Chromatin Structure Predicts Epigenetic Therapy Responsiveness in Sarcoma

This article is featured in Highlights of This Issue, p. 219

Epigenetic modifying agents such as DNA-methyltransferase inhibitors (DNA-MI) and histone deacetylase inhibitors (HDACI) are independently Food and Drug Administration approved for the treatment of hematopoietic malignancies such as myelodysplastic syndromes and cutaneous T-cell lymphomas, respectively (1). However, there is no obvious mechanistic link between epigenetic modification and either of these two tumor types that would necessarily exclude other cancer subtypes from responding on the basis of a similar concept of epigenetic modulation, be it global or specific. Yet, early clinical trial work with these agents in solid tumors has not been as successful as in hematopoietic tumors (2). To further rationally develop a therapeutic strategy for epigenetic therapy in solid tumors, we focused on sarcomas. We chose sarcomas, as these solid tumors share many developmental features with hematologic malignancies: (i) both are mesodermally derived (3, 4); (ii) both are believed to arise from developing progenitor cells, such as mesenchymal stem cells and hematopoietic stem cells, respectively (5–9); and (iii) subtypes of both harbor characteristic gene fusions, such as BCR-ABL in chronic myelogenous leukemia (10) and COL1A1-PGFRa in dermatofibrosarcoma protuberans (4).

As for many other solid tumors, the state of genomic methylation of sarcomas has been partially examined. Synovial sarcomas and osteosarcomas have both been reported to have global changes in DNA methylation; the former most likely due to the activity of its characteristic fusion-gene product, SYT-SSX (7, 11), and the latter as part of a complex karyotype pattern (12). Uterine leiomyosarcomas have characteristic suppression of BRCA1 via DNA promoter methylation (13). In addition, DNA promoter methylation of CDKN2A (p16), MST1, MST2, and RASSF1A in various other sarcoma subtypes has been reported (14, 15). However, as is apparent from this brief summary, rarely has the DNA methylation pattern been consistent across sarcoma subtypes. Global histone acetylation patterns have not yet been examined.

Consistent with the notion that reversal of DNA-methylation silenced characteristic tumor suppressor genes may result in therapeutic yield, both DNA-MIs 5-azacytidine (vidaza) and 5-aza-2′-deoxycytidine (decitabine) have been tried in preclinical sarcoma studies. While in all reported cases the addition of a DNA-MI could reverse expression of the initially DNA-methylation suppressed gene, the effect of that reversal was not examined when studied in rhabdomyosarcomas (16), found to result in minor apoptosis when studied in osteosarcomas (17), or extremely dependent on the time of exposure and choice of cell line (18) even within the same sarcoma subtype. In contrast, preclinical data indicate that treatment of multiple sarcoma subtypes with multiple HDACIs shows significant cellular growth inhibition at (depending on HDACI) doses therapeutically active in hematologic malignancies (19–24). Combination DNA-MI/HDACI preclinical studies have shown marked synergism for the combination in some specific sarcoma cell lines and/or mouse models (25–27). However, there remains a significant lack of consistency of the response even within cell lines from the same subtype of sarcoma, which must be overcome for this therapy to develop more effectively at the clinical level.

In this report, we extend the field of epigenetic therapy to solid tumors focusing on sarcomas. We formally explore the potential therapeutic effect of HDACIs and DNA-MIs alone and in combination on a large sarcoma cell line panel and build on the observed responses to (a) identify a gene signature that predicts epigenetic synergism in sarcomas and (b) propose a mechanism by which this synergism is selectively mediated in specific situations.

A complete description, characterization, and authentication of all cell lines used (including primary source from which they were obtained, either American Type Culture Collection or primary human sarcoma tissue) within this article were recently reported by us (28). Briefly, MG63, U2OS, HT1080, SAS2, Sw1353, and SW982O were obtained from the American Type Culture Collection and authenticated by short tandem repeat analysis: MFH and LS141 from Sam Singer (Memorial Sloan-Kettering Cancer Center) and SKUT1, SKUT1B, LMS, and AX from Eva Hernando (New York University) as described (28). Once thawed, all cell lines are routinely tested for Mycoplasma and authenticated through conformity to their baseline analysis of cell morphology monitoring and growth curve analysis as defined by us (28) upon original acquisition.

