Spatial transcriptomics reveals a molecular tumor budding signature in head and neck cancer

Abstract Background Tumor budding (TB) is a histopathological feature associated with poor prognosis across multiple cancer types, including head and neck squamous cell carcinoma (HNSCC). Tumor buds represent the earliest traceable local invasion and are considered the before origin of minimal residual disease, local recurrence, and metastasis. However, the molecular processes underlying this phenomenon in HNSCC remain incompletely characterized, particularly with regard to intratumoral gene expression heterogeneity. Methods We performed whole-transcriptome spatial transcriptomics on tissue sections from Human Papillomavirus (HPV) negative HNSCCs, sampling distinct regions of interest encompassing tumor buds, tumor bulk from budding and non-budding tumors, and adjacent stroma. Differential gene expression analyses led to the development of a 28-gene tumor budding signature (TBS) that separated tumor buds from all other tissue types. The TBS was validated using bulk RNA-seq (TCGA-HNSC), single-cell RNA-seq, and independent spatial transcriptomics datasets. Associations of the TBS with responses to drugs were evaluated in pharmacogenomic datasets and findings were further investigated in a 3D invasion model including MEK inhibition. Results Tumor buds exhibited distinct transcriptional programs with upregulation of epithelial-mesenchymal transition markers and extracellular matrix remodeling genes. The TBS effectively identified tumor buds in the spatial transcriptomics dataset (AUC = 0.97), separated budding and non-budding tumors in the bulk RNA-seq TCGA-HNSC dataset (AUC = 0.8), and predicted overall survival in the latter dataset (HR = 1.54, p = 0.02). Analyses of single-cell and spatial transcriptomics datasets confirmed TBS expression primarily in malignant cells, its association with a hybrid epithelial-mesenchymal state, its expression predominantly at the leading edges of tumors, and its induction via subtypes of epidermal growth factor receptor activities. Pharmacogenomic analysis revealed that TBS-high squamous cell carcinoma cell lines were sensitive to MEK inhibitors, a finding validated in a 3D model of early local invasion. Conclusions Integrated spatial-molecular profiling established a molecular bud biomarker facilitating the quantification of TB in both tumor tissues and cultured cells. Insights from the analysis of diverse datasets contribute to a better understanding of the molecular mechanisms underlying tumor invasion, improved risk stratification, and the development of new therapeutic approaches in HNSCC.