Runx2 transcriptome of prostate cancer cells: insights into invasiveness and bone metastasis

1Runx2 together with Runx1 and Runx3 comprise the Runx class of transcription factors, defined by their highly homologous Runt-related DNA-binding domain. As heterodimers with Cbfß, Runx proteins bind to cognate DNA elements with the consensus nucleotide sequence 5'-ACCACA in the promoters/enhancers of their target genes. The three Runx proteins coordinate proliferation and differentiation of various cell types. Runx1 is important for hematopoiesis; Runx2 is pivotal in osteogenesis; and Runx3 is critical for neurogenesis, thymopoiesis, and maintenance of the gastric epithelium. While promoting specific cellular phenotypes, Runx proteins have evolved to inhibit cell proliferation. Runx3 is a bona fide tumor suppressor as down-regulation of its promoter by hypermethylation contributes to the development of gastric cancer. Ablation of Runx1 activity leads to leukemia and disruption of Runx2 results in deregulated cell proliferation and immortalization. Paradoxically, Runx2 is also implicated in carcinogenesis. In a mouse screen for c-Myc-collaborating oncogenes, MLV-induced leukemia occurred most frequently when the provirus integrated into the Runx2 locus resulting in its ectopic expression. It was suggested that Runx2 initially provides the cells with a survival advantage, and its anti-mitogenic activity is counteracted by the CD2-Myc transgene present in the mouse model used for this screen. Therefore, Runx2-mediated tumorigenesis likely requires additional loss of check-point genes such as Trp53 or improper regulation of an oncogene such as c-Myc.

2Runx2 has been extensively studied in the context of osteoblastogenesis from mesenchymal progenitors, where as a master regulator it stimulates the expression of various bone matrix components such as osteocalcin and bone sialoprotein (BSP). Runx2-/- mice die soon after birth due to the lack of differentiated osteoblasts and thus a mineralized skeleton. Runx2 haploinsufficiency in humans causes the rare skeletal disorder Cleidocranial Dysplasia. In search for hints to explain the high predilection of prostate and breast cancer to metastasize to bone, investigators have noticed ectopic expression of Runx2 and some of its target genes in biopsies from advanced tumors and their derivative cell lines. In a mouse model of PCa, conditional deletion of Pten in prostate epithelial cells resulted in the development of tumors with progressive increase in Runx2 expression. Among the osteomimetic properties of prostate and breast cancer cells are expression of the Runx2 target genes MMP9, BSP and VEGFA, as well as induction of mineralization.

3In addition to promoting osteoblast differentiation, Runx2 drives the expression of osteoclastogenic signals, both in osteoblasts and in the PC3 bone metastasis-derived PCa cell line. PC3 cells robustly express Runx2, and its silencing decreased their osteoclastogenic property in vitro and their growth within the bone microenvironment in vivo. Runx2 also promotes metastatic aspects not necessarily related to bone. Invasion of PC3 cells through Matrigel™, a basement membrane-like preparation, decreased after Runx2 silencing, and its ectopic expression in mammary epithelial cells increased their proliferation and disrupted their normal acinar organization. An oncogenic role for Runx2 has also been suggested in tumors that do not exhibit high predilection to bone, including pancreatic ductal adenocarcinoma and thyroid papillary carcinoma.

4Whereas Runx2 is being increasingly recognized as a pro-metastatic factor, little is known about the underlying transcriptional programs. To establish gene regulatory networks downstream of Runx2 in aggressive PCa, we analyzed gene expression in response to Runx2 in the C4-2B PCa cell line. These cells are castration-resistant derivatives of the androgen-dependent LNCaP cells, and serve as a model for the aggressive stage of bone metastatic PCa. Although C4-2B cells express Runx2 at levels higher than LNCaP cells, these levels are far lower than those observed in PC3 cells or osteoblasts. We therefore engineered a C4-2B sub-line that allowed us to profile gene expression after induction of Runx2 with Doxycycline to levels seen in PC3 cells. Remarkably, the most significant changes were the up-regulation of genes implicated in cancer progression and cellular movement, and the down-regulation of genes involved in cell cycle progression. Consistent with these changes in gene expression, Runx2 enhanced PCa cell invasiveness and inhibited their proliferation.