Detailed information on cell lines, growth conditions, and properties have all been previously described by us (28). MTS assays: On day 0, a total of 5,000 cells in 100 μL were plated per well in 96-well plate. At 24 hours, drug (or solvent as control) was added in increasing concentrations and/or varying combinations to a total of 200 μL of media per well in triplicates. On day 4, 80 μL of media was removed from each well and 20 μL of CellTiter 96 Aqueous One Solution Assay (Promega) was added to each well. The plate was incubated at 37°C for 1 hour and then read at 490 nm on a plate reader. Cell viability was calculated by dividing treated well absorbance by control well absorbance, with control arbitrarily defined at “1” or 100%. romidepsin, LBH589, and belinostat were kindly provided by Owen O'Connor (New York University). Valproate and decitabine were purchased from Sigma and SAHA from Toronto Research Chemicals.

One million sarcoma cells in 250 μL PBS were resuspended in 250 μL Matrigel (BD Bioscience) and introduced subcutaneously in the flank of NOD/SCID mice (Jackson Labs). Treatments were initiated when tumors reached a palpable nodule (4–5 mm in greatest dimension or approximately 20 mm² in cross-sectional area). Mice were treated on the following schedule: M, W, F via intraperitoneal injections for approximately 3 weeks, SAHA: 50 mg/kg, valproate: 300 mg/kg, decitabine: 1 mg/kg. Experiments were deemed concluded when the untreated mouse xenograft controls reached the largest cross-sectional diameter of 10 to 12 mm (or a tumor cross-sectional area of 100 mm²).

The mice were randomized to no treatment (control) or treatments as indicated in the results. Each experimental group consisted of 5 animals. The entire experiment as depicted in the article was carried out 3 times; thus, 15 animals were included per arm in the final analysis depicted in Figure 2. The size of tumors were measured individually twice per week with microcalipers. Tumor size was calculated as a cross-sectional area defined as length × width. The individual relative tumor size was calculated as the ratio of the initial tumor at which therapy was initiated to the tumor size on a given day. Statistic analyses were conducted using one-way ANOVA for comparison of tumor growth in different treatment groups.

All mice were sacrificed within 24 hours of final treatment in accordance with Columbia University Animal Welfare and the Institutional Animal Care and Use Committee policy under IRB protocol AAAD9669, and all tumors were surgically excised. Tumors were fixed in 10% formalin, embedded in paraffin, and stained with hematoxylin and eosin as previously described (28). To quantitatively assess the qualitative immunohistochemical findings, 3 pathologists with significant sarcoma expertise (Carlos Cordon-Cardo, MD, PhD; Mireia Castillo-Martin, MD, PhD; and Fabrizio Remotti, MD, Columbia University) independently scored in all a minimum of 3 sections from each xenograft for percentage of positively staining cells for each marker. Hematoxylin and eosin was graded as percentage cellularity per visualized field. Averages of all were scores obtained and plotted.

All protein gels were run as sequential immunoprecipitation-immunoblotting (IP-IB), using the same antibody for both steps to increase sensitivity of detection as previously described (29). Antibodies: AV40323 (Sigma); anti-CUGBP2 (Abcam), mouse anti-RHOJ monoclonal antibody. Immunohistochemistry (IHC) was conducted as previously described by us using the following antibodies: Ki67 (Dako; Cat # M7240), p21/waf1 (Calbiochem; Cat # OP64), and AcH3 (Cell Signaling; Cat # 9677S).

Total RNA was extracted from the indicated cell lines using an RNA extraction kit (Qiagen) according to the manufacturer's instructions. One microgram of RNA was transcribed into cDNA using Super-Script III First Strand Synthesis System for RT-PCR (Invitrogen). To assess the expression levels of CUGBP2, RHOJ, SLC2A13, and SMPDL3A, quantitative RT-PCR (qRT-PCR) reactions containing 500 ng cDNA, QuantiTect SYBR Green PCR mix (Qiagen), and corresponding primers were carried out on a Stratagene MX3000P QPCR System with ROX reference dye (Stratagene). Expression levels of the abovementioned genes were normalized to expression levels of the β-actin control gene per sample. All experiments were carried out twice and each time in triplicate. The t test was used to assess the significance of differences in expression between indicated treatments. Primers sequences were as follows: CUGBP2: CCTTCCACAGGAGTTTGGAGA and TCAGTAAGGTTTGCTGTCGTT (30); RHOJ: GCTACGCCAACGACGCCTTC′ and TTGAGCTCGGGGACCCATTC (31); and β-actin as previously described by us (6). Chromatin immunoprecipitation (ChIP) was carried out as previously described by us (6), using the following p21 promoter primers: 5′-ACCAAC GCAGGCGAGGGACT-3′ and 5′-GGT GTCTAGGTGCTCCAGGT-3′ (32).

