In ESCC, aberrant DNA methylation is the most mature epigenetic hallmark. Large-scale epigenomic studies have identified widespread methylation alterations that correlate with patient outcomes^[23-25]. Genome-wide hypomethylation often affects repetitive elements, including long interspersed nucleotide sequences (LINE-1). This pattern has been associated with chromosomal instability and poor prognosis after curative tumor resection^[24,26]. Moreover, global methylation loss coexists with localized promoter CpG island hypermethylation, which can silence regulatory loci. One example is hypermethylation-mediated silencing of the tumor suppressor SEMA3B and its antisense lncRNA SEMA3B-AS1, which has been associated with ESCC progression and unfavorable survival^[27].In addition, multi-region methylation profiling has revealed both intertumor and intratumor methylation heterogeneity in ESCC. This heterogeneity correlates with lymph node metastasis, suggesting that subclonal epigenetic diversification may accompany selection of more aggressive tumor-cell populations^[28].

Research on DNA methylation biomarkers in ESCC is also expanding rapidly. For instance, CDH1 promoter hypermethylation is observed in 43% of ESCC patients and is associated with recurrence of early-stage ESCC^[29]. Also, the association between ESCC and promoter hypermethylation of DNA damage repair genes, such as MGMT and MLH1, has been well studied^[30,31]. Additionally, targeted bisulfite sequencing has generated multi-gene methylation panels for ESCC detection and identified several genomic regions that were able to act as biomarkers, including regions covering STK3 and ZNF418^[32]. Locus-specific hypermethylation, including FGF5 promoter methylation, has been reported as a candidate marker of sensitivity to definitive chemoradiotherapy^[33]. These findings highlight their clinical potential, but careful validation across independent cohorts, sample types, and treatment settings remains necessary.

Histone-modifying enzymes and structural chromatin states provide further mechanistic insight into ESCC pathogenesis. The available evidence is more pathway-specific than that for DNA methylation, but several studies support its functional relevance. The histone demethylase JMJD3 promotes ESCC pathogenesis through epigenetic regulation of MYC proto-oncogene (MYC)^[34]. Histone modifiers can also cooperate with transcriptional regulators to repress expression of non-coding RNAs. For example, SOX4 recruits EZH2 and Histone deacetylase 3 (HDAC3) to the miR-31 promoter, thereby silencing miR-31 expression and promoting an invasive cellular phenotype^[35]. In addition, super-enhancer remodeling contributes to oncogenic transcription in ESCC. Active H3K27 acetylation (H3K27ac) can activate the lncRNA CCAT1, which coordinates SPRY4 and HOXB13 expression, resulting in ESCC migration and proliferation^[36].

At the architectural level, chromatin remodeling complexes appear to be disrupted early during squamous carcinogenesis. Somatic mutations in SWI/SNF complex subunits have been reported as early events in esophageal squamous carcinogenesis^[37]. Furthermore, loss or downregulation of the core SWI/SNF subunit ARID1A is associated with infiltrative growth in established ESCC tumors^[38]. These structural changes usually intersect with lineage-specific transcriptional programs. The squamous transcription factors TP63, SOX2, and KLF5 form a core regulatory circuitry that binds open, hyper-accessible chromatin regions and reshapes enhancer usage^[39]. Chromatin accessibility profiling has also identified hyper-accessible enhancers and enhancer RNAs in ESCC cells, including a putative enhancer-derived enhancer RNA (eRNA) for Epidermal Growth Factor Receptor (EGFR)^[40].