5To establish a C4-2B cell line that conditionally expresses Runx2, we employed the recently described lentivirus-based pSLIK vector system, which allows tight Doxycycline (Dox)-inducible, RNA PolII-mediated transcription of a gene of interest. C4-2B cells were transduced with Flag-tagged Runx2-encoding lentiviruses (Figure 1A), resulting in the C4-2B/Rx2dox subline. As control, we established the C4-2B/Rx2-Mdox subline, where Dox treatment induced expression of the transcriptionally inactive Flag-Runx2-M (Figure 1A). Western blot analysis with anti-Flag antibodies confirmed roughly equal expression levels of the wild type and mutant Runx2 proteins, which were strictly and dose-dependently regulated by Dox (Figure 1B). RT-qPCR analysis revealed that the Dox treatment increased Runx2 mRNA by ~20-fold compared to its endogenous levels, and that the induced level was comparable to that observed in the PC3high sub-line (Figure 1C). Western analysis using anti-Runx2 antibodies indicated that the level of endogenous Runx2 protein was negligible in untreated C4-2B cells, and that Dox induced expression of the exogenous Runx2 to the levels normally found in osteoblasts (Figure 1D).

6The transcriptional activity of Dox-induced Runx2 was initially assessed using luciferase reporter assay (Figure 1E). In the reporter plasmid 6XOSE2-Luc, luciferase expression was controlled by six copies of the osteoblast-specific element 2 (OSE2) from the Runx2-regulated osteocalcin (OC) gene promoter. In the absence of Dox, 6XOSE2-luc activity was indistinguishable from the background luciferase activity observed without any cell extract, suggesting lack of endogenous Runx2 activity (Figure 1E). The luciferase reporter was strongly stimulated by WT but not by the mutant form of Runx2 (Figure 1E). As shown in Figure 1F, Runx2 also stimulated transcription of its endogenous target genes Bone Sialoprotein (BSP) and Matrix Metalloprotease-9 (MMP9). These genes were not stimulated in the Dox-treated C4-2B/Rx2-Mdox cells (data not shown). Interestingly, Runx2 did not significantly enhance the expression of OC (Figure 1F) and Alkaline Phophatase (ALP; data not shown), although these genes are strongly stimulated by Runx2 in osteoblasts. This observation reflects cell type-dependent Runx2-mediated transcriptional control, and is consistent with the results of Yeung et al., who demonstrated that in PC3 cells the OC promoter is responsive to the transcription factors AP-1 and SP1, but not Runx2. To identify Runx2-regulated genes and pathways in advanced PCa cells in an unbiased manner, we subjected C4-2B/Rx2dox cells to global gene expression profiling.

7C4-2B/Rx2dox cells were subjected to microarray gene expression analysis after one- and two- days of treatment with either Dox or vehicle in biological quadruplicates (a total of 16 samples). Of 24,526 probes represented in the microarray, 532 genes showed ≥2-fold increased expression and 378 genes showed ≥2-fold decreased expression with high statistical significance (p < 0.008) on either day of treatment (see additional file 1). RT-qPCR analysis of 50 representative genes conformed to the microarray data (Table 1 and additional file 2).

8An unsupervised hierarchical analysis of these 910 up- and down-regulated genes resulted in a clear separation between the Dox-treated and control samples (Figure 2). The variation among the biological quadruplicates was small, indicating the overall robustness of the methodology utilized. Gene clusters showing changes in expression pattern with respect to time and Dox treatment were clearly discernable. In general, changes observed on day 1 of treatment were maintained or intensified by day 2.

