Cross-layer multiomic and digital pathology analysis identifies a malignant keratinization state linked to immune exclusion in cervical squamous carcinoma

Abstract

Immune exclusion contributes to heterogeneous benefit from immunotherapy in cervical squamous carcinoma, but the malignant epithelial state most closely associated with this phenotype and its tissue- and morphology-level correlates remain unclear. We investigated whether a lesion-grade-associated malignant epithelial state was linked to immune-excluded tissue architecture and could be translated across transcriptomic and pathology modalities. We integrated single-cell RNA-seq discovery (GSE208653), spatial transcriptomic evaluation (GSE208654), bulk RNA-seq translation in primary squamous TCGA-CESC tumors, external whole-tumor evaluation in CGCI-HTMCP-CC, and whole-slide H&E analysis of 259 slides from 250 TCGA patients. External immune-focused datasets, a local neoadjuvant immunotherapy-treated cervical squamous carcinoma cohort, and a representative pilot whole-section multiplex immunofluorescence were used as supportive layers. A basal-squamous stress keratinization (BSK) program was the malignant epithelial state most consistently associated with the cross-sectional normal–HSIL–squamous carcinoma spectrum. Across four spatial sections, BSK showed a section-consistent core-boundary-shell organization comprising a BSK-rich tumor core, a stromal-myeloid boundary, and a more peripheral lymphoid shell. In primary squamous TCGA-CESC tumors, this biology was translated most clearly into an epithelial-exclusion bulk state associated with fibro-myeloid niche enrichment and weaker engagement of inflamed/dysfunctional CD8 T-cell programs. Patient-level out-of-fold morphology scores from matched TCGA H&E slides correlated positively with epithelial exclusion, supporting a detectable histologic correlate within the matched pathology arm. In a local 18-patient neoadjuvant immunotherapy-treated cohort, H&E-derived morphology scores were associated with postoperative pathological response grade, providing exploratory clinical-pathology support rather than predictive validation. The exclusion-centered ordering was directionally preserved in CGCI-HTMCP-CC and aligned with stromal/EMT/TGFβ, angiogenesis, and more moderate gMDSC-related programs. BSK marks an exclusion-associated cervical squamous carcinoma state that is spatially organized, measurable in bulk transcriptomes, and partially reflected in routine histology. These findings provide a human-data-derived translational framework for future immune-access stratification and prospective biomarker testing but do not establish BSK as a causal driver or validated predictor of immunotherapy response.