Aberrant Retinoblastoma (RB)-E2F Transcriptional Regulation Defines Molecular Phenotypes of Osteosarcoma*

Background: Gene expression signatures define prognostically significant osteosarcoma phenotypes. Results: Deregulation of the RB-E2F pathway establishes more aggressive phenotype. Inhibitors of DNA and chromatin remodeling promote comparable transcriptional changes as genetic restoration of RB. Conclusion: Aberrant RB-E2F pathway alters epigenetic landscape and biological behavior of osteosarcoma. Significance: Epigenetic remodeling regulated by RB-E2F gives rise to patterns of gene expression that are associated with different biological behavior and progression of osteosarcoma. We previously identified two distinct molecular subtypes of osteosarcoma through gene expression profiling. These subtypes are associated with distinct tumor behavior and clinical outcomes. Here, we describe mechanisms that give rise to these molecular subtypes. Using bioinformatic analyses, we identified a significant association between deregulation of the retinoblastoma (RB)-E2F pathway and the molecular subtype with worse clinical outcomes. Xenotransplantation models recapitulated the corresponding behavior for each osteosarcoma subtype; thus, we used cell lines to validate the role of the RB-E2F pathway in regulating the prognostic gene signature. Ectopic RB resets the patterns of E2F regulated gene expression in cells derived from tumors with worse clinical outcomes (molecular phenotype 2) to those comparable with those observed in cells derived from tumors with less aggressive outcomes (molecular phenotype 1), providing a functional association between RB-E2F dysfunction and altered gene expression in osteosarcoma. DNA methyltransferase and histone deacetylase inhibitors similarly reset the transcriptional state of the molecular phenotype 2 cells from a state associated with RB deficiency to one seen with RB sufficiency. Our data indicate that deregulation of RB-E2F pathway alters the epigenetic landscape and biological behavior of osteosarcoma.

We previously identified two distinct molecular subtypes of osteosarcoma through gene expression profiling. These subtypes are associated with distinct tumor behavior and clinical outcomes. Here, we describe mechanisms that give rise to these molecular subtypes. Using bioinformatic analyses, we identified a significant association between deregulation of the retinoblastoma (RB)-E2F pathway and the molecular subtype with worse clinical outcomes. Xenotransplantation models recapitulated the corresponding behavior for each osteosarcoma subtype; thus, we used cell lines to validate the role of the RB-E2F pathway in regulating the prognostic gene signature. Ectopic RB resets the patterns of E2F regulated gene expression in cells derived from tumors with worse clinical outcomes (molecular phenotype 2) to those comparable with those observed in cells derived from tumors with less aggressive outcomes (molecular phenotype 1), providing a functional association between RB-E2F dysfunction and altered gene expression in osteosarcoma. DNA methyltransferase and histone deacetylase inhibitors similarly reset the transcriptional state of the molecular phenotype 2 cells from a state associated with RB deficiency to one seen with RB sufficiency. Our data indicate that deregulation of RB-E2F pathway alters the epigenetic landscape and biological behavior of osteosarcoma.
Osteosarcoma is a genetically complex, heterogeneous disease that occurs naturally in humans and dogs (1)(2)(3). In the past decade, the molecular basis of osteosarcoma has received significant attention, and a number of recurring chromosomal aberrations and changes in gene expression have been identified (1)(2)(3)(4)(5). However, these findings have not yet translated into significant improvements in disease prognosis or outcome (2, 6 -8), placing osteosarcoma among the "most wanted" for new and effective therapies (9).
The ability to prospectively identify patients whose tumors have distinct gene expression profiles (molecular phenotypes) associated with clinical outcomes may offer insights to develop new therapeutic strategies adapted to tumor behavior. The conservation of disease mechanisms between canine and human osteosarcoma supports using the former as a comparative model to achieve this goal (2, 5, 10 -12). Previously, we identified a gene signature consisting of ϳ250 genes that stratified canine osteosarcoma into subgroups predictive of patient outcome (5,13). The gene signature and its prognostic value were conserved in tumors from human osteosarcoma patients (5). Tumors from patients with longer survival (henceforth called "molecular phenotype 1") were characterized by decreased expression of genes associated with G 2 /M transition and DNA damage-induced cell cycle checkpoints. Conversely, decreased expression of genes associated with microenvironment interactions was observed in tumors from patients with shorter survival (henceforth called "molecular phenotype 2"). In the present study, we characterized mechanisms that are causally related to these distinct molecular phenotypes. Specifically, we show that deregulation of the RB-E2F 3 pathway is a major feature of molecular phenotype 2 tumors and that restoration of RB in cells from these tumors resets gene expression to a state comparable with that seen in tumors from patients with longer survival.