9We next employed the Ingenuity Pathway Analysis (IPA™) platform to indentify disease pathways, as well as molecular and cellular functions associated with Runx2-regulated genes. The analysis suggested 'cancer' as the disease most significantly associated with both the up- and the down-regulated gene groups (Table 2). A total of 248 genes, half from each group, had highly significant cancer related function (see additional file 3). Additionally, the up-regulated genes were strongly associated with genetic disorders, inflammatory responses, and gastrointestinal diseases (Table 2A). Among the most significant molecular and cellular functions, cellular movement, cell death, cellular growth, and proliferation were associated with the up-regulated genes, whereas cell cycle, cell death, cellular assembly and DNA replication functions were associated with the down-regulated genes (Table 2B).

10The major functions reported for the up-regulated genes belonged to cancer progression (Table 2). Importantly, these genes encode transcriptional regulators, cytoskeletal components, signaling molecules and peptidases, which have been implicated in tumor metastasis (Table 1). The transcription factors Sox9 and SNAI2, and the extracellular-matrix (ECM) protein LCN2, all major regulators of epithelial-to-mesenchymal transition (EMT), were up-regulated by ~4-fold after one day and by > 6-fold after two days of Runx2 induction, and their upregulation was confirmed by RT-qPCR (Table 1). However, the functional significance of these EMT markers requires further investigation in light of the unexpected increase in E-cadherin mRNA (see additional file 4). Runx2 also enhanced the expression levels of multiple transcripts encoding matrix modifying peptidases (Table 1). These included MMP9, a known Runx2 target in BCa cells and aspartyl proteases with fibronectin degrading activities such as Prolactin Induced Protein (PIP) and Pepsinogen (PGC). The latter two showed a rapid ~10-fold increase within 24 hours (Table 1) and PIP exhibited the highest change (79-fold) in response to Runx2 on day 2 (Table 1 and additional file 2). PIP protein in the C4-2B/Rx2dox culture supernatant was below detectable levels under control conditions, but was readily detected after induction of Runx2 (Figure 3). We also found increased transcript levels for Cystatin 7 (CST7), S100A4 and SMAD3, with a mild ~3-fold increase on day 1, but a robust > 20-fold increase on day 2 (Table 1). These genes function as metastasis promoters. Interestingly, S100A4 and SMAD3 physically interact to potentiate cancer cell invasiveness.

11Runx2 also up-regulated genes involved in cellular movement and cytoskeleton remodeling (Table 1). SH3PXD2A, which was up-regulated by 7-fold on day 2, is a scaffold protein involved in the formation of invadopedia, which are matrix digesting, actin rich, short lived protrusions observed in osteoclasts and cancer cells. Runx2 up-regulated by > 9-fold the transcripts for Nav2, a scaffold protein crucial for actin cytoskeleton remodeling. Other genes that were up-regulated by > 3-fold, with known roles in actin cytoskeleton dynamics included ESPN, which interacts with the Src-homology 3 (SH3) adaptor proteins to regulate cytoskeletal actin functions; MAP1B, known to maintain cytoskeletal integrity; LIMA1, which cross-links actin monomers; and PTK9L (a.k.a. twinfiln), which sequesters ADP-actin monomers in the cytoplasm and delivers them to sites of rapid actin-filament assembly.

