Deep learning pathomics integrated with single-cell transcriptomics identifies MGAT4A-associated “lock-and-load” dysfunction signatures of CD8+ tissue-resident memory T cells in endometrial cancer

Aberrant glycosylation contributes to cancer progression, but its histopathological and immune significance in endometrial cancer (EC) remains unclear. This study aimed to identify glycosylation-associated pathomic features as prognostic biomarkers and to explore how these alterations are associated with immune dysfunction and with computational indicators of immune escape and possible immunotherapy resistance in EC. We integrated pathomics with bulk and single-cell transcriptomics. Machine learning linked image features to N-glycan biosynthesis. Immune states, cell-cell communication, differentiation trajectories, and virtual MGAT4A knockout were analyzed computationally. Therapeutic response was predicted using TIDE and oncoPredict; no measured clinical outcomes were available. The optimal model stratified survival in the training and internal validation cohorts. The resulting glycosylation-associated pathomics RiskScore independently predicted prognosis and correlated with aggressive clinicopathological features. High-risk tumors exhibited upregulated cell cycle pathways, an immune-cold microenvironment, and computational indicators of checkpoint blockade resistance. MGAT4A was associated with poor prognosis and enriched in terminally differentiated CD8 + tissue-resident memory T cells (Trm). MGAT4A expression marked a putative “lock-and-load” dysfunction state: an inferred transcriptional pattern in which Trm cells remain armed with cytotoxic effector molecules (“loaded”) yet simultaneously display retention- and restraint-associated programs (“locked”), leaving them spatially trapped and functionally restrained. Virtual knockout generated regulatory hypotheses. A “Trapped Dysfunction Ratio” was associated with poor survival. This study identifies a glycosylation-associated CD8 + Trm dysfunction signature in EC. Glycosylation-associated pathomics features are associated with poor prognosis and with computational indicators of immune escape and possible immunotherapy resistance. MGAT4A represents a candidate biomarker requiring experimental and external validation.