Multi-gene DNA methylation profiles of tumor suppressor genes for prognostic prediction in gastric cancer

Abstract

Background

This study aimed to identify prognostic biomarkers for gastric cancer (GC) by analyzing the methylation status of multiple tumor suppressor genes.

Methods

Using the Epi-TOP™ methylation detection system, we analyzed 51 genes in 169 matched tumor and adjacent normal tissue samples. Methylation levels were quantified as Percent Methylated Reference (PMR) in tumor (PMR-T) and normal (PMR-N) tissues; the differential methylation (PMR-D) was also calculated.

Result

Tumor tissues exhibited significantly higher DNA methylation levels than matched normal tissues across 51 tumor suppressor genes (all p < 0.001). Clustering analysis based on PMR-T identified four epigenetic subtypes associated with known molecular classifications (epithelial–mesenchymal transition (EMT) and microsatellite instability–high (MSI-H)) and overall survival (p = 0.030). In contrast, clustering based on PMR-N showed no significant association with molecular subtypes or survival outcomes, suggesting limited prognostic relevance. Two prognostic gene panels were constructed: one PMR-T–based panel (ALX, BMP3, CDKN2A, MINT25, PTGDR) and another PMR-D-based panel (ADCYAP1, SOCS1, SEPTIN9, CDKN2B). Both panels independently predicted overall survival in multivariate Cox regression. The PMR-D panel demonstrated stronger prognostic performance (hazard ratio (HR) = 0.329, p = 0.002), while the PMR-T panel also demonstrated significant prognostic value (HR = 0.512, p = 0.012), highlighting that tumor methylation profiles alone may provide meaningful survival predictions for patients with GC.

Conclusion

This study demonstrates that tumor-specific DNA methylation changes, particularly when evaluated using multi-gene panels can enhance prognostic stratification in GC. These findings support the potential use of methylation-based biomarkers for personalized management of GC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-15620-3.

Introduction

Gastric cancer (GC) remains a major global health concern, ranking as the fifth most common malignancy and the third leading cause of cancer-related mortality worldwide [1]. Despite a gradual decline in incidence, overall survival remains unsatisfactory [2, 3]. Current prognostic indicators based on conventional clinicopathological variables, including the TNM staging system, have clinical limitations in accurately predicting patient outcomes, as they do not fully capture the biological heterogeneity of GC [4]. This limitation has driven increasing interest in molecular biomarkers that more directly reflect tumor biology. Among these, DNA methylation has emerged as a promising biomarker due to its relative stability, cancer-specific patterns, and potential clinical applicability.

Epigenic alterations–particularly aberrant DNA methylation–is a well-established hallmark of carcinogenesis [5–7]. In GC, promoter hypermethylation of several tumor suppressor genes has been reported and linked to adverse outcomes [8–10]. However, most studies have focused on single-gene candidates, which may insufficiently represent the complexity of the GC epigenetic landscape. Recent studies across cancer types suggests that multi-gene molecular panels, rather than single-markers, provide superior prognostic and diagnostic performance [11–13]. Consistent with this, methodological standards established through methylation-panel studies in other malignancies, including breast cancer, emphasize the advantages of multi-gene epigenetic signatures for early detection and patient stratification [14]. In addition, pan-cancer analyses of circulating cell-free DNA methylation demonstrate that methylation alterations are consistent across tissue and liquid biopsy contexts, underscoring their translational potential [15]. Despite these advances, the application of such multi-gene methylation panels in GC remains limited. Given the intricate epigenetic alteration characteristic of GC, such multi-gene panels may hold promise for improving prognostic evaluation in this malignancy.

In this study, we aimed to identify prognostic biomarkers for GC by analyzing the methylation status of multiple tumor suppressor genes. Unlike the conventional bisulfite polymerase chain reaction (PCR) methods, we used a novel technique called Epi-TOP methylation detection. This method employs a modified peptide nucleic acid (PNA) probe with higher binding affinity for methylated cytosine than for unmethylated cytosine [16, 17]. Using this approach, we compared matched tumor and adjacent normal tissue samples from patients with GC to screen for methylation markers associated with patient survival and to evaluate their overall prognostic significance. Based on this analysis, we aimed to propose a panel of methylation markers with potential prognostic utility for clinical application in GC management.

