruz); for detection of mouse Ink4a (p16^Ink4a), we used M156 antibody (Santa Cruz) and for human Ink4a DCS-50.1 (Abcam); for detection of Arf (p19^Arf), we used R562 antibody (Abcam); for p53, we used CM5 (Novocastra); for Bmi1, we used Clone F6 antibody (Millipore); and for β-actin, we used mouse monoclonal antibody AC-15 (Sigma). For secondary antibodies, we used horseradish peroxidase (HRP)–linked anti-rabbit or HRP–linked anti-mouse (DAKO), both at a 1:2,000 dilution. Detection was conducted by chemiluminescence using enhanced chemiluminescence (ECL; Amersham).

For immunofluorescence, cells were fixed with 4% paraformaldehyde for 15 minutes and washed with PBS supplemented with 0.2% Triton X-100 and 1% fetal calf serum, for 5 minutes at 4°C. Subsequent to blocking for 1 hour with PBS and 1% fetal calf serum, cells were incubated with P-H3 (Chemicom) antibody for 2 hours. Secondary antibodies were from Jackson ImmunoResearch. Nuclear DNA was stained with a 4% PBS-buffered paraformaldehyde solution containing 10 μg/mL 4′6-diamidino-2-phenylindole (DAPI; Sigma).

Chromatin immunoprecipitation (ChIP) assays were conducted according to standard procedures (see Supplementary Information). DNA fragments were immunoprecipitated with antibodies against Sox9 (Chemicon), H3K4me3, H3K9me3, and H3K27me3 (Millipore). The immunoprecipitated DNA was extracted and subjected to PCR amplification with primers directed against the proximal promoter of Bmi1 or Sox9, respectively.

Murine tissues were dissected, fixed in 10% buffered formalin (Sigma), and embedded in paraffin. Three-micrometer thick sections were stained with hematoxylin and eosin (H&E). Antibodies used for immunohistochemical analysis included rabbit polyclonal anti-Sox9 (1:200 dilution; Millipore); a DAKO automated system was used for the immunostaining after antigen retrieval in a PTLink (DAKO).

Human tissue microarrays (TMA) were obtained from the CNIO Tumor Bank Unit. Human colorectal carcinoma samples and data from patients were provided by the Basque Biobank for Research-OEHUN (http://www.biobancovasco.org) and 2 independent transcriptome profiling data sets previously described (26, 27). The raw data are accessible through the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) public repository for microarray data (accession numbers GSE25071 and GSE24550). Samples were processed following standard operation procedures with appropriate ethical approval.

The array-comparative genomic hybridization (CGH) data were recently generated (28). Here, the NimbleGen Human Whole-Genome Array CGH Analysis v1 (Roche Diagnostics) was applied, which provides measurements from 385,000 unique genomic loci.

For 3-methylcholanthrene–induced carcinogenesis, 3- to 6-month-old mice received a single intramuscular injection, in one rear leg, of 40 μL of 3-methylcholanthrene (Sigma) dissolved in sesame oil (Sigma) at a concentration of 25 μg/μL. Mice were observed on a daily basis until tumors reached 12 mm in diameter. For nude mice, neoplastically transformed MEFs or human cells were harvested with trypsin/EDTA and resuspended in PBS. Cells (1 × 10⁶) were injected subcutaneously into both flanks of nude mice (8-week-old). Mice were observed on a daily basis for up to 40 days, and external calipers were used to measure tumor diameters at the indicated time points from which tumor volume was estimated.

Data are presented as mean values ± SEM with the number of experiments (n) in parentheses. Unless otherwise indicated, statistical significance (P values) was calculated using the Student t test. Asterisks (*, **, and ***) indicate statistical significance (P < 0.05, P < 0.01, and P < 0.001, respectively).