One mechanism of RB-dependent gene regulation is through changing chromatin structure (14,15). Thus, we hypothesized that the RB-E2F pathway might be functionally restored by pharmacologic alteration of DNA and chromatin structure. We recently reported that DNA methyltransferase (DNMT) and histone deacetylase (HDAC) inhibitors had synergistic and selective cytotoxicity effects against human and canine osteosarcoma cells (16). Here, we show that treatment using DNMT inhibitor (zebularine) with the HDAC inhibitor (vorinostat) was sufficient to alter the transcriptional state of molecular phenotype 2 cells to one resembling that seen with active RB.

Experimental Procedures
Cell Lines and Cell Culture-Canine and human osteosarcoma cell lines and the Jurkat T-cell leukemia line were established and maintained as previously described (5,16,17). OSCA-8, OSCA-32, and OSCA-40 cells were modified to stably express GFP and firefly luciferase (Luc) for in vivo experiments (18). Fluorescence in situ hybridization was used to determine the number of GFP/Luc copies in the cell lines. Morphologic appearance, doubling time, and routine viability assays were used to confirm that growth properties of the derivative cell lines were comparable with those of the parental cell lines. Luciferase activity in the parental cell lines and the GFP/Luc modified cells was measured in vitro with the dual luciferase reporter assay system (Promega, Madison, WI) (19) using a Wallac 1420 microplate reader (PerkinElmer Life Sciences). Firefly luciferase was normalized to Renilla luciferase.
Expression Vectors and Transfections-A pGL3 luciferase reporter encoding Luc downstream from a 515-bp AURKB promoter was a kind gift of Dr. Masashi Kimura (Gifu, Japan) (20). The 515-bp sequence contains full AURKB promoter activity. Constructs encoding wild type, N-terminal truncated RB (WT RB) or a cyclin-dependent kinase (CDK)-insensitive, N-terminal truncated mutant (PSM 7-LP) RB were provided by Dr. Erik S. Knudsen (Dallas, TX and San Diego, CA) (21). Expression vectors encoding wild type p16 or p21 have been described (19,22). pGL4.73 hRenillaLuc/SV40 vector was purchased from Promega, and empty CMV-Neo-Bam vector was purchased from Addgene (Cambridge, MA). Expression vectors were mixed with a 1/100 molar equivalent of hRenillaLuc/ SV40 vector in 20 l of supplemented SE solution (Lonza, Basel, Switzerland). These mixtures were added to 200,000 cells, which were then transfected using the Lonza 4D Nucleofector. The reactions were optimized to achieve Ͼ80% viability, and transfection in all cell lines was monitored by GFP expression. Luciferase activity was measured using the dual luciferase reporter assay system.
Site-directed Mutagenesis-PCR site-directed mutagenesis of the [GGCGGG] E2F binding site in the AURKB promoter was done using the mutagenic primers reported (20) and the GENEART site-directed mutagenesis system (Life Technologies, Inc.).
RNA Preparation and Real Time Quantitative RT-PCR (qRT-PCR)-RNAs were prepared using the miRVANA kit (Life Technologies, Inc.), and cDNA was synthesized from total RNA using a miScript reverse transcription kit. cDNAs were quantified using the miScript SYBR Green PCR kit (Qiagen) and the 7500 Real Time PCR system (Applied BioSystems, Foster City, CA) protocol. Previously published primer sequences were used (24). GAPDH was used for normalization, and relative levels of mRNA were established using the ⌬⌬Ct method.
