Abstract Background The study aimed to analyze aberrantly methylated genes, relevant pathways and transcription factors (TFs) in osteosarcoma (OS) development. Methods Based on the DNA methylation microarray data [35]GSE36002 that were downloaded from GEO database, the differentially methylated genes in promoter regions were identified between OS and normal samples. Pathway and function enrichment analyses of differentially methylated genes was performed. Subsequently, protein-protein interaction (PPI) network was constructed, followed by identification of cancer-associated differentially methylated genes and significant differentially methylated TFs. Results A total of 1379 hyper-methylation regions and 169 hypo-methylation regions in promoter regions were identified in OS samples compared to normal samples. The differentially hyper-methylated genes were significantly enriched in Neuroactive ligand-receptor interaction pathway, and Peroxisome proliferator activated receptor (PPAR) signaling pathway. The differentially hypo-methylated genes were significantly enriched in Toll-like receptor signaling pathway. In PPI network, signal transducers and activators of transcription (STAT3) had high degree (degree=21). MAX interactor 1, dimerization protein (MXI1), STAT3 and T-cell acute lymphocytic leukemia 1 (TAL1) were significant TFs enriched with target genes in OS samples. They were found to be cancer-associated and hyper-methylated in OS samples. Conclusion Neuroactive ligand-receptor interaction, PPAR signaling, Toll-like receptor signaling pathways are implicated in OS. MXI1, STAT3, and TAL1 may be important TFs involved in OS development. Keywords: Differentially methylated gene, Osteosarcoma, Protein-protein interaction, Transcription factor 1. Introduction Osteosarcoma (OS) is the most common primary malignant bone tumor in children and adolescents. More than 15% of patients with OS develop metastases, frequently to lung [36][1]. For patients with metastasis or recurrence, the long-term survival rate is less than 20% [37][2], [38][3]. Understanding of the molecular mechanisms of OS would provide a basis for developing new therapeutic strategies. It has been demonstrated that genetic alternations in the status of DNA methylation count among the most common molecular alterations in human neoplasia [39][4]. DNA methylation usually results in obstruction of the promoter region, hampering gene transcription and causing gene silencing [40][5]. Quite a few studies have reported findings related to DNA methylation in OS. It has been reported that methylation of frizzled-related proteins (SFRPs) may promote Wnt signaling pathway, thereby enhancing OS cell invasion [41][6]. Hyper-methylation of p14ADP-ribosylation factor (ARF) and estrogen receptor 1 (ESR1) have been found in OS as well, and may have implications in the prognosis of OS patients [42][7]. Moreover, Lu et al. [43][8] have reported that iroquois homeobox 1 (IRX1) enhances OS metastasis and may be a potential molecular marker. Recent study [44][9] suggests that promoter hyper-methylation of reversion-inducing cysteine-rich protein with Kazal motifs (RECK) is a causative factor in metastasis of OS. Despite some advances have been achieved in this field, certain mechanisms underlying OS remain largely unknown. Microarray analysis has been applied to identify the gene alterations and screen potential targets in human OS cell lines [45][10]. Kresse et al. [46][11] integrated genomewide genetic and epigenetic profiles from the EuroBoNeT panel, a European Network of Excellence on bone tumors ([47]http://www.eurobonet.eu), of 19 human osteosarcoma cell lines based on microarray technologies, and deposited the DNA methylation dataset in the Gene Expression Omnibus (GEO) data repository (accession number [48]GSE36002). In their study, they have comprehensively analyzed the relationships of DNA copy number, DNA methylation and mRNA expression in OS. Additionally, they screened out the differentially methylated genes and performed functional enrichment analysis. However, the interaction between differentially methylated genes, and the classification of these genes have not been analyzed. Since methylation of CpG islands in promoter regions is a mechanism for inactivating genes, we estimated the differentially methylated genes in promoter regions between OS and normal samples from [49]GSE36002 in this study. Pathway and functional enrichment analyses of differentially methylated genes was then performed. Subsequently, protein-protein interaction (PPI) network was constructed to analyze the interactions between differentially methylated genes. Furthermore, the classifications of differentially methylated genes, including tumor suppressor (TS) genes, oncogenes and TFs were investigated. 