Background The tumor microenvironment (TME) is the milieu of cells and molecules surrounding the tumor. Single cell methods have been used to identify subpopulations of cells that have pro- or anti-tumor properties, and modulating these is critical to immunotherapy. However, there is limited information on the spatial organization of these subpopulations, which determines how they signal and their therapeutic potential. Single cell spatial computational analysis is needed to describe the interactions of the TME and their effect on patient outcomes. Hodgkin’s Lymphoma presents a unique TME due to the sparse distribution of Hodgkin’s Reed-Sternberg tumor cells. Though it is difficult to disentangle signaling in Hodgkin’s tumor cells from the immune TME, Hodgkin’s is receptive to checkpoint inihibitors among lymphomas and insights gleaned from the Hodgkin’s TME could inform immunotherapies across lymphomas.

Methods Here we apply Imaging Mass Cytometry (IMC), a technology to perform ~40-plex protein analysis with 1 micron resolution in tissue, to study a cohort of 260 matched samples at diagnosis and after relapse from 90 patients with relapsed/refractory Hodgkin’s Lymphoma. We developed a computational pipeline to perform cell phenotyping, spatial analysis, and biostatistics, to describe tumor architecture and propose putative biomarkers of Hodgkin’s clinical response and relapse. A novel feature of the pipeline is to quantify proteins and spatial analysis on the same numerical scale for each cell, to generate hybrid biomarkers.

Results We analyzed over 7 million cells for their phenotype and spatial organization. We use IMC to describe spatial features of the tumor that correlate to clinical outcomes. We identify proteins such as CXCR5 that correlate to survival in spatial contexts, and we describe spatial reorganization from diagnosis to the relapsed tumor as it relates to survival and relevant clinical factors such as MHC expression and EBV infection. A significant conceptual advance was to use spatial metrics to perform ‘digital biopsies’, a selection of tumor regions comparable across patients. We validated multiple existing biomarkers in the literature using our data set and proposed novel biomarkers using IMC data.

Conclusions Spatial analysis of the HL microenvironment revealed composite features of the TME that predict clinical outcomes. These features cannot be described using single cell tools or low-plexed imaging, and represent a truer picture of HL biology. The pipeline developed here can be applied to other spatial protein data for biomarker discovery and analysis.