Inhibition of DNA Methylation and of Histone Deacetylation-Canine OSCA-40, OSCA-78, and OSCA-32 cells were cultured in the presence of 1 M suberoylanilide hydroxamic acid (SAHA/vorinostat; Cayman Chemical, Ann Arbor, MI) and 10 M zebularine (Zeb; Sigma-Aldrich) as previously described (16) Chromatin Immunoprecipitation-ChIP assays were performed using the ChIP-IT Express kit (Active Motif, Carlsbad, CA). Briefly, cells were cross-linked in culture medium containing 1% formaldehyde, lysed, and then sheared to an average size of 250 -500 bp by sonication in shearing buffer using a Branson sonicator (Thomas Scientific, Swedesboro, NJ). ChIP was performed by incubating 25 g of chromatin/reaction with protein G magnetic beads and 5 g of anti-E2F1 antibody purchased from Abcam (catalog no. ab112580; Cambridge, MA), anti-human RB antibody (catalog no. OP66-100UG; EMD Millipore), or control IgG overnight at 4°C. Immunoprecipitated chromatin was purified by magnetic separation, and proteins were digested with proteinase K and enrichment of E2F1 sequences. To amplify the GGGCGG (CDE site) sequence of the human AURKB (AC135178.13) promoter, the following primers were used: 5Ј-GAGCCAATGGGAACTAGGCA (forward) and 5Ј-CCCTGGCCAAGGACTTTTCA (reverse). To amplify the TTTCCAGCCAAT E2F binding site in canine AURKB (NC_ 006587.3), the following primers were used: 5Ј-TTGGGTCC-CAAGGTCTACGT (forward) and 5Ј-AGGCCCTTTCAAAT-CTCCCG (reverse). To amplify the CGGCGCTAAA E2F binding site in canine CHEK1 (NC_006587.3), the following primers were used: 5Ј-TTGGGTCCCAAGGTCTACGT (forward) and 5Ј-AGGCCCTTTCAAATCTCCCG (reverse). For all primer pairs, PCR was performed at 60°C, annealing temperature for 40 cycles. For each sample, fold enrichment of target sequence in ChIP samples versus negative control was calculated by the ⌬Ct method. All ChIP reactions were performed in duplicate. The data represent the means Ϯ S.D. of fold enrichment.
Gene Expression Profiling-Hybridization to canine 4 ϫ 44, 000 microarray chips (Agilent Technologies, Santa Clara, CA) was done as described at the University of Minnesota Genomics Center (5,18). Probe signal levels were quantile-normalized and summarized as previously described (5) (data archive submitted to the Gene Expression Omnibus). Two group t tests were done to determine differentially expressed genes.
Identification of Transcriptional Regulators-The ingenuity pathway analysis (IPA) suite (Ingenuity Systems, Redwood City, CA) was used to identify potential driver upstream transcriptional regulators responsible for gene signatures or differentially expressed genes. IPA upstream regulator analysis is based on prior knowledge of predictable effects between transcriptional regulators and their target genes stored in the Ingenuity Knowledge Base. IPA provides two statistical measures: the p value and regulation Z score to detect potential upstream transcriptional regulators. First, the p value was calculated based on how many known targets of each transcriptional regulator were present in the gene signature. Second, the known effect (repression or activation) of a transcriptional regulator on each target gene was compared with the observed changes in gene expression in the signature. A Z score was calculated from the concordance of the known effects of transcriptional regulators and the observed changes in gene expression. A Z score of Ͼ2 indicated activation of the transcriptional regulator, whereas a Z score of ϽϪ2 indicated repression of the transcriptional regulator. The predicted upstream regulators were limited to those known to be a "transcriptional regulator" or a "group." DNA Motif Identification-The hg19_genes_2012-03-09 GTF file (University of California, Santa Cruz Genome Browser) was used for retrieval of Ϫ1000 to ϩ1 nucleotide regions relative to the predicted ATG translation start site for each ORF (25,26). The 5Ј promoter sequences of 143 genes of the G 2 /M cell cycle transition and DNA damage cluster and 108 genes of the microenvironment interactions cluster (5) were available for motif discovery using the Multiple Expectation Maximization for Motif Elicitation Suite (version 4.9.0) in the Galaxy platform (27). Motifs with zero or one occurrence in each promoter and a length of between 5 and 10 nucleotides were identified.
Orthotopic Model of Canine Osteosarcoma Cell Lines-Procedures using laboratory animals were done according to the guidelines and under the supervision, of the University of Minnesota Institutional Animal Care and Use Committee (protocol 1207A17293). Six-week-old (ϳ20 grams) female athymic nude mice (NCr-nu/nu; NCI, National Institutes of Health, Fredrick, MD) anesthetized with xylazine (10 mg/kg) and ketamine (100 mg/kg) were injected intratibially with OSCA-8, OSCA-32, or OSCA-40 cells (10 5 per mouse). Buprenorphine (0.075 mg/kg q.8 h) was used for pain control over the first 24 h, and Tylenol administered in the water was used as needed for pain control thereafter. Routine tumor end points (ill thrift, or a tumor reaching 1 cm in the largest diameter for any animal in a group) or the inability to control pain or discomfort (visible lameness or difficulty moving in the cage) triggered termination of the experiment and humane euthanasia of the mice for that group. Tumor growth was monitored using caliper measure-ments and in vivo imaging as described (18). For histological confirmation, tumors were collected immediately upon sacrifice, fixed in 10% neutral buffered formalin, and evaluated grossly and histologically by board-certified veterinary pathologists (Masonic Cancer Center Comparative Pathology Shared Resource Core).