2. Materials and methods 2.1. DNA methylation microarray data The [50]GSE36002 DNA methylation microarray data [51][11] were downloaded from GEO database in National Center of Biotechnology Information (NCBI) ([52]http://www.ncbi.nlm.nih.gov/gds/), based on the platform of Illumina Human Methylation 27 BeadChip (Illumina Inc., California, USA). [53]GSE36002 dataset is comprised of 25 samples: 19 osteosarcoma cell lines samples and 6 normal control samples (four normal bone samples and two osteoblast cultures). 2.2. Identification of differentially methylated genes DNA methylation microarray data were processed with simple scaling normalization using Lumi package ([54]http://bioconductor.org/packages/release/bioc/html/lumi.html) [55][12], [56][13] in Bioconductor [57][14]. Subsequently, the differentially methylated regions (DMR) between osteosarcoma and normal samples were identified using methyAnalysis with M-value>1 and false discovery rate (FDR)<0.01. M-values were calculated by the formula: [MATH: Mvalue=log2methylated probe intensityunmethylated probe intensity :MATH] (1) FDR method is also called Benjamini and Hochberg (BH) method [58][15], used to adjust p value. In detail, the original p values of all genes were ranked in descending order. The maximum p value was assigned as n, and the minimum was assigned as 1. The adjusted p value (FDR) was calculated as followed: [MATH: FDR=originalpvalue*(n/i) :MATH] (2) where n represents the number of all genes; i represents the ith p value (from minimum to maximum). The identified DMR were performed gene annotation to screen the differentially methylated sites located in promoter region. Dkhil et al. [59][16] have reported that most of the promoters display the changes of DNA methylation in their Ups-regions, which are between +500 and +2000 bp upstream from the transcription start site of genes. Therefore, in consideration of numerous genes in the dataset, the promoter region was defined as 2000 bp upstream of the transcription start site in this study. 2.3. Functional enrichment analysis of differentially methylated genes Functional enrichment analyses included Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, Reactome pathways and Gene Ontology (GO) terms analyses. The differentially methylated genes were performed functional enrichment analysis using GOFunction ([60]http://www.bioconductor.org/packages/release/bioc/html/GOFunction. html) in R package. P value<0.05 and count (number of significantly enriched genes)≥2 were used as thresholds. 2.4. PPI network construction PPI analysis can provide new insights into protein function, and uncover the generic organization principles of functional cellular networks [61][17]. Therefore, we constructed PPI network to further analyze the functions of differentially methylated gene. The Search Tool for the Retrieval of Interacting Genes (STRING) database ([62]http://www.mybiosoftware.com/pathway-analysis/7789) provides the information of both experimental and predicted interactions [63][18]. With this tool, the differentially methylated gene pairs with significant interactions (combine score>0.9) were identified and used for construction of PPI network which was visualized using Cytoscape ([64]http://cytoscape.org/) [65][19]. In the network, the nodes with higher degrees were defined as hub nodes. 2.5. Gene classification analysis Based on tumor suppressor (TS) genes database [66][20] and tumor associated genes (TAG) database [67][21], the known TS genes and oncogenes were selected from the obtained differentially methylated genes. 2.6. TF analysis On the basis of the TF-target gene pairs in Encode [68][22], [69][23], the differentially methylated genes corresponding to the differentially methylated TFs were identified. Then according to the obtained TF-target gene interaction pairs, significant TFs (p value<0.01) were screened using hypergeometric analysis. The formula of hypergeometric distribution was shown as follows: [MATH: P(X=k)=CMkCNMnkCNnk< /mi>{0,1,2,m} :MATH] (3) where N represents the total number of genes; n represents the number of target genes regulated by TF; M represents the number of differentially methylated genes; k represents the number of differentially methylated genes regulated by TF. 3. Results 3.1. Identification of differentially methylated genes In the study, a total of 2845 differentially methylated loci were identified in OS samples relative to normal samples, including 1379 hyper-methylation regions and 169 hypo-methylation regions in the promoter region. Additionally, the numbers of hyper- and hypo-methylated loci in enhancers are shown in [70]Table 1, and in exons and other regions are shown in [71]Table 2. Table 1. The numbers of hyper- and hypo-methylated loci in enhancer region. Enhancer Hyper Hypo Total TRUE 43 163 206 FALSE 1490 1071 2561 [72]Open in a new tab Table 2. The numbers of hyper- and hypo-methylated loci in exon, promoter and other regions. Regions Hyper Hypo Total 1stExon 57 489 546 3'UTR 3 8 11 Body 84 619 703 IGR 12 25 37 Promoter 1379 169 1548 [73]Open in a new tab UTR: untranslated regions; IGR: intergenic region. 