RNA from the indicated cell lines were hybridized to HG U133 Plus 2.0 oligonucleotide arrays per standard protocols of the Columbia Genomics Core Facility (described in detail in ref. 33). Class-comparison analysis using 2-sided Student's t tests identified mRNAs that were differentially expressed between indicated samples (P < 0.05). Cluster analysis was done with Cluster 3.0 (http://linus.nci.nih.gov/BRB-ArrayTools.html) and displayed using TreeView program (http://rana.lbl.gov/EisenSoftware.htm). Raw data will be deposited in the public repository. Class prediction analysis was done using BRB Array Tools as described in the Supplementary Data.

IC₅₀ values were calculated independently by both BioDataFit and CompuSyn software. Data are presented as mean ± SD (in vitro) or mean ± SE (in vivo), and significance was assessed with Student's t test. Differences in measurements were considered significant at P < 0.05.

To explore the potential for different classes of epigenetic agents to synergize, we treated a 13-sarcoma cell line panel with 5 HDACIs (i.e., romidepsin, LBH589, belinostat, SAHA, and valproate) in the presence or absence of 1 μmol/L of the DNA-MI decitabine. [Refer to two recent reviews by Mai and colleagues and Mercurio and colleagues (34, 35) for structural and developmental history of these drugs.] Our 13 sarcoma cell lines were previously comparatively characterized and defined by us to have similar proliferative patterns (28), thus a priori discounting the possibility that differences in observed growth inhibitory effects may be attributed to differential proliferative patterns of the cell lines. All 13 sarcoma cell lines treated with the indicated HDACI exhibited little inter–cell line variation. IC₅₀ values as calculated independently using BioDataFit and CompuSyn software were as follows: romidepsin IC₅₀ ≈ 20 nmol/L; LBH589 ≈ 50 nmol/L; melinostat ≈ 1 μmol/L; SAHA ≈ 5 μmol/L; valproate ≈ 25 mmol/L using MTS as a cell growth assay and are in agreement with the IC₅₀ values of these agents for other cancer cell lines (Supplementary Table 1). These results were independently confirmed using crystal violet staining of treated cells (data not shown). Interestingly, treatment with decitabine or another DNA-DI, 5-azacytidine (vidaza; data not shown), did not result in any detectable growth inhibition at doses as high as 10-fold higher than those necessary to ensure global DNA demethylation (Supplementary Fig. 1).

Four of the 13 cell lines (MG63, MFH, SKUT1, and U2OS) showed clear synergism (compare bar pairs, Fig. 1A) for all HDACIs when combined with 1 μmol/L decitabine, further referred to as epigenetically synergistically responsive (ESR) cell lines, whereas 4 other cell lines did not (HT1080, SKUT1B, LMS, and LS141; Fig. 1A) as observed from the fractional decrease in cell growth comparing each HDACI at a given dose in the absence or presence of decitabine. Five additional cell lines (OSA, RH30, SW872, HOS, and SKNEP) showed variable synergistic responses (data not shown). In all cases, the observed synergism was confirmed using median effect equation of Chou via CompuSyn Software (refs. 36–38; data not shown).

In each case, and regardless of synergistic activity, HDACI and DNA-MI activity was independently confirmed (see Supplementary Figs. 1 and 2 and accompanying text).The enhancement of HDACI-mediated growth inhibition by the “cytotoxically” inactive agent decitabine meets the formal definition of synergism (37). Finally, treating all sarcoma cell lines with the indicated HDACI in the absence and presence of decitabine showed that on average there is a statistically significant decrease in the HDACI IC₅₀ (ranging from 20% to 80% depending on the HDACI) in the presence of decitabine (data not shown). This analysis provides rationale to suppose that even an unselected group of patients may benefit from such treatment.

To examine whether the decrease in cellular growth observed in the presence of epigenetic agents was cytostatic or cytotoxic, we carried out crystal violet staining (data not shown), viability (trypan blue exclusion), and cell-cycle and caspase activity assays on all cell lines using the 5 HDACIs (at a wide spectrum of doses) in the presence and absence of decitabine. Our results indicate that at the IC₅₀ of each respective HDACI primarily corresponds to a G₁ arrest, while at doses higher than the IC₅₀, significant apoptosis is observed (Supplementary Figs. 3–5 and accompanying text).