Statistical Analysis-Graphs were created using Prism (version 5.0; GraphPad Software, Inc., La Jolla, CA). The results are presented as the means Ϯ S.D. Student's two-tailed t test was used to assess significance. p values Ͻ 0.05 were considered significant.

Deregulation of the RB-E2F Pathway Is Associated with Molecular Phenotype That Predicts Worse Clinical Outcomes-
We anticipated that one or few upstream transcriptional regulators were likely responsible for the previously published, observed expression changes that segregate osteosarcoma samples into two distinct molecular phenotypes predictive of tumor behavior and outcome ( Fig. 1A) (5). To identify potential candidates, we used upstream regulator analysis within the IPA suite. The direction of gene expression changes in the molecular phenotype 2 samples (shorter median survival times) was consistent with inactive RB and p53 tumor suppressor genes ( Table 1, activation Z scores of Ϫ3.801 and Ϫ3.791, respectively). Other significant, predicted altered regulators included E2F transcription factors and chromatin remodelers: E2F-1, E2F-2, E2F-3, E2F4, E2F6, SMARCB1, KDM5B, and HDAC1. E2F4 was the most significantly altered transcriptional regulator of the gene signature (Table 1). However, because E2F4 up-regulates some of these genes and down-regulates others, a direction of activity (Z score) could not be determined.
An important role for E2F regulation of the target genes in the gene signature became more evident when we searched for conserved DNA response elements in 5Ј upstream promoter sequences. More than 70% of the 5Ј upstream promoter sequences of the genes comprising the prognostic signature contained the E2F consensus binding motif sequence, CCAG-GCTGG (data not shown). The CCAGGCTGG sequence was present in 106 of 143 promoters of genes in the G 2 /M cluster (E value 8.1E-105) and in 80 of 108 promoters of the genes associated with microenvironment interactions (E value 5.0E-32) (5). Importantly, the 9 base pair dyad sequence also is one of the most common sequence motifs in promoters of E2F4 target genes (28).
Orthotopic Xenografts Using Cell Lines Derived from Molecular Phenotype 1 and Molecular Phenotype 2 Osteosarcomas Recapitulate Their Clinical Behavior-To test the tumorigenic potential and outcome of cells from tumors from each molecular phenotype and to establish suitable cell lines for downstream functional studies, we evaluated tumor growth in orthotopic xenografts. The pattern of outcomes and tumor behavior was maintained in vivo: the OSCA-32 cell line, which was derived from a molecular phenotype 1 tumor, progressed more slowly and generated less local bone destruction than the OSCA-40 and OSCA-8 cell lines, which were derived from dogs with molecular phenotype 2 (shorter median survival times) tumors (Fig. 1B). Microscopic findings for these tumors were consistent with clinical outcome. The OSCA-32 tumor cells showed relatively well differentiated tumor cells laying down osteoid seams in an orderly fashion (Fig. 2, A and C); in contrast, OSCA-40 had highly anaplastic cells embedded in a poorly organized osteoid matrix and extensive areas of necrosis (Fig. 2, B and D).
Characterization of Representative Cell Lines of Each Osteosarcoma Phenotype-The results from the orthotopic xenografts supported the use of these cell lines to elucidate pathogenetic mechanisms responsible for the biological behavior of the two molecular phenotypes of osteosarcoma. Deregulation of E2F transcriptional activity could result from direct or indirect mechanisms upon loss of or reduced RB function. This loss of or reduced RB function, in turn, might be due to mutations that decrease or eliminate RB-1 expression or that render the RB protein inactive; however, RB is a component of a complex pathway, and its activity can be influenced by several regulatory factors (29). The steady state levels of RB protein were reproducibly lower in cell lines derived from molecular phenotype 2 tumors (OSCA-40, OSCA-78, and OSCA-8) as compared with those seen in the cell line derived from a molecular phenotype 1 tumor (OSCA-32) (Fig. 3A).
