Spatial localization of epithelial-mesenchymal transition and immune niches in non-small cell lung cancer via interpretable gene set activity analysis.

e20556 Background: The tumor microenvironment (TME) in non-small cell lung cancer (NSCLC) is highly heterogeneous, with specific spatial niches driving metastasis and drug resistance. Epithelial-mesenchymal transition (EMT) is a critical process in progression, yet identifying exactly where EMT occurs within the tissue architecture remains challenging with standard tools. We utilized GESSO, a novel spatially informed analysis framework, to dissect the functional architecture of the NSCLC microenvironment, aiming to resolve spatially confined EMT programs and distinct immune cell niches that standard methods fail to localize. Methods: We analyzed high-resolution NanoString CosMx spatial transcriptomics data from human NSCLC tissue (55,365 single cells). To quantify gene set activity, we applied GESSO (code available at https://github.com/YMa-lab/GESSO), a spatially informed computational framework we developed that integrates spatial coordinates with gene expression to infer spatially coherent activity scores. We focused on characterizing the spatial distribution of TGF-beta-induced EMT programs and immune cell signatures. To validate the precision of these biological findings, we benchmarked the spatial fidelity of these signatures against standard activity scoring methods (AUCell, ssGSEA, GSDensity). Results: The analysis revealed that EMT is not a diffuse phenomenon but is strongly localized to a specific spatial niche at the interface between the tumor core and the desmoplastic stroma. Within this interface, the analysis identified a potent EMT signature driven by the coordinated upregulation of mesenchymal markers COL1A1 and FN1, and the invasion-associated metalloproteinase MMP2. Importantly, activity within this specific spatial niche was positively correlated with tumor pseudotime (Pearson r = 0.57), suggesting a direct link between this localized stromal interaction and tumor progression stages. Additionally, the framework resolved distinct, spatially separated immune microenvironments that standard methods blurred: specifically isolating macrophage niches enriched for CD68 and C1QB, and T-cell activation niches enriched for CD2 and CCL5. In comparative validation, standard methods failed to resolve these specific immune niches or the confined EMT interface, producing diffuse signals that obscured these critical microanatomical structures. Conclusions: This study demonstrates that key drivers of NSCLC progression, including EMT and immune activation, are organized into distinct spatial niches rather than occurring uniformly across the tumor. By precisely mapping these functional interfaces, we provide evidence of spatially coordinated stromal-tumor interactions that may serve as localized targets for disrupting metastatic progression.