Pediatric acute lymphoblastic leukemia (PALL) is the most common childhood cancer. While survival rates have improved dramatically over decades, better biological understanding is needed to identify reliable biomarkers for diagnosis, prognosis, and treatment monitoring. Two molecular features are consistently altered in PALL: DNA methylation patterns and gene expression levels.
DNA methylation is an epigenetic modification - a chemical tag added to DNA without changing the underlying genetic sequence. It typically occurs when a methyl group is attached to cytosine bases in the DNA, often at CpG sites. Abnormal methylation patterns in cancer can silence tumor suppressor genes (hypermethylation) or activate oncogenes (hypomethylation).
While individual studies of either methylation or gene expression can identify cancer-associated changes, analyzing them in isolation misses the complex interactions between epigenetic regulation and gene activity. Integrating both datasets within a network framework has the potential to reveal more robust and biologically meaningful cancer signatures.
This study uses a protein-protein interaction (PPI) network approach to integration. In a PPI network, proteins are nodes and connections represent functional interactions between them. Proteins that are highly connected - so-called hubs - tend to be critical regulators that influence many biological processes. Hubs that are disrupted in cancer are excellent candidates for biomarkers or therapeutic targets.
A key innovation in this study is the use of Methyl-IT, a novel signal detection and machine learning approach for analyzing whole genome bisulfite sequencing (WGBS) data. WGBS is the gold standard for methylation analysis - it sequences the entire genome after chemical treatment that converts unmethylated cytosines, allowing methylated sites to be identified at single-base resolution.
Methyl-IT treats the detection of differentially methylated positions as a signal-in-noise problem, using statistical signal detection theory and machine learning to distinguish true cancer-associated methylation changes from normal biological variability. This approach provides higher sensitivity and specificity than previous methods.
The analysis identified differentially methylated genes (DMGs) - genes where methylation patterns differ significantly between leukemia patients and healthy controls - and differentially expressed genes (DEGs) - genes showing altered expression levels. Additionally, a critical set of DEG-DMGs was identified: genes that are simultaneously differentially methylated AND differentially expressed, suggesting their expression is directly regulated by methylation changes.
Protein-protein interaction networks were built for each gene set using the STRING database, a comprehensive resource of known and predicted protein interactions, visualized and analyzed using Cytoscape software. Network centrality measures (degree, betweenness centrality, closeness centrality) were used to identify hub genes within each network.
The analysis identified 4,795 differentially methylated genes (DMGs) in PALL patients, including 3,338 protein-coding genes. Of the 2,360 differentially expressed genes (DEGs) reported in the original leukemia study used as a reference dataset, 75.2% (1,774 genes) were also identified as DMGs - a remarkable concordance suggesting that methylation changes are a major driver of the gene expression alterations seen in leukemia.
Network enrichment analysis of both the DMG and DEG sets revealed 80% overlap in the cancer pathways they targeted. The most enriched pathways included Pathways in Cancer (the general cancer pathway in the KEGG database), PI3K-Akt signaling, Ras signaling, Rap1 signaling, MAPK signaling, JAK-STAT signaling, Wnt signaling, and focal adhesion. These are core oncogenic signaling pathways that drive cell proliferation and survival.
A Bland-Altman concordance analysis confirmed statistically significant agreement between the pathway targeting patterns of methylation changes and gene expression changes, with a Lin's concordance correlation coefficient of 0.71 and a Kendall coefficient of concordance of 0.81. This quantifies the remarkable alignment between epigenetic and transcriptional dysregulation in leukemia.
Among the 254 identified DMGs that overlap with the COSMIC Cancer Gene Census - a curated database of confirmed cancer-related genes - 112 were simultaneously differentially expressed (DEG-DMGs). This highly confident set represents genes where both layers of evidence (methylation and expression) converge on the same cancer-relevant target.
K-means clustering of the DMG protein interaction network divided it into three clusters, with the most important cluster - the main hub subnetwork of 46 highly connected genes - showing the highest network centrality measures. Hub genes in this network included well-established cancer regulators such as NOTCH1, RAC1, PIK3CD, BCL2, and EGFR.
These hub genes are biologically significant. NOTCH1 is a key developmental signaling protein frequently mutated in T-cell ALL. BCL2 is the master anti-apoptotic protein targeted by the drug venetoclax. EGFR is an important growth factor receptor targeted by multiple approved cancer drugs. Finding these genes as methylation-regulated hubs in leukemia provides molecular context for their importance.
The DEG-DMG network - genes simultaneously altered at both the expression and methylation level - was similarly organized into hub subnetworks. Enrichment analysis of these DEG-DMG hubs confirmed upregulation of the same core cancer pathways, providing an even more confident set of candidate biomarkers than either dataset alone.
A particularly interesting finding was the stochastic-deterministic relationship between methylation and expression changes: genes with larger methylation changes were probabilistically associated with larger expression changes. This is not a perfectly linear relationship but a statistical trend, consistent with the known complexity of epigenetic regulation where methylation is one of several factors controlling gene expression.
DNA methylation has several properties that make it attractive as a source of cancer biomarkers. Unlike gene expression, which fluctuates in response to many environmental and physiological signals, methylation marks tend to be more chemically stable and reproducible. A methylation change that is consistently present in cancer samples is likely to be a reliable diagnostic or prognostic marker.
The concordance between methylation changes and gene expression changes observed in this study further validates methylation biomarkers: when both layers of molecular evidence point to the same gene or pathway, the signal is more robust and less likely to be a technical artifact. The DEG-DMG set represents the most reliable candidates for biomarker development.
The Methyl-IT approach used here detected a much larger and more significant methylation signal than previously reported in PALL studies using conventional methods. This improvement in sensitivity - achieved by treating methylation change detection as a signal detection problem rather than a simple statistical comparison - means that previously overlooked methylation events can now be identified and studied.
The network-based approach also provides important biological context for individual biomarker candidates. A methylation change at a hub gene has different implications than one at a peripheral gene: the hub gene change is more likely to have widespread downstream effects on cancer biology, making it both a better biomarker and a more attractive therapeutic target.
This study demonstrates that integrating methylation and gene expression data through network analysis provides substantially deeper insight into leukemia biology than either dataset analyzed in isolation. The concordance between the two data types validates the biological signals and identifies the most confident cancer-relevant targets.
The identified hub genes and their associated pathways represent a prioritized list for future leukemia biomarker development and drug target validation studies. Genes that are simultaneously hypermethylated, underexpressed, and central nodes in cancer-relevant interaction networks are particularly compelling candidates for diagnostic markers or therapeutic targets.
By applying WGBS - the highest-resolution methylation measurement technology - and the sensitive Methyl-IT detection approach, this study achieves a level of methylation signal resolution in cancer genes not previously reported for leukemia. This greater sensitivity means that subtle but biologically important methylation changes can now be captured and studied.
Looking forward, the approach described here provides a generalizable workflow for integrated multi-omics biomarker discovery that could be applied to other cancer types and other combinations of molecular data. As the cost of genome-wide sequencing continues to fall, such integrative analyses will become increasingly feasible for clinical application.