To test the hypothesis that functional deregulation of RB-E2F transcriptional regulation was causally related to the osteosarcoma molecular phenotypes, we measured the effect of RB protein abundance on expression levels of AURKB (aurora kinase B), a target gene that was among those most highly expressed in molecular phenotype 2 samples (5). As shown in Fig. 3B, differential expression of AURKB between molecular phenotypes 1 and 2 was maintained; molecular phenotype 1 cells had lower expression levels of AURKB, whereas molecular phenotype 2 cells had higher expression levels, providing the rationale to next assess the activity of a AURKB-Luc reporter construct (20). The AURKB 515 base pair minimal promoter alone showed basal activity in all four osteosarcoma cell lines that was consistent with observed endogenous AURKB transcript abundance (Fig. 3C). Reporter activity was not further enhanced when we used an AURKB promoter that included 1000 base pairs upstream from the transcriptional start site (20), indicating that the full complement of activity for this promoter in osteosarcoma cells was contained within the minimal promoter sequence. Ectopic RB Partially Rescues the Effects of RB-E2F Deregulation in Vitro-The minimal AURKB promoter reporter does not contain the CCAGGCTGG sequence or the canonical E2F binding DNA motif (TTTCCCGC). However, it contains the cell cycle-dependent element (CDE; GGGCGG) that is responsive to E2F-mediated transcriptional activation (20,30). To evaluate the role of RB-E2F1, we used both the AURKB-Luc reporter and a CDE mutant AURKB-Luc reporter.
Canine osteosarcoma cell lines were co-transfected with AURKB reporter and an N-terminally truncated, RB pocket protein domain (WT RB) or the same construct containing seven mutations that render the protein insensitive to CDK inactivation (PSM-7LP RB). Introduction of a CMV empty vector control (21) did not alter the activity of the AURKB reporter, whereas ectopic expression of RB inhibited AURKB promoter activity in the four cell lines (Fig. 4, A and B). However, only the CDK-insensitive RB construct decreased the activity of the AURKB reporter in OSCA-32 cells, and the effect was modest when compared with that observed in the OSCA-40, OSCA-78, and OSCA-8 cell lines, where both RB plasmids showed approximately equal repression (Fig. 4B).
Repression by ectopic PSM-7LP RB was significantly attenuated in each of the four cell lines when we used a CDE mutant AURKB promoter, although partial repression was still observed in the molecular phenotype 2 cell lines (Fig. 4C). ChIP analysis confirmed that endogenous canine E2F1 binds to the CDE in the AURKB reporter and that E2F1 had lower affinity for the mutant CDE site (Fig. 4C).
The effect of ectopic PSM-7LP RB to suppress the AURKB reporter also was rapidly saturable in OSCA-32 cells, consistent with the presence of active endogenous RB (Fig. 4D). In contrast, ectopic PSM-7LP RB showed dose-dependent suppression in OSCA-40, OSCA-78, and OSCA-8 cells (Fig. 4D), as would be predicted by absence of endogenous functional RB.
Ectopic RB Displaces E2F1 from Endogenous Promoters-Given the repressive effect of ectopic PSM-7LP RB on the ectopic AURKB vector in cells representing the two molecular phenotypes of osteosarcoma, we investigated whether these effects were relevant and reproducible in the endogenous context. Ectopic PSM-7LP RB protein was detectable in all of the cell lines (Fig. 5A), and the presence of ectopic RB consistently reduced transcript abundance of AURKB and three other genes (AURKA, BUB1B, and TOP2A) that were part of the signature that identified osteosarcoma molecular phenotypes (5) GAPDH transcript abundance was not affected by the presence of ectopic PSM-7LP RB (Fig. 5B).
To further validate the importance of the E2F-dependent interactions in the study context, we next confirmed that E2F1 was bound to the endogenous TTTCCAGCCAAT motif in the canine AURKB promoter and that its presence on the promoter contributed to transcription of AURKB. ChIP assays (Fig. 6A) showed greater enrichment for E2F1 bound to endogenous AURKB promoter in OSCA-8 cells (molecular phenotype 2) than in OSCA-32 cells (molecular phenotype 1). Furthermore, E2F1 binding to the promoter in both molecular phenotypes was reduced in the presence of ectopic PSM-7LP RB. Quantitative assessment of AURKB transcript abundance by qRT-PCR was consistent with what was observed in our ChIP data, showing that ectopic PSM-7LP RB caused a greater magnitude of reduction of AURKB transcript in OSCA-8 cells versus OSCA-32 cells (data not shown). We observed a similar effect upon analysis of E2F1 binding to the CGGCGCTAAA motif of the endogenous CHEK1 promoter (Fig. 6B), illustrating that ectopic PSM-7LP RB was repressing the binding of E2F1 protein to genes associated with G 2 /M progression and the DNA damage checkpoint.
