食管鳞状细胞癌（esophageal squamous cell carcinoma，ESCC）是一种全球高致死性恶性肿瘤，患者五年生存率不足30%。其临床异质性显著，表现为预后差异大、药物反应不一，传统基于组织病理学的分类方法难以指导精准治疗。尽管免疫检查点抑制剂（如PD-1/L1抗体）和CDK4/6抑制剂（如帕博西尼）展现出潜在疗效，但缺乏可靠的生物标志物导致患者筛选困难。近年研究发现，增强子表观遗传调控在肿瘤发生发展中起关键作用，但食管鳞状细胞癌中增强子动态调控网络及其临床转化价值仍不明确。本研究拟通过整合多组学数据解析增强子-靶基因调控网络，构建分子分型模型，旨在为食管鳞状细胞癌精准诊疗提供新策略。

本研究纳入310例未经治疗的食管鳞状细胞癌患者的全基因组重亚硫酸盐测序（whole genome bisulfite sequencing，WGBS）数据、转录组测序（RNA sequencing，RNA-seq）数据以及对应的临床信息的数据。通过构建增强子甲基化、增强子RNA（enhancer RNA，eRNA）及靶基因三者的调控网络，进而采用支持向量机（support vector machine，SVM）、加权基因共表达网络（weighted gene co-expression network analysis，WGCNA）、最小绝对收缩和选择算子（least absolute shrinkage and selection operator，LASSO）回归和随机森林四种机器学习方法，从148个候选靶基因中鉴定出12个核心靶基因。通过多因素Cox风险回归模型构建了增强子去甲基化调控基因评分（enhancer demethylation-regulated gene score，EDRGS），并以中位值为界分为EDRGS高组和EDRGS低组这两个亚型。EDRGS模型在训练集（HRA003107）和独立验证集（TCGA-ESCC、GSE53622）中均表现出较高的分类效能。预后分析显示，EDRGS高组的食管鳞癌患者总生存期显著缩短，且与TNM分期、淋巴结转移显著相关。更重要的是，EDRGS不仅局限于食管鳞癌，在其他9种癌症中，EDRGS同样能够作为一个有效的预后预测标志物。

分子机制解析表明：（1）EDRGS高组的患者呈现“免疫热但免疫抑制”表型：单细胞RNA测序（single-cell RNA sequencing，scRNA-seq）结果显示CD8⁺ T细胞、自然杀伤细胞浸润增加，但伴随PD-1、LAG3、TIM3等免疫检查点分子高表达以及免疫抑制性的肿瘤相关成纤维细胞（cancer-associated fibroblast，CAF）的富集；EDRGS高组患者的anti-PD-1治疗响应率显著高于EDRGS低组患者，并且EDRGS预测anti-PD-1治疗响应的效能优于PD-L1表达；（2）EDRGS低组的患者以细胞周期激活为特征：基因集合富集分析（gene set enrichment analysis，GSEA）结果显示EDRGS低组患者显著富集到G1/S转换、CDK4/6信号通路，并且CDK4以及CDK6表达显著升高；体外实验证实，EDRGS低组细胞对CDK4/6抑制剂帕博西尼敏感性显著增强。泛癌种分析进一步验证了EDRGS的跨癌种价值：在肾透明细胞癌（kidney renal clear cell carcinoma，KIRC，n = 181）、膀胱尿路上皮癌（bladder urothelial carcinoma，BLCA，n = 348）等13种癌症中，EDRGS高组患者免疫治疗响应更佳；而在胰腺癌（pancreatic adenocarcinoma，PAAD）、乳腺癌（breast invasive carcinoma，BRCA）等9种癌症队列中，EDRGS低组患者对CDK4/6抑制剂的响应更佳。

本研究通过整合增强子甲基化、eRNA以及靶基因的三元调控网络与机器学习方法，构建了食管鳞状细胞癌分子分型模型——EDRGS。该模型不仅实现了预后分层，还可精准预测免疫治疗与CDK4/6抑制剂的疗效，为ESCC的个体化治疗提供了可靠工具。其跨癌种适用性进一步提示，EDRGS可能成为多种实体瘤精准治疗的通用生物标志物。

Esophageal squamous cell carcinoma (ESCC) is a globally lethal malignancy with a five-year survival rate below 30%. Its marked clinical heterogeneity, reflected in divergent prognoses and variable drug responses, renders conventional histopathology-based classification systems inadequately guide precision therapy. Although immune checkpoint inhibitors (such as PD-1/L1 antibodies) and CDK4/6 inhibitors (such as palbociclib) demonstrate therapeutic potential, the absence of reliable biomarkers hinders patient stratification. Recent studies highlight the critical role of enhancer-mediated epigenetic regulation in tumorigenesis; however, the dynamic enhancer regulatory networks and their clinical implications in ESCC remain elusive. This study aims to integrate multi-omics data to delineate enhancer-target gene regulatory networks and construct a molecular subtyping model, providing novel strategies for precise diagnosis and treatment of ESCC.

This study included whole-genome bisulfite sequencing (WGBS) data, transcriptome sequencing (RNA-seq) data, and corresponding clinical information of 310 untreated ESCC patients. By constructing the regulatory network among enhancer methylation, enhancer RNA (eRNA), and target genes, and then using four machine learning methods, namely support vector machine (SVM), weighted gene co-expression network analysis (WGCNA), least absolute shrinkage and selection operator (LASSO) regression, and random forest, 12 core target genes were identified from 148 candidate genes. An enhancer demethylation regulatory gene score (EDRGS) was constructed through a multivariate Cox regression model, and it was divided into EDRGS-high and EDRGS-low subtypes based on the median value. The EDRGS model showed high classification efficiency in the training set (HRA003107) and independent validation sets (TCGA-ESCC and GSE53622). Prognostic analysis revealed that ESCC patients in the high-EDRGS group exhibited significantly shortened overall survival (OS) and showed significant associations with advanced TNM stages and lymph node metastasis. Notably, the prognostic utility of EDRGS extended beyond esophageal squamous cell carcinoma, serving as a valid prognostic biomarker across nine other cancer types.

Molecular mechanism analysis showed that: (1) The EDRGS-high subtype presented an "immune-hot but immune-suppressive" phenotype: Single-cell RNA sequencing (scRNA-seq) results showed an increase in the infiltration of CD8⁺ T cells and natural killer cells, but was accompanied by high expression of immune checkpoint molecules such as PD-1, LAG3, and TIM3, as well as the enrichment of immunosuppressive cancer-associated fibroblasts (CAFs); the anti-PD-1 treatment response rate of the EDRGS-high subtype was significantly higher than that of the EDRGS-low group, and the efficacy of EDRGS in predicting anti-PD-1 treatment response was better than PD-L1 expression; (2) The EDRGS-low subtype was characterized by cell cycle activation: Through gene set enrichment analysis (GSEA), the G1/S phase transition and Cyclin-dependent kinase 4/6 (CDK4/6) signaling pathway were significantly enriched, and the expressions of CDK4 and CDK6 were significantly increased; in vitro experiments confirmed that EDRGS-low cells were significantly more sensitive to palbociclib. Pan-cancer analysis further verified the cross-cancer value of EDRGS: In 13 types of cancers such as kidney clear cell carcinoma (KIRC, n = 181) and bladder urothelial carcinoma (BLCA, n = 348), the EDRGS-high group had a better response to immunotherapy; while in 9 cancer cohorts such as pancreatic cancer (PAAD) and breast cancer (BRCA), the efficacy of CDK4/6 inhibitors in the EDRGS-low group was significantly improved.

This study is the first to construct an ESCC molecular subtyping model, EDRGS, by integrating the regulatory network of enhancer methylation, eRNA, and target genes with machine learning methods. This model not only achieves prognostic stratification but can also accurately predict the efficacy of immunotherapy and CDK4/6 inhibitor treatment, providing a reliable tool for the individualized treatment of ESCC. Its cross-cancer applicability further suggests that EDRGS may become a universal biomarker for precision medicine of multiple solid tumors.