We first analyzed SOX9 levels by immunohistochemistry on a multitumor TMA containing human cancers derived from a diverse range of tissues. Compared with normal control tissue, SOX9 staining showed more intense and widespread staining in all cancer types, with the exception of lymphomas. According to frequencies of SOX9-positive tumors, cancers can be broadly classified as highly positive (brain, pancreatic, colorectal, lung, and prostate cancers), medium (sarcomas, thyroid, and breast cancers), and low (kidney cancers; Supplementary Table S1). The staining was generally observed in the nucleus with varying intensities among cells, and similar results were obtained using 2 different primary antibodies against SOX9. To obtain more compelling evidence, we decided to analyze tumor TMA of specific tissues. More than 70% of lung, ovary, colon, prostate, pancreatic, neurofibroma, malignant peripheral nerve sheath tumor (MPNST), and medulloblastomas displayed moderate to robust expression of SOX9 (Supplementary Table S2 and Fig. S1). This may reflect the concept that cancer cells exhibit some features corresponding to the stem/progenitor cells of their tissues of origin and it is notable that SOX9 has been associated with the stem cell compartments of most of the corresponding normal tissues (see Introduction). Quantification of the number of cancers containing more than 25% of SOX9-positive cells using an automated scanning microscope and computerized image analysis system (Ariol SL-50; Genetix) generated essentially identical results (Supplementary Table S2 and Fig. S2). Together, these data show that SOX9 is frequently expressed in many types of human cancers, and particularly in those derived from tissues where SOX9 plays a role in their development and homeostasis.

To further characterize SOX9 expression in human cancer, we decided to focus our attention in colorectal cancers. Colorectal cancer arises through an ordered sequence of genetic and epigenetic events in the adenoma–carcinoma development. It typically begins with the formation of noninvasive benign adenomatous polyps, which ultimately leads to the development of invasive adenocarcinomas with metastatic capacity (29). Our studies showed significantly higher levels of SOX9 in both adenomas and adenocarcinomas than in normal colon tissue (P = 0.0001; Fig. 1A and B; Supplementary Table S3). In particular, we observed SOX9 overexpression in 18 of 24 (75%) adenomas and 91 of 110 (83%) adenocarcinomas. The frequency of SOX9 overexpression was equivalent at protein and mRNA levels reflecting increased transcription. Analysis of SOX9 mRNA levels in colorectal adenoma and adenocarcinoma samples revealed that SOX9 was upregulated by an average 2- and 3.5-fold in colon adenomas and adenocarcinomas, respectively, compared with nontumor colonic mucosa (Fig. 1C). Furthermore, when we analyzed SOX9 levels in relation to tumor stage, we found that higher levels of SOX9 mRNA were associated with advanced tumor stage (Fig. 1D). High levels of SOX9 were found both in adenocarcinoma of the colon and rectum (Supplementary Fig. S3). From high resolution array-CGH data, we found that some adenocarcinoma samples (7%) had low copy number gain at the distal part of chromosome 17, including the SOX9 gene (Fig. 1E, Supplementary Fig. S4). The gain is observed at low level and in few samples, implying that copy number change is a plausible but not exclusive mechanism for the increase of SOX9 expression in colorectal adenocarcinoma. Consistent with the tissue data, SOX9 was generally highly expressed in human colorectal cancer cell lines, with levels ranging from 1.5 to 30 times higher than in normal colonic mucosa (Fig. 1F and G, Supplementary Fig. S3). In particular, SOX9 expression was remarkably higher in SW620 than in SW480 (Fig. 1F and G), 2 cell lines established from lymph node metastasis and its primary colon adenocarcinoma, respectively, from the same patient, further supporting the role of SOX9 in tumor progression and malignancy in colorectal cancers.

Finally, we examined the expression of Sox9 in colorectal cancers in mice. Immunohistochemistry revealed strong expression of Sox9 in colitis-associated colon tumors obtained from mice treated with azoxymethane/dextran (Fig. 1H). In summary, these findings confirm that SOX9 is expressed at a high level in human and murine colorectal cancers and cancer-derived cell lines and that SOX9 expression correlates with tumor progression in this type of cancer. Our findings support the use of SOX9 as a potential tumor and prognostic marker in colorectal cancer.

