Toward Precision Oncology for HPV-Associated Head and Neck Cancer (HNC): Multi-Omics Analysis and Machine Learning

Abstract

Precision oncology refers to treating cancers based on the characteristics of each patient. This approach requires analyzing and understanding specific tumor information from patients, aiming to improve treatment outcomes and minimize side effects. Over the past decade, HPV-associated head and neck squamous cell carcinoma (HPV(+) HNSCC) has gradually increased, surpassing cervical cancer as the most prevalent HPV-associated cancer in the U.S. and UK. Since HPV(+) HNSCC in the oropharynx has a better 5-yr survival rate compared to HPV-negative HNSCC (~80% vs ~50%), de-escalation treatment options are being actively investigated in HPV-associated HNSCC. However, there are still no established molecular biomarkers to identify and stratify patients by risk as required for precision oncology. HPV-associated HNSCC can be classified into two subtypes, one characterized by features such as a strong immune response and mesenchymal differentiation (IMU), and another dominated by keratinization and PICK3A mutations (KRT). IMU has been shown to be associated with better survival outcomes and has potential as a biomarker for de-escalation treatment. In Chapter 2, we further characterized these subtypes with DNA methylation data, by integrating gene expression, genomic instability and DNA hydroxymethylation data to perform a multi-omics analysis. We found overall hypermethylation in IMU compared to KRT, and demonstrated that nearly all previously observed hypermethylated sites in HPV(+) versus HPV(-) HNSCC are solely due to IMU. The significantly higher methylation of transposable elements in IMU may serve as a potential prognostic biomarker. The IMU/KRT subtypes have been consistently validated by multi-omics data in multiple studies; however, no standardized method exists to subtype new HPV+ HNSCC tumors. In Chapter 3, we introduce a machine learning (ML)-based classifier and webtool based on RNA-seq data that reliably subtypes HPV+ HNSCC tumors using the IMU/KRT paradigm and highlights the importance of these subtypes in HPV+ HNSCC. We extended our subtyping studies to 219 HPV(+) HNSCC patients with RNA-seq data and identified 21 relevant carcinogenic pathways and clinicodemographic variables associated with the subtypes. In Chapter 4, we focused on another potential biomarker for risk stratifying patients for precision oncology: HPV integration. HPV integration is a process where the human papillomavirus (HPV) DNA inserts into the host genome and it is associated with carcinogenesis and tumor progression in both cervical cancer and HNSCC, but its relationship with clinical outcomes remains unconfirmed. We investigated the consequences of HPVint in both human and HPV characteristics by analyzing 261 HPV-associated HNSCC bulk and single-cell RNA-seq samples from five cohorts, and DNA HPVint events from 102 HPV+ participants in two of those cohorts. We revealed an oncogenic network based on the recurrent HPV integration genes in HNSCC and classified HPVint-positive patients by HPV RNA features, showing that subsets of HPVint-positive HNSCC patients have worse clinical outcomes. In summary, this work serves to advance precision oncology in HPV(+) HNSCC by exploring its DNA methylation profiles, performing multi-omics analysis based on subtypes, developing a machine learning (ML) classifier for future tumor subtyping, and identifying potential clinical outcome-associated biomarkers related to HPV integration.