Background The tumor microenvironment (TME) contains an elaborate mixture of varied cell phenotypes, whose spatial organization plays an important but incompletely understood role in disease outcome and response to therapy. Neighborhood Analysis is a promising new spatial analysis method used to extract broad features from a spatial dataset and summarize recurring regions that may be important in the TME [Schürch et al 2020, Griffin et al 2021, Patel et al 2019, Oyler-Yaniv et al 2018].^1–4 However, it remains unclear which neighborhood sizes provide the best insights. Schürch et al 2020 used 19 nearest neighbors plus self, which is a small spatial neighborhood. Our group has previously used a circular area with radius of 75 microns, which is a larger spatial neighborhood. Primarily, we aim to explore how neighborhood size can affect neighborhood analysis results and interpretations. Secondarily, we propose to explore other new features in neighborhood analysis.

Methods The Human Tumor Atlas Network’s (HTAN) efforts to establish a cancer atlas 3-dimensionally and over disease progression presents the opportunity to mine spatial datasets across cancer types and explore parameters used in neighborhood analysis. As part of the HTAN consortium, DFCI focused on three diseases to investigate with paired sequencing and spatial analyses: metastatic breast, melanoma, and colorectal cancers. Patient FFPE samples were stained and imaged using Akoya Biosciences Phenoptics mIF platform. Two panels were used: a macrophage-oriented panel including CD3, CD68, Tumor marker (SOX10 or Cytokeratin), PDL1, CD163, and Ki67; and a T cell-focused panel including CD8, PDL1, PD1, FOXP3, and Tumor marker. mIF images were thresholded and phenotyped using inForm software from Akoya Biosciences. Pythologist python package [https://github.com/dfci/pythologist] was used to ingest inForm outputs and to perform neighborhood analysis.

Results A pilot comparison of ‘r’=45 microns (r45) and ‘r’=75 microns (r75) revealed that r45 neighborhoods contained approximately 1/3 the number of cells of r75 neighborhoods and were more susceptible to one cell of a certain phenotype influencing that entire neighborhood cluster.

Conclusions Primarily, we postulate that smaller neighborhoods may lead to more granular results, perhaps describing small locales better, but leading to less interpretable overall conclusions than larger neighborhoods. Secondarily, we hypothesize that new features measured from neighborhoods may be used to identify boundaries in different tissue regions such as the tumor/margin/stroma borders. By exploring parameters and features for neighborhood analysis, we aim to improve interpretability of these analyses, to provide new features to use, and to strengthen rationale for picking certain neighborhood sizes in future spatial studies.

Acknowledgements This work is supported through the Human Tumor Atlas Network, grant U2CCA233195.

References

1. Schürch, Christian M., et al. Coordinated Cellular Neighborhoods Orchestrate Antitumoral Immunity at the Colorectal Cancer Invasive Front. Cell, vol. 182, no. 5, 3 Sept. 2020, https://doi.org/10.1016/j.cell.2020.07.005.

2. Griffin, Gabriel K., et al. Spatial Signatures Identify Immune Escape via PD-1 as a Defining Feature of T-Cell/Histiocyte-Rich Large B-Cell Lymphoma. Blood, vol. 137, no. 10, 2021, pp. 1353–1364., https://doi.org/10.1182/blood.2020006464.

3. Patel, Sanjay S., et al. The Microenvironmental Niche in Classic Hodgkin Lymphoma Is Enriched for CTLA-4- Positive T-Cells That Are PD-1-Negative. Blood, 2019, https://doi.org/10.1182/blood.2019002206.

4. Oyler-Yaniv, Alon and Krichevsky, Oleg. Imaging Cytokine Concentration Fields Using PlaneView Imaging Devices. Bio Protoc. 2018 Apr 5; 8(7): e2788. doi: 10.21769/BioProtoc.2788

Pythologist python package https://github.com/dfci/pythologist

Ethics Approval The HTAN study is approved by the DFCI Institutional Review Board as protocol 18–452.