Enhanced proliferation, immortalization, and neoplastic transformation are important stages in the conversion of a cell with normal growth into one that is malignant. Thus, regulators of these cellular phenomena play a major role in tumorigenesis. To test whether SOX9 may be involved in these processes, we first analyzed the biologic consequences of altering Sox9 expression levels in primary MEFs derived from 2 lines of transgenic mice that provide gain and loss of Sox9 function, respectively. In particular, we used mice carrying a conditional Sox9-IRES-EGFP transgene (Z/Sox9) in which Cre recombinase–mediated excision of a stop cassette results in the stable expression of Sox9 and enhanced GFP (EGFP; ref. 30). MEFs were established from embryos hemizygous for a β-actin Cre transgene alone (referred to here as wt) or hemizygous for both Z/Sox9 and β-actin Cre (referred to as Z/sox9tg) and we confirmed elevated levels of Sox9 protein and mRNA (Fig. 2A and B). Of note, the level of overexpression of Sox9 in the transgenic MEFs (30-fold higher) is within the range of overexpression observed in human colorectal cancer cell lines (Fig. 1F), suggesting that the levels of overexpression achieved with the transgene are of biologic relevance. To study the effect of Sox9 gene inactivation, we established MEFs from embryos homozygous for a conditional null allele of Sox9 (Sox9^flox/flox) that also carried a broadly expressed Cre-ERT2 transgene (pCAGGS-ERT2iCre; ref. 31). Cre-mediated excision of Sox9 after treatment with 4-hydroxy tamoxifen (4-OHT) significantly diminished the basal levels of Sox9 present in normal MEFs (Fig. 2C and D). However, Sox9 expression was not completely lost reflecting inefficiency in the Cre-mediated deletion. Expression of other Sox gene family members (Sox2, Sox8, and Sox10) was not affected by gain or loss of Sox9 indicating no cross-talk or compensatory changes (Supplementary Fig. S5). Having established these cellular models of gain and loss of Sox9, we analyzed their proliferation. Z/sox9tg cells exhibited a higher rate of proliferation than wt cells (Fig. 2E), whereas Sox9^Δ/Δ MEFs grew slower than their corresponding controls (Fig. 2F). Next, we examined whether Sox9 regulates cell-cycle progression by fluorescence-activated cell-sorting analysis. The numbers of Z/sox9tg cells in S-phase were higher and those in G₁ lower than in wt cells (Fig. 2G). Conversely, the number of Sox9^Δ/Δ MEFs in S-phase was reduced compared with controls (Fig. 2G). Together, our results support the idea that Sox9 regulates proliferation in primary cells controlling the progression of the cell cycle from G₁ to S-phase.

Having established that SOX9 promotes proliferation in MEFs, we moved to a more relevant cellular setting by analyzing human colorectal cancer cells. We focused on HCT116 and SW620 because they express very high levels of SOX9 (Fig. 1G). Sublines were derived after stable introduction of 2 independent SOX9 shRNAs. We obtained similar results but we present the protein and mRNA data obtained with the most efficient shRNA (sh1; Fig. 2H and I). Downregulation of SOX9 levels resulted in a reduction in proliferation as measured by cell growth curves and phospho-histone H3 (p-H3) immunostaining (Fig. 2J and K). No significant differences were observed in apoptosis in HCT116 cells (data not shown). These observations suggest that SOX9 contributes to the proliferation of human colorectal cancer cells.

Next, we analyzed the effect of Sox9 in cell senescence. This is a fundamental mechanism triggered in primary cells after a finite number of cell divisions or in response to oncogenic stress that limits cellular life span and constitutes a barrier against cellular immortalization in vitro and in vivo (32). In contrast, cancer cells are essentially immortal if cultured under optimal conditions. Z/sox9tg and wt MEFs were cultured using a 3T3 serial passage protocol designed to study senescence. Strikingly, Z/sox9tg MEFs bypassed senescence and continued proliferating, whereas control MEFs entered senescence at around passage 3 to 4 (Fig. 3A). Consistent with this result, wt cells seemed enlarged and with flat morphology, typical of senescent cells, meanwhile Z/sox9tg exhibited normal fibroblast morphology (Fig. 3B). Reduced proliferation and subsequent senescence of primary cells is partly due to the progressive accumulation of Ink4A and Arf tumor suppressors (p16^Ink4a and p14^ARF/p19^Arf; p14 when referred to the human protein and p19 when referred to the murine protein) encoded in the Cdkn2a locus (33). This prompted us to ask whether Sox9 may contribute to proliferation and senescence modulating Ink4a and Arf expression. Importantly, early-passage Z/sox9tg MEFs displayed decreased basal levels of Ink4a and Arf, whereas the opposite was true for early-passage Sox9^Δ/Δ MEFs (Fig. 3C, Supplementary Fig. S6). It is well established that Arf stabilizes p53, another major regulator of senescence (33). In accordance with Arf expression, Z/sox9tg MEFs contained lower p53 basal expression (Fig. 3C). In addition, the expression of Ink4b and DcR2, established markers of senescence (34), was inversely correlated with Sox9 levels (Supplementary Fig. S6).

