Osteosarcoma is the most common primary malignant bone cancer in children and adolescents. It originates from mesenchymal stem cells and exhibits osteoblastic differentiation, meaning the cancer cells resemble the bone-forming cell lineage. Despite its relative rarity, with an incidence of approximately one to three cases per million people annually worldwide, it carries a disproportionately heavy burden because of its predilection for the young. Surgery combined with chemotherapy has improved outcomes substantially over the past several decades, but a hard ceiling has appeared: overall survival has plateaued, and roughly 30 to 40 percent of patients still experience progressive metastasis within five years of diagnosis and ultimately die from the disease.
The biological bottleneck: The stagnation in survival is not for lack of clinical effort. Rather, the underlying molecular drivers of osteosarcoma, particularly the gene expression changes and non-coding RNA regulatory mechanisms that distinguish malignant bone cells from normal ones, remain incompletely characterized. Without a clearer molecular map, it is difficult to identify which genes or pathways represent the most tractable diagnostic or therapeutic targets.
microRNAs as regulatory actors: One class of molecules that has attracted considerable attention is microRNAs (miRNAs), short non-coding RNA sequences approximately 22 nucleotides in length. miRNAs regulate gene expression post-transcriptionally by binding to the 3'-untranslated region (3'-UTR) of messenger RNAs (mRNAs), typically causing mRNA degradation or translational repression. Depending on which mRNAs a given miRNA silences, it can function either as an oncogene (promoting tumor growth by suppressing tumor suppressors) or as a tumor suppressor (limiting growth by silencing oncogenes). Several miRNAs have already been implicated in osteosarcoma. For example, miR-497 can activate the cell cycle inhibitor P21 by suppressing the MAPK/Erk signaling pathway and thereby promotes apoptosis in osteosarcoma cells. Protein-coding genes have also emerged as potential biomarkers: SPARCL1 is silenced in osteosarcoma through epigenetic methylation of its promoter, and restoring SPARCL1 expression inhibits metastasis in cell and animal models; NRP2 is overexpressed in osteosarcoma tissues and cell lines and correlates with poor patient survival.
This paper by Pan and colleagues, published in Hereditas in 2018, takes a systematic bioinformatics approach to map the landscape of differentially expressed genes and miRNAs in osteosarcoma, with the goal of constructing a comprehensive miRNA-mRNA negative regulatory network. By integrating two publicly available GEO datasets, the authors aimed to reveal new candidate biomarkers and therapeutic targets that had not previously been identified in experimental studies.
The authors built their analysis on two paired datasets from the Gene Expression Omnibus (GEO), a publicly accessible repository of microarray and sequencing data. Both datasets were deposited by Namlos et al. in 2011 and share the same panel of cell lines, making them ideally suited for integrated analysis. The gene expression dataset (GSE28424) and the miRNA expression dataset (GSE28423) each profile the same 19 osteosarcoma cell lines (the case group) against 4 normal bone samples (the control group). Using identical biological specimens across both datasets eliminates the confounding variability that arises when gene and miRNA data come from different cohorts.
Differential expression calling: Raw data from both datasets were normalized using the preprocessCore R Bioconductor package, which applies quantile normalization to reduce systematic technical variation between arrays. Differentially expressed genes (DEGs) and differentially expressed miRNAs (DEMIs) were then identified using the limma package, a widely used linear-models framework for microarray data. The cutoff criteria were stringent: adjusted P-value below 0.01 (after Benjamini-Hochberg multiple-testing correction) and an absolute log2 fold change exceeding 2, corresponding to a four-fold difference in expression between osteosarcoma and normal bone.
Functional annotation: DEGs were submitted to DAVID (Database for Annotation, Visualization and Integrated Discovery) for gene ontology (GO) enrichment and KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway analysis. GO analysis covers three orthogonal dimensions: cellular component (CC, where the gene product is located), biological process (BP, what process it participates in), and molecular function (MF, what biochemical activity it carries out). A significance threshold of P less than 0.05 was applied to all enrichment tests.
