Osteosarcoma (OS) is a rare bone tumor most frequently occurring in children and young adolescents. It is characterized by severe genomic instability and extensive mutations that make identifying a single genetic root cause nearly impossible. Frequent alterations include mutations and deletions in tumor suppressors TP53, RB1, and CDKN2A, as well as amplification of the MYC locus. Despite decades of research, the molecular pathogenesis remains poorly characterized, and almost all patients receive adjuvant chemotherapy regardless of individual molecular profiles.
MicroRNAs are small non-coding RNA molecules that act primarily at the post-transcriptional level. They are incorporated into the RNA Induced Silencing Complex (RISC) and bind to 3'-UTR sequences of target mRNAs, leading to degradation or translational inhibition. Individual microRNAs can regulate hundreds of target genes, and approximately 60% of human genes are predicted to be regulated by multiple microRNAs cooperatively. Their dysregulation is linked to cancer through both oncogenic and tumor-suppressor mechanisms.
Transcription factors (TFs) and microRNAs do not act independently. They frequently form co-regulatory motifs, the most common being the feedforward loop (FFL), where a TF regulates both a microRNA and the microRNA's target gene. This coordinated regulation creates complex multi-layered networks governing cell behavior. This study is the first to systematically analyze microRNA and TF co-regulatory networks specifically in the context of OS cell proliferation, integrating experimental data with computational predictions.
The study analyzed seven authenticated OS cell lines divided into two groups by proliferation assay. Four cell lines (MG-63, MNNG/HOS, SJSA-1, U2-OS) showed high proliferative activity with a doubling time under 10 hours; three cell lines (HOS-58, SaOS-2, ZK-58) showed low proliferation. Additionally, MNNG/HOS, U2-OS, and SJSA-1 displayed extensive migratory capabilities, making them potentially relevant to OS metastasis biology.
The workflow proceeded through five steps: (1) divide OS cell lines by proliferative activity and perform microRNA and mRNA expression analysis; (2) integrate computational microRNA target predictions with differentially expressed mRNAs to identify functional target genes; (3) cluster target genes by functional similarity using fuzzy c-means clustering (FCM) on Gene Ontology (GO) terms; (4) assemble 3-node and 4-node feedforward loop motifs using transcription factor binding site (TFBS) information; and (5) merge significant motifs into co-regulatory networks for analysis. Statistical thresholds were applied at each step, with FDR less than 0.05 for microRNA differential expression and FDR less than 0.2 for network motif significance.
MicroRNA expression analysis used FDR less than 0.05 and |log2 fold change| at least 1 as criteria for differential expression. For mRNA targets, a looser threshold of p-value less than 0.05 and |log2 FC| at least 0.7 was applied without multiple testing correction, acknowledging that microRNA regulation can cause subtle expression changes. Target gene enrichment within the differentially expressed gene list was confirmed by hypergeometric test and validated by permutation analysis across 1,000 shuffled samples.
Expression analysis identified 17 differentially expressed microRNAs between high and low proliferative OS cells. Of these, 9 were down-regulated and 8 up-regulated in high-proliferative cells. The most significantly down-regulated microRNA was miR-181a (log2 FC = -2.28, FDR = 0.000013), followed by miR-9-3p (log2 FC = -2.01), miR-181d (log2 FC = -1.82), miR-138 (log2 FC = -1.81), and miR-214 (log2 FC = -1.59). Up-regulated microRNAs included miR-130a (log2 FC = 2.77, FDR = 0.007), miR-21-5p (log2 FC = 2.31), and miR-155 (log2 FC = 2.05).
After testing for target gene enrichment, 5 microRNAs were excluded (miR-92b, let-7f, miR-9-3p, miR-151-5p, miR-100) as their target associations were not statistically significant. The remaining 12 proliferation-related microRNAs regulated a combined set of 474 target genes. Hierarchical clustering of these target genes perfectly separated high and low proliferative OS cell samples. Importantly, miR-138 was exclusively identified in this study as proliferation-related in OS, not in any prior literature.
Functional clustering by GO biological process terms using FCM produced two clusters: Cluster C1 (172 members) enriched for metabolic processes including protein modification, nucleic acid metabolism, and carbohydrate metabolism; and Cluster C2 (212 members) enriched for signal transduction pathways controlling proliferation, differentiation, apoptosis, and migration. A further 90 target genes lacked GO annotation. Both clusters reflect how cancer cells adapt fundamental cellular processes for sustained growth.