12The expression of SDF-1 and its receptor CXCR7 was enhanced by > 5-fold based on the microarray analysis and by > 20-fold based on the RT-qPCR results (Table 1). Runx2-induced SDF-1 protein was also detectable in the culture supernatant (Figure 3). SDF-1 signaling is critical for homing of hematopoietic cells to the bone marrow space and their survival in this environment. Within one day, Runx2 also increased by 10-fold the mRNA for BSP (Figure 1F), whose abundant expression by bone metastatic tumor cells facilitates their attachment to the bone matrix. Once settled in the bone microenvironment, the metastatic cells secrete regulatory molecules that stimulate bone turnover. Remarkably, Runx2 enhanced the expression of the osteoclastogenic cytokine CSF2 by > 50-fold within 48 hours (Table 1). This presumably occurred by direct binding of Runx2 to the CSF2 promoter. Runx2-mediated induction of CSF2 in PCa cells likely contributes to the increased bone turnover in bone metastatic sites, similar to the role of this cytokine in breast cancer bone metastasis. CSF2 production by tumor cells may also contribute to accumulation of macrophages, inflammatory T cells, and cytokines that exacerbate morbidity and mortality. Two additional Runx2-up-regulated genes associated with osteoclast function are SPHK1, a kinase responsible for the production of sphingosine 1 phosphate (S1P), and S1P receptor 3 (S1P3) a.k.a. EDG3 (Table 1). Production of S1P in the bone microenvironment promotes bone resorption by chemotactically attracting osteoclast precursors. The SPHK1/S1P/S1P3 axis plays additional roles in cancer progression, including cell growth, migration, angiogenesis, and resistance to chemotherapy. Notably, Runx2 was the only gene differentially up-regulated in chemotherapy-resistant versus -sensitive osteosarcoma tumors. Adding to this, Runx2 also repressed the expression of GDF-15, an osteoclastogenesis inhibitor (Table 1). This repression was mild on day 1 (2-fold), but by day 2 GDF-15 was the most repressed gene in response to Runx2 (8-fold; Table 1 and additional file 2). Thus, Runx2-mediated alterations in gene expression may contribute to both the predilection of PCa to bone and the subsequent pathological increase in bone turnover, which further fuels growth of the metastatic tumors.

13Runx2 has been implicated in promoting angiogenesis by stimulating VEGFA expression during bone development as well as during tumorigenesis. In C4-2B/Rx2dox cells, Runx2 increased VEGFA mRNA by 4-fold (Table 1) and the presence of VEGF in the cell culture supernatant was detectable only after Dox treatment (Figure 3). Additionally, our study revealed a 32-fold upregulation of the VEGFA co-receptor Syndecan-2 (SDC2; Table 1). SDC2, which is also a Runx2 target in osteoprogenitor cells, is a member of the heparan sulfate proteoglycans family, and is also implicated in cell adhesion and communication. Interestingly, VEGFA can functionally synergize with SDF-1 to promote neoangiogenesis in vivo. Our microarray analysis also revealed Runx2-mediated induction of the angiogenic EDN-2 gene (Table 1). Endothelins and VEGFA are secreted by PCa cells to stimulate angiogenesis as well as differentiation of neighboring osteoblasts in the bone microenvironment.

14Because Runx2 enhanced the expression of multiple extracellular enzymes involved in ECM degradation (Table 1), we initially tested by in-gel zymography the presence of proteases in the supernatant of Dox-treated C4-2B/Rx2dox cultures. The results demonstrated that Runx2 induced several gelatin-degrading proteins, in particular one with a molecular weight of ~140 kDa, the identity of which remains to be determined (Figure 4A). We further investigated whether Runx2 stimulates invasion of C4-2B/Rx2dox cells through Matrigel™, a tissue basement membrane-like preparation containing laminin, type IV collagen, heparan sulfate proteoglycans and entactin. For convenient and accurate assessment of cells that successfully invade through the Matrigel™ membrane, we transduced the C4-2B/Rx2dox cells with a lentivirus constitutively expressing luciferase. Parallel trans-wells that do not contain Matrigel™ were employed as migration controls. Cells were incubated in the respective chambers in the presence or absence of Dox, and the relative migration or invasion capacity was assessed. Runx2 expression led to a 2.3-fold decrease in cell migration, but a 4.3-fold increase in invasion through Matrigel™, i.e., a ~10-fold increase in invasiveness after adjustment for the reduced cell migration (Figure 4B). The increased invasiveness was further confirmed in an independent experiment by histological staining (Figure 4C). In parallel experiments, expression of Runx2-M in C4-2B cells showed no significant effects on either migration or invasion (data not shown). Thus, expression of transcriptionally active Runx2 is sufficient to enhance the tissue invasion potential of C4-2B cells.