Next, we evaluated the consequences of altering Sox9 in response to oncogene-induced senescence. High levels of oncogenic H-RasV12 result in the activation of a protective response that limits proliferation and enhances resistance to neoplastic transformation in primary MEFs (35). In contrast to wt MEFs, Sox9-overexpressing cells were able to proliferate and generate colonies even in the presence of H-RasV12 (Fig. 3D). The Ink4a/Arf locus is one of the most important defensive mechanisms against oncogenic stress. In particular, H-RasV12 upregulates both Ink4a and Arf (35). We hypothesized that Sox9 could modulate their expression in response to this oncogene. The levels of Ink4a and Arf were lower in Z/sox9tg cells in the presence of H-RasV12 than in wt cells, although H-RasV12 was still capable of upregulating Ink4a and Arf in Z/sox9tg cells (Fig. 3E, Supplementary Fig. S6). Conversely, Sox9^Δ/Δ MEFs showed higher Ink4a and Arf levels in response to H-RasV12 (Fig. 3E, Supplementary Fig. S4). Moreover, the expression of other established markers of senescence, such as p53, Ink4b or DcR2, inversely correlated with Sox9 in cells transduced with H-RasV12 (Fig. 3E, Supplementary Fig. S6). To study the contribution of the members of the Ink4a/Arf locus underlying the proliferation and senescence phenotypes seen in cells with altered levels of Sox9 expression, we compared Ink4a/Arf-null MEFs that had been retrovirally transduced with a plasmid expressing Sox9 or the empty vector. We found no difference in their rates of proliferation (Supplementary Fig. S7), revealing that the role of Sox9 in proliferation and senescence is dependent, at least in part, on Ink4a and Arf. In summary, our results suggest that Sox9 is a novel regulator of senescence in vitro, and its activation is sufficient to escape both serial passage and oncogene-induced senescence, partly through modulating the expression of Ink4a and Arf tumor suppressors.

It is well known that senescence occurs in vivo as a barrier against tumor progression. Indeed, tumor senescent cells are prevalent in a variety of premalignant tumors whereas progression to malignancy requires evading senescence. This paradigm has been well characterized in the K-RasG12V conditional mouse model, where low-grade premalignant lung adenomas are strongly positive for senescence markers and high-grade invasive malignant adenocarcinomas are essentially negative (34). In this context, immunohistochemical analysis of Sox9 expression revealed remarkable high levels of expression exclusively in lung adenocarcinomas, whereas normal tissue and premalignant adenomas were for the most part Sox9-negative (Fig. 3F). Moreover, the most intensely Sox9-positive cells were observed in morphologically more malignant and aggressive areas of the adenocarcinomas, particularly at their growth edge and in endobronchial invasions. This observation suggests the existence of an inverse correlation between the expression of Sox9 and senescence markers in vivo and further indicates that Sox9 expression is associated with tumor progression and malignancy.

Extensive cultivation of primary wt MEFs usually favors the amplification of rare preexisting mutant cells that overgrow senescent cells. Loss of the Cdkn2a locus or loss of p53 are the most characteristic events responsible for the immortalization of MEFs. In agreement with the role of Sox9 as a negative regulator of Cdkn2a, 3 analyzed immortal Z/Sox9tg MEFs retained functional p53 (data not shown). Established cultures of such immortal (passage 15) cells were found to express higher levels of Sox9 than those at early passage (passage 1; Fig. 4A and C). This prompted us to analyze Sox9 expression in neoplastically transformed cells. For this, we retrovirally transduced primary wt and Z/sox9tg MEFs with a combination of E1a and H-RasV12 oncogenes (E1a/Ras). Transformed cells contained increased Sox9 expression compared with nontransformed, with the highest levels in Z/sox9tg cells (Fig. 4B and C). This is likely to reflect increased transcription and not protein stability, as we found that, compared with normal control MEFs, Sox9 mRNA levels were augmented by around 10- and 15-fold in immortal and transformed wt cells and between 40 and 70 times, respectively, in Z/sox9tg cells (Fig. 4D and E). In contrast, SOX9 expression was significantly decreased in Sox9^Δ/Δ transformed cells (Fig. 4F). These data show that endogenous levels of Sox9 increase with progressive stages of tumorigenesis, such as spontaneous immortalization and neoplastic transformation, which emphasizes the likely physiologic significance of Sox9 in these processes. Moreover, the levels of Sox9 in Z/sox9tg cells at early passage were comparable with endogenous Sox9 levels at immortal and neoplastic stages in wt cells reinforcing the concept that the degree of overexpression achieved with the transgene is of biologic relevance.