Our preliminary analysis did not reveal any differences in either baseline global histone acetylation or global DNA-methylation comparing ESR to non-ESR cell lines (Supplementary Figs. 1 and 2). To explore the possibility that gene-specific, rather than global, epigenetic modification may account for the observed effect, we sought to address whether the ESR cell lines may have in general a more active or open chromatin, thus being more accessible to epigenetic agents. Because our data indicated that DNA-MIs were augmenting an HDACI-mediated decrease in cellular growth, we focused on examining the status of p21, a well-studied cell-cycle inhibitor known to be responsive to HDACIs (32), in ESR and non-ESR cell lines. First, we specifically assessed whether p21 associated with the presence of the classical “histone 3 lysine 4 trimethyl” (H3K4me) active mark by carrying out ChIP using an anti-H3K4 antibody to precipitate nucleosomes and assessed for the presence of the p21 promoter (32). As could be seen in Figure 1B, only the p21 promoter in the ESR cell lines was marked by active H3K4me, indicative of an open chromatin pattern. We also examined the repressive marks H3K9 and H3K27 (39, 40). We found mixed results in that there does not seem to be a direct relationship between either of these two repressive marks and the pattern of ESR versus non-ESR cell lines at the p21 promoter (data not shown).

Because HDACIs can active the p21 promoter, we further sought to determine the acetylation status of the p21 promoter. As could be seen in Figure 1B, a similar ChIP experiment using an antibody directed against acetylated histone H4 (anti-H4Ac) indicated that at baseline the p21 promoter is not acetylated in either ESR or non-ESR cell lines (Fig. 1B, fourth pair of panels from the top). Treatment of the 8 cell lines with the HDACI SAHA (at the previously determined IC₅₀ for 24 hours) and subsequent ChIP using an anti-H4Ac antibody indicated that acetylated H4 now associated with the p21 promoter but only in the ESR cell lines (Fig. 1B, fifth pair of panels from the top). Finally, treatment of all 8 cell lines with SAHA and decitabine similarly resulted in association of only the ESR cell lines of p21 with acetylated histones (Fig. 1B, sixth pair of panels from the top). As a control decitabine alone could not induce changes in the association of p21 with the acetylated histone H4 (Fig. 1B, third pair of panels from the top).

A quantitative analysis comparing SAHA-treated versus SAHA/decitabine-treated ESR cell lines in terms of relative amount of acetylated histones associated with p21 (Fig. 1C) indicated that the combination therapy can dramatically increase the amount of acetylated histones associated with the p21 promoter. These observations indicate that mechanistically a baseline active or open chromatin structure at specific genes and/or their promoters facilitates epigenetic agents from activating those promoters and thus explains the preferential upregulation of specific genes in specific cell lines and the resultant phenotypic effect via treatment of epigenetic agents.

To assess whether our in vitro observations translated to the in vivo setting, we chose tumorigenic cell lines that, based on our in vitro data, were either an ESR cell line (SKUT1) or non-ESR cell line (SKUT1B). Both of these are morphologically similar leiomyosarcoma cell lines (28) that also share a high degree of gene expression similarity (discussed later) but experimentally differ greatly in epigenetic responsiveness. NOD/SCID mice were subcutaneously inoculated with 1 million cells. Treatments were initiated with SAHA, valproate, or decitabine alone and SAHA or valproate in combination with decitabine when tumors reached a palpable nodule (4–5mm in greatest dimension). Mice were treated on the following schedule: M, W, F via intraperitoneal injections for 3 weeks, SAHA: 50 mg/kg, valproate: 300 mg/kg, decitabine: 1 mg/kg. Mice were observed until the untreated mouse xenograft controls reached the largest cross-sectional diameter of 10 to 12 mm (or a tumor cross-sectional area = 100 mm²).