15The IPA™ analysis of Runx2 down-regulated genes indicated their strong association with cell cycle and proliferation-related functions (Table 2B). Many of these down-regulated genes, as well as several up-regulated genes, shed light on the well-established anti-proliferative activity of Runx2. Most striking was the > 19- and > 8-fold up-regulation of RASD1 and DUSP1, respectively (Table 1). RASD1 belongs to the Ras superfamily of G-proteins, and its expression in breast cancer suppressed cell growth. DUSP1, a.k.a. MAP kinase phosphatase 1 (MKP1), is a dual specificity (Thr/Tyr) protein phophatase with anti-proliferative properties. Among the most important cell cycle regulatory genes inhibited by Runx2 was c-Myc, with a ~3-fold decrease at the mRNA level (Table 1) and a corresponding significant decrease at the protein level (Figure 3). In line with the down-regulation of c-Myc, the mRNA encoding its cell cycle promoting targets E2F2 and CDK2 were also down-regulated (Table 1). CDK2 protein was decreased below detectability (Figure 3). To further characterize effects of Runx2 on PCa cell proliferation, we first validated by RT-qPCR the changes in the transcript levels of RASD1, DUSP1, c-Myc and E2F2 in the day-2 samples (Figure 5A). Next, we tested the effect of Runx2 on C4-2B cell proliferation by performing MTT assays every 48 hours after Dox-mediated Runx2 induction. Runx2 significantly restrained cell proliferation (Figure 5B). By contrast, the transcriptionally-inactive Runx2-M did not affect proliferation (Figure 5C). Thus, Runx2 restrains PCa cell proliferation via its transcriptional activation property.

16To delineate the anti-proliferative effect of Runx2, we tested its influence on apoptosis and cell cycle progression. Apoptosis was measured using succinyl-AMC, a fluorogenic Caspase-3 substrate. Expression of Runx2, but not Runx2-M, resulted in a transient 40% increase in apoptosis on day 4 (Figure 5D), but this could not account for the dramatic inhibition of cell proliferation (Figure 5B). Instead, fluorescence activated cell-sorting (FACS) analysis revealed a 2.2-fold decrease in the fraction of cells in the S/G2/M phases of the cell cycle within 24 hours of Dox treatment, and this effect persisted on day 2 (Figure 5E). Notably, the cell cycle inhibition preceded any change in apoptosis, indicating that Runx2 restrained C4-2B cell proliferation by inhibiting the G1/S phase transition of the cell cycle.

17The prometastatic but anti-mitogenic properties of Runx2 in PCa cells suggest that it may initially facilitate metastasis, after which it must be degraded or antagonized (e.g. by co-repressors) to allow cell proliferation and tumor growth. Therefore, we examined if the anti-mitogenic effect of Runx2 was reversible by withdrawing Dox from cultures after 48 hours of treatment (Figure 5F). Dox withdrawal led to substantial clearance of Runx2 within two days, and undetectable Runx2 levels after four days (Figure 5G). This resulted in resumption of cell cycle progression (Figure 5H) and cell proliferation (Figure 5I). Thus, the Runx2-regulated gene networks induce reversible cellular quiescence by blocking the G1/S phase transition of the cell cycle.

