Osteosarcomas are highly complex and heterogeneous tumors. A large number of genes can be mutated though few are recurrent. Rather, these tumors have aneuploidy, chromothripsis, and uncontrolled cell cycle, suggesting defects in DNA repair mechanisms and/or epigenetic instability [9]. Nevertheless, the high diversity of genomic alterations may share similar functional modules related to events such as DNA damage, response to stress, epigenetic processes, mitosis and cellular motility [28].
The reversibility of epimutations turns them into interesting targets for the development of therapies. In osteosarcomas cell lines, the demethylating agent decitabine (5-aza-dC, a DNMT1-inhibitor) can promote the recovery of gene expression silenced by hypermethylation, resulting in reduced cell growth [29] or increased sensibility to chemotherapeutic agents at concentrations within therapeutic levels [30]. Therefore, DNA methylation profiles may directly point to robust disrupted mechanisms in osteosarcomas.
The genome-wide methylation characterization of 28 osteosarcomas showed a significant epigenetic heterogeneity, as shown by a coefficient of variation (CV = 0.7), much higher than other solid tumors (~ 0.08 in Ewing sarcoma; ~0.03 in glioblastoma) [20, 31]. It is an expected finding as these tumors usually present a high mutational and chromosomal alteration burden [2], particularly compared to tumors with few genetic alterations such as Ewing sarcoma [20]. It could indicate that the high DNA methylation heterogeneity is an intrinsic event of osteosarcoma tumorigenesis or that is related to the high number of cell types found in these tissues. This second hypothesis is supported by the high CV in control bone samples (0.73), what raises the possibility that DNA methylation is intrinsic of the bone/cell composition and not related to the tumorigenesis itself. We had one exception among our samples that grouped with the bone samples in the unsupervised clusterization; this sample also had few SNVs and no detectable CNAs. At a closer look at the clinical and pathological parameters, this case was not different from the others, suggesting that this biological behavior may be explained by the intratumor heterogeneity, with the sample used for the experiments being epi- and genetically not representative of the tumor as a whole.
Osteosarcomas displayed a global hypomethylation with a pattern of CpG islands hypermethylation, similar to other cancers [32, 33]. Global hypomethylation is usually related to the activation of oncogenes and increase of genomic instability, while hypermethylation of CpG islands usually results in the inactivation of tumor suppressors and other regulatory elements [34]. Consistently, among the genes contained within hypermethylated DMRs, we identified 17 tumor suppressors, from which six were also affected by alternative inactivation mechanisms such as deletions (DLEC1, GJB2, HIC1, MIR149, PAX6, and WNT5A). Silencing by aberrant promoter methylation of cell proliferation regulators and genomic stability maintainers such as DLEC1 and HIC1 is a recurrent event in a variety of cancers [35–38].
We hypothesize that the preferential hypermethylation pattern at CpG islands has a connection with DNMT3B gains in osteosarcomas and genetic perturbations related to H3K27me3 and PRC2. DNMT1 and DNMT3A also had extra copies in some samples. DNMT1 overexpression is a recurrent event across osteosarcomas, with a negative impact on the expression of several tumor suppressor genes [30]. DNMT3A and DNMT3B seem to have both redundant and exclusive targets; while DNMT3A remains active in adults [39], expression of DNMT3B catalytic isoforms decreases during cell differentiation [40]. There is evidence that DNMT3B induces DNA methylation of tumor suppressor genes such as RASSF1, CDH1, CDKN1A and CDKN2A in colorectal cancer [41], gliomas [42], bladder cancer [43], prostate cancer [44], acute myeloid leukemia [45], esophageal squamous cell carcinomas [46], breast adenocarcinoma and lung carcinoma [47]. Ectopic DNMT3B expression results in widespread CpG island hypermethylation in several cancer types available in TCGA [48]. Although we did not analyze gene expression, it is likely that there is a correlation with this variable DNA methylome.
DNMT3B-induced methylation coincides with H3K27me3 and results in methylation of the surrounding CpG island DNA, affecting PRC2 recruitment, a mechanism involved in the tumorigenesis of gliomas [49]. The involvement of H3K27me3 in osteosarcomas is strengthened by the fact that 87% of the tumors from young patients present imprinting defects at the 14q32-locus, an enriched region for both the active marker H3K4me3 and the silence marker H3K27me3 [50].
An enrichment of hypomethylated DMRs was reported at 6p22, where several histone-coding genes are located. Samples exhibited both gains and losses events encompassing 6p22, reinforcing the genomic instability. Interestingly, only gains of 6p/6p22 were reported as a recurrent event in a variety of cancers, including osteosarcomas [51, 52], with impact in tumor progression, aggressiveness and metastatic potential [53]. In agreement with the hypomethylation found in 6p22, a cluster of over-expressed genes in 6p has been reported [54].
DMRs often overlapped with CNAs in our analysis. It is possible that at least part of the DMRs' hypomethylation was induced by the unbalance between copy number gains and the reestablishment of original DNA methylation patterns. Both mutational processes are associated with high genomic instability [3, 9, 34]. Furthermore, several genes related to DNA damage response and repair, such as BAP1, BRCA1, BRCA2, PALB2, PARP1, PTEN, RB1, SETD2, and TP53, presented copy number losses or loss-of-function mutations (but no differential methylation), although these pathways were not enriched in our analyses. TP53 is not a recurrent target for epigenetic regulation, but changes in the pathways or genes modulated by its activity have already been reported [9]. Thus, further studies are necessary to disclose the joint role of DNA methylation and chromosomal alterations in the tumorigenesis of osteosarcomas, mainly targeting the gene expression patterns associated with these changes.
Methylation data from 19 OS cell lines revealed 328 genes with at least one differentially expressed CpG site located at the promoter region [54]. From these, 62 genes overlapped with our list of DMR-genes. However, considering the technical differences between the approaches (Illumina 27K and 450K Beadchip arrays), we decided to compare biological processes instead of individual genes. The enriched biological processes within the DMR-genes suggested the involvement of mechanisms associated with all steps of cellular differentiation, similar to other studies [29, 54] and consistent with the period of onset of the disease [1, 55]. Osteogenic lineages generated by the induction of differentiation from adipose-derived stem cells present a DNA methylation pattern similar to that of osteocytes, with the majority of CpG sites that change DNA methylation located at promoter-associated CpG islands [56]. Therefore, dysregulation of DMR-genes, as we described, might be associated with the interruption of the differentiation process. Accordingly, among the biological processes, skeletal development, represented by 28 genes (Table 2), was enriched within the DMRs.
We reported alterations of TCA anaplerotic substrates and a very high glycolysis rate in the peripheral blood of these same patients [27]. Alterations in intermediates of the TCA cycle (histidine and glutamine, which are 2-OG precursors; valine as a succinyl-CoA precursor; and phenylalanine and tyrosine, that can be metabolized to fumarate) can interfere with TET1 activity [57], what ultimately could result in the global DNA hypomethylation found in osteosarcomas.
We emphasize that only frozen samples with more than 70% of the cells of interest were selected for this study because the ratio of contamination of normal cells affects the data of Infinium HumanMethylation 450. This criterion resulted in the main limitation of this study, which was the lack of paired normal and tumor samples. This prevented us from safely determining the spectrum of somatic alterations, methylation, or mutation. To overcome these, we used normal bone tissues in the methylation analysis and applied restrictive data filtering criteria, excluding any variant present in population databases.