As seen in Figure 2A, the combination of HDACIs with decitabine was much more effective than any of the single-agent treatments and furthermore that the combination treatment was more effective than single-agent treatment to a statistically significant degree (see the table insert Fig. 2A). Final analysis of xenografted tumors and comparative analysis by weight confirmed the results obtained from cross-sectional measurements (Fig. 2A, right). Similar results, especially the benefit of double epigenetic agent treatment, were not observed for SKUT1B-derived xenografts (Fig. 2B). Final pathology showed no tumor necrosis in any samples, suggesting a cytostatic rather than cytotoxic effect of the treatments (hematoxylin and eosin staining; Supplementary Fig. 6) in agreement with our in vitro observations at IC₅₀ doses. Ki67, a reliable marker of sarcoma cell proliferation (see Supplementary Fig. 7 and the text) decreased in response to both valproate and SAHA and in both cell lines. However, when SAHA or valproate was combined with decitabine, Ki67 staining cells were rarely found in SKUT1 xenografts by day 5 and were completely absent on day 21. In contrast, SKUT1B cells continued to have Ki67-positive staining cells even at day 21 of treatment, although those levels were significantly reduced as compared with untreated samples (Fig. 2C, compare SKUT1 vs. SKUT1B day 5 and day 21 yellow (SAHA + decitabine) and light blue (valproate + decitabine) bars, and Supplementary Fig. 7, green and red boxes). Interestingly, in both SAHA- and valproate-treated xenografts, the acetylation of histone H3 was notably increased by the addition of decitabine, but more so in SKUT1 as opposed to SKUT1B xenografts (Fig. 2C and Supplementary Fig. 8, compare SKUT1 vs. SKUT1B day 5 and day 21 (yellow SAHA + decitabine) and light blue (valproate + decitabine) bars.)

To further explore the mechanism by which SKUT1 cells are preferentially growth inhibited in the presence of HDACIs/DNA-MIs in our xenografts, we once again turned our attention to p21. We noted that p21 was upregulated by both HDACIs (but not by decitabine) and only in SKUT1 cells (Fig. 2E and Supplementary Fig. 9, purple box). Furthermore, only SKUT1 cells could show a synergistic response, as noted by further increase in p21 in the presence of combined HDACI/DNA-MI (Fig. 2E and Supplementary Fig. 9; compare green and red boxes).

To identify genes and/or a gene pattern that may a priori predict which sarcoma cell lines show a synergistic response to HDACIs and DNA-MIs, we extracted total cellular RNA from the 4 ESR cell lines (i.e., MG63, MFH, SKUT1, and U20S) and the 4 non-ESR cell lines (i.e., HT1080, SKUT1B, LMS, and LS141) and labeled and hybridized the RNA to Affymetrix U133.0 Plus 2.0 Arrays in conjunction with the Columbia Genomics Core. Unsupervised hierarchical clustering showed no inherent synergistic related clustering pattern of the 8 cell lines (Fig. 3A). We then used multiple predictive algorithms (Supplementary Table 2) to identify genes that not only associate strongly with the ESR (cell lines shown in green)/non-ESR (cell lines shown in red) phenotype but may also identify untested samples for appropriate synergism. A 2-gene set was identified consisting of CUGBP2 and RHOJ. [Of note using Significance Analysis of Microarrays (SAM) comparing ESR cell lines with the non-ERS cell lines identified 4 genes to be significantly differentially expressed: CUGBP2, RHOJ (as identified by the predictive algorithms), and SLC2A13, and SMDDPL3A. Differences in the latter two could not be validated either by RT-PCR or immunoblot analysis (data not shown) and not further pursued.]

Unsupervised hierarchical clustering using the 2-gene predictive set could cluster the samples by the experimentally determined phenotype—ERS/non-ERS (Fig. 3B). We went on to confirm that CUGBP2 and RHOJ are significantly differentially expressed between ESR and non-ESR cell lines. using RT-PCR and immunoblot analysis (Fig. 3C and D). Similar analysis of the cell lines having a mixed response (data not shown) to combined HDACI/DNA-MI therapy (i.e., SKNEP, SW872, HOS, OSA143, and RH30) showed a lack of differential expression of these genes between the samples and the absence of the inverse patterns of CUGBP2 and RHOJ (Supplementary Fig. 10A and B) that seemed to define the ESR versus the non-ESR cell lines.