18Our study is the first to provide genome wide analysis of Runx2-regulated genes in PCa cells. Although we focused on the bone metastasis-derived C4-2B cells, similar responses to Runx2 were observed in the parental lymph node-derived LNCaP cells (see additional file 5) as well as in the unrelated bone metastatic 22RV1 PCa cells (see additional file 6). Furthermore, in PC3high and PC3low cells with high and low levels of Runx2, respectively, the expression of six randomly selected Runx2 up-regulated genes from the present study correlated with Runx2 expression (see additional file 7). Together, these results suggest that our observations are relevant to various stages of PCa progression. Among the 910 genes that Runx2 up- or down-regulated by ≥2 fold, IPA™ identified 248 genes related to cancer with high statistical significance (p < 0.003; Table 2 and additional file 3). The IPA™ analysis, as well as survey of literature on gene expression profiling in osteoblasts and fibroblasts, further suggested that the Runx2-regulated gene network in PCa (Figure 6) bears little resemblance to its targets in mesenchymal cells. In fact, only five of the cancer-related genes in our study have been previously reported as Runx2 targets (MMP13, CEBPA, VEGF, SMAD3, and SMAD6) and only six others are Runx1 and/or Runx3 targets (SKP2, CSF2, IL3, ACHE, IGFBP3, and HIPK2; Figure 6). The remaining 234 genes are therefore novel Runx2-regulated genes related to cancer in general and to metastasis in particular. The extensive Runx2-regulated cancer-related gene network highlights Runx2 as a viable target for the diagnosis, prognosis and treatment of PCa.

19Runx2, traditionally known for its master regulatory roles in the chondro-osteoblast lineage, is emerging as a prometastatic transcription factor. The Runx2 transcriptome in C4-2B cells documents gene networks that control multiple aspects of metastasis. Potentially contributing to local invasion and dissemination are the genes known to function in EMT, motility and ECM degradation. Additionally, the prometastatic function of Runx2 likely involves its target genes SDF-1, CXCR7 and BSP, which promote homing and attachment to bone. We also discovered Runx2 targets such as CSF2 and SPHK1, osteoclast activators that likely contribute to the most important alteration that occurs in the bone microenvironment in response to PCa metastasis, namely enhanced bone turnover. During this process, bone matrix components such as TGFß, BMPs and calcium ions are released and further fuel tumor growth and bone microenvironment modifications. The regulation of SPHK1 by Runx2 probably potentiates additional aspects of the cancer phenotype, including angiogenesis (a function assisted by the Runx2 targets VEGFA and EDN2) and drug resistance.

20The anti-mitogenic activity of Runx2 is consistent with the slow growth of PCa tumors, and may contribute to drug resistance. We imagine that future anti-Runx2 drugs will be administered along with traditional chemotherapy to eliminate cells that regain proliferative capacity. Interestingly, Runx2 physically and functionally interacts with the receptors for androgens and estrogens. Since these receptor proteins are targeted by many drugs for prostate and breast cancer, it is important to investigate their effects on Runx2-regulated transcription. Moreover, development of selective estrogen and androgen receptor modulators may benefit from consideration of their effects on Runx2 and expression of its target genes reported in the present study.

21C4-2B cells were obtained from ViroMed Laboratories (Minneapolis, MN). LNCaP and 22RV1 cells were from ATCC (Rockville, MD, USA). PC3 cells were also obtained from ATCC, but propagated for several years in either our laboratory (PC3low) or that of USC's Dr. Pradip Roy-Burman (PC3high). The cells were maintained in RPMI-1640 medium supplemented with 10% Tet System Approved FBS from Clontech, CA, USA. Hygromycin B was purchased from Invitrogen, Carlsbad, CA, USA and added to the growth medium at 50 μg/ml. Doxycycline from Calibiochem, La Jolla, CA, USA was used at 0.25 ug/ml unless otherwise stated, and an equal volume of distilled water was used as vehicle control. Mouse ANTI-FLAG® M2 monoclonal antibody for was purchased from Sigma, St Louis, MO, USA. Mouse anti-Runx2 was from Invitrogen, Carlsbad, CA, USA. The anti-PIP antibody (ab 62363) was purchased from abcam Inc., Cambridge, MA; and anti-GAPDH (V-18), anti-TGFß-RIII (Sc 28975), and anti-SDF-1 (Sc-28876) antibodies were purchased from Santa Cruz Biotechnology Inc, Santa Cruz, CA, USA. The mouse monoclonal anti-Myc and anti-VEGFA antibodies were developed and kindly provided by USC's Dr. Prakash S. Gill and Dr. Young Hong, respectively. The mouse monoclonal anti-Tubulin antibody, developed by Dr. Charles Walsh, was obtained from the Developmental Studies Hybridoma Bank under the auspices of the NICHD and The University of Iowa, Department of Biological Sciences, Iowa City, USA.