Materials and methods

Study design and sample collection

A total of 169 patients who underwent radical gastrectomy for GC at Seoul National University Hospital (SNUH) between 2009 and 2021 were included. Formalin-fixed, paraffin-embedded (FFPE) tissue specimens—including both tumor and adjacent matched normal tissues—were obtained from the Department of Pathology, SNUH. Clinical and pathological data, including patient age, sex, histologic subtype, and cancer stage (according to the 8th edition of the American Joint Committee on Cancer staging system), were retrospectively collected from electronic medical records and pathology reports. Overall survival (OS) was defined as the time from the date of surgical resection to the date of death or last follow-up and was recorded accordingly.

Molecular subtypes were assigned according to a previously published, algorithm-based gastric cancer classification framework that applies a hierarchical and mutually exclusive strategy [18]. Samples were sequentially classified as EBV-positive, MSI-H (among EBV-negative tumors), EMT-like (within EBV-negative and microsatellite-stable tumors), and finally according to p53 status. Consequently, each sample was assigned to only one molecular subtype.

This study was approved by the Institutional Review Board of Seoul National University Hospital (IRB No. H-2106-215-1231).

Sample preparation and DNA extraction

For each patient, five consecutive 5-µm-thick sections were obtained from a single FFPE tissue block. Genomic DNA was then extracted separately from the macrodissected tumor and normal tissues using the Panamax™ FFPE DNA Extraction Kit (Panagene, Daejeon, Korea), following the manufacturer’s instructions. The concentration and purity of the extracted DNA were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, MA, USA). All DNA samples were stored at − 20 °C until further downstream analyses, including DNA methylation profiling.

DNA methylation testing

Methylation analysis was conducted using the Epi-TOP™ Tumor Suppressor Assay/MPP assay (Seasun Biomaterials, Daejeon, Korea) [16, 19], which evaluates the relative methylation levels of 62 tumor suppressor genes previously reported to exhibit differential methylation patterns in solid tumors. Methylation PCR was conducted using this assay, and 51 genes with complete data (excluding those with missing values) were included in the final analysis. All assays were performed using validated, commercially available primer sets under standardized conditions, with appropriate no-template and positive controls included in each run.

Data interpretation involved comparing the cycle threshold (Ct) values of the internal control and target genes in clinical specimens. Percent Methylated Reference (PMR) was calculated using the formula:

∆∆Ct was defined as:

The Zymo > 90% fully methylated DNA standard was used as the reference to minimize biological variability and served as a reliable, consistent positive control for methylation detection [20]. Methylation positivity was defined as a PMR value of 10 and was applied uniformly to both tumor and matched normal tissue samples. PMR values for normal and tumor tissues were denoted as PMR-N and PMR-T, respectively. Furthermore, the difference in methylation levels between tumor and matched normal tissues was calculated and defined as PMR-D (PMR difference).

For score calculation, methylation status was dichotomized using panel-specific thresholds. For the PMR-T, markers with PMR values greater than 10 were considered methylated and assigned a value of 1. For the PMR-D panel, markers with PMR-D values greater than 0 were considered methylated. Composite scores were calculated by summing the dichotomized methylation statuses across all markers included in each panel.

Tissue microarray (TMA) construction and immunohistochemistry (IHC)

FFPE tissues were used to construct tissue microarray (TMA) blocks (Superbiochips Laboratories, Seoul, Republic of Korea) [21]. For each case, a representative tumor area was selected, and a 2 mm core was extracted and arrayed into recipient blocks. Immunohistochemical (IHC) staining was performed for p53 (dilution 1:1000; DAKO, Santa Clara, CA, USA) and E-cadherin (dilution 1:800; BD Biosciences, San Jose, CA, USA) using the Benchmark XT autostainer (Ventana Medical Systems, Tucson, AZ, USA), following the manufacturer’s protocols. For E-cadherin, altered expression was defined as either loss of membranous staining or the presence of aberrant cytoplasmic staining in ≥ 30% of tumor cells. For p53, altered expression was defined as either complete loss of nuclear staining or strong, diffuse nuclear positivity.