We subsequently screened an additional 5 sarcoma cell lines (CHSA-CS, SAOS2, SW982, SW1353, and AX) from our collection (28), attempting to identify sarcoma cell lines that have inverse patterns of CUGBP2 and RHOJ. SAOS2 was identified as a cell line that had relatively high levels of CUGBP2 and low levels of RHOJ at both the RNA and protein levels, suggesting that it might be a cell line that would respond to synergistic epigenetic treatment; SW982 showed the inverse pattern, suggesting a pattern similar to non-ESR cell lines (Fig. 4A and B). To formally assess the responsiveness of SAOS2 and SW982 to combinations of HDACIs and DNA-MIs, we treated these two cell lines with the 5 HDACIs (as outlined for Fig. 1) either alone or in combination with decitabine (at 1 μmol/L). SAOS2 showed a decrease in IC₅₀ values of HDACIs in the presence of decitabine whereas SW982 did not (Fig. 4C and D). Taken together, these data suggest that inverse levels CUGBP2 and RHOJ can correlate sarcoma cell lines that are either ESR or non-ESR.

Next, we assessed the changes to CUGBP2 and RHOJ following epigenetic therapy. CUGBP2 decreased and RHOJ increased in response to combined epigenetic therapy (Supplementary Fig. 10C and D); however, interestingly, this reversal occurred only in the original ESR cell lines. To assess whether CUGBP2 and RHOJ were markers or effectors of synergistic response, we downregulated CUGBP2 in SAOS2 cells and RHOJ in SW982 cells using a siRNA approach; in essence, reversing the ratio of CUGBP2 and RHOJ levels in these cell lines. Treating each cell line with siRNAs targeting either CUGBP2 or RHOJ could downregulate the intended target while not affecting the nonintended target, thus confirming specificity of the siRNAs (Supplementary Fig. 11A and B). Treating the siRNA pretreated cells with 5 HDACIs alone and in combination with decitabine yielded almost identical results (Supplementary Fig. 11C and D) to those seen in the absence of siRNA treatment. The lack of change to epigenetic treatment in growth inhibition following the reversal of the CUGBP2 and RHOJ ratios in these cell lines suggests that these genes are markers rather than mediators of the epigenetic response.

To assess whether our findings are clinically relevant, we examined the expression profiles CUGBP2 and RHOJ in a large publicly available sarcoma gene expression database (International Genomics Consortium; https://expo.intgen.org/geo/home.do). As seen in Figure 5A, 13 of 44 (29.5%) primary human sarcomas (bottom dashed box) have a CUGBP2/RHOJ ratio comparable with those seen in the ESR cell lines. Furthermore, an IHC analysis of CUGBP2, RHOJ, and H3K4 levels of an extensive tissue microarray (previously described by us in refs. 6, 39) shows that a high CUGBP2/low RHOJ ratio is detected in 23 of 41 (56.5%) high-grade sarcomas (Fig. 5B, asterisks) and correlates tightly with the exception of two cases (Fig. 5B, marked “@”) with detectable H3K4. These data suggest that combined HDACI/DNA-MI treatment is likely to be most effective for the more aggressive, high-grade sarcomas.

Our data suggest that (a) HDACIs are potential potent cytostatic drugs in the treatment of sarcomas; (b) a 2-gene (CUGBP2, RHOJ) signature can predict the response of sarcoma cells to synergistic (i.e., HDACI/DNA-MI) epigenetic treatment; high CUGBP2/low RHOJ = responsive; low CUGBP2/high RHOJ = nonresponsive; (c) HDACI/DNA-MI–mediated synergism is mediated via concurrent activation of the CDKI p21 and subsequent downregulation of CDK4 and G1 arrest; (d) mechanistically, cell line–specific synergism is dependent on a gene-specific, not global, baseline DNA and histone epigenetic modification pattern; and (e) combined HDACI/DNA-MI would be most effective for the more aggressive, high-grade sarcomas.