A 3-node feedforward loop (FFL) consists of a microRNA, a TF, and a commonly regulated target gene, where the TF also activates or represses the microRNA. A 4-node FFL extends this by adding a secondary TF target gene that interacts with the primary microRNA target in a protein interaction network. Both motif types were tested for non-random association using the hypergeometric test at FDR less than 0.2, applied separately to C1 and C2 functional clusters.
For the C1 metabolic network: 10 microRNAs and 29 TFs formed non-random pairs, with 85 primary co-regulated target genes in 3-node motifs and an additional 213 secondary targets in 4-node motifs, totaling 46 significant microRNA-TF pairs. For the C2 signaling network: 10 microRNAs and 35 TFs formed non-random pairs, with 117 primary targets in 3-node motifs and 550 secondary targets in 4-node motifs, totaling 70 significant microRNA-TF pairs. Co-expressed gene pairs co-regulated by the same microRNA-TF combination showed significantly higher Pearson correlation than random pairs (p-value less than 2.2 x 10^-16, Kolmogorov-Smirnov test).
The co-regulatory networks were built by joining all significant microRNA-TF co-regulatory relationships. Node degree distributions were highly right-skewed, with few high-degree hub nodes and many low-degree nodes. Average node degrees for microRNAs and TFs in C1 were 19 each; in C2 they were 25 (microRNAs) and 49 (TFs). Approximately 50% of target gene hubs in both networks were TFs themselves, reflecting the hierarchical self-regulatory architecture of transcriptional control in cancer.
Among all microRNA-TF combinations tested, miR-9-5p and SP1 showed the highest number of co-regulated target genes in both functional clusters, making them the most prominently identified co-regulation pair in OS cell proliferation. Hub microRNAs and TFs common to both networks were miR-214, CREB1, SP1, and ZIC2, suggesting core regulatory roles independent of metabolic or signaling context.
In the C1 (metabolic) network, miR-9-5p and SP1 co-regulated target genes including the TFs NFKB1, NFKB2, RELA, RELB, and BCL3, as well as their inhibitors NKRF, NFKBIA, and TNIP2, collectively activating or blocking NFKB target gene expression. NFKB1 is an experimentally validated target of miR-9-5p. In the C2 (signaling) network, their co-regulated targets included cadherins (CDH1, CDH2, DSC2), cell adhesion molecules (FN1, ITGA6, ITGB1, VCL), and calcium signaling receptors, implicating focal adhesion processes.
Walktrap algorithm subnetwork analysis identified 6 modules in each network. Module C2.1 was particularly notable: it contained the hubs miR-138, MYC (node degree 83), and SIN3A (node degree 57), and was associated with negative regulation of osteoblast differentiation (enrichment score greater than 3.9). Members included cell cycle regulators CDK6, CCND1, CCND3, CDKN1A, CDKN2C, all within the RB1 pathway. SIN3A is an experimentally validated target of miR-138 and can bind RB1 to repress E2F4 targets.
The authors integrated their findings into a mechanistic model of OS cell proliferation: in high-proliferative OS cells, miR-9-5p, miR-138, and miR-214 are down-regulated, releasing their targets from repression. Direct target genes that increase include CDK6, E2F4, HES1, ITGA6, NFKB1, NOTCH1, and SIN3A. CDK6 phosphorylates RB1, releasing the RB1/SIN3A/SKI/NCOR1/HDAC repressor complex from E2F4 target genes. Activated NOTCH1 induces HES1 and sustains NFKB signaling. Unbound CTNNB1 translocates to the nucleus through cadherins, inducing further proliferative gene expression.
These convergent signals drive expression of proliferation-promoting outputs including CCND1, FN1, and additional microRNAs up-regulated in high-proliferative cells, while repressing growth-arrest genes. The model therefore identifies miR-9-5p, miR-138, and miR-214 as potential therapeutic entry points, with their restoration potentially halting the proliferative cascade by simultaneously suppressing multiple oncogenic pathways.
The complete microRNA and TF co-regulatory networks derived in this study are publicly available at http://www.complex-systems.uni-muenster.de/co_networks.html. This open-access resource enables the broader research community to generate and test hypotheses about OS pathogenesis. Future priorities include validating these network predictions in primary patient samples and investigating whether microRNA mimic or antagomir therapies targeting the identified hub regulators can alter OS cell behavior in vivo.