Histone modifications play an important role in the regulation of gene expression. Methylation of histone 3 lysine 4 (H3-K4me) is associated with gene activation whereas trimethylation of H3-K9 and H3-K27 (H3K9me3 and H3K27me3) are linked with gene silencing (36). To examine whether Sox9 overexpression is associated with histone modifications on the promoter region, we conducted ChIP analysis against those histone marks in E1a/Ras-transformed cells. We observed an enrichment of the gene activation mark histone H3 lysine 4 trimethylation (H3K4me3), whereas the repressive marks H3K9me3 and H3K27me3 completely disappeared from the Sox9 promoter in transformed MEFs (Fig. 4G). These results show that Sox9 upregulation in neoplastic cells correlates with chromatin changes, thus indicating another mechanism for its upregulation.

To further substantiate the oncogenic properties of Sox9, we explored the behavior of cells by altering Sox9 in neoplastic transformation. To test whether Sox9 collaborated with oncogenes to trigger cell transformation, we used wt and Z/sox9tg MEFs transduced with E1a/Ras. Importantly, Z/sox9tg MEFs gave rise to 5 times more foci than wt cells (Fig. 4H). In contrast, Sox9^Δ/Δ MEFs formed 3 times fewer foci in response to E1a/Ras (Fig. 4I). Because the ability of cells to grow independently of adhesion is a feature of cancer cells, we analyzed whether the growth of E1a/Ras transformed cells was anchorage-independent by plating cells in soft agar. Similar results were obtained with Z/sox9tg generating more colonies (Fig. 4J). Together, our findings indicate that Sox9 deregulation promotes proliferation, inhibits senescence, and collaborates in malignant transformation in primary MEFs. Finally, to investigate the oncogenic activity of SOX9 in human colorectal cancer cells, we addressed its ability to promote anchorage-independent cellular growth using the same soft agar assay. We first increased SOX9 expression by transiently transfecting a plasmid carrying Sox9-IRES-EGFP in SW480 cells. Cells overexpressing SOX9 formed higher number of colonies than cells carrying the empty vector (Fig. 4K). In contrast, the growth of SW620 and HCT116 in soft agar was inhibited by 50% with SOX9 knockdown (Fig. 4L). In summary, our observations show that SOX9 modulates phenotypes associated with cancer in human colorectal cancer cells.

