Response to induction chemotherapy and identification of tumor heterogeneity through a comprehensive multi-omics profiling of HPV-associated oropharyngeal squamous cell carcinoma.

Abstract

e18053 Background: The response of neoadjuvant chemotherapy (NAC) can guide personalized treatment de-intensification in HPV-associated oropharyngeal squamous cell carcinoma (OPSCC). Comprehensive molecular characterization based on NAC response is crucial for accurate selection of candidates of treatment de-escalation and further guidance of individualized therapeutic approaches. Methods: Multi-omic data from biopsy samples of HPV-associated OPSCC cases were leveraged, including clinical, transcriptomic, proteomic, and digital pathology profiles in this study. All patients received NAC and radical radiotherapy/chemoradiotherapy at Fudan University Shanghai Cancer Center. Correlation between NAC response and multi-omics features was analyzed. Through integrative analysis, we presented a comprehensive multi-omic landscape of NAC-sensitive and NAC-resistant subgroups. Combined transcriptomic and proteomic analyses identified differences in gene expression and tumor microenvironment between two subgroups. A prediction model for NAC response was fitted upon pathology features and identified differentially expressed genes, which was further validated in an external validation cohort from four additional centers. Furthermore, novel tumor subtypes were identified based on NAC-associated multi-omics features and activated pathways, which were also validated in the external validation cohort. Results: NAC response was significantly associated with prognosis. HPV-associated OPSCC exhibited phenotype-specific tumor heterogeneity regarding response to NAC. NAC-sensitive HPV-associated OPSCC showed strong enrichment in anti-tumor immune responses, while resistant tumors showed regulation of cytoskeletal structures featured by high expression of keratins and intermediate filaments, etc. Resistant and sensitive tumors harbored distinct immune infiltration features and ecosystems. Machine learning-based multi-omic predicting model for NAC response showed satisfactory area under curve (AUC) in both training cohort and external validation cohort. Combination of proteomic and transcriptomic data revealed 4 distinct molecular clusters of HPV-associated OPSCC, which was validated in the external cohort. These clusters differed in transcriptional regulatory network, tumor immune microenvironment, and clinical outcomes, plausibly corresponding to different mechanisms of chemoresistance. Conclusions: Integrative multi-omics analysis revealed molecular heterogeneity of HPV-associated OPSCC in relation to NAC response. Multi-omic profiles can function as biomarkers for NAC response prediction. Molecular subtypes of HPV-associated OPSCC represent distinct mechanisms of chemoresistance, hopefully providing insights for precision treatment strategies in future.