Our conclusion that a 2-gene signature can predict epigenetic agent synergism is based on the following 4 points: (i) gene expression comparisons of cell lines that were experimentally determined to exhibit HDACI/DNA-MI synergism with those that did not identify CUGBP2 and RHOJ as 2 of the 4 differentially regulated genes and the only 2 genes that also display the inverse pattern at both the qRT-PCR and immunoblot levels; (ii) gene expression analysis of cell lines that showed mixed responses to combination therapy did not exhibit the CUGBP2/RHOJ inverse pattern; (iii) a screen of additional sarcoma cell lines identified one cell line with high CUGBP2/low RHOJ, and one cell line with low CUGBP2/high RHOJ; both cell lines responded to synergistic epigenetic treatment as predicted, thus validating our initial observations; and (iv) in vivo treatment of xenografted cell lines displaying responsiveness/nonresponsiveness to synergistic treatment responded similarly in vivo. Furthermore, our data would suggest that the inverse pattern of CUGBP2/RHOJ is a predictive marker of response rather than the mediator of that response. This conclusion is based mainly on our observations that siRNA knockdown strategies either against CUGBP2 in an attempt to downregulate CUGBP2 in a cell line (SAOS2) that is initially responsive to epigenetic synergism in order to make it nonresponsive; or vice versa, siRNA against RHOJ, in an attempt to make a cell line that is initially nonresponsive to epigenetic treatment become sensitive, did not result in the intended effect. It is, of course, possible that reversing the ratios is not sufficient and what is really needed is absolute levels; in other words, not only does CUGBP2 need to be reduced but also RHOJ needs to be exogenously overexpressed in SAOS2 cells to make these cells nonresponsive to epigenetic synergy. Functionally, relatively little is known about these genes. RHOJ is a poorly studied RAS homologue gene family member (40) and CUGBP2 is the CUG triplet-repeat, RNA-binding protein 2 involved in RNA splicing (30, 41). Ingenuity pathway analysis did not identify any potential link, direct or remote (data not shown). At this point, the lack of functional association between these two genes, and the fact that neither one alone could even subtly change the responsive pattern to epigenetic treatment, argues against these two genes serving as mediators of the epigenetic synergism.

From a mechanistic point of view, our data indicate that HDACI/DNA-MI–mediated synergism is mediated via concurrent activation of the CDKI p21. Our data directly show that p21 is upregulated via HDACIs and occurs in only those cells that have an open or active chromatin structure (as identified by the H3K4 mark). In cells not having this active mark, HDACIs seem to be incapable of upregulating p21. Upregulation of p21 seems to be more effective in decreasing CDK4 activity and likely explains the growth inhibitory effect observed in our MTS assays. Simultaneous treatment with both HDACIs and DNA-MIs increases the overall CDKI level compared to those needed to significantly inhibit CDK4 activity (and possibly others) and lead to G₁ arrest.

Perhaps, most interesting fact is that our observations explain the mechanism underlying the specific cell line preferential response to combined epigenetic therapy. Up to this point, it is possible that the observed synergism of HDACIs and DNA-MIs may be explained by the independent and additive upregulation of p16 and p21. However, our observations that combined HDACI/DNA-MI treatment could specifically increase the amount of acetylated histones associated with the p21 promoter suggest (Fig. 1) that HDACIs and DNA-MIs synergize mechanistically on the promoter of their targets (as previously reviewed by us in ref. 42) and by others in refs. 43, 44. Our observation that this occurs only in cell lines with open or active chromatin structure further leads to the mechanistic understanding that cell lines with chromatin configurations permissive or accessible to epigenetic modifying agents are the cell lines/samples that are likely to benefit from this treatment.

Finally, through an examination of primary human sarcomas, we show that a high CUGBP2/low RHOJ pattern is present in the majority of high-grade sarcomas and correlates strongly with H3K4 markings, suggesting that the most aggressive tumors would be the ideal targets of combined HDACI/DNA-MI therapies.

In conclusion, in this report, we not only describe the cytostatic effect of common epigenetic agents on sarcomas but also identify genes and propose a mechanism of synergistic action that accounts for the preferential response of specific sarcoma cell lines. We believe that our data add to our understanding of epigenetic therapy and would be appropriate to be used in designing clinical trials using combination epigenetic therapies, as the majority of primary high-grade sarcomas have this gene signature. Although there have been several reports considering genes that might predict responsiveness to either HDACIs or DNA-MIs, to the best of our knowledge, this is the first report of genes and a mechanistic understanding that may predict responsiveness to combined epigenetic therapy. Our data further suggest that scrutiny of chromatin structure may itself be used to predict epigenetic responsiveness and that perhaps chromatin configuration itself may be further modifiable to make unresponsive tumors responsive. Although our work has centered on sarcomas, there is no clear reason why these observations may be not be equally applicable to other solid tumors.

No potential conflicts of interest were disclosed

The authors greatly acknowledge Carlos Cordon-Cardo, MD, PhD, for his critical reading, review, and general support of this work. We also thank Mireia Castillo-Martin, MD, PhD, and Fabrizio Remotti, MD, for assistance with validation of pathology review.

This work was carried out in the laboratory of Igor Matushansky and supported by grants to I. Matushansy from the NIH/NCI, Columbia's Irving Institute's Clinical and Translational Science Award, and the Gerstner Development Award.

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.