The tumor microenvironment architecture correlates with risk of recurrence in head and neck squamous cell carcinoma
We performed highly multiplexed microscopy using CODEX on 9 surgical specimens of patients with stage II-IV HPV negative squamous cell carcinoma explants of the oral cavity and oropharynx (HNSCC). We selected and validated a panel of 30 antibodies that allows the discrimination of the main leukocyte subsets infiltrating the tumor as well as auxiliary markers for tumor, vascular, and stromal. Fluorochrome conjugated reported oligonucleotides were hybridized on 12-cycle experiments with the Akoya PhenoCycler connected to a BZ-X700 microscope. Images were automatically acquired on 121 contiguous 20X fields (5.6mm x 4.2mm, resolution 377.47 nm/pixel) with 7 z-slices (1.5 µm/z pitch) and 30% tile overlap, with 10ms, 250 ms, 350 ms, and 350 ms acquisition time for DAPI, AF488, Cy3, and cy5 filters, respectively. The resulting 40,656 images per specimen were processed with the Akoya’s PhenoCycler™ Instrument Manager (version 1.3) and CODEX Processor (version 1.7.2) for segmentation using the default parameters. Cell phenotyping was performed in R using unsupervised clustering, followed by manual curation of clusters based on marker expression, morphology, and tissue localization. Cellular neighborhoods were identified implementing in R the neighborhood analysis from the NOLAN lab (https://github.com/nolanlab/NeighborhoodCoordination). We identified 14 conserved, distinct cellular neighborhoods that constitute a collection of components characteristic of the HNSCC infiltrating tumor microenvironment. This study highlights the power of CODEX immune fluorescence in immuno-monitoring, reveals the extraordinary complexity of the tumor microenvironment of HNSCC patients, and supports the notion that TLS1 and long-lived plasma cells are important determinants of antitumor immunity and tumor progression. HNSCC_results.csv table contains coordinates of all segmented objected, expression profiles, cell types and cellular neighborhood annotations.
Steps to reproduce
Raw, unprocessed imaging data is available upon request from the authors. Segmentation was performed using CODEX Processor (version 1.7.2) with default parameters. The annotated, spatial data provided here can be loaded R or used with the NOLAN lab Phyton scripts (https://github.com/nolanlab/NeighborhoodCoordination) for further interrogation and analysis. The processed R data files are available upon request by the authors.