Protein-protein interaction network: The STRING database (version 10) was used to build a protein-protein interaction (PPI) network from the DEG list. Only interactions with a combined confidence score of 0.7 or higher were retained, representing high-confidence physical or functional associations backed by multiple lines of evidence. The network was visualized and analyzed in Cytoscape using the NetworkAnalyzer plug-in to extract topological properties, and the MCODE plug-in (Molecular Complex Detection) to find densely connected gene clusters within the network.
miRNA target prediction and network integration: Target genes for each DEMI were predicted using TargetScan, which identifies mRNAs whose 3'-UTR sequences are complementary to the seed region of a given miRNA. A cumulative weighted context++ score of -0.4 or below was required to retain a predicted interaction, selecting only those with a relatively high predicted efficacy of silencing. Finally, the predicted target list was cross-referenced against the DEG list; only miRNA-mRNA pairs in which the two molecules changed in opposite directions in the osteosarcoma data were kept, producing a bona fide negative regulatory network that was visualized in Cytoscape.
Applying the cutoff criteria to GSE28424 yielded 346 differentially expressed genes (DEGs) between the 19 osteosarcoma cell lines and 4 normal bone samples. Of these, 43 were up-regulated and 303 were down-regulated in osteosarcoma, a pronounced asymmetry suggesting widespread gene silencing rather than broad transcriptional activation. The gene showing the greatest up-regulation was CBS (cystathionine-beta-synthase), with a log2 fold change of 3.44 and an adjusted P-value of 4.06 x 10^-5. CBS is an enzyme in the transsulfuration pathway that produces hydrogen sulfide and has been reported as up-regulated in colorectal cancer, multiple myeloma, and bladder cancer, where its silencing inhibits tumor growth. Notably, no prior study had linked CBS to osteosarcoma specifically, making it a potentially novel candidate biomarker for this tumor type.
Top up-regulated genes: Beyond CBS, the next most significantly up-regulated genes included TMSL8 (log2FC 3.35), PSAT1 (3.20), PHGDH (3.07), ASNS (2.91), TUBB3 (2.73), UBE2C (2.62), PBK (2.51), and LARP6 (2.48). Several of these belong to serine and one-carbon metabolism pathways (PSAT1, PHGDH, ASNS), consistent with the well-recognized metabolic reprogramming in cancer cells that favors biosynthetic precursor production.
Top down-regulated genes: The most dramatically suppressed gene was HBB (hemoglobin beta), with a log2FC of -7.79 and adjusted P of 5.88 x 10^-34. HBA1 (-7.33), MMP9 (-6.34), CD74 (-5.84), S100A8 (-5.83), VWF (-5.64), HLA-DRA (-5.54), and LYZ (-5.43) rounded out the most significantly reduced transcripts. The preponderance of immune-related and extracellular matrix genes in this down-regulated set foreshadowed the functional enrichment results: immune system suppression is a prominent feature of the osteosarcoma transcriptome relative to normal bone.
miRNA landscape: The parallel analysis of GSE28423 identified 90 differentially expressed miRNAs, of which 58 were down-regulated and 32 were up-regulated. The most dramatically altered miRNA was hsa-miR-451 (log2FC -15.39, adjusted P = 2.80 x 10^-7), followed by hsa-miR-144 (-9.27), hsa-miR-142-3p (-8.33), hsa-miR-223 (-7.52), hsa-miR-126 (-6.35), and hsa-miR-142-5p (-6.12). The two most up-regulated miRNAs were hsa-miR-9 (log2FC 6.04) and hsa-miR-18a (6.03), both of which have established oncogenic roles in multiple cancer types.
Gene ontology enrichment of the 346 DEGs across all three GO axes revealed a consistent picture: the transcriptomic differences between osteosarcoma and normal bone are dominated by changes in immune function and the extracellular environment. This finding is scientifically coherent because osteosarcoma, like many solid tumors, evolves in an immunologically complex microenvironment where escaped tumor cells progressively suppress immune surveillance.
Cellular component (CC): The largest enriched GO categories in the CC dimension were extracellular exosome (120 genes), extracellular space (77 genes), extracellular region (72 genes), and extracellular matrix (20 genes). The prominence of these categories indicates that many of the differentially expressed genes encode secreted or membrane-associated proteins that shape the tumor microenvironment, consistent with the known role of the extracellular matrix and paracrine signaling in osteosarcoma invasion and metastasis. The cytosol (95 genes) and cell surface (25 genes) were also substantially enriched, and MHC class II protein complex contributed 8 genes, reflecting immune-related subcellular compartments.