Microsatellite instability (MSI)

Microsatellite instability (MSI) status was determined using the National Cancer Institute-recommended panel of five microsatellite markers: BAT-26, BAT-25, D5S346, D17S250, and D2S123. Genomic DNA was extracted from FFPE tumor tissues and their corresponding normal tissues. PCR amplification of the selected loci was performed, and the resulting products were analyzed using an ABI 3730 DNA Analyzer (Applied Biosystems, Foster City, CA, USA). Samples were classified as microsatellite instability-high (MSI-H) if instability was detected in two or more markers, and as microsatellite stable (MSS) if only one or none of the markers showed instability [22].

Epstein–Barr virus (EBV) in situ hybridization

Detection of Epstein–Barr virus (EBV) was performed using in situ hybridization for EBV-encoded small ribonucleic acids (RNAs) (EBERs). TMA sections were processed and hybridized using the INFORM EBER Probe (Ventana Medical Systems, Tucson, AZ, USA) on the Benchmark XT autostainer (Ventana Medical Systems), following the manufacturer’s protocol. EBV positivity was defined by the presence of distinct nuclear staining in tumor cells, as observed under light microscopy.

Statistical analysis

All statistical analyses were conducted using R software (version 4.4.0, R Development Core Team, Vienna, Austria). Differences in DNA methylation levels between tumor and matched normal tissues were assessed using paired statistical tests, such as the paired t-test or Wilcoxon signed-rank test, depending on data distribution. Receiver operating characteristic (ROC) curve analysis was conducted to evaluate the diagnostic performance of each methylation marker, with the area under the curve (AUC) and 95% confidence intervals (CIs) calculated using the pROC package. Unsupervised K-means clustering was applied to classify samples based on methylation profiles. The optimal number of clusters was determined using the Elbow method with elbow points observed at K = 3 and K = 4, which were therefore considered appropriate choices (Fig. S1). Survival analysis was conducted using Kaplan–Meier curves and the log-rank test. For markers showing statistically significant associations with survival, multivariate analysis using the Cox proportional hazards model was performed to identify independent prognostic markers. Model performance was internally validated using bootstrap resampling (1,000 iterations), and optimism-corrected C-indices were calculated. Associations between methylation clusters and clinicopathological characteristics were evaluated using the chi-square test, Fisher’s exact test, or Kruskal–Wallis test, as appropriate. A two-sided p-value < 0.05 was considered statistically significant.

Results

Differential DNA methylation in tumor versus normal tissues

To evaluate differences in DNA methylation patterns between tumor and normal tissues, we analyzed the PMR values ​​of 51 tumor suppressor gene markers in 169 tumor tissues and their corresponding normal tissues using the Epi-TOP™ tumor suppressor gene analysis/MPP system. Tumor tissues exhibited significantly higher PMR values ​​than normal tissues (p < 0.001; Fig. S2). An overview of DNA methylation patterns across all tumor suppressor genes and samples is shown in a heatmap (Fig. S3). In addition, gene-specific heatmaps illustrating methylation differences between tumor and matched normal tissues are provided to facilitate detailed comparison at the individual gene level (Fig. S4). Based on these global methylation patterns and survival analysis, markers were selected to construct the PMR-T panel, and their tumor-associated methylation patterns are presented in Fig. 1. The PMR-T panel comprises five tumor suppressor gene markers (ALX, BMP3, CDKN2A (p14), MINT25, and PTGDR). These panel-associated genes demonstrated consistently elevated methylation levels in tumor tissues compared with matched normal tissues (all p < 0.05). In addition, ROC curve analysis of the PMR-T model confirmed high AUC values, indicating strong discriminatory performance between tumor and normal tissues. Detailed methylation profiles of all analyzed genes are provided in Table S1.

Fig. 1.