Next, we explored the hypothesis that Sox9 is relevant to tumor formation and growth in vivo. First, we studied whether the increased sensitivity to neoplastic transformation observed in Z/sox9tg cells in vitro translates into an increased tumorigenicity in vivo. Consistent with this idea, pools of E1a/Ras-transformed Z/sox9tg cells gave rise to tumors at notably higher rates than E1a/Ras-transformed wt cells when transplanted in nude mice (Fig. 5A, Supplementary Fig. S8). To further substantiate whether Sox9 contributes to tumor growth, we repeated the same experiment with pools of Sox9^flox/flox and Sox9^Δ/Δ MEFs. Conversely, fewer than 20% of mice injected with E1a/Ras-transformed Sox9^Δ/Δ cells generated tumors at day 10 compared with 45% with control cells, and the average tumor volume was also markedly reduced (Fig. 5B, Supplementary Fig. S8). These results further support the idea that Sox9 possesses oncogenic properties and facilitates tumorigenesis in vivo. To provide further evidence for the contribution of Sox9 to tumorigenesis in vivo, we took advantage of mice carrying both Z/Sox9 and pCAGGS-ERT2iCre transgenes (Z/Sox9c). Groups of Z/Sox9c⁺ and Z/Sox9c⁻ mice (i.e., receiving 4-OHT to activate the transgene or ethanol as a control, respectively) were injected with the carcinogen 3-methylcholanthrene to evaluate their susceptibility to fibrosarcoma, a cancer type which reveals strong staining of Sox9 (Fig. 5C). In this context, Z/Sox9c⁺ mice were notably more susceptible to cancer formation than their Z/Sox9c⁻ littermates (Fig. 5D). In particular, 6 months after 3-methylcholanthrene treatment, 100% of Z/Sox9c⁺ mice developed fibrosarcoma compared with only 55% of Z/Sox9c⁻ control mice. Moreover, the median survival of Z/Sox9c⁺ (4.5 months) was significantly decreased compared with control mice (5.9 months; log-rank test, P = 0.03; Fig. 5D). Finally, to evaluate whether the protumorigenic role of SOX9 could be extended to human cells, we conducted similar analysis using human colorectal cancer cell lines. We used HCT116 and SW620, in which we have previously shown (see above) that knockdown of SOX9 impairs cell proliferation. These results were paralleled by an attenuation of tumorigenicity (Fig. 5E and F, Supplementary Fig. S8). In summary, our findings support the notion that Sox9 facilitates tumor formation and progression in vivo and indicates a functional involvement of SOX9 in the tumor establishment and/or development of colorectal human cancer.

With respect to the mechanisms underlying the phenotypes seen in cells with altered levels of Sox9 expression, we have seen that Sox9 activity modulates the expression of Ink4a and Arf tumor suppressors (see Fig. 3). The Polycomb group protein Bmi1, a well-established oncogene, plays an important role in proliferation, senescence, and carcinogenesis through its ability to repress the transcription of the Ink4a/Arf locus (37). We hypothesized that Sox9 might regulate Ink4a/Arf expression via Bmi1. We therefore analyzed Bmi1 expression in Sox9 gain- and loss-of-function MEFs finding that Sox9 overexpression in Z/sox9tg MEFs resulted in increased Bmi1 compared with controls whereas Sox9^Δ/Δ MEFs had lower levels (Fig. 6A). Sox9 is known to function as a transcriptional activator in several contexts, and consistent with this, the analysis of Bmi1 mRNA in early-passage MEFs indicated that Sox9 modifies its levels in a manner that parallels the changes in protein (Fig. 6B). These results suggest that Sox9 exerts its effect on proliferation and senescence, at least in part, through regulation of Bmi1-Ink4a/Arf expression.

With respect to transformed cells, Bmi1 paralleled Sox9 upregulation in response to E1a/Ras-induced transformation, with Z/sox9tg MEFs presenting the highest levels of Bmi1 (Fig. 6C and D) and Sox9^Δ/Δ MEFs the lowest (Fig. 6E and F). Next, we tested the possibility that Sox9 modulates Bmi1 expression by direct transcriptional control. To test this, we conducted ChIP assays that revealed readily detectable Sox9 bound to the Bmi1 promoter in Z/sox9tg but not in wild-type control MEFs (Fig. 6G). In addition, ChIP assays also showed that Sox9 interacts with the Bmi1 promoter in transformed cells (Fig. 6H). Our data show that Bmi1 is a novel direct downstream target of Sox9 in primary and transformed MEFs and may therefore represent a key mechanism mediating the role of Sox9 in tumorigenesis.

It has been previously reported that high levels of BMI1 are linked with progression and poor prognosis in colorectal carcinoma (38, 39). The results in MEFs prompted us to address whether SOX9 may contribute to colorectal carcinogenesis through BMI. First, we found that BMI1 expression mirrors SOX9 expression in human colorectal adenocarcinomas. Thus, the same tumor regions that strongly expressed SOX9 also contain high levels of BMI1 (Fig. 7A). Analysis of mRNA levels revealed that SOX9 upregulation correlated with increased BMI1 levels in colon adenocarcinomas compared with nontumor colonic mucosa (Fig. 7B). Furthermore, the correlation analysis showed a significant association between SOX9 and BMI1 expression in colorectal adenocarcinomas (Fig. 7C, Supplementary Fig. S9). In addition, SOX9 mRNA levels correlated negatively with ARF (Supplementary Fig. S9). Next, we extended this association to human colorectal cancer cell lines. BMI1 was strongly expressed in SW620, with high levels of SOX9, and it was weak in SW480, with low levels of SOX9 (Fig. 7D). Conversely, INK4a expression is very high in SW480 and low in SW620. These results show that regulatory interactions between SOX9, BMI1, and INK4a/ARF are likely to be critical during the initiation and progression of colorectal cancers.