Biological process (BP): The biological process categories were dominated by innate and adaptive immune response terms. Immune response (35 genes), innate immune response (26 genes), and inflammatory response (26 genes) were among the top enriched BP terms. Additional immune categories included defense response to Gram-positive bacterium (12 genes), antigen processing and presentation (9 genes), and antigen processing and presentation via MHC class II (7 genes). Platelet degranulation (12 genes), muscle filament sliding (9 genes), and oxygen transport (6 genes) were also enriched, the platelet degranulation pathway being particularly noteworthy because activated platelets coat tumor cell surfaces and shield them from immune recognition.
Molecular function (MF): The MF enrichment highlighted binding-related activities: actin binding (17 genes), heparin binding (13 genes), heme binding (11 genes), protease binding (10 genes), collagen binding (9 genes), peptide antigen binding (7 genes), and oxygen binding (7 genes). MHC class II receptor activity (6 genes) and oxygen transporter activity (6 genes) were additional enriched terms. These binding categories align with the structural and immune-regulatory functions of the extracellular matrix proteins that appear down-regulated in osteosarcoma.
KEGG pathway enrichment: KEGG analysis confirmed the inflammatory immune response theme. The top pathways were Phagosome (19 genes), Staphylococcus aureus infection (15 genes, a phagosome-activating pathway), Cell adhesion molecules (CAMs, 15 genes), Systemic lupus erythematosus (14 genes, heavily dominated by HLA and complement genes), and Asthma (11 genes). The convergence of multiple immune-related KEGG pathways onto the DEG list strongly supports the conclusion that osteosarcoma development involves disruption of immune recognition and inflammatory regulation in bone tissue.
A protein-protein interaction (PPI) network was constructed from all 346 DEGs using the STRING database, retaining only interactions with a combined confidence score of 0.7 or above. PPI networks represent the web of physical and functional associations between proteins in a cell, and topologically central genes (so-called "hub" genes) in such networks tend to be functionally critical, often encoding proteins that coordinate signaling cascades, structural complexes, or metabolic hubs. Disruption of hub genes frequently has outsized downstream effects compared to perturbation of peripheral genes.
Network topology analysis: The authors used the Cytoscape NetworkAnalyzer plug-in to characterize four key topological properties of the network: degree distribution (how many connections each node has), proximity to center, average clustering coefficient, and shortest path distribution. Importantly, the degree distribution followed a power-law pattern, meaning a small number of highly connected nodes dominated the network while most nodes had few connections. This architecture is characteristic of "scale-free" or "small-world" networks observed in biological systems, and it validates the relevance of focusing on hub genes as candidate biological drivers.
Top 10 hub genes: Using a degree cutoff of 10 or more connections as the selection criterion, 25 genes were designated as hub genes. The top 10, ranked by connectivity degree, were: TYROBP (Tyrosine kinase-binding protein, an immune-signaling adaptor expressed on myeloid and NK cells), HLA-DRA (MHC class II alpha chain, central to antigen presentation), VWF (von Willebrand Factor, a platelet adhesion and hemostasis protein), PPBP (Pro-platelet basic protein, a platelet chemokine), SERPING1 (Serpin G1/C1-inhibitor, a complement and coagulation regulator), HLA-DPA1 (another MHC class II component), SERPINA1 (alpha-1 antitrypsin, a protease inhibitor), KIF20A (a kinesin motor protein involved in cytokinesis), FERMT3 (kindlin-3, critical for integrin activation in hematopoietic cells), and HLA-E (a non-classical MHC class I molecule involved in NK cell regulation). The clustering of multiple HLA and immune checkpoint-related proteins in the top hub gene list reinforces the centrality of immune dysregulation in osteosarcoma.
Seed genes from module analysis: MCODE analysis at a score cutoff of 5 or above identified 4 densely connected clusters within the PPI network. The most significant cluster contained 11 nodes and 55 edges. Each cluster contained one seed gene: SEPP1 (selenoprotein P, involved in antioxidant defense), CKS2 (cyclin-dependent kinase regulatory subunit 2), TCAP (telethonin, a muscle-related structural protein), and BPI (bactericidal permeability-increasing protein, an innate immunity factor). None of these four seed genes had been previously reported in the context of osteosarcoma, making them candidates for further experimental validation.