Fig. 1. Heatmaps showing DNA methylation patterns of markers selected for the PMR-T panel in 169 paired gastric tumor and adjacent normal tissues. The PMR-T panel consists of five tumor suppressor gene markers (ALX, BMP3, CDKN2A (p14), MINT25, and PTGDR). Receiver operating characteristic (ROC) curves illustrate the discriminatory performance of the five markers, with performance represented by the area under the curve (AUC) values. Box plots compare median PMR values between tumor and normal tissues for the five differentially methylated genes (***p < 0.001, **p < 0.01, *p < 0.05)

Epigenetic classification by PMR-T in relation to subtypes and prognosis

To further classify the epigenetic profiles of tumor samples, K-means clustering was performed using PMR-T values of 51 tumor suppressor genes, resulting in four distinct methylation-based clusters. These clusters were defined based on the degree of tumor-specific gene hypermethylation observed in the heatmap (Fig. 2A): Group 1 had the fewest hypermethylated genes, with the degree of hypermethylation progressively increasing through Groups 2 and 3, and Group 4 exhibiting the highest level of gene hypermethylation. These clusters showed significant associations with established molecular subtypes of GC (p < 0.001). Specifically, Groups 1 and 3 were enriched for the EMT-like subtype, while Groups 2 and 4 were predominantly associated with the MSI-H subtype (Fig. 2B).

Fig. 2.

Clustering of DNA methylation profiles and their association with molecular subtypes and clinical outcomes. A Heatmap of methylation profiles on PMR Tumor values for 51 tumor suppressor genes in gastric cancer samples, revealing four distinct methylation-based clusters. The degree of tumor-specific gene hypermethylation increased progressively from Group 1 to Group 4. B Bar plots displaying the distribution of gastric cancer molecular subtypes across the four methylation clusters. Each sample was assigned to a single molecular subtype using a hierarchical classification approach. C Kaplan–Meier survival curves comparing overall survival within the four methylation-based clusters, and evaluated using the log-rank test

Survival analysis revealed significant differences in overall survival across the four clusters (log-rank p = 0.030) (Fig. 2C). Clinicopathological characteristics further supported the distinct features of each group. Group 4 was associated with a higher frequency of the MSI-H subtype, poorly differentiated (PD) histology, and a lower incidence of perineural invasion. In contrast, Group 1 was predominantly characterized by the EMT-like subtype and a higher frequency of poorly cohesive carcinoma (PCC) (Table 1).

Table 1.

Clinicopathological characteristics of each group, according to the classification by PMR-T

Group	1 N=(54)	2 (N = 40)	3 (N = 32)	4 (N = 23)	p-value
Sex					0.222
Female	14 (25.9%)	13 (32.5%)	5 (15.6%)	9 (39.1%)
Male	40 (74.1%)	27 (67.5%)	27 (84.4%)	14 (60.9%)
Age					0.264
< 65	25 (46.3%)	15 (37.5%)	16 (50.0%)	6 (26.1%)
>= 65	29 (53.7%)	25 (62.5%)	16 (50.0%)	17 (73.9%)
Subtype					< 0.001
EBV	2 (3.7%)	3 (7.5%)	1 (3.1%)	4 (17.4%)
EMT	25 (46.3%)	6 (15.0%)	12 (37.5%)	1 (4.3%)
MSI	2 (3.7%)	15 (37.5%)	1 (3.1%)	13 (56.5%)
p53-Negative	7 (13.0%)	6 (15.0%)	8 (25.0%)	2 (8.7%)
p53-Positive	18 (33.3%)	10 (25.0%)	10 (31.2%)	3 (13.0%)
Histologic					0.045
WD	1 (1.9%)	1 (2.5%)	1 (3.1%)	1 (4.3%)
MD	18 (33.3%)	21 (52.5%)	14 (43.8%)	9 (39.1%)
PD	13 (24.1%)	10 (25.0%)	9 (28.1%)	10 (43.5%)
mixed	0 (0.0%)	3 (7.5%)	0 (0.0%)	0 (0.0%)
PCC	16 (29.6%)	5 (12.5%)	5 (15.6%)	1 (4.3%)
mucinous	4 (7.4%)	0 (0.0%)	0 (0.0%)	1 (4.3%)
papillary	2 (3.7%)	0 (0.0%)	3 (9.4%)	1 (4.3%)
Stage					0.202
1	3 (5.6%)	6 (15.0%)	1 (3.1%)	4 (17.4%)
2	19 (35.2%)	10 (25.0%)	11 (34.4%)	6 (26.1%)
3	26 (48.1%)	24 (60.0%)	16 (50.0%)	12 (52.2%)
4	6 (11.1%)	0 (0.0%)	4 (12.5%)	1 (4.3%)
Lauren					0.254
diffuse	25 (46.3%)	8 (20.0%)	12 (37.5%)	8 (34.8%)
intestinal	25 (46.3%)	28 (70.0%)	18 (56.2%)	12 (52.2%)
mixed	4 (7.4%)	4 (10.0%)	2 (6.2%)	3 (13.0%)
Border					0.543
expanding	3 (5.6%)	4 (10.3%)	3 (9.4%)	3 (13.0%)
infiltrative	51 (94.4%)	35 (89.7%)	28 (87.5%)	20 (87.0%)
not applicable	0 (0.0%)	0 (0.0%)	1 (3.1%)	0 (0.0%)
Lymphatic					0.346
not identified	10 (18.5%)	14 (35.0%)	9 (28.1%)	6 (26.1%)
present	44 (81.5%)	26 (65.0%)	23 (71.9%)	17 (73.9%)
Venous					0.385
not identified	30 (55.6%)	26 (65.0%)	15 (46.9%)	15 (65.2%)
present	24 (44.4%)	14 (35.0%)	17 (53.1%)	8 (34.8%)
Perineural					0.021
not identified	18 (33.3%)	20 (50.0%)	12 (37.5%)	16 (69.6%)
present	36 (66.7%)	20 (50.0%)	20 (62.5%)	7 (30.4%)