Evidence for non-coding RNAs and m⁶A RNA modifications in ESCC is growing, but it remains immature. Interestingly, some ncRNAs not only act as downstream effectors, but also influence chromatin-modifying complexes or DNA methylation directly. The lncRNA POU3F3, for example, promotes DNA methylation in ESCC cells^[41]. Similarly, circIMMP2L promotes malignant progression by securing nuclear retention of the transcriptional corepressor CtBP1 and modulating downstream epigenetic modifications^[42]. Moreover, m⁶A RNA modification extends another epigenetic layer to ESCC cell-state control. The key point is not simply the presence of m⁶A marks, but their dynamic deposition and removal on mRNA transcripts. The m⁶A demethylase ALKBH5 suppresses esophageal cancer malignancy by regulating primary microRNA biogenesis and RAI1 expression^[43]. METTL14, a critical m⁶A methylase, participates in a positive feedback loop with miR-99a-5p and TRIB2 in radioresistant ESCC cell populations. This circuit recruits HDAC2, sustains a cancer stem cell phenotype, and promotes radioresistance^[44]. These molecule-specific mechanisms broaden the biological view of ESCC plasticity. However, the breadth of clinical validation available for RNA methylation biomarkers remains inadequate.

EAC has a different epigenetic context from ESCC. Its epigenetic landscape is closely linked to the inflammatory and metaplastic trajectory of Barrett’s esophagus. In this glandular lineage, widespread DNA hypomethylation occurs early in the premalignant metaplastic stage. It may also synergize with focal gene amplification during progression toward adenocarcinoma^[45]. Importantly, DNA methylation profiling has delineated molecular subclasses across Barrett’s esophagus and EAC^[46]. This classification becomes more informative when methylation profiles are integrated with transcriptomic and genomic data^[47]. One subtype-associated event is hypermethylation-mediated downregulation of NDRG4, a tumor-suppressive gene in EAC^[48]. Meanwhile, chromatin-level studies suggest that global accessibility shifts can establish a hybrid stomach-intestinal chromatin state that underlies metaplastic cellular reprogramming in human Barrett’s tissue^[49].

Overall, DNA methylation and localized chromatin states provide the strongest epigenetic evidence base for EAC progression. Direct evidence for EAC-specific histone modifiers, lncRNA scaffolds, and m⁶A RNA-modification pathways remains comparatively sparse. This imbalance matters for interpretation. ESCC and EAC share broad epigenetic mechanisms, but they deploy them within distinct lineage and developmental contexts. This distinct lineage context underscores the need to evaluate epigenetic mechanisms in ESCC and EAC independently, despite their shared molecular machinery.

Clinical translation remains limited and differs by histological subtype. No epigenetic biomarker or epigenetic drug is established in the routine management of ESCC. The DNA methylation panels, LINE-1 methylation, and FGF5 methylation remain candidate tissue markers that require prospective assay validation. In Barrett’s esophagus and EAC, non-endoscopic cell collection coupled to methylated VIM and CCNA1 testing has progressed. The EsoGuard assay has been analytically validated as a CLIA/CAP laboratory-developed test using samples collected with the FDA-cleared EsoCheck device^[50]. However, analytical validation alone does not establish population-level clinical utility. The 2022 American College of Gastroenterology (ACG) guideline found insufficient evidence to recommend routine biomarker panels. Therefore, positive molecular tests still require a confirmatory endoscopy^[51]. In addition, azacitidine priming has entered early-phase evaluation in resectable gastric and esophageal adenocarcinoma. However, neither DNA methyltransferase nor histone deacetylase inhibitor is used as standard treatment for ESCC or EAC^[52]. Future studies should prioritize prospective validation in well-defined risk populations, particularly for ESCC-specific methylation markers, where tumor heterogeneity and the lack of standardized assays remain major barriers to clinical implementation.

DNA methylation is the best-characterized epigenetic alteration in GC. Two broad patterns are repeatedly described: focal CpG island promoter hypermethylation and global hypomethylation. As observed in EC, promoter hypermethylation can silence tumor-suppressive or DNA-repair genes, whereas global hypomethylation often affects repetitive elements and may contribute to chromosomal instability and retrotransposon activity^[56-58]. Notably, genome-wide methylation studies have identified methylation-defined GC subgroups, indicating that DNA methylation functions both as a gene-specific regulatory event and as a tumor-classification feature^[55,59].

The EBV-positive and MSI subtypes illustrate two clinically recognizable methylation patterns. EBV-positive GC shows extensive CpG island hypermethylation, a feature that helps distinguish it from other molecular subtypes^[55]. In MSI tumors, promoter methylation of mismatch-repair genes, particularly MLH1, leads to mismatch-repair deficiency and microsatellite instability^[55]. These subtype-linked methylation patterns are useful because they link epigenetic alterations to established molecular classifications rather than treating GC as a single disease.