3.2. Functional enrichment analyses of differentially methylated genes Pathway enrichment analysis revealed that the differentially hyper-methylated genes were significantly enriched in KEGG pathways including Pathways in cancer, Bladder cancer, Neuroactive ligand-receptor interaction, and Peroxisome proliferator-activated receptor (PPAR) signaling pathways ([74]Table 3). The differentially hypo-methylated genes were significantly related to Cytokine-cytokine receptor interaction and Toll-like receptor signaling pathways ([75]Table 4). With regard to Reactome pathways, the differentially hyper-methylated genes were primarily enriched in Gastrin-CREB signaling pathway via PKC and MAPK, and G alpha (q) signaling events pathway, while the differentially hypo-methylated genes were mainly enriched in Defensins, and Alpha-defensins pathways ([76]Table 5). For GO function, the differentially hyper-methylated genes were predominately enriched in developmental process and anatomical structure development ([77]Table 6), while the differentially hypo-methylated genes were primarily enriched in extracellular region ([78]Table 7). Table 3. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (top 10) of the differentially hyper-methylated genes. KEGG ID Name Gene count P value 5200 Pathways in cancer 40 0.002712 4080 Neuroactive ligand-receptor interaction 34 0.004135 3320 PPAR signaling pathway 12 0.007642 4974 Protein digestion and absorption 13 0.009805 4512 ECM-receptor interaction 13 0.014496 5219 Bladder cancer 8 0.014911 4920 Adipocytokine signaling pathway 11 0.015917 5414 Dilated cardiomyopathy 13 0.022575 4916 Melanogenesis 14 0.025452 4340 Hedgehog signaling pathway 9 0.028947 [79]Open in a new tab Gene count represents the number of gene enriched in the pathway. Table 4. The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (top 10) of the differentially hypo-methylated genes. KEGG ID Name Gene count P value 4740 Olfactory transduction 10 0.004797 5320 Autoimmune thyroid disease 3 0.015042 4060 Cytokine-cytokine receptor interaction 7 0.016163 4620 Toll-like receptor signaling pathway 4 0.018845 4950 Maturity onset diabetes of the young 2 0.025692 830 Retinol metabolism 3 0.026061 4622 RIG-I-like receptor signaling pathway 3 0.034039 982 Drug metabolism - cytochrome P450 3 0.036526 4140 Regulation of autophagy 2 0.045367 [80]Open in a new tab Gene count represents the number of gene enriched in the pathway. Table 5. The reactome pathways (top 5) of the differentially hyper- and hypo-methylated genes. Reactome ID Name Gene count P value Differentially hyper-methylated genes 881907 Gastrin-CREB signaling pathway via PKC and MAPK 34 6.26E−06 416476 G alpha (q) signaling events 29 8.10E−05 163685 Integration of energy metabolism 19 0.000224877 500792 GPCR ligand binding 51 0.000330922 1474244 Extracellular matrix organization 35 0.000489987 Differentially hypo-methylated genes 1461973 Defensins 7 4.84E−07 1462054 Alpha-defensins 3 0.000113 1592389 Activation of Matrix Metalloproteinases 4 0.000245 381753 Olfactory Signaling Pathway 11 0.000577 1461957 Beta defensins 4 0.000602 [81]Open in a new tab Gene count represents the number of gene enriched in the pathway. Table 6. The gene ontology (GO) of the differentially hyper-methylated genes. GO categories Name Gene count P value BP Developmental process 483 1.11E−16 Anatomical structure development 436 2.22E−16 System development 387 4.44E−16 Cell differentiation 323 2.00E−15 CC Extracellular region part 141 1.79E−10 Plasma membrane part 211 1.14E−09 Proteinaceous extracellular matrix 60 1.29E−09 Extracellular matrix 64 1.22E−08 MF Sequence-specific DNA binding 92 3.03E−08 Sequence-specific DNA binding RNA Polymerase II transcription factor activity 46 8.37E−07 Voltage-gated cation channel activity 28 9.45E−07 Sequence-specific DNA binding transcription factor activity 118 1.16E−06 [82]Open in a new tab BP: biological process; CC: cellular component; MF: molecular function; Gene count represents the number of gene enriched in GO term. Table 7. The gene ontology (GO) of the differentially