Abbreviations: MSI Microsatellite instability, EMT Epithelial–mesenchymal transition, EBV Epstein-Barr virus, WD Well differentiated, MD Moderately differentiated, PD Poorly differentiated, PCC poorly cohesive carcinoma

Epigenetic clustering by PMR-N and PMR-D

To determine whether the methylation pattern of normal tissue or the difference between tumor and normal tissues influences tumor classification, additional K-means clustering analyses were performed using the PMR-N and the PMR-D of 51 tumor suppressor genes. The optimal number of clusters was determined using the Elbow method, yielding three clusters for both analyses. In the clustering analysis of PMR-N, Group 1 had the fewest hypermethylated genes, while Group 3 exhibited the highest methylation levels (Fig. 3A). Despite these differences, no significant association was observed between the clusters and the molecular subtypes of GC (p = 0.699). Similarly, there was no statistically significant difference in overall survival (p = 0.922; Fig. 3B).

Fig. 3.

Clustering analyses on methylation levels of normal tissues (PMR-N) and tumor-normal differences (PMR-D). A Heatmap of methylation profiles in PMR-N based on K-means clustering of 51 tumor suppressor genes, identifying three clusters with varying degrees of baseline methylation. B Bar plots displaying the distribution of gastric cancer molecular subtypes across the PMR-N–based clusters, and Kaplan–Meier survival curves comparing overall survival within of the PMR-N clusters, and evaluated using the log-rank test. C Heatmap of PMR-D, revealing three clusters with distinct levels of tumor-specific hypermethylation. D Bar plots displaying the distribution of gastric cancer molecular subtypes across the PMR-D–based clusters, and Kaplan–Meier survival curves of the PMR-D based clusters, and evaluated using the log-rank test

Clustering analysis based on PMR-D, which reflects the methylation difference between tumor and normal tissues, also identified three groups with varying degrees of tumor-specific hypermethylation. Group 1 showed the least differential methylation, while group 3 had the most hypermethylated genes (Fig. 3C). This clustering showed a significant association with molecular subtypes (p = 0.003). Groups 1 and 2 were predominantly characterized by the EMT-like subtype, while group 3 was primarily associated with the MSI-H subtype. Nevertheless, survival analysis revealed no statistically significant differences in overall survival across the PMR-D clusters (p = 0.267; Fig. 3D).

Survival prediction using multiple DNA methylation markers

The epigenetic classification described above was based on 51 methylation markers, which may be too numerous for practical clinical application. Therefore, we constructed survival panels using selected markers and evaluated the prognostic utility of multiple DNA methylation markers. For PMR-T values, methylation was defined as PMR > 10. Among the markers with p-values < 0.2 in Kaplan–Meier survival analysis, five markers (ALX, BMP3, CDKN2A (p14), MINT25, and PTGDR) were selected. Each methylated marker was assigned one point, and samples were scored accordingly (range: 1–4). Based on the score distribution, patients were classified into a low-score group (1–2) and a high-score group (3–4). Survival analysis revealed a significant difference in overall survival between the two groups (log-rank p = 0.005, Fig. 4A), and multivariate Cox regression confirmed the panel as an independent prognostic factor after adjusting for clinical covariates (hazard ratio (HR): 0.512; 95% CI: 0.304–0.862, p = 0.012) (Fig. 4B).