Having established the landscape of differentially expressed coding genes and miRNAs separately, the authors integrated both layers to construct a biologically meaningful regulatory network. Target genes for all 90 DEMIs were first predicted computationally using TargetScan, a tool that identifies mRNA 3'-UTR sequences complementary to each miRNA's seed region. From the full predicted target list, only genes that also appeared in the DEG list and showed an expression direction opposite to their targeting miRNA were retained. This "opposite trend" filter is the key quality criterion: a miRNA that is up-regulated in osteosarcoma should suppress its target mRNAs, so only cases where the miRNA went up while the mRNA went down (or vice versa) were accepted as putative regulatory relationships.
Network composition: The filtering process yielded 35 DEMIs (25 up-regulated and 10 down-regulated in osteosarcoma) regulating 78 DEGs (67 down-regulated mRNAs matched to up-regulated miRNAs, and 11 up-regulated mRNAs matched to down-regulated miRNAs). These interactions formed 89 miRNA-mRNA pairs, and the complete network was visualized in Cytoscape.
miR-9 and its targets: miR-9 was the most significantly up-regulated miRNA among the filtered DEMIs, with a log2FC of 6.04 in osteosarcoma vs. normal bone. TargetScan predicted two down-regulated target genes for miR-9 in this dataset: CEACAM6 (Carcinoembryonic antigen-related cell adhesion molecule 6) and PDK4 (Pyruvate dehydrogenase kinase 4). CEACAM6 is known to participate in cell adhesion and immune evasion in several epithelial cancers, while PDK4 regulates the switch between oxidative phosphorylation and glycolysis. The suppression of these targets by an up-regulated miR-9 could contribute to metabolic reprogramming and immune escape in osteosarcoma.
miR-210 as the highest-connectivity node: miR-210 had the highest connectivity in the miRNA-mRNA network, targeting 12 differentially expressed genes: ANGPTL4, AQP1, ARHGAP25, BTG2, CD247, DNASE1L3, LYL1, P2RY8, SH2D3C, SRL, STAT5A, and TNFRSF1B. This breadth of targeting puts miR-210 in a position to simultaneously regulate angiogenesis (ANGPTL4), water transport (AQP1), T cell signaling (CD247, STAT5A), and several immune regulation genes. The importance of miR-210 in osteosarcoma is supported by clinical evidence showing that elevated miR-210 expression correlates with large tumor size, poor response to preoperative chemotherapy, positive metastasis, and reduced overall and progression-free survival.
miR-96 targets hub genes: Among the predicted targets of miR-96 were two of the PPI hub genes, HLA-DPA1 and TYROBP. Both are involved in immune recognition and myeloid cell signaling, respectively. The convergence of a single miRNA on multiple hub genes suggests that miR-96 may have a particularly amplified effect on the immune landscape of osteosarcoma. Additionally, several mRNAs were predicted as common targets of multiple miRNAs: CSF1R (colony-stimulating factor 1 receptor, a macrophage differentiation gene and one of the PPI hub genes) was a shared target of both miR-149 and miR-22, indicating that its suppression may be reinforced by two independent miRNA mechanisms.
The authors devoted substantial discussion to the biological significance of the seed genes and the most connected miRNAs identified in their analysis, contextualizing these findings against the broader cancer biology literature and highlighting their novelty in osteosarcoma specifically.
SEPP1 (selenoprotein P): SEPP1 is the principal selenium transport protein in humans and has an endogenous antioxidant function. It catalyzes the reduction of harmful lipid hydroperoxides, protecting cells from oxidative damage. Reduced SEPP1 expression has been documented in colon tumors, prostate tumors, and prostate cancer cell lines, and several single nucleotide polymorphisms (SNPs) in the SEPP1 gene have been associated with increased risk of colorectal adenomas. The consistent down-regulation of SEPP1 across multiple tumor types, now including osteosarcoma through bioinformatics evidence, suggests that loss of selenium-mediated antioxidant protection may be a common facilitator of malignant progression. No experimental work had previously implicated SEPP1 directly in osteosarcoma.