Helicobacter pylori infection provides another gastric cancer-specific link between inflammation and DNA methylation. Aberrant methylation can be detected in H. pylori-infected noncancerous gastric mucosa, supporting an epigenetic defect model in which histologically non-malignant tissue already carries cancer-associated methylation changes^[60]. This effect has practical implications. Methylation-based markers have been tested to stratify the risk of primary GC after H. pylori eradication, including in high-risk individuals who remain under endoscopic surveillance^[61]. This epigenetic defect model separates GC-specific DNA methylation biology from a purely tumor-centered view. The methylation signal in noncancerous mucosa may reflect previous inflammatory injury, persistent epithelial remodeling, or selection of altered epithelial clones. This explains why risk remains measurable after H. pylori clearance: although eradication removes the inflammatory stimulus, it does not necessarily erase the accumulated methylation changes^[60,61].

At the gene level, promoter hypermethylation has been reported in cell-cycle regulators, DNA-repair genes, adhesion-related genes, and apoptosis-associated genes, including CDKN2A/p16INK4a, MLH1, MGMT, CDH1, and CASP8^[62-66]. LINE-1 hypomethylation in upper gastrointestinal cancers has been associated with increased retrotransposition and tumor-specific insertions, providing a plausible connection between global hypomethylation and genomic instability^[56]. In GC models, DNA hypomethylation can also promote invasive behavior by regulating SP1 binding at the CDCA3 promoter^[67]. Taken together, abnormal DNA methylation patterns in GC and noncancerous gastric mucosa shed light on its clinical applications.

Compared with DNA methylation, research on histone modifications in GC has focused on specific mechanisms. For instance, EZH2, the catalytic subunit of PRC2, mediates H3K27 trimethylation and drives gastric tumorigenesis by silencing tumor suppressor genes^[68-71]. Furthermore, EZH2 can interact with the DNA methylation machinery, suggesting that polycomb-mediated repression and promoter methylation often cooperate to silence target genes^[69].

Other chromatin regulators appear to act in more restricted settings. Inhibition of the lysine demethylase KDM4C induces senescence in TP53-mutant GC cells, suggesting that dependency on histone demethylases may vary by genetic background^[72]. Chromatin remodeling defects show a similar subtype bias. The BAF subunit ARID1A is frequently altered in subsets of GC, especially MSI and EBV-associated tumors, and TCGA identified ARID1A among the most recurrent chromatin-regulatory alterations^[55,73]. These findings indicate that chromatin remodeling in GC is highly subtype-specific, rather than a general process across all tumors.

Non-coding RNA studies in GC cover many molecules and phenotypes, so the evidence is less uniform than that for DNA methylation. Several miRNAs have been linked to drug response, tumor progression, or biomarker potential. For example, miR-27a has been associated with HIF-1α-related multidrug resistance, whereas miR-215/192 and miR-148a-3p have been reported in GC progression and biomarker studies^[74-76]. LncRNAs and circRNAs have also been widely studied. For instance, HOTAIR is linked to cisplatin resistance and histone-mark switching during the epithelial-mesenchymal transition^[77]. Additionally, MALAT1 promotes tumor progression via a miRNA-associated axis, while circAKT3 and circPVT1 are implicated in cisplatin resistance, proliferation, and prognostic stratification^[78-81]. However, compared with DNA methylation alterations, evidence supporting ncRNA-based mechanisms in GC remains less consistent.

RNA editing provides a distinct post-transcriptional mechanism. ADAR-mediated A-to-I RNA editing is altered in GC and has been associated with disease progression and prognosis, with ADAR1 overexpression and ADAR2 downregulation contributing to an editing imbalance in tumor tissues^[82]. CPEB3 has also been reported to suppress GC progression partly through regulation of ADAR1 mRNA localization and ADAR1-mediated RNA editing^[83].

m⁶A RNA modification is an emerging area in GC research. METTL3 has been reported to promote GC proliferation, migration, epithelial-mesenchymal transition, and MYC-associated transcriptional programs^[84,85]. The m⁶A reader YTHDF1 can promote gastric carcinogenesis by enhancing translation of FZD7 and activating Wnt/β-catenin signaling^[86]. These findings support a role for m⁶A-mediated post-transcriptional control, but most data still come from mechanistic studies rather than prospective clinical cohorts.

