Single-cell spatial analysis of the tumor immune microenvironment in NSCLC primary tumors and brain metastases.

8585 Background: Brain metastases (BrM) are associated with poor prognosis in non-small cell lung cancer (NSCLC). Although immune checkpoint inhibitors (ICIs) demonstrate intracranial activity, intracranial progression remains common, especially in patients with driver-negative NSCLC. To define mechanisms of intracranial immune resistance, we applied imaging mass cytometry to profile the tumor immune microenvironment (TIME) in BrM and primary lung tumors. Methods: Forty-four patients with advanced driver-negative NSCLC were included; 39 (89%) were treated with ICIs. A 40-marker antibody panel enabled site-specific immune phenotyping of primary lung tumors and BrM, with all samples obtained prior to initiation of systemic immunotherapy. Three regions of interest per sample (two intratumoral, one peritumoral) were analyzed using a Hyperion imaging mass cytometer. Single-cell segmentation was performed using Mesmer with immune phenotypes defined by marker expression. Spatial neighborhoods were identified using unsupervised k-means clustering of nearest-neighbor-derived local cellular compositions. Results: Compared with primary lung tumors, BrM demonstrated significantly reduced CD4⁺ and CD8⁺ T cells (p < 0.001), B cells (p = 0.017) and endothelial markers (p < 0.001). Despite reduced overall CD4⁺ infiltration, BrM exhibited higher proportions of regulatory T cells (Tregs; p = 0.017), Ki67⁺CD4⁺ T cells (p = 0.002) and Ki67⁺ Tregs (p = 0.017). BrM were enriched for CD206-high immunosuppressive macrophages, whereas primary tumors demonstrated predominance of CD11c-high inflammatory myeloid cells (p = 0.006). Spatial analysis demonstrated increased proximity of CD8⁺ T cells (p = 0.039) and Tregs (p = 0.020) to M2-like macrophages in BrM. Neighborhood analysis revealed site-specific immune organization, with primary tumors enriched for adaptive immune-dominant niches containing mixed CD4⁺, CD8⁺ and B-cell populations. In contrast, BrM were characterized by myeloid-T-cell interface neighborhoods marked by close spatial coupling of T cells with macrophages and monocytes. In primary tumors, durable immunotherapy benefit (≥24-month PFS and OS) was associated with significantly higher CD8⁺ and CD4⁺ T cells, B cells, dendritic cells, monocytes and macrophages. In BrM, no consistent immune correlates of long-term benefit were observed. Conclusions: NSCLC BrM exhibit a distinct TIME characterized by depleted adaptive immune infiltration, CD206-high macrophage dominance and myeloid-T-cell spatial organization. While primary lung tumors display adaptive immune signatures predictive of durable ICI benefit, BrM lack consistent immune correlates of response. These site-specific compositional and spatial differences provide a mechanistic framework for the attenuated and less durable intracranial responses to immune checkpoint blockade.