Morphology-derived spindle-cell gene signature in HPV+ HNSCC is linked to pEMT, molecular subtype, HPV integration, and recurrence-free survival

Spindle-shaped epithelial cells are frequently observed in HPV+ head and neck squamous cell carcinoma (HNSCC) cell lines and at invasive fronts in tumors, but how this morphology relates to transcriptional programs, aggressive phenotypes, and spatially organized niches remains unclear. This study aimed to understand the spindle phenotype by deriving a morphology-linked gene signature and testing its association with partial epithelial-mesenchymal transition (pEMT), HPV integration (HPVint) into the host tumor genome, tumor molecular subtypes (immune-strong, IMU vs highly keratinized, KRT), and patient recurrence-free survival (RFS), and by mapping this program in situ using spatial transcriptomics. We quantified cell morphology in two HPV+ HNSCC cell lines using cell aspect ratio, roundness, and circularity, and used bulk RNA-seq (n = 12) to identify genes differentially expressed in spindle-enriched conditions. We derived a 45-gene spindle-cell signature (SCS) and computed SCS and pEMT scores in 236 HPV+ HNSCC tumors from four cohorts: UM_FF (18), UM_FFPE (62), HVC (83), and TCGA (73). In parallel, we generated Nanostring CosMx 6k-plex spatial transcriptomics from four HPV+ tonsil tumors, clustered cells into major compartments, and projected SCS/pEMT scores and HPV gene expression onto tissue coordinates to characterize spatial niches. In HNSCC tumors, SCS scores are highly correlated with pEMT scores (Pearson's r = 0.72) despite sharing only two genes, indicating this signature robustly captures a pEMT-like program. KRT tumors had markedly higher SCS scores than IMU tumors in the pooled analysis ( p = 1.3 × 10 -9 ). Across four cohorts, SCS scores tended to be higher in HPVint+ than HPVint- tumors, with significant differences in three cohorts ( p range 0.016-0.045) but not the HVC ( p = 0.21), suggesting cohort-specific heterogeneity. Taken together, these patterns firmly link the spindle program to molecular subtype and suggest an association with HPVint. Survival analysis of 70 University of Michigan patients shows that patients with low SCS scores have significantly better RFS than those with high scores (Log-rank test, p = 0.014). We are incorporating CosMx data with these patterns to test the hypothesis that high SCS epithelial states preferentially occupy tumor-stromal interfaces and cancer-associated fibroblast-rich regions and are associated with distinct local immune architecture, defining a spatially restricted invasive niche. These analyses support that SCS captures an invasive state enriched in KRT and HPVint+ HNSCC and associated with poorer RFS. By connecting cell shape, transcriptional state, viral features, and spatial microenvironment, this study suggests morphology-derived biomarkers and spatial niches that may refine risk stratification and guide targeted therapeutic strategies in HPV-associated HNSCC.