Importantly, we did not see any difference in E2F1 protein abundance in the cells transfected with either the ectopic PSM-7LP RB or CMV plasmids, suggesting that ectopic RB was not simply reducing E2F1 protein levels. PSM-7LP RB also was not found in complexes bound to the E2F1 response elements in the endogenous AURKB promoter as determined by ChIP in OSCA-8 cells using an antibody that specifically recognized the ectopic human RB protein (data not shown). We similarly did not find evidence of complex formation between ectopic PSM- 7LP RB and endogenous E2F1 in co-immunoprecipitation assays; yet we also did not see a quantitative difference in the amount of immunoprecipitated E2F1 protein in cells from either molecular phenotype transfected with PSM-7LP RB or with the CMV empty vector, suggesting that ectopic PSM-7LP RB did not compete with or sterically hinder binding of the anti-E2F1 antibody (data not shown). Thus, we favor the interpretation that displacement of E2F1 from the AURKB promoter was an indirect effect of RB.
Ectopic RB Alters the Transcriptional Landscape of Molecular Phenotype 2 Osteosarcoma Cells toward One Resembling Molecular Phenotype 1-The selective effects of RB on molecular phenotype 2 osteosarcoma cells led us to hypothesize that restoration of RB in these cells would shift the genome-wide transcriptional state in these cells to one resembling that of molecular phenotype 1 cells. Genome-wide expression pro-filing highlighted 78 genes that were differentially expressed in OSCA-78 cells transfected with PSM-7LP RB versus the CMV control; these genes were significantly associated with functions of DNA replication, metabolism of DNA, and binding of chromatin ( Table 2). The most significant canonical pathway identified by IPA was cell cycle control (data not shown). Importantly, active RB, and inactive E2F1 and E2F1 were among the most significant predicted transcriptional regulators associated with the differential expression of these 78 genes (Table 3). In addition, IPA yielded a number of predicted inactivated oncogenes that have been shown to be aberrantly expressed in osteosarcoma, including MYC (31), as well as transcriptional regulators associated with DNA damage repair and the mitotic checkpoint, including TBX2 (32) ( Table 3). Intriguingly, the gene expression profiles resulting from restored RB activity in these cells reflected an apparent recovery of activity for the TP53 tumor suppressor gene ( Table 3).
In U2-OS cells, p16 protein is silenced by methylation of the promoter (33). Loss of p16 can lead to unrestrained activation of CDK4/6 and may impair function of RB at the G 0 /G 1 transition checkpoint, during G 1 progression and at cell cycle exit (re-entry to G 0 ). We hypothesized that p16 deficiency would not be equivalent to complete loss of RB, as other mechanisms of control are operative during progression through the S phase and the G 2 /M phase (35,36). Therefore, we tested whether CDK inhibition occurred in p16-deficient U2-OS cells upon RB activation. First, we confirmed that U2-OS cells do not express p16 protein by Western blot (Fig. 7C). We then determined whether RB was completely inactivated in U2-OS cells by culturing them under conditions of serum deprivation, which leads to growth arrest. Within 72 h of serum withdrawal, RB was predominantly present in the active, faster migrating (hypophosphorylated) form, indicating that, despite silencing of p16, CDKs can still be inhibited in U2-OS cells (Fig. 7D).
HDAC and DNMT Inhibitors Alter Genome Wide Gene Expression in Molecular Phenotype 2 Cells to a Status Associated with a Functional RB-E2F Regulatory Network (Molecular Phenotype 1)-RB protein is known to associate with HDACs and DNMTs (14,15). We reasoned that the absence of func- tional RB would hinder the function of these chromatin-remodeling enzymes and that RB activity might be restored through pharmacological modulation. To assess the effects on transcription caused by treatment with DNMT and HDAC inhibitors, we did genome-wide expression profiling of molecular phenotype 2 cells treated with Zeb and SAHA. Fig. 8 (A and  B) shows that in molecular phenotype 2 cells (OSCA-78) cells, the transcriptional state of the defining prognostic signature   NOVEMBER 20, 2015 • VOLUME 290 • NUMBER 47

Aberrant RB-E2F Defines Molecular Phenotypes
shifted to a state that resembled molecular phenotype 1 (OSCA-32) after treatment with Zeb and SAHA (5). Next, we examined whether the observed effect was associated with displacement of E2F1 from relevant promoters. ChIP showed that the amount of E2F1 bound to the endogenous AURKB (Fig. 8C) promoter was decreased by ϳ90% in molecular phenotype 1 cells and by 95% in molecular phenotype 2 cells, after treatment with these drugs. Similarly, the amount of E2F1 bound to the endogenous CHEK1 promoter decreased by ϳ80% in both molecular phenotype 1 and molecular phenotype 2 cells after treatment (Fig. 8C).
