Epigenetic genes were obtained from the literature reported by yan et al. (Zhang et al., 2022). A total of 990 epigenetic genes were summarized in the study (Supplementary Table S1).

The research data were obtained from the TCGA database, downloaded through the UCSC Xena platform (http://xenabroswer.net/hub), comprising 8,739 pan-cancer transcriptomic profiles. The dataset ID is “EB++AdjustPANCAN_IlluminaHiSeq_RNASeqV2.geneExp.xena”. The analysis included pan-cancer data from 32 types of solid tumors, excluding samples of acute myeloid leukemia. To ensure the reliability and generalizability of the model, We divided the samples into a training cohort and a testing cohort by randomly assigning 70% of the samples to the training cohort and the remaining 30% to the testing cohort.

The signature construction followed these steps: (1) Based on the training set, we performed univariate Cox regression analysis to screen for prognostic genes significantly associated with survival (p < 0.0001, HR ≤ 0.7, HR ≥ 1.3). (2) We then used the Least Absolute Shrinkage and Selection Operator (LASSO) regression algorithm to perform feature selection, reducing the risk of overfitting. (3) Finally, we constructed the final predictive model using a multivariate Cox proportional hazards regression model.

The epigenetic score was trained using all cancer types collectively. The epigenetic score was calculated as follows: Epigenetic Score = Σ(Expression level of gene i × Corresponding regression coefficient). Specifically, Epigenetic Score = 0.147*ACTB+0.372*AP2A1+0.307*ASXL1+-0.14*BAHD1+0.104*BCDIN3D+0.193*BRD4+0.095*CDYL+0.179*DDX17+-0.097*DDX24+-0.278*DDX5+0.079*DHX35 + 0.128*DHX8+0.101*ENY2+0.117*FKBP1A + -0.269*FTSJ1+-0.33*HDGF + -0.196*KDM4B + -0.264*L3MBTL2+-0.183*MEPCE+0.081*PAK2+0.293*PHC2+-0.141*PHF7+-0.177*SETD3+-0.17*SETDB2+-0.158*SETMAR+0.134*SIRT7+0.065*SP140L+0.299*SUPT7L + -0.114*TADA2B+0.17*UBE2A+0.175*UCHL5+-0.103*USP7+-0.45*YTHDC1. Patients were divided into high epigenetic score and low epigenetic score groups based on their epigenetic score: samples with a standardized epigenetic score (z-score) greater than 0 were defined as the high epigenetic score group, while those with a standardized epigenetic score less than or equal to 0 were defined as the low epigenetic score group.

The nomogram was constructed using the nomogram function from the rms R package, setting the linear predictor (Linear Predictor) and survival probabilities (1-year, 3-year, 5-year, and 10-year) as output variables, with the maximum score set to 100 points.

The ssgsea method was used to evaluate the activities of key biological processes.The ssgsea was performed using the GSVA R package. The gene sets for glycolysis, EMT, DNA repair, angiogenesis, and inflammatory response were obtained from www.gsea-msigdb.org, including HALLMARK_ANGIOGENESIS, HALLMARK_DNA_REPAIR, HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION, HALLMARK_INFLAMMATORY_RESPONSE, and WP_GLYCOLYSIS_IN_SENESCENCE. The gene set for cell cycle was derived from previous study (Sanchez-Vega et al., 2018).

We used the limma R package to analyze the differentially expressed genes (DEGs) between the high-risk and low-risk groups, applying the following selection criteria: log2 fold change absolute value is greater than or equal to 1 and adjusted false discovery rate (FDR) p-value less than 0.05.

All statistical analyses were performed using R software (version 4.2.2). Kaplan-Meier analysis was used to evaluate patient survival, and decision curve analysis (DCA) was used to assess the clinical utility of the model. Time-dependent receiver operating characteristic (ROC) curve analysis was used to compare the predictive performance of the ES and the nomogram model (area under the curve, AUC). All statistical tests were two-sided, and p-values less than 0.05 were considered statistically significant. In univariate Cox analysis, HR > 1 is considered risk factors, while HR < 1 is considered protective factors.

We established a systematic workflow for constructing the epigenetic-related signature of pan-cancer (Figure 1). In the initial screening stage, we performed univariate Cox analysis on 990 epigenetic-related genes and identified 59 candidate genes (p < 0.0001, HR ≤ 0.7, HR ≥ 1.3) Supplementary Table S2). To further optimize feature selection, we conducted LASSO regression analysis on these candidate genes using the TCGA pan-cancer training set (Figures 2A, B). Through stepwise Cox proportional hazards regression modeling, we ultimately identified 33 key epigenetic-related genes with important prognostic value, and established an epigenetic score based on the standardized expression levels of these genes.