Next, we checked whether SOX9 regulates BMI1 and INK4a expression in colorectal cancer cells. Indeed, SOX9 ectopic overexpression increased BMI1 levels accompanied by decrease in INK4a, whereas SOX9 knockdown diminished BMI1 expression and increased INK4a (Fig. 7E). Paralleling the results observed in MEFs, SOX9 regulated BMI1 expression at transcriptional level in colorectal cancer cells (Fig. 7F and G). To test the role of BMI1 and INK4a in SOX9 regulation of cancer cells, we stably inactivated BMI1 and overexpressed INK4a in SW620 cells, which possess high levels of SOX9. The shRNA-mediated inhibition of BMI1 and the restoration of INK4a significantly impaired proliferation and the number of colonies in soft agar (Fig. 7H and I), whereas SOX9 levels were unchanged (Fig. 7J and K). Together, these data provide direct experimental support for the role of SOX9 in cancer by regulating the expression of BMI1 and INK4a/ARF.

SOX9 is a critical gene for embryonic development and tissue homeostasis in several organs, with a role in cancer not totally understood yet. Our results show that its role in human cancer is context dependent. We found SOX9 overexpressed in a wide range of human cancers, particularly, in tissues where it plays critical roles in their development and in stem/progenitor cells such as colon, ovary, pancreas, prostate, lung, peripheral nerves, and brain (Supplementary Tables S1 and S2). Nevertheless, our results extend these findings to fibrosarcomas, breast, and thyroid cancers (Supplementary Table S1), where SOX9 is not known to be involved during normal physiology. On the other hand, SOX9 seems to be unaffected in lymphoma and kidney cancers and it has been reported to behave as a tumor suppressor in some melanoma cases (16), although its expression is upregulated in others (40–42). In addition to ourselves, others have also observed SOX9 overexpression in cancers from tissues such as prostate, lung, skin carcinomas, and brain (12–15) indicating that de novo or increased expression of SOX9 in cancer is likely to be positively selected rather than a bystander event.

In line with this idea, our results show that SOX9 is overexpressed in colorectal cancers and exhibits important functions in the regulation of colorectal cancer cells. The high levels might reflect the origin of the tumors from the expansion of SOX9-positive stem/progenitor cells. The increased SOX9 expression detected in early premalignant adenomas together with the role for Sox9 in stem/progenitor cells in the intestinal epithelium (21) and the evidence that transformation of such cells is the initiating step in colorectal tumorigenesis (43) support this hypothesis. Alternatively, the high levels of SOX9 could reflect genetic or epigenetic alterations and/or other types of deregulation. We identified altered SOX9 chromosomal copy number in colorectal cancers. These might not be exceptional cases. Amplification of 17q24.3, the chromosomal region containing SOX9, has been found in cancers such as prostate, MPNSTs, neuroblastoma, medulloblastoma, breast, and ovarian cancer (44–48), all of them cancer types in which we have also identified SOX9 as highly expressed (Supplementary Tables S1 and S2). These results support the idea that SOX9 copy number change might be relevant for a subgroup of samples among several malignancies. At the same time, a few cases of SOX9 duplication are reported in the literature that are the cause of sex reversal and brachydactyly anonychia (49, 50), but to our knowledge, these individuals have no developed cancer. However, this is consistent with ectopic expression of Sox9 in mice, for which tumor incidence is not increased in comparison with controls (51), but together with a transforming agent, activated oncogene, or event (such as chronic injury), higher levels of Sox9 significantly increase the rate of developing tumors. SOX9 expression in cancer might also be due to loss of repression. The changes in chromatin modifications that we found are consistent with altered upstream regulation. In line with this idea, SOX9 expression in colon cancer cell lines is controlled by Wnt signaling (19), a pathway in which constitutive activation generally initiates colorectal cancers (43).

