1199 Nested spatial units: an organizing architecture describing heterogeneity in the tumor microenvironment and clinical response to immunotherapy

<h3>Background</h3> The tumor microenvironment (TME) is an important cellular ecosystem. However, immense structural and compositional heterogeneity presents a significant challenge in distinguishing conserved properties of TME spatial organization from qualities that are tumor-specific. Through the application of statistical interference models to a diverse database of spatial transcriptomics tumor datasets, we aimed to identify properties of the TME that are biologically conserved and clinically meaningful for patient outcomes to immune checkpoint blockade (ICB) therapy. <h3>Methods</h3> We created a database of 96 spatial transcriptomic (ST-seq, 10X Visium) datasets spanning twelve different types of solid tumors. We then developed and applied an interpretable statistical learning framework capable of using the transcriptional content in these data for predicting spot spatial locations in the TME. By modeling the components of our framework as biologically meaningful objects (ST-seq spots) rather than as abstractions as would be used in other machine learning models, we were able to directly identify both common and diverse modules of TME spatial biology. We then used non-negative matrix factorization to classify tumors using these common spatial modules into signatures of spatial biology at the tumor-wide level. Finally, we applied this spatial signature analysis to diagnostic tumor samples from 16 metastatic non-small cell lung cancer (NSCLC) patients who went on to receive frontline ICB therapy. We applied survival analysis to determine whether our spatial signature, as well as other biomarkers such as PD-L1, was predictive of progression-free survival after ICB treatment. <h3>Results</h3> We found a conserved spatial TME architecture comprised of nested multicellular groups we term ‘Nested Spatial Units’ (NSUs). NSUs reflect three tumor-agnostic interacting modules of intratumoral heterogeneity – tumor proliferation, immune infiltration, and cell adhesion. Variations in NSU biology defined signatures of TME spatial biology that we term ‘SpatialTypes’. SpatialType classification of sixteen patients with metastatic NSCLC using pre-treatment samples effectively stratified response to ICB therapy while other markers – PD-L1 status and bulk transcriptional information – did not. Responders were associated with spatially heterogenous activation of multiple T cell checkpoints, while non-responders were associated with spatially organized collagen expression. <h3>Conclusions</h3> Our findings reveal a conserved spatial architecture of the TME that is biologically and clinically significant, motivating further study into understanding collective TME organization and the development of NSU-based spatial biomarkers for clinical application in cancer immunotherapy.