Clinical use of epigenetic information in GC remains indirect. MSI status is incorporated into clinical biomarker assessment and treatment selection, whereas EBV status is mainly used for molecular or pathological classification in selected settings. These features are generally assessed by immunohistochemistry, MSI testing, sequencing, or Epstein-Barr Virus Early RNA (EBER) in situ hybridization rather than by methylation assay^[87]. The extensive CpG island hypermethylation of EBV-positive tumors and MLH1-associated methylation in MSI tumors are therefore biologically and taxonomically important but are not routine stand-alone clinical tests. RIMS1 methylation measured in non-neoplastic antral and body biopsies is undergoing prospective validation for risk stratification after H. pylori eradication^[61]. However, external validation, assay standardization, and incorporation into surveillance algorithms are still required. Overall, DNA methylation currently represents the most clinically advanced epigenetic approach in GC, whereas histone modifications, ncRNAs, and m⁶A-related biomarkers remain largely investigational. Further prospective studies with standardized assays and clinically relevant endpoints are required before these emerging markers can be incorporated into routine clinical practice.

Aberrant DNA methylation is a hallmark of CRC, manifesting both as genome-wide hypomethylation and as locus-specific hypermethylation of promoter CpG islands. The CIMP defines a distinct subset of CRCs characterized by simultaneous hypermethylation of multiple loci. In a large cohort analysis of 295 sporadic CRCs, Toyota and colleagues first identified CIMP-high tumors, accounting for approximately 20% of cases^[89]. Subsequent studies have established the clinical relevance of CIMP. For instance, CIMP-high tumors are associated with BRAF mutations, proximal colon location, and high-level MSI^[90]. In terms of prognosis, CIMP-high status generally predicts worse overall survival^[91,92]. Notably, this prognostic effect is possibly context-dependent. Some studies reported that CIMP-high confers higher mortality among microsatellite-stable cancers, but not MSI-high cancers^[93].

Beyond global CIMP status, hypermethylation of individual tumor suppressor gene promoters occurs early in colorectal tumorigenesis. A comprehensive analysis including colorectal adenomas and carcinomas showed that promoters of APC, p16/CDKN2A, and RASSF1A are already methylated in adenomas at frequencies comparable to those seen in carcinomas^[94]. Even in aberrant crypt foci, the earliest histologically identifiable lesions, hypermethylation of APC can be detected and correlates with reduced gene expression^[95]. Similarly to other gastrointestinal tumors, methylation of DNA repair gene promoters, including MGMT, MLH1, and MSH2, has been linked to CRC carcinogenesis^[94,96]. Genome-wide methylation sequencing has further expanded the list of frequently methylated genes in CRC, including NDRG4, BMP3, MFAP2, APOL3, RNASEL, SFRP1, and SFRP2, all of which distinguish cancer from normal tissue with high sensitivity^[97,98]. To clarify their clinical use, combining machine learning and integrated analysis of TCGA DNA methylation signatures with comorbidity patterns revealed that a three-gene panel of ADHFE1, ADAMTS5, and MIR129-2 achieved the best CRC classification performance among ten candidate genes^[99]. In terms of prognosis, p16/CDKN2A promoter hypermethylation is present in about one-third of tumors and is associated with loss of p16 expression, thus correlating with shorter patient survival^[95,100]. Collectively, these studies establish DNA methylation as an early and heterogeneous feature of colorectal tumorigenesis and provide the biological basis for molecular classification and biomarker development.