Mechanistically, we observed that Sox9 activity promotes proliferation and its upregulation is sufficient to bypass senescence in primary MEFs. Furthermore, SOX9 regulates proliferation and facilitates tumor growth in colorectal cancer cells. It has been shown that Sox9 promotes proliferation and prevents differentiation in intestinal stem cell/progenitors, acting downstream of Wnt signaling (20, 22). These results support the idea that the mechanisms governing self-renewal and differentiation are overlapping with those regulating proliferation and senescence and are consistent with a role for SOX9 in early stages of colorectal tumorigenesis. Furthermore, we found that SOX9 levels increase with advanced tumor stages in colorectal cancers and it cooperates with activated K-Ras, an oncogene frequently found mutated during colorectal cancer development (52), to facilitate transformation and tumor progression. It is noteworthy that high levels of SOX9 are associated with lower survival and adverse prognosis (17), and amplification of the 17q24.3 region correlates with the transition from primary colorectal cancer to metastasis (53). Collectively, these data indicate that SOX9 is important in colorectal cancer progression and support its use as a potential diagnostic and prognostic marker in colorectal cancers.

BMI1 is a key oncogene implicated in enhancing proliferation, stem cell self-renewal, evading senescence, and favoring multiple cancer types partly through its ability to repress the transcription of INK4a and ARF (54). Nevertheless, BMI1 plays also a role in tumorigenesis through an INK4a/ARF-independent mechanism (55, 56). We found that Sox9 directly binds the Bmi1 promoter, it regulates BMI1 expression in MEFs and colorectal cancer cells, and its expression is associated with BMI1 levels in primary colorectal adenocarcinomas. Moreover, BMI1 inactivation partly rescues the phenotypes seen in cancer cells with high levels of SOX9. These data indicate that SOX9 oncogenic activity is achieved, at least in part, by its ability to regulate the expression of BMI1 and show that SOX9–BMI1 interactions are likely to be critical events in colorectal tumor development and progression. In agreement with this, it is well known that Bmi1 is a stem/progenitor cell marker involved in the transformation of stem/progenitor cells in several tissues including intestinal epithelium (54, 57). High levels of BMI1 have been observed in benign adenomas and malignant carcinomas, and its expression has been linked with progression, lower survival, and poor prognosis in colorectal cancers (38, 39, 58). Moreover, both genes, SOX9 and BMI1, have been linked to tumorigenic epithelial-to-mesenchymal transition, a mechanism causing cancer metastasis, and that occurs in the most advanced colorectal cancers (59, 60). Finally, we provide evidence that some of the Sox9–Bmi1 functions are dependent on Ink4a and Arf, whose reduced expression has been associated with BMI1 overexpression in human colorectal cancers (61). The relevance of this pathway may be extended to other cell types and tissues where these proteins are involved in their normal physiology and deregulated in cancers. In summary, our observations provide direct experimental support for the oncogenic activity of SOX9, its relevance at different stages of tumorigenesis, and the mechanisms involved, particularly in colorectal cancers.

No potential conflicts of interest were disclosed.

The authors thank the Biological Services Staff at NIMR, Maribel Muñoz for excellent mouse colony management, and Manuel Morente and Maria Jesus Artigas in the Tumor Bank Unit at the CNIO. They also acknowledge the patients enrolled in this study for their participation and the Basque Biobank for Research-OEHUN for its collaboration.

A. Matheu was supported by a postdoctoral fellowship from the Human Frontier Science Program. M. Collado is funded by the “Ramon y Cajal” Program from the Spanish Ministry of Science (MICINN). L. Manterola is a Sara Borrell fellow from Carlos III Health Institute. Work in the laboratory of R. Lovell-Badge is funded by the UK Medical Research Council (U117512772) and an NIH Quantum Grant. Work in the laboratory of M. Serrano is funded by the CNIO and by grants from the MICINN (SAF2005-03018 and OncoBIO-CONSOLIDER), from the European Research Council, and from the “Marcelino Botin” Foundation. K.S.E. Cheah is supported by the Research Grants Council and University Grants Council of Hong Kong (HKU 4/05C and AoE/M-04/04).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.