Cellular and spatial continuum of TAM functional states governs immune exclusion and dictates immune checkpoint resistance in melanoma

Background Immune checkpoint inhibitor (ICI) therapy has transformed melanoma treatment, yet many patients either fail to respond or develop resistance after an initial response. The cellular and spatial determinants within the tumor microenvironment (TME) that underlie these divergent ICI therapy outcomes remain poorly defined. Here, we aimed to characterize the immune composition and spatial organization of the TME associated with ICI response and resistance. Methods We integrated public scRNA-seq data (N = 32) with in-house spatial transcriptomics and imaging mass cytometry from an independent melanoma cohort (N = 6) to achieve high-resolution spatial profiling of the tumor microenvironment. This multimodal approach allowed us to map cellular heterogeneity within discrete spatial neighborhoods. To validate our findings, we constructed a scRNA-seq derived signature-matrix and deconvoluted a bulk RNA-seq cohort (N = 334). Clinical associations were assessed using Kaplan–Meier survival analysis, Cox hazards models, and a machine-learning classifier. Results We identified a distinct tumor-associated macrophage (TAM) state, termed lipid-associated TAMs (LA-TAMs), characterized by conserved lipid metabolic programs and tumor-specific adaptations in glycolysis under hypoxic conditions. Deep spatial multi-omics profiling revealed divergent immune-architectures associated with treatment outcomes. In Non-Responders, LA-TAMs colocalized with sparsely distributed TIM3⁺ exhausted CD8⁺ T cells within an immune-excluded TME. In contrast, Responder TMEs harbored interferon-primed CCL4 ⁺ TAMs (IFN-TAMs) spatially associated with Granzyme B⁺ activated CD8⁺ T cells, reflecting an immune-active microenvironment. These findings define contrasting macrophage-centered immune neighborhoods linked to ICI response. Interestingly, patients with LA-TAM–enriched neighborhoods were consistently non-responders or progressed during follow-up, irrespective of the presence of IFN-TAM neighborhoods. Lastly, we constructed a single-cell–derived signature matrix that estimates TAM fractions from bulk RNA-seq data and revealed LA-TAM abundance as a robust, independent predictor of poor ICI response and survival in a validation cohort. Conclusions Our study defines the spatial- and metabolic-landscapes of TAM-heterogeneity in melanoma and identifies LA-TAM neighborhoods as a key determinant of ICI resistance. Although patients span a continuum of TAM-centered immune states, the presence of LA-TAM overrides the beneficial effects of IFN-TAM–associated immune activation. Finally, our proposed Melanoma-TME–derived signature-matrix provides a translatable framework to robustly predict ICI response and resistance.