The clinical application of DNA methylation biomarkers has been extensively validated in non-invasive settings. For plasma-based screening, a prediction model including five blood leukocyte DNA methylation markers for detecting colorectal neoplasms reached excellent performance with an area under the curve (AUC) of 0.85, which can further improve to 0.89 by combining with a lifestyle risk score^[101]. More recently, circulating tumor DNA (ctDNA) methylation haplotypes have emerged as powerful prognostic tools. The ColonES assay, developed by a multicenter study, detects methylation haplotypes across 191 genomic regions, predicting CRC with 86% sensitivity and advanced adenoma with 79% sensitivity^[102]. Importantly, patients with high preoperative ctDNA methylation scores had significantly shorter relapse-free and overall survival^[102]. Additionally, methylation of BCAT1 and IKZF1 in plasma ctDNA frequently occurred in CRC and rapidly reversed after surgical resection, thus can inform adequacy of surgical resection^[103]. Using a single-tube multiplex qMSP assay for six methylation markers, Zhu and colleagues reported a specificity of 98.2% for CRC detection with sensitivity of 67.5 %^[104]. Strikingly, highly methylated ctDNA levels were significantly associated with shorter relapse-free and overall survival^[104]. Collectively, these findings suggest DNA methylation is critical to CRC generation and development and already displays remarkable potential in clinical diagnosis.

Although the roles of histone modifications in CRC are not as well developed as DNA methylation, their dysregulation contributes to every stage of CRC pathogenesis. Regarding tumor initiation, the H3K36 methyltransferase SETD2 acts as a tumor suppressor, whose mutation is frequent in CRC^[105]. In a conditional knockout mouse model crossed with Apc^Min/+ mice, SETD2 loss increased the number of intestinal polyps by 3.5-fold and promoted progression to adenocarcinoma^[106]. For CRC progression, the H3K9 methyltransferase SUV39H2 is significantly elevated in tumor tissues, and its high expression correlates with distant metastasis and poor survival^[107]. Conversely, the H3K9 demethylase JMJD2C is overexpressed in liver metastases compared to primary tumors^[108]. Mechanistically, JMJD2C binds the MALAT1 promoter, reduces repressive H3K9me3 and H3K36me3 marks, and therefore upregulates MALAT1 expression, which in turn enhances β-catenin signaling and promotes metastasis in mouse models^[108]. An emerging modification, histone lactylation, links metabolism to epigenetics. In KRAS-mutant CRC tissues, lactate accumulation drives histone H3 lysine 9 lactylation (H3K9la). ChIP-seq revealed that H3K9la is enriched at the GRAMD1A promoter, a cholesterol transporter gene. Knockdown of GRAMD1A suppresses proliferation and migration in KRAS-mutant cells, identifying H3K9la as a key mediator of KRAS-driven tumorigenesis^[109].

Chemoresistance is a major clinical challenge, and histone acetylation plays a central role. HDAC3 mediates 5-fluorouracil resistance by directly deacetylating p21, thereby promoting its ubiquitination and degradation^[110]. Treatment with the HDAC3 inhibitor restored p21 acetylation and stability and re-sensitized resistant cells to 5-FU^[110]. Similarly, HDAC6 is overexpressed in chemoresistant CRC cells, and its selective inhibitor ACY1215 synergizes with chemotherapy to increase apoptosis and inhibit proliferation^[111,112]. Besides HDACs, histone acetylase p300 is elevated in a substantial proportion of CRCs, acting as an oncogene through enhancing H3K27 acetylation and activating β-catenin^[113].

Chromatin accessibility dynamics have been mapped at single-cell resolution. Using single-cell ATAC-seq on 48 polyps, 27 normal tissues, and 6 CRCs, researchers identified a continuum of chromatin remodeling from normal stem cells through polyps to cancer, characterized by progressive loss of differentiation-associated transcription factor motifs and gain of proliferation-associated motifs^[114]. In the context of metastasis, CRC cells lose colon-specific accessible sites and instead gain accessibility at HNF4A binding sites^[115]. Overall, histone modification and chromatin accessibility are critical epigenetic pathways in CRC, spanning tumorigenesis, proliferation, metastasis, and drug resistance. Although these findings indicate the clinical potential of therapy targeting histone modifications, there is a lack of solid evidence to support the application of enzyme- or pathway-specific inhibitors.

LncRNAs and circRNAs function as key regulators of gene expression in CRC. HOTAIR is one of the most extensively studied lncRNAs in CRC. HOTAIR expression was higher in tumors from patients who developed liver metastasis, and its overexpression increased cell invasion significantly^[116]. Mechanistically, HOTAIR binds to PRC2 and promotes H3K27me3-mediated silencing of metastasis suppressor genes^[116]. Another lncRNA, MALAT1, is elevated in metastatic CRC and facilitates tumor progression by upregulating PI3K/Akt/mTOR and Wnt pathways^[117,118]. In addition, circular RNAs, characterized by their covalently closed structure and high stability, often act as microRNA sponges. CircCCDC66 is upregulated more than 8-fold in CRC tissues; its knockdown reduces tumor proliferation, migration, and invasion^[119]. In the context of drug resistance, hsa_circ_0020095 is markedly elevated in cisplatin-resistant CRC cells, and it sponges miR-487a-3p to upregulate SOX9, accelerating cell proliferation and drug resistance^[120].

In terms of RNA modification, an integrative analysis of 583 CRC patients from TCGA and Gene Expression Omnibus (GEO) cohorts, examining 20 m⁶A regulators, identified two distinct clusters with significantly different overall survival^[121]. Furthermore, m⁶A-associated gene signatures were constructed and validated as independent prognostic predictors^[121]. Additionally, METTL3 is elevated in nearly two-thirds of tumors and predicts poor survival^[122]. Mechanistically, METTL3 methylates the 3’UTR of MYC mRNA, increasing its stability and protein levels^[123,124]. These findings support a role for m⁶A-associated RNA regulation in CRC biology, but therapeutic translation remains at an early stage.

Overall, epigenetic dysregulation is a driving force in colorectal cancer pathogenesis. DNA methylation signatures provide the most abundant insights in tumor pathology, serving as robust biomarkers for early detection, prognosis, and treatment prediction. Moreover, they have been applied in clinical settings with acceptable accuracy and sensitivity. However, more cautious validation across multiple centers and distinct molecular subtypes is still necessary. Histone modifications and chromatin accessibility orchestrate critical processes underlying tumor initiation, progression, metastasis, and acquired drug resistance. Importantly, although some histone modification-based therapies have shown potential in CRC clinical treatment, they still need large-scale validation in patients across tumor stages and subtypes before clinical implementation. Additionally, it is not entirely clear how individual pathways function and how they form a network in this process. Finally, ncRNAs, including HOTAIR, MALAT1, and circCCDC66, function as key regulators of chromatin modification and miRNA sponging. Emerging RNA modifications m⁶A add post-transcriptional control layers that influence prognosis and metabolism.

CRC has the most established clinical use of epigenetic testing among the cancers reviewed. For stool-based screening, multitarget DNA/Fecal Immunochemical Test (FIT) assays incorporate methylated DNA markers and are included among recommended screening strategies, while positive results require diagnostic colonoscopy^[125,126]. For blood-based screening, the methylated SEPT9 test achieved ~75% sensitivity at 87% specificity for CRC detection^[127]. Nonetheless, it has a restricted US regulatory indication for average-risk adults aged 50 years or older who have not completed other recommended screening, and it should not be presented as a replacement for first-line screening or colonoscopy^[128]. In MLH1/PMS2-deficient tumors, MLH1 promoter methylation is used as a cascade test, along with BRAF V600E, to support a sporadic origin and triage germline evaluation for Lynch syndrome^[129]. A cell-free DNA assay incorporating genomic and epigenomic signals received US approval in 2024, although its sensitivity for advanced precancerous lesions was substantially lower than for CRC^[130,131]. By contrast, ctDNA methylation assays for postoperative monitoring and inhibitors of DNA methyltransferases or histone deacetylases still require validation. A phase II trial of pembrolizumab plus azacitidine showed limited activity in refractory metastatic CRC^[132]. Overall, DNA methylation represents the most clinically mature epigenetic layer in CRC, with established applications in screening and molecular classification, whereas histone modifications, chromatin remodeling, ncRNAs, and RNA modifications remain primarily at the investigational stage.