hypomethylated genes. GO categories Name Gene count P value BP Keratinocyte differentiation 9 3.54E−07 Epidermal cell differentiation 10 7.98E−07 Keratinization 6 1.88E−06 Defense response 29 3.62E−06 CC Extracellular region 50 3.73E−11 Extracellular space 23 1.18E−06 Extracellular region part 27 1.81E−06 Extracellular region 50 3.73E−11 MF Serine-type endopeptidase activity 8 2.01E−05 Serine-type peptidase activity 8 5.12E−05 Serine hydrolase activity 8 5.56E−05 Monooxygenase activity 6 8.15E−05 [83]Open in a new tab BP: biological process; CC: cellular component; MF: molecular function; Gene count represents the number of gene enriched in GO term. 3.3. PPI network construction In this study, a PPI network with 442 nodes and 693 PPI pairs was constructed ([84]Fig. 1). In this network, adenylate cyclase 2 (ADCY2) (degree=34), proopiomelanocortin (POMC) (degree=29), signal transducers and activators of transcription (STAT3) (degree=21), neuropeptide Y (NPY) (degree=20), and somatostatin (SST) (degree=18) with degrees more than 17 were regarded as hub nodes. The sub-network modules that contained the five hub nodes are shown in [85]Fig. 2A–E. Fig. 1. [86]Fig. 1 [87]Open in a new tab A protein-protein interaction network of differentially methylated genes. Red nodes stand for differentially hyper-methylated genes, while green nodes stand for differentially hypo-methylated genes. Fig. 2. [88]Fig. 2 [89]Open in a new tab The networks that contained five hub genes of adenylate cyclase 2 (ADCY2), proopiomelanocortin (POMC), signal transducers and activators of transcription (STAT3), neuropeptide Y (NPY), and somatostatin (SST). Red nodes stand for differentially hyper-methylated genes, while green nodes stand for differentially hypo-methylated genes. 3.4. Gene classification analysis Among the identified differentially hyper-methylated genes, 27 were found to be oncogenes, 78 were TS genes and 22 were tumor-associated genes (oncogenes or TS genes). The identified differentially hypo-methylated genes contained one oncogene, five TS genes and one tumor-associated gene. 3.5. TF analysis A total of 76 TFs that were significantly enriched in target genes were identified ([90]Supplementary material Fig. 1). Among these TFs, MAX interactor 1, dimerization protein (MXI1), STAT3, and T-cell acute lymphocytic leukemia 1 (TAL1) were also cancer-associated TFs. Interestingly, STAT3 was a hub node in the PPI network. Furthermore, the methylation levels of the three TFs were up-regulated in OS samples relative to control samples ([91]Fig. 3). Fig. 3. [92]Fig. 3 [93]Open in a new tab The methylation levels of MXI1, STAT3 and TAL1 in all samples. Horizontal axis represents 25 samples, and vertical axis represents the M value of methylation level of MXI1, STAT3 and TAL1. 4. Discussion OS is a rare malignant tumor with a high tendency of metastasis. Aberrant methylation of tumor-related genes in the promoter region is associated with human tumors [94][24]. In the current study, a total of 2845 differentially methylated loci were identified in OS samples compared to normal samples, including 1379 hyper-methylation regions and 169 hypo-methylation regions in the promoter region. Although so many differentially methylated loci were identified, not all of them may involve in OS progression. Further analyses revealed that the differentially hypo-methylated genes were significantly enriched in Toll-like receptor signaling pathway, while differentially hyper-methylated genes were significantly enriched in Neuroactive ligand-receptor interaction pathway, and PPAR signaling pathway. Additionally, ADCY2 and POMC had the top two highest degrees in the PPI network. Moreover, MXI1, STAT3, and TAL1 were significant TFs enriched with target genes, and they were also tumor-associated TFs. Toll-like receptors play important roles in the innate immune system, especially in inflammatory response, which is considered to be an important epigenetic factor contributing to neoplasia and tumor progression [95][25], [96][26]. Triggering of toll-like receptor 4 (TLR4) on metastatic breast cancer cells has been found to promote adhesion and invasive migration of tumor cells [97][27]. There is evidence that inflammatory cytokines, such as TNF-α and IL-1 are required for promoting the tumorigenesis of osteosarcoma [98][28]. The present study found that the differentially hypo-methylated genes in OS samples were related to Toll-like receptor signaling pathway. These findings led to a speculation that the differentially hypo-methylated genes and the Toll-like receptor signaling pathway might be involved in OS-related inflammation. A recent microarray analysis