In addition, we identified 1047 statistically significant differentially expressed genes (p Ͻ 0.05 and average fold change of 2.0) in untreated and treated OSCA-78 cells. These genes clustered into two groups (data not shown). The first group of genes, which was significantly associated with functions of cell cycle progression and proliferation, was down-regulated in treated cells. The second group, which was significantly associated with functions of cellular organization, maintenance, and cell-cell interactions, was up-regulated in treated cells. As a single group, these genes were significantly associated with functions related to cell cycle, proliferation, and cancer functions (data not shown).
The IPA transcription factor module was used to predict upstream regulators of the 1047 genes that were differentially expressed between cells treated with DNMT and HDAC inhibitors and those that were not. RB was a predicted upstream transcriptional regulator (Table 4) and was predicted as being active in treated cells. The predicted activity of other upstream transcriptional regulators for these 1047 genes paralleled that observed in molecular phenotype 1 cells (Table 1), as well as that seen upon ectopic reintroduction of RB into molecular phenotype 2 cells (Table 2). SMARCB1 was predicted as being activated in treated cells, whereas E2F1, MYC, and FOXM1 were predicted as being inactivated (Table 4).

Discussion
Here we provide insight into mechanisms that account for a prognostic gene signature that reduces the heterogeneity associated with osteosarcoma. As shown in our model Fig. 9, the signature allowed us to group the disease into two subgroups (molecular phenotype 1 and molecular phenotype 2) that differ in their biological behavior; i.e. time to progression and clinical outcomes (5).
We observed that molecular phenotype 1-and molecular phenotype 2-derived xenograft tumors recapitulated the gross and histologic features of spontaneous canine osteosarcoma.

TABLE 2 Gene function enrichment analysis after RB is ectopically restored in molecular phenotype cells
Genome-wide expression profiling was used to compare OSCA-78 cells transfected with CDK-insensitive RB (PSM-7LP RB) to OSCA-78 cells transfected with CMV empty vector. Genes with a p value of Ͻ0.05 and an average fold change of 1.53 were identified for further analysis. IPA was applied to identify biological functions associated with differentially expressed genes in molecular phenotype 2 osteosarcoma RB restored cells. More importantly, we show that molecular phenotype 2 xenograft tumors appeared to be phenotypically more aggressive than molecular phenotype 1, exhibiting more rapid growth at the primary tumor site and a greater propensity for pulmonary metastasis.

Functions annotation
We determined that critical transcriptional regulators of this evolutionarily conserved signature responsible for these two osteosarcoma phenotypes are in the RB-E2F regulatory pathway. In molecular phenotype 2 osteosarcoma (which represents tumors from patients with worse prognosis), the RB-E2F pathway is dysfunctional. As a consequence, E2F-regulated genes including those involved in G 2 /M transition and DNA damage-induced cell cycle checkpoints are up-regulated, and microenvironment-interacting genes are down-regulated.
Our results show that molecular phenotype 2 cells are more responsive to RB restoration than molecular phenotype 1 cells. The observation that ectopic expression of constitutively active RB resulted in transcriptional repression of E2F targets and other genes associated with cell cycle control and DNA replication is consistent with previously defined mechanisms (15). However, assessments of individual components of the RB pathway have not always correlated with event free or overall survival (37,38). Loss of heterozygosity of RB has been proposed as an indicator of poor prognosis in human osteosarcoma patients (39), but our work is the first

Regulators of genes whose expression changes in molecular phenotype 2 osteosarcoma cells after RB is ectopically restored
Genome-wide expression profiling was used to compare OSCA-78 cells transfected with CDK-insensitive RB (PSM-7LP RB) to OSCA-78 cells transfected with CMV empty vector. Genes with a p value of Ͻ0.05 and an average fold change of 1.53 were identified for further analysis. IPA was applied to identify upstream transcriptional regulators specific to restoration of RB. A Z score of Ͼ2 indicates activation, whereas a Z score of ϽϪ2 indicates inactivation. . Differential RB-E2F activity associated with molecular phenotypes 1 and 2 is independent of p16. A, the transcriptional activity of the AURKB minimal promoter in human U2-OS osteosarcoma cells (RB wt ) and SAOS-2 osteosarcoma cells (RB mutant ) was determined using dual luciferase assays as described in Fig. 1D. to establish a direct link between a conserved, prognostic genome-wide gene expression signature and deregulation of the RB-E2F pathway in osteosarcoma (5). More specifically, current models of the RB-E2F pathway do not consistently account for expression levels of genes during the G 2 /M phase of cell cycle (40,41); yet many of the genes that are overexpressed in molecular phenotype 2 osteosarcoma, including AURKB, are associated with the G 2 /M transition. Recent data suggest that, unlike canonical E2F response elements that operate primarily in G 1 /S, the binding of E2F to the CDE is stronger in G 2 /M (30). One interesting finding from our study was that restoration of RB compensated for other common genetic alterations (e.g. TP53 and MYC) associated with osteosarcoma. The finding suggests that RB loss or deregulation can modify tumor behavior and disease progression downstream of oncogenes that are often altered in osteosarcoma. Although some of these effects could be due to direct modulation of E2F activity, it also is likely that indirect changes in the epigenetic landscape that are established by functional RB contribute to these effects. The interpretation that functional RB utilizes mechanisms that are independent of E2F binding is consistent with the observation that mutating the CDE sequence did not completely abrogate reduced AURKB activity upon reintroduction of a functional RB gene to molecular phenotype 2 cells. Additional support for our interpretation was the observation that mutating the CDE sequence also did not completely abolish E2F1 binding and that we did not see evidence of RB binding to the E2F-responsive elements in the AURKB promoter.