CKS2 (cyclin-dependent kinase regulatory subunit 2): CKS2 interacts with CDK1 and CDK2, the master kinases governing cell cycle progression through the G2/M and S phases. In prostate cancer, aberrant CKS2 expression promotes cell proliferation and suppresses programmed cell death. In papillary thyroid carcinoma, miR-26a and miR-7 modulate tumor growth by targeting CKS2. In colorectal cancer, reducing CKS2 decreases cell viability, triggers apoptosis, and induces cell cycle arrest. The PPI network placed CKS2 as the seed gene of one of the four MCODE clusters, and TargetScan predicted it as a target of miR-513c. If this interaction holds in osteosarcoma cell lines, pharmacological or RNA-based strategies to boost miR-513c could reduce CKS2 and thereby impair cell cycle progression in the tumor.
miR-210 and osteosarcoma dedifferentiation: miR-210 is frequently described as a hypoxia-responsive miRNA, induced under low-oxygen conditions common in the tumor core. Elevated miR-210 in osteosarcoma is associated with more aggressive disease. Beyond mere correlation, at least one validated direct interaction exists: miR-210 suppresses NFIC, a transcription factor that normally restrains TGF-b1-induced osteosarcoma dedifferentiation, and this suppression can be recapitulated by treating the human osteosarcoma cell line MNNG/HOS with synthetic miR-210. The 12 additional target genes identified here (ANGPTL4, AQP1, BTG2, CD247, STAT5A, and others) represent a largely uninvestigated set of candidate mediators through which miR-210 may exert its broad oncogenic effects, and any of these pairs could constitute a novel therapeutic axis worth pursuing experimentally.
TCAP and BPI: TCAP (telethonin or titin-cap) is a sarcomeric Z-disk protein involved in muscle filament organization. Its expression is dysregulated in several tumor types, though its mechanistic role in cancer remains unclear. BPI (bactericidal permeability-increasing protein) is part of the innate immune system, normally secreted by neutrophils to kill Gram-negative bacteria. Aberrant expression of both proteins has been linked to multiple malignancies, but, consistent with the pattern for SEPP1 and CKS2, no prior literature had described their roles in osteosarcoma at the time of this publication.
The authors are candid about the principal limitation of their work: every finding reported is computational. The differentially expressed genes and miRNAs, the hub genes, the seed genes, the miRNA-mRNA regulatory pairs, and all the inferred biological functions are derived entirely from microarray data analysis and database-based predictions. No cell culture experiments, animal models, or patient tissue samples are presented to confirm that any of these molecules actually drives osteosarcoma pathogenesis rather than merely correlating with it.
Small comparison group: The underlying datasets compare 19 osteosarcoma cell lines to only 4 normal bone samples. While this imbalance is common in studies using GEO public data and the paired nature of the datasets strengthens internal consistency, the 4-sample control group limits statistical power and increases the risk that the normal bone signature does not fully represent the diversity of normal skeletal tissue. Additionally, cell lines are inherently different from primary tumors: they have been adapted to grow in culture for extended periods, may have accumulated additional mutations, and do not capture the stromal, vascular, and immune compartments of the in vivo tumor microenvironment.
TargetScan predictions as hypothesis generators: miRNA target prediction from sequence complementarity is a computationally tractable but imperfect approach. TargetScan estimates the probability that a miRNA will bind a given 3'-UTR based on seed sequence matching and several contextual features, but it is well established that many predicted targets are not actually regulated in a particular cell type or disease context. The 89 miRNA-mRNA pairs in this network should be regarded as hypotheses requiring experimental confirmation, such as luciferase 3'-UTR reporter assays and RNA immunoprecipitation, rather than proven regulatory relationships.
Translational gap: Even experimentally validated miRNA-mRNA regulatory pairs in cell lines face a substantial translation gap before they inform clinical care. Demonstrating that a miRNA suppresses a target gene in a luciferase assay does not automatically confirm that modulating the miRNA changes tumor growth in an animal model, nor that the same pathway is active in primary patient tumors across different osteosarcoma subtypes or stages. The authors note that the pathogenesis of key miRNAs and genes identified here needs to be validated through both in vivo and in vitro experiments as the immediate next step.
Future directions: The network constructed here provides a prioritized list of candidate nodes for experimental follow-up. The most actionable near-term experiments would include: silencing or overexpressing miR-210 in osteosarcoma cell lines and measuring the expression of its 12 predicted target genes; testing whether CKS2 overexpression confers resistance to apoptosis-inducing treatments and whether miR-513c rescues this effect; and examining CBS expression in primary osteosarcoma patient biopsies to assess its diagnostic potential. Longer-term goals would include in vivo xenograft or orthotopic osteosarcoma models to test whether modulating these regulatory nodes alters tumor growth, invasion, or metastatic spread.