To elucidate the roles of various epigenetic modifications in cancer progression, we categorize epigenetic genes based on their functions into two groups: protein modification-related genes and RNA modification-related genes. To gain deeper insights into the distinctive characteristics of 33 epigenetic associated biomarkers, we conducted a comprehensive analysis comparing their expression profiles across tumor and corresponding normal tissue specimens. Our investigation aimed to elucidate the molecular variations in these critical biomarkers between cancerous and non-cancerous tissue environments. We discovered that protein modification-related, including HDGF, SIRT-7, UBE2A, and UCHL5, elevated expression in tumor tissues. Similarly, the RNA modification-related genes DHX35, DHX8, ENY2, and FTSJ1 also upregulated in most tumor tissues. These findings suggest that epigenetic gene expression undergoes significant alterations in tumor tissues compared to normal tissues, indicating that epigenetic dysregulation may be a critical mechanism underlying tumor initiation (Figure 2C).

Notably, through univariate Cox analysis (Figure 2D), we found that these epigenetic-related genes did not exhibit clear protective or risk characteristics in most cancer types. This result suggests that epigenetic regulation may play a more complex role in the tumor progression process, rather than a simple promoting or inhibiting relationship. This complexity also highlights the necessity of further investigating the underlying mechanisms of epigenetic regulatory networks in tumor development.

Our systematic analysis of ES across 32 different cancer types revealed significant variations in epigenetic profiles among tumors from different organ origins. Nineteen cancer types, including glioblastoma (GBM), mesothelioma (MESO), esophageal carcinoma (ESCA), cholangiocarcinoma (CHOL), and uterine carcinosarcoma (UCS) and other types exhibited relatively high epigenetic scores. In contrast, 11 malignant neoplasms, such as kidney chromophobe (KICH), prostate adenocarcinoma (PRAD), thyroid carcinoma (THCA), testicular germ cell tumors (TGCT), thymoma (THYM), pheochromocytoma and paraganglioma (PCPG), breast invasive carcinoma (BRCA), uterine corpus endometrial carcinoma (UCEC), kidney renal papillary cell carcinoma (KIRP), uveal melanoma (UVM), and adrenocortical carcinoma (ACC), displayed lower epigenetic scores.

We analyzed the differences in epigenetic scores across different types of tumors (Supplementary Figure S1). Our research findings indicate that in most cancer types, such as bladder cancer (BLCA), breast cancer (BRCA), colon cancer (COAD), gastrointestinal cancer (KIH), Kidney Inferred Carcinoma (KIRC), Kidney Papillary Cell Carcinoma (KIRP), liver cancer (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), TGCT, and thyroid cancer (THCA), there is a significant upward trend in epigenetic scores as tumor staging increases. This finding suggests a significant correlation between tumor progression and epigenetic characteristics. However, in some tumor types, the epigenetic scores did not show a significant increase with disease progression, which may be related to insufficient clinical samples or missing information. Furthermore, in tumor types such as cervical cancer (CESC) and endometrial cancer (UCEC), the epigenetic scores also exhibit a gradual increase with higher tumor grades. This further supports the close relationship between tumor progression and epigenetic mechanisms.

Further analysis revealed that the epigenetic score held significant prognostic value across most cancer types (Supplementary Figure S2). In the majority of tumor types, the high epigenetic score acted as a risk factor and was significantly associated with patients’ progression-free interval (PFI), overall survival (OS), and disease-specific survival (DSS). Notably, the high epigenetic score was identified as a risk factor for PFI across all cancer types, with only THYM showing a protective effect in OS analysis and diffuse large B-cell lymphoma (DLBC) and PCPG exhibiting protective roles in DSS analysis. Interestingly, even though KIRC and brain lower grade glioma (LGG) had moderate epigenetic scores, the epigenetic score remained a strong risk factor for predicting the prognosis of these two cancer types (Figure 3A).

To assess the prognostic reliability of our epigenetic score, we segmented patients in the TCGA training cohort into high epigenetic score group and low epigenetic score group. Our analysis revealed that individuals in the high epigenetic score group demonstrated significantly reduced Disease-Specific Survival (DSS), Overall Survival (OS), and Progression-Free Interval (PFI) compared to the low-risk cohort (Figure 3B). These findings were subsequently validated in the independent testing cohort, which consistently demonstrated that the high-risk group experienced markedly inferior survival trajectories (Figure 3C). These results not only confirm the reliable prognostic prediction capability of our epigenetic score but also suggest its potential broad applicability across pan-cancer.

To translate the epigenetic score into a practical clinical tool, we developed an integrative prediction model. This model was presented in the form of a nomogram (Figure 4A), which not only incorporated the epigenetic score but also integrated key clinical features such as age and tumor type, aiming to provide clinicians with an intuitive and comprehensive prognostic assessment tool.

