Document resource
Background Intra- and inter-tumor heterogeneity in head and neck cancers can lead to phenotypic diversity that makes cancer prognosis difficult to assess accurately. The heterogeneity that lies within the tumour-stroma interactions are poorly understood. There is an unmet clinical need to develop phenotypic signatures that can be predictive of clinical progression and therapeutic outcomes.Methods In this study, we applied our SpaceIQ platform to identify a subset of key markers characteristic of the complex tumour microenvironment features that can be used to derive a predictive test for a patient’s likelihood of cancer recurrence. Specifically, we applied cell segmentation and unbiased cell typing on spatial proteomics data (43 cores of 67-plex mIF panel across 15 pre-treated patients with 5 non-recurring status) yielding 23 cell types. Recurrence-specific microdomains, which are spatial arrangements of cells that are differentially expressed between the two response groups, were derived from pointwise mutual information between the unbiased cell types. Further, a de-plexing algorithm was implemented to obtain a low-dimensional proxy representation for each of the unbiased cell types involved in the microdomains based on their biomarker profiles. Finally, a prognostic modeling strategy, based on de-plexed representation, employing a spatial score between cell types and an estimated mean biomarker expression as predictive features to the recurrence outcome.Results We identified several spatial interactions associated with non-recurrence: (1) spatial interaction between M2 macrophages (unbiased cell type 21111) with anti-inflammatory Tregs (unbiased cell type 22200) was predictive of recurrence outcome (in-sample/CV AUC: 0.99/0.78); (2) spatial interaction between b-Catenin activated immune cells in stroma (unbiased cell type 11121) with mature NK cells (unbiased cell type 21212) was predictive of recurrence outcome (in-sample/CV AUC: 0.81/0.67). (3) spatial interaction between glutamine-dependent cancer-associated fibroblast, CAFs (unbiased cell type 11110) with antigen-presenting CAFs, apCAFs (unbiased cell type 12122) was predictive of recurrence outcome (in-sample/CV AUC: 1.0/0.84) ( figure 1).Conclusions We uncovered key spatial signatures of tumour non-recurrence in HNSCC patients involving interactions between M2 macrophage and Tregs, stromal immune cells and mature NK cells, and the heterogeneous stromal CAFs. These spatial interactions demonstrate the complexity of the TME affected by Wnt/b-Catenin signaling regulation on ECM remodeling, fibrotic processes, and immune response, in additional to potential glutamine-dependent facilitation of epithelial tumor invasion by CAFs. Through the SpaceIQ platform, we were able to de-plex a 67-plex IF panel to a few key predictive biomarkers involving macrophages, Tregs, NK cells, and glutamine-dependent CAFs yielding good predictive performance for recurrence outcome in HNSCC patients.Abstract 69 Figure 1(A) unbiased cell typing extracts numerically stable, spatially distinct and biologically interpretable recursive cell types; (B) microdomain families emerge from spatial domain PMI network; (C) microdomains shaping cancer outcomes