reported that Neuroactive ligand-receptor interaction pathway and PPAR signaling pathway were associated with OS metastasis [99][29]. Moreover, it has been established that PPAR agonists can suppress OS proliferation and growth, and induce apoptosis [100][30]. Results of this study suggest that the roles of Neuroactive ligand-receptor interaction pathway and PPAR signaling pathway in OS may be partly due to hyper-methylation of genes. It has been reported that PPI network can organize all protein-coding genes into a large network which provides a better understanding of the functional organization of the proteome [101][17]. In the present study, PPI analysis found that ADCY2 and POMC had the top two highest degrees in the constructed PPI network. Genome-wide studies have shown that deletion of a hub protein is more likely to be lethal than deletion of a non-hub protein. As a result, hub nodes may play more important roles in OS than the other nodes. Presently, we focused on the top two nodes with higher degrees for discussion. Interestingly, they have been reported to be implicated in the progression of OS. Recently, Sun et al. [102][31] performed gene expression profiling analysis of OS cell lines, and found that ADCY2 was involved in purine metabolism pathway in OS. POMC has been found in osteoclasts [103][32], which is a precursor of β-endorphin. Baamonde et al. [104][33] reported that endogenous β-endorphin involved in the initial stages of murine OS. Taken together, our study may further suggested the role of ADCY2 and POMC in OS. STAT3 is a member of STAT protein family and plays a critical role in cell growth and apoptosis [105][34]. Increasing studies have established that STAT3 activation promotes the development of OS [106][35], [107][36]. STAT3 was found to be hyper-methylated in OS samples in the present study, suggesting that STAT3 might play a role in OS. It has been reported that STAT3 had an oncogenic or a tumor suppressor role in human brain tumor depending on the mutational profile of the tumor [108][37]. It indicates that STAT3 may also play dual roles in promoting or discouraging the development of OS. Moreover, STAT3 was a hub node (degree=21) in the PPI network of this study, suggesting that the role of STAT3 in OS might be associated with its interactions with other proteins. MXI1 is a member of the MAD gene family, which plays a key role in regulating cell proliferation and growth [109][38], [110][39]. It is located at 10q24-25, where loss of heterozygosity occurs in several human cancers, including endometrial cancers, prostate tumors, gliomas, renal cell carcinomas, and small-cell lung cancers [111][40]. Importantly, numerous experiments have verified that MXI1 functions as a tumor suppressor gene via negatively modulating the promoter of oncogene c-Myc, which is involved in the control of cell proliferation and apoptosis [112][41], [113][42], [114][43], [115][44]. To our knowledge, the role of MXI1 in OS remains unclear. The study revealed hyper-methylation of MXI1 in OS. In light of these results, we speculate that MXI1 may inhibit progression of OS by compromising cell proliferation and inducing apoptosis. TAL1 is a basic helix-loop-helix TF required for blood cell development [116][45]. In the study of Kresse et al. [117][46], TAL1 was found to be significantly differentially methylated between OS and normal control samples. In agreement with the study above, the present study also revealed that it was a differentially methylated gene and a tumor-associated TF. Therefore, TAL1 might be an important TF associated with OS. In this study, some limitations have to be acknowledged. First, the sample size was a little small. Second, no experiment was conducted to validate the expression level of these methylated genes, including MXI1, STAT3 and TAL1. Therefore, further studies with more samples and experimental validations are needed to validate our results. Taken together, microarray analysis of epigenetic alterations revealed that Neuroactive ligand-receptor interaction pathway, PPAR signaling pathway and toll-like receptor signaling pathway may be implicated in tumorigenesis of OS. MXI1, STAT3, and TAL1 might be critical TFs in development of OS. Results of the study provide more enlightening insights concerning the association of DNA methylation with tumorigenesis of OS. More studies are warranted to verify these results. Competing interests The authors declare that they have no competing interests. Authors contribution Jie Xu, Deng Li and Ruofan Ma Wrote Manuscript, Designed Research, Zhiqing Cai, Yingbin, Zhang Yulin Huang and Baohua Su help Analyzed Data. Acknowledgements