To validate the reliability of the model, we conducted multi-dimensional performance evaluations. The calibration curve analysis showed that the model’s predictions of 3-year and 5-year overall survival were highly consistent with the actual observations (Figure 4B), confirming the accuracy of the predictions. In terms of predictive efficacy, the model achieved AUC values of 0.76 and 0.77 in the training and testing sets, respectively (Figure 4C), indicating its stable and reliable predictive capability. Further, decision curve analysis (DCA) demonstrated that the integrated model generated positive net benefits across various decision thresholds (Figure 4D). Although the integrated model did not show statistically significant advantages over using the epigenetic score alone, its provision of a multi-dimensional risk assessment is of great clinical relevance. For instance, we found that patients with glioblastoma (GBM) and pancreatic adenocarcinoma (PAAD) had significantly lower expected survival rates compared to those with testicular germ cell tumors (TCGT) and prostate adenocarcinoma (PRAD). This risk stratification based on multiple factors not only helps clinicians identify high-risk patients but also provides a scientific basis for developing personalized treatment strategies.

In summary, the integration model of epigenetic scores and clinical features can significantly improve the accuracy of prognostic prediction and provide clinicians with a reliable decision-support tool. These findings emphasize the important value ofepigenetic score in modern precision oncology.

The development of malignant tumors is a complex biological transformation process. During tumor progression, normal cells may acquire a series of key characteristics, including increased glycolysis, EMT, sustained cell proliferation and so on. To deeply explore the intrinsic connection between epigenetic regulation and tumor biological features, we employed the ssgsea algorithm to systematically quantify the activities of cancer-related pathways in pan-cancer.

Through comprehensive analysis of pan-cancer data, we found that the epigenetic score exhibited significant positive correlations with key tumor biological processes. Specifically, the epigenetic score was positively correlated with the expression of genes related to glycolysis (R = 0.27, p < 0.001), EMT (R = 0.3, p < 0.001), and cell cycle (R = 0.49, p < 0.001), DNA repair (R = 0.17, p < 0.001), angiogenesis (R = 0.29, p < 0.001), inflammation (R = 0.43, p < 0.001) (Figure 5). This funding reveals that tumors with elevated epigenetic scores are associated with enhanced metabolic, cellular, and microenvironmental characteristics, including increased glycolytic metabolism, more aggressive EMT, higher cell cycle proliferation, improved DNA repair mechanisms, enhanced angiogenesis, and heightened inflammatory response. These features indicate an aggressive phenotype.

In further cancer type specific analyses, we observed interesting differences. In most tumor types, the expression of EMT-related genes was positively correlated with the epigenetic score, but in a few tumor types, such as cholangiocarcinoma (CHOL), melanoma (SKCM), and uterine carcinosarcoma (UCS), this relationship was negatively correlated (Supplementary Figure S3). The expression patterns of genes related to glycolysis, cell cycle, DNA repair, angiogenesis, inflammatory response also showed complex variability, with most tumor types exhibiting positive correlations between gene expression and epigenetic score (Supplementary Figures S3–5). It is worth noting that the correlation between epigenetic scores and tumor biological characteristics varies across different types of cancer. Different tumor types exhibit unique epigenetic regulation patterns, and this heterogeneity reflects the complexity and individual differences of epigenetic regulation in pan-cancer.

Our comprehensive analysis revealed that tumors characterized by elevated epigenetic scores demonstrated more aggressive and invasive biological properties. A critical research objective was to determine whether the biological mechanisms underlying the high-risk group were intrinsically linked to tumor invasion processes. To elucidate these potential correlations, we systematically divided the samples into two distinct groups based on their epigenetic score stratification and conducted an extensive functional enrichment analysis. This investigation encompassed a multifaceted approach, including KEGG pathway annotations, biological process (BP) characterizations, cellular component (CC) assessments, and molecular function (MF) evaluations.

In the high risk group, we found a series of functions enriched that are relevant to tumor invasion (Figure 6A). For molecular functions, we observed enrichment of tubulin binding, microtubule binding, cytokine activity, extracellular matrix structural constituent, and growth factor binding. For biological processes, we found enrichment of organelle fission, nuclear division, chromosome segregation, mitotic nuclear division, and nuclear chromosome segregation. Similarly, for cellular components, we saw enrichment of spindle, condensed chromosome, chromosome, and centromeric region. Furthermore, for KEGG pathways, we identified enrichment of cytokine-cytokine receptor interaction, cell cycle, IL-17 signaling pathway, viral protein interaction with cytokine and cytokine receptor, and pertussis.

In contrast, in the high risk group, we observed downregulation of functions related to metabolic regulation and differentiation, including KEGG pathways such as Cushing’s syndrome, estrogen signaling pathway, insulin secretion, thyroid hormone synthesis, and cortisol synthesis and secretion; molecular functions like ligand-activated transcription factor activity and nuclear receptor activity; cellular components such as neuronal cell body, ion channel complex, cation channel complex, and potassium channel complex; and biological processes including sex differentiation, male sex differentiation, branching morphogenesis of an epithelial tube, phenol-containing compound metabolic process, and thyroid hormone generation (Figure 6B).

In summary, tumors with higher epigenetic scores exhibited functional characteristics associated with cytoskeleton remodeling, cell cycle, cell division, and cell invasion, while those with lower epigenetic scores tended to be enriched for metabolic regulation and differentiation related functions. These findings provide new insights into the role of epigenetic regulation in tumor progression.