Fig. 4.

Prognostic significance of a methylation-based survival panel constructed using selected tumor suppressor genes. A, B Kaplan–Meier survival curves comparing overall survival between low-risk (1–2 methylated markers) and high-risk (3–4 methylated markers) groups, based on PMR-T values of five selected markers (ALX, BMP3, CDKN2A (p14), MINT25, and PTGDR), with risk stratification assessed by univariate and multivariate Cox regression analyses. C, D Kaplan–Meier survival curves based on PMR-D values using four selected markers (ADCYAP1, SOCS1, SEPTIN9, and CDKN2B (p15)), with patient groups classified into low (0–2 markers) and high (3–4 markers) methylation categories, and prognostic significance evaluated by univariate and multivariate Cox regression analyses.

Similarly, a panel was developed based on the PMR-D, where methylation was defined as PMR > 0. Among markers with p-values < 0.1, four markers were selected (ADCYAP1, SOCS1, SEPTIN9, and CDKN2B (p15)). Patients were scored according to the number of methylated markers (range: 0–4) and categorized into low (0–2) and high (3–4) groups. This panel also demonstrated statistically significant differences in overall survival between the groups (log-rank p < 0.001, Fig. 4C) and retained significance in multivariate analysis as an independent prognostic factor (HR: 0.329, 95% CI: 0.162–0.666, p = 0.002) (Fig. 4D). PMR-T panel achieved an optimism-corrected C-index of 0.60, whereas the paired PMR-D panel showed a higher optimism-corrected C-index of 0.64. For both panels, the optimism was negligible, indicating robust internal performance (see Table S2).

Discussion

In this study, we analyzed the DNA methylation profiles of tumor suppressor genes in GC by comparing tumor tissues with their corresponding matched normal tissues. Our results revealed that PMR values were significantly elevated in tumor samples, aligning with the widespread hypermethylation typically observed in the cancer epigenome [8]. Notably, genes such as CD1D, ADCYAP1, NPTX2, PENK, and MSC exhibited significantly increased methylation levels in tumor tissues and demonstrated strong discriminatory power in ROC analyses, underscoring their potential as diagnostic biomarkers for GC.

Methylation-based classification using K-means clustering of PMR-T values identified four distinct epigenetic clusters with varying levels of tumor-specific hypermethylation. These clusters correlated with known GC molecular subtypes, with Groups 1 and 3 were predominantly EMT-like subtype, and Groups 2 and 4 were predominantly MSI-H subtype, despite sharing the same molecular subtype, these groups exhibited significant differences in clinical behavior. In particular, Group 1 (EMT-like predominant) had the worst prognosis, frequent peritoneal invasion, and was characterized by PCC, whereas Group 3 (also EMT-like predominant) had the best prognosis, highlighting heterogeneity within the EMT-like subtype. Recent studies have emphasized that cancer cell plasticity enables dynamic and reversible cell-state transitions underlying such phenotypic diversity, with DNA methylation acting as a key epigenetic regulator of these processes [23]. In this context, the methylation-defined clusters observed in our study–especially within the EMT-like subtype–may reflect epigenetic states associated with phenotypic flexibility of GC cells rather than static prognostic identities. To investigate the contribution of normal tissue methylation, additional cluster analysis was performed using PMR-N. Distinct methylation patterns were evident across the three groups; however, clustering based on normal tissue methylation showed no correlation with molecular subtypes or overall survival. This suggests that, while methylation patterns in normal tissues are diverse among patients with GC, they have a relatively limited impact on molecular classification or prognostic stratification in GC.

