Assessing the Tumor Immune Landscape Across Multiple Spatial Scales to Differentiate Immunotherapy Response in Metastatic Non-Small Cell Lung Cancer

Ashley P Tsang,Santhoshi N Krishnan,Joel N Eliason,Jake J McGue,Angel Qin,Timothy L Frankel,Arvind Rao

DOI: https://doi.org/10.1016/j.labinv.2024.102148

Abstract:Although immune checkpoint inhibitor-based therapy has shown promising results in non-small cell lung cancer patients with high programmed death-ligand 1 expression, not all patients respond to therapy. The tumor microenvironment (TME) is complex and heterogeneous, making it challenging to understand the key agents and features that influence response to therapies. In this study, we leverage multiplex fluorescent immunohistochemistry to quantitatively assess interactions between tumor and immune cells in an effort to identify patterns occurring at multiple spatial levels of the TME. To do so, we introduce several computational methods novel to a data set of 1,269 multiplex fluorescent immunohistochemistry images from a cohort of 52 patients with metastatic non-small cell lung cancer. With the spatial G-cross function, we quantify the degree of cell interaction at an entire image level, where we see significantly increased activity of cytotoxic T cells and helper T cells with epithelial tumor cells in responders to immune checkpoint inhibitor-based (P = .022 and P < .001, respectively) and decreased activity of T-regulatory cells with epithelial tumor cells compared with nonresponders (P = .010). By leveraging spatial overlap methods, we define tumor subregions (which we call the tumor "periphery," "edge." and "center") and discover more localized immune-immune interactions influencing positive response, including those between cytotoxic T cells and helper T cells with antigen presenting cells in these subregions specifically. Finally, we trained an interpretable deep learning model that identified key cellular regions of interest that most influenced response classification (area under the curve = 0.71 ± 0.02). Assessing spatial interactions within these subregions further revealed new insights that were not significant at the whole image level, particularly the elevated association of antigen presenting cells and T-regulatory cells with one another in responder groups (P = .024). Altogether, we demonstrate that elucidating patterns of cell composition and interplay across multiple levels of spatial analyses can improve our understanding of the TME and better differentiate patient responses to immunotherapy.