{
  "abstract": "Background Immune checkpoint inhibitors (ICIs) have shown clinical efficacy across multiple tumor types, including melanoma, non-small cell lung cancer (NSCLC), head and neck squamous cell carcinoma (HNSCC), and triple-negative breast cancer (TNBC), particularly in tumors with high PD-L1 expression. 1–4 While tumor-intrinsic characteristics differ across cancer types, immune programs may exhibit strong cross-cancer similarity, indicative of shared immune-modulatory mechanisms. We hypothesized that conserved features of the tumor microenvironment, especially from immune cell populations such as CD45+ leukocytes and CD68+ macrophages, could serve as tumor agnostic predictors of ICI response across solid tumors.Methods We collected spatial transcriptomic data from seven cohorts spanning four cancer types (n = 314): melanoma; n=100, NSCLC; n=117, TNBC; n=42, and HNSCC; n=55. Each cancer type included two independent ICI-treated cohorts, except for HNSCC. We extracted spatially resolved gene expression data from CD45+, CD68+, and CK/S100B+ (tumor) compartments. K-means clustering was used to assess cross-cancer similarity of cell-type-specific profiles. Pairwise cosine similarity scores were computed from gene expression profiles per compartment, and within- versus between-cancer similarities were compared using the Wilcoxon rank-sum test. Predictive gene expression signatures were trained on one cohort per cancer type and half of the HNSCC cohort using a LASSO-based predictor trained on differentially expressed genes (50% train/test split).Results Clustering of gene expression profiles revealed clear separation between tumor and immune compartments, with cell types forming distinct, compartment-specific groupings across cancer types ( figure 1A). K-means clustering revealed that clusters were primarily defined by cell type and, to a lesser extent, by cancer type (Figure 1B). Tumor and CD45 compartments showed significantly higher gene expression similarity within the same cancer type compared to across cancer types (tumor: p < 2.2×10-16, CD45: p = 4.8×10-4), whereas CD68 did not show significance (p = 0.31), suggesting that CD68 gene expression profiles are more conserved across cancer types. Gene expression signatures from each compartment predict immunotherapy outcomes across cancer types achieving AUCs of 0.95 (95% CI: 0.93–0.97) for CD68, 0.64 (95% CI: 0.56–0.71) for CD45, and 0.92 (95% CI: 0.89–0.95) for tumor in training; and 0.81 (95% CI: 0.74–0.86), 0.63 (95% CI: 0.55–0.72), and 0.65 (95% CI: 0.57–0.92) respectively in testing (p < 0.01 for all) (figure 2).Conclusions These results support our hypothesis that compartment-specific gene expression, particularly CD68 subset, may allow generation of a spatially informed tumor agnostic gene signature that can predict immunotherapy outcomes across diverse cancer types.References Coulton A, et al. Using a pan-cancer atlas to investigate tumour associated macrophages as regulators of immunotherapy response. Nature Communications 2024;15(1):5665.Ding X, et al. Network-based transfer of pan-cancer immunotherapy responses to guide breast cancer prognosis. NPJ Systems Biology and Applications 2025;11(1):4.Wei C, et al. Tumor-associated macrophage clusters linked to immunotherapy in a pan-cancer census. NPJ Precision Oncology 2024;8(1):176.Yuan M, et al. HAPIR: a refined Hallmark gene set-based machine learning approach for predicting immunotherapy response in cancer patients. NPJ Precision Oncology 2025;9(1):1-11.Ethics Approval Resected tissue samples were collected retrospectively with approval from the Yale Human Investigation Committee protocol #9505008219. Written informed consent, or waiver of consent in some cases, was obtained from Yale cohort patients with the approval of the Yale Human Investigation Committee. For the neoadjuvant clinical trial ( NCT02489448), Ethical approval was obtained from the Yale Human Investigations Committee (HIC; Yale University, New Haven, CT; HIC no. 1409014537)Abstract 515 Figure 1Clustering of spatial transcriptomic profiles across three compartments in four cancer types. (A) t-SNE plot of gene expression, colored by compartment: Tumor (green gradient), CD45+ (yellow gradient), CD68+ (red gradient). (B) t-SNE with k-means clustering of all ROIs (2/patient), annotated by top two cell types per clusterAbstract 515 Figure 2Predictive performance of compartment-specific gene signatures. AUCs with 95% CIs for gene signatures from CD45, CD68, and tumor compartments in training (A) and testing (B) cohorts. Dots show AUCs; lines indicate 95% CIs. Significance: ***p < 0.001, **p < 0.01 (Mann-Whitney)",
  "authors": [
    {
      "affiliations": [
        "Yale School of Medicine, Department of Pathology, New Haven, CT, USA"
      ],
      "name": "Thazin N Aung"
    },
    {
      "affiliations": [
        "Purdue University College of Veterinary Medicine, West Lafayette, IN, USA"
      ],
      "name": "Diba Yaghoubi"
    },
    {
      "affiliations": [
        "University of Queensland, Brisbane, QLD, Australia"
      ],
      "name": "Arutha Kulasinghe"
    },
    {
      "affiliations": [
        "Yale School of Medicine, Department of Pathology, New Haven, CT, USA",
        "Yale School of Medicine, Department of Medical Oncology, New Haven, CT, USA"
      ],
      "name": "Kurt A Schalper"
    },
    {
      "affiliations": [
        "Yale Cancer Center, New Haven, CT, USA",
        "Yale School of Medicine, Department of Internal Medicine, New Haven, CT, USA"
      ],
      "name": "Harriet M Kluger"
    },
    {
      "affiliations": [
        "Yale School of Medicine, Department of Medical Oncology, New Haven, CT, USA"
      ],
      "name": "Lajos Puzstai"
    },
    {
      "affiliations": [
        "Yale University, Department of Molecular Biophysics and Biochemistry, New Haven, CT, USA"
      