Furthermore, our findings support the development of prognostic panels based on epigenetic alterations, particularly through multi-gene methylation combinations. The PMR-T panel—composed of ALX1, BMP3, CDKN2A (p14), MINT25, and PTGDR—includes genes involved in cell cycle regulation and developmental pathways [24–28]. The concurrent epigenetic silencing of these genes may amplify oncogenic signaling in GC. Similarly, the PMR-D panel—comprising ADCYAP1, SOCS1, SEPTIN9, and CDKN2B (p15) —highlights the prognostic relevance of tumor-specific methylation changes, particularly those reflecting disruptions in cell signaling, cytokine regulation, and tumor suppressor pathways [29–32]. Notably, both panels demonstrated independent prognostic significance in multivariate Cox regression analysis, regardless of clinical covariates. The PMR-D panel exhibited stronger prognostic performance; however, the PMR-T panel also showed significant predictive value. These results emphasize that even tumor-derived methylation profiles alone can offer meaningful survival prediction in patients with GC. Beyond their prognostic utility, DNA methylation biomarkers may also contribute to therapeutic development and decision-making. Biomarker-driven approaches have been shown to accelerate drug development by shortening the track between biological insight and clinical evaluation [33]. Accordingly, the multi-gene methylation panel proposed in our study may provide informative clues about disease biology and potential treatment sensitivity, supporting the design and prioritization of therapeutic strategies.

Traditionally, DNA methylation analysis has relied on bisulfite treatment [34]. However, bisulfite-based methods often have low diagnostic accuracy and reproducibility due to DNA template degradation during the conversion process [35]. To address these limitations, we employed a novel technology called Epi-TOP methylation detection, which uses a modified PNA probe. This PNA probe can distinguish with single-base resolution without requiring prior conversion treatments such as bisulfite. Methylation levels are evaluated by comparing the Ct values of the target gene and a control gene unaffected by methylation [19].

The results of this study are promising; however, some limitations should be acknowledged. First, the study was conducted as a retrospective analysis at a single center, which may introduce selection bias and limit the generalizability of the findings. Second, the absence of an independent external validation cohort constrains our ability to confirm the prognostic performance of the methylation panels across diverse patient populations and clinical settings. Although external validation in independent, multicenter cohorts is required prior to clinical implementation, internal bootstrap validation demonstrated that the prognostic performance of the PMR-T and PMR-D panel was robust and not attributable to overfitting. Future studies involving large, multicenter prospective cohorts are essential to validate the robustness, reproducibility, and clinical utility of the proposed biomarkers.

Conclusions

In conclusion, we demonstrate that tumor-specific DNA methylation alterations—particularly when assessed through multi-gene panels—hold significant prognostic value in GC. Two independent methylation panels, based on PMR-T and PMR-D, effectively stratified patient prognosis, with the PMR-D panel exhibiting superior predictive performance. These findings suggest that methylation-based biomarkers may enhance risk stratification and support personalized treatment strategies in GC.

Abbreviations

GC

Gastric cancer

PMR

Percent Methylated Reference

PMR-T

Percent Methylated Reference for tumor tissue

PMR-N

Percent Methylated Reference for normal tissue

PMR-D

PMR difference (difference in PMR between tumor and matched normal tissue)

ROC

Receiver operating characteristic

AUC

Area under the curve

EMT

Epithelial–mesenchymal transition

MSI-H

Microsatellite instability–high

PCC

Poorly cohesive carcinoma

OS

Overall survival

HR

Hazard ratio

CI

Confidence interval

FFPE

Formalin-fixed paraffin-embedded

AJCC

American Joint Committee on Cancer

SNUH

Seoul National University Hospital

IRB

Institutional Review Board

PNA

Peptide nucleic acid

Ct

Cycle threshold

Authors’ contributions

Conceptualization of the research study was achieved by SKN, CB and HSL. Investigation and methodology were the responsibility of CB and EBK. Data curation was the responsibility of SKN, YK and HSL. Formal analysis and visualization were the responsibility of SKN, JP and YK. Resources were acquired by SHK, DJP , HJL and HKY. Writing-original draft preparation was conducted by SKN. Writing-review and editing was conducted by HSL. Supervision was performed by HSL. Project administration was the responsibility of HSL. Funding acquisition was the responsibility of HSL. All authors have read and approved the final manuscript.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (Ministry of Science and ICT) (No. RS-2024-00337984).

Data availability

The data generated in the present study may be requested from the corresponding author.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with protocols approved by the IRB of SNUH (IRB number: H-2106-215-1231) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants before participation.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.