Upstream regulator
An important finding from our study was that treatment with DNMT and HDAC inhibitors was sufficient to supplant RB-E2F pathway function in molecular phenotype 2 cells. Treatment with these inhibitors altered genome wide gene expression in molecular phenotype 2 cells to a status associated with a functional RB-E2F regulatory network (as seen in molecular phenotype 1). Moreover, we show that treatment with DNMT and HDAC inhibitors achieved comparable transcriptional changes as genetic restoration of RB, albeit through regulation of somewhat different gene sets. Still, these gene sets were concentrated in or near control nodes for overlapping biochemical pathways, which is not entirely unexpected, given the expectation that chromatin-remodeling enzymes would have a broader effect to modulate gene expression.
The observation that p16-deficient U2-OS cells maintained at least a partially functional RB, possibly through compensation of other CDK inhibitors like p21, indicates that by itself, the status of p16 cannot explain the RB-dependent heterogeneity of osteosarcoma. Specifically, our findings show that RB-E2F pathway are still able to regulate genes associated with the G 2 /M transition in U2-OS cells, probably through inhibition of the S phase and G 2 /M CDKs. Nevertheless, CDKN2A deletion or silencing could contribute to deregulation of the E2F pathway in osteosarcoma (42). Not surprisingly, CDKN2A was a predicted transcriptional regulator of the prognostic gene signature. The CDKN2A locus was recently linked to osteosarcoma risk in dogs, and the risk allele is "fixed" in certain breeds like Rottweilers and Irish Wolfhounds (43). This finding illuminates the need to investigate an explanation for why canine osteosarcoma is so often a highly aggressive disease (5). It is widely accepted that RB inactivation is not necessary for the development of osteosarcoma but rather accelerates its development and progression (44,45). Here, we show that the integrity of the RB-E2F pathway is mechanistically associated with the biological behavior of tumor cells derived from spontaneous canine osteosarcoma in vitro and in vivo and that the RB-E2F molecular regulatory network extends to human osteosarcoma cells. Specifically, our data suggest that alternative treatment options that create a state analogous to that seen with functional RB could improve outcomes in osteosarcoma patients, especially those with the worst prognoses. Our data provide support for further evaluation of the mechanistic role of RB-E2F pathway in chromatin remodeling and the contribution of the epigenetic landscape in osteosarcoma pathogenesis.  . Working model of RB-E2F regulated gene expression in the two molecular phenotypes of osteosarcoma. In molecular phenotype 1 osteosarcoma (left side), functional RB down-regulates E2F activity and restricts expression of genes associated with cell cycle progression through its interaction with the E2F DNA binding sequences. In molecular phenotype 2 osteosarcoma (right side), RB is absent or nonfunctional and cannot form stable RB-E2F complexes, leading to deregulation of E2Fresponsive targets (indicated by black shapes in figure). Restoration of RB or treatment with DNMT and HDAC inhibitors shifts the transcriptional state of genes associated with molecular phenotype 2 osteosarcoma (rapid progression and worse prognosis) to a transcriptional state associated with functional RB and molecular phenotype 1 osteosarcoma (less rapid progression and better prognosis).

TABLE 4 Regulators of genes whose expression significantly changes in molecular phenotype 2 osteosarcoma cells after treatment with DNMT and HDAC inhibitors
Genome-wide expression profiling was used to compare OSCA-78 cells treated with DNMT and HDAC inhibitors to untreated OSCA-78 cells. Genes with a p value of Ͻ0.05 and an average fold change of 1.53 were identified for further analysis. IPA was applied to identify upstream transcriptional regulators specific to treatment with DNMT and HDAC inhibitors. A Z score of Ͼ2 indicates activation, whereas a Z score of ϽϪ2 indicates inactivation.