Background
Spatial transcriptomics platforms produce immense datasets, measuring hundreds to thousands of genes across up to 1 million single cells in a single tissue section. For investigations of clinical outcomes like response to immunotherapy, this data richness presents as not a windfall but a quagmire. To facilitate immuno-oncology research, we have devised algorithms for automatically measuring the fundamental units of the tumor-immune interaction. Given data from a single tumor profiled with the CosMxTM Human Universal Cell Characterization panel, our algorithms output dozens of relevant, human-intelligible variables, which we propose as ideal outputs for multi-tumor comparisons.
Methods
Our first set of algorithms is knowledge-driven, measuring outputs that are already known to be important. These include variables like CD8 T-cell abundance, total activating vs. suppressive signaling around T cells, the balance of macrophages with activating vs. suppressive phenotypes, CD8 invasiveness into the tumor bed, interferon signaling and antigen presentation activity in tumor cells exposed to immune cells, tumor accessibility to immune attack, vascularization, and presence of tertiary lymphoid structures.
A second set of algorithms is data-driven. We identify modules of immune-signaling genes with tendencies to be expressed in the same locations and quantify these modules across the space of a tumor. An example output quantifies hotspots of a module consisting of CCL5, CCL8, CXCL9, CXCL10, CD48, SEMA4D, and SIGLEC1.
Results
We apply these algorithms to >20 tumors, including liver cancer, NSCLC, RCC, and melanoma, obtaining >50 spatial signature scores describing each tumor. We then demonstrate how this tissue-level data matrix can be used to predict tumors’ clinical features.
Conclusions
In summary, our spatial signatures – currently >50, with more under development – measure tumor attributes fundamental to anti-tumor immunity and immune evasion. We propose them as a core set of variables for describing tumors and for predicting patient’s response to immunotherapies.