22The cDNA encoding mouse Runx2 (MASN isoform, type-2), which is 97% identical to human type-2 Runx2, was amplified using pcDNA3.0-Runx2 as template and the Flag epitope was inserted during the PCR amplification (see primers in additional file 8). The Flag-Runx2 cDNA was initially cloned into the SpeI/MfeI-digested lentiviral entry vector pEN_TmiRc3 (ATCC® catalog: MBA-248), and the resulting plasmid was recombined using Gateway® LR Clonase® II enzyme mix (Invitrogen, Carlsbad, CA, USA) with the pSLIK (single lentivector for inducible knockdown) destination vector carrying a hygromycin resistance gene (ATCC® catalog: MBA-237). The entry and destination vectors were kindly provided by USC's Dr. Elizabeth Lowler (Childrens Hospital Los Angeles). The DNA-binding mutant Runx2-M was constructed by site directed mutagenesis of two arginine residues at positions 265 and 268 (GRSGRGK) known to contact DNA in the crystal structure. Lentiviral plasmid for constitutive Luciferase expression pCCL-c-MNDU3c-Luc-PGK-eGFP was kindly provided by USC's Dr. Michael Kahn at the Zilkha Neurogenetic Institute.

23For packaging, the lentiviral expression plasmids were cotransfected by the calcium chloride method into HEK293T cells along with helper plasmids pMD.G1 and pCMVR8.91. Culture media containing viral particles were harvested after 48-72 hours and used for transduction of C4-2B cells in the presence of 8 μg/ml Polybrene (Millipore Corp., MA, USA). After infection with the lentiviruses, the transduced cells were selected with 50 μg/ml of Hygromycin.

24Transient transfection and Luciferase assays were performed essentially as described by Khalid et al.. Briefly, 25,000 cells were plated in a 24-well plate 48 hours prior to transfection using Invitrogen's Lipofectamine™ LTX reagent according to manufacturer's instructions. Cells were harvested and subjected to luciferase assay using Luciferase Assay system from Promega, Madison, WI, USA.

25Gene expression profiling was performed using the BeadChip™ platform (Illumina) and chip reference 8, version 3 for humans, which contains 24,526 gene probes and 664 negative control probes. Details of the raw data processing and analysis are provided in the additional file 9. Briefly, after background correction, the normalized expression intensities for all probes were subjected to a two-way analysis of variance (ANOVA) and the resulting p-values were adjusted for multiple testing using the Benjimini-Hochberg method. The differentially expressed probes were further investigated using the Ingenuity Pathways Analysis package (IPA™; http://www.ingenuity.com) to identify biological functions and disease categories that are significantly enriched among the differentially expressed genes. Right-tailed Fisher's exact test as implemented in the IPA software was used to calculate a p-value for the probability of each network to be enriched for Runx2-regulated genes due to chance alone. The microarray data has been deposited to the GEO database with the accession code GSE24261.

26C4-2B/Rx2dox cells were cultured in 10 cm culture dishes to 80% confluence, washed 3 times with RPMI-1640 and treated with Dox in 10 mL of RPMI-1640 without FBS for 24 hrs. For zymography the 25 μl of the conditioned media from Dox treated or control cells were analyzed by 8% acrylamide gels containing 0.1% w/v gelatin After electrophoresis, the gel was washed with 2.5% Triton X-100 and incubated in the developing buffer (500 mM Tris-HCl, pH 7.8, 2 M NaCl, 50 mM CaCl2) overnight to induce gelatin lysis. Gel was stained by Coomasie blue-250. For western blot analysis of conditioned media, 250 μl of the supernatant was TCA-precipitated and subjected to reducing SDS-PAGE analysis using standard procedures.

27Invasion through Matrigel™ was assessed by incubating 20,000 luciferase-expressing cells at the top of 24 well BD BioCoat™ Growth Factor Reduced Matrigel™ chamber (BD Biosciences, MA, USA). Migration was assessed using BD BioCoat™ Control Inserts. Cells that migrated or invaded to the bottom compartment were visualized by Diff-Quick™ staining kit (Baxter, MN, USA) or subjected to luciferase assays for quantification. Invasion index was defined as the percentage of cells that invaded through Matrigel™ over those that migrated under the same conditions but without the Matrigel™.

28Cultures on different days were subjected to MTT assay (Sigma, St Louis, MO) to measure viable cells in culture. For cell cycle analysis, 2 × 105 to 1 × 106 cells were harvested, washed twice with PBS (1 mL) at room temperature and stored in absolute ethanol (4 mL) for at least 24 hours. Pelleted cells were rehydrated in 5 mL PBS for 15 minutes, followed by staining with 1 mL of a propidium iodide (PI, Calibiochem, La jolla, CA, USA) solution containing 3 μM PI in incubation buffer (100 mM Tris, pH 7.4, 150 mM NaCl, 1 mM CaCl2, 0.5 mM MgCl2, 0.1% Nonidet® P-40). The cell suspension (1 mL) was subjected to fluorescence-activated cell sorting (FACScaliber, Becton Dickinson, MA, USA) and each cell was assigned to the G1, S, G2 or M phase of the cell cycle based on the PI intensity and using the Multicycle v3.0 software (Phoenix Flow Systems, San Diego, CA, USA). To assess apoptosis, cells were lysed in caspase assay buffer containing 50 mM HEPES (pH 7.5), 100 mM NaCl, 2 mM EDTA, 0.1% CHAPS, 10% sucrose, and 5 mM DTT. Aliquots of crude cell lysate (50 μg protein) were incubated with the caspase-3 substrate Ac-DEVD-AMC (EMD/Calbiochem, La Jolla, CA, USA) at 37°C for 30 min and the caspase-3 activity was quantified by flow fluorimetry with excitation at 380 nm and emission at 440 nm using Victor3V™ from PerkinElmer, Shelton, CT, USA.

29Between 1 × 105 and 2 × 105 cells were washed once with PBS and lysed with 200 μL of incubation buffer [100 mM Tris (pH 7.4), 500 mM NaCl, 1 mM CaCl2, 0.5 mM MgCl2, 0.1% Nonidet® P-40] supplemented with Complete™ protease inhibitor mix (Roche Diagnostics, Indianapolis, IN, USA). Aliquots of 40 μg cell lysate were mixed with an equal volume of Laemmli buffer and proteins were resolved by 10% SDS-PAGE, transferred to Amersham Hybond™-P PVDF (GE Healthcare, Piscataway, NJ, USA) membranes, and visualized using respective antibodies and the Western Lightning™ Plus-ECL kit (PerkinElmer Inc, Waltham, MA, USA) followed by exposure to X-ray film (ISCBioExpress®, Kaysville, UT, USA).

30Total RNA was isolated using Aurum Total RNA kit (Bio-Rad Laboratories Inc., Hercules, CA, USA) following the manufacturer's recommendations and 1 μg was reverse transcribed using the Superscript III kit (Invitrogen, CA, USA). The cDNA was subjected to real-time PCR amplification using iQ SYBR Green Supermix and a Opticon™2 real time PCR machine from Bio-Rad, Hercules, CA. The sequences of primers for amplification of the cDNA of interest and the control GAPDH are listed in the additional file 8.

31GEO accession code for the microarray data: GSE24261

32The authors declare that they have no competing interests.

33Conceived and designed the experiments: SKB and BF. Performed the experiments: SKB, OK, YG, AEG, SS and DJP. Bioinformatics analysis: RRS and DM. Analyzed data and wrote the paper: SKB, GAC and BF. All authors have read and approved the final manuscript.