A stress granule-related gene signature for prognostic stratification of esophageal squamous cell carcinoma

Esophageal squamous cell carcinoma (ESCC) currently lacks robust molecular biomarkers for precise prognostic stratification, which limits the ability to tailor therapeutic strategies to individual patient risk. Stress granules (SGs) are dynamic cytoplasmic assemblies that promote tumor cell survival under adverse microenvironmental conditions, yet their systematic prognostic value in ESCC remains unexplored. We aimed to construct a robust stress-granule-related genes (SGRGs) prognostic signature by integrating multi-cohort transcriptomics, single-cell RNA sequencing, and ensemble machine learning. Differentially expressed SGRGs were identified by intersecting transcriptomic profiles with 844 curated SGRGs. Prognosis-associated genes were screened via univariate Cox regression and refined through least absolute shrinkage and selection operator (LASSO), random forest, and support vector machine (SVM) algorithms to build a risk model, independently validated in TCGA-ESCC and GSE44021. Immune infiltration was characterized by CIBERSORT, quanTIseq, and xCell; immunotherapy response by Tumor Immune Dysfunction and Exclusion (TIDE); drug sensitivity by Genomics of Drug Sensitivity in Cancer (GDSC). scRNA-seq with pseudotime and ligand-receptor analyses dissected cell-type-specific expression and microenvironmental crosstalk. In vitro assays in two ESCC cell lines (KYSE-150 and TE-1) validated the biological relevance of the top contributor gene. Eight core genes ( CFTR, KCNMA1, MAPK14, PCSK2, PIK3R1, PTK2, STX11, UNC13B ) were identified from 242 differentially expressed SGRGs. The derived risk score stratified patients across all cohorts (training AUC = 0.72; external AUCs=0.71-0.76) and was confirmed as an independent prognostic factor. High-risk patients exhibited an immunosuppressive, stroma-enriched microenvironment (M2 macrophage polarization, CD8⁺ T cell depletion) and reduced predicted immunotherapy benefit. Differential drug sensitivities to EGFR and JAK inhibitors were observed between risk groups. PTK2 was the leading contributor to the risk model, consistent with its Cox coefficient and SHAP value, and showed preferential upregulation in the malignant epithelial subpopulation Malignant_C1, with its expression progressively increasing along pseudotime. Ligand-receptor analysis revealed that myCAFs communicate with Malignant_C1 through collagen-integrin signaling pathways, potentially linking stromal activation to PTK2 -mediated focal adhesion signaling in tumor cells. PTK2 knockdown suppressed ESCC proliferation, migration, and invasion in vitro in two ESCC cell lines. This study established a robust SGRG-based prognostic model for ESCC with cross-cohort generalizability, offering a clinically translatable tool for risk stratification. The identified stromal-epithelial communication patterns, particularly those converging on PTK2 -expressing malignant subsets, warrant further investigation as candidate therapeutic avenues.