Head and neck squamous cell carcinoma (HNSCC) is a disease known to have poor outcomes with a mortality rate as high as 300,000 deaths per year. Despite advances in therapy, many patients fail to respond or experience subsequent recurrence. Better understanding of the cellular and molecular landscapes of the disease and their association with clinical phenotypes such as HPV status, smoking history, metastases, and survival could improve patient stratification and treatments.

In this retrospective study, we applied highly multiplexed CODEX immunofluorescence imaging to a well-characterized clinical cohort of 86 HNSCC patients. A total of 308 core biopsy specimens from HNSCC laryngeal and pharyngeal primary sites, with a subset of paired specimens from nodal metastases in patients with recurrent disease, were collected to generate tissue microarrays. An antibody panel targeting 36 biomarkers was developed to characterize key cellular, structural and functional proteins involved in HNSCC. CODEX imaging captured hallmarks of HNSCC such as large pleomorphic tumor cells (fig1a), keratin pearls, and stratified squamous epithelium and revealed diverse tumor immune landscapes (fig1b) across clinical phenotypes. Images were quantified utilizing a novel image analysis pipeline including quality control filters which excluded cells from downstream analysis on the basis of automatically detected artifacts (e.g. doublets, fig1c) to generate a final dataset of over 2 million single cells. Unsupervised clustering enumerated 15 distinct cell clusters including 5 tumor subclusters, 8 immune subclusters (e.g. CD4+ T cells, CD8+ T cells, CD20+ B cells, macrophages, fibroblasts), and 2 tissue subclusters (e.g. blood vessels). Biologic and spatial features including protein expression and cellular frequencies, interactions and neighborhoods were computed and exploratory analyses were performed to assess their association with clinical phenotypes such as primary tumor site, smoking/alcohol consumption, HPV status, TNM stage, and survival. This dataset recapitulated many known associations in HNSCC such as the link between HPV status, immune infiltration, and survival. Importantly, novel associations were also identified between the tumor subtypes identified during unsupervised computational analysis and their clinical outcomes. Tumor subtypes were stratified on the basis of expression of Sialyl-LewisX (CD15), Ki67, and podoplanin. While CD15 and podoplanin are known to play a role in a number of tumor types, little work has been done regarding their characterization in HNSCC. Interestingly, we observed that patients enriched for podoplanin+ tumor cell clusters had significantly worse prognosis than those enriched for CD15+ tumor cell clusters. In patients with improved overall survival at 5 years, immune cell interactions were enriched with CD15+ tumor subclusters and not podoplanin+ tumor subclusters, suggesting potential differences in immune exclusion and suppression between these subtypes. To further investigate this hypothesis, we examined tumor immune neighborhoods and identified a tumor cellular neighborhood enriched in B cells that was strongly associated with overall survival [fig1d, Kaplan-Meier, log rank p-value 0.00340]. We believe this work shows a novel infrastructure and analytic pipeline for refined biological subtyping of HNSCC that could inform development of future adjuvant immunotherapies, identify prognostic biomarkers, and ultimately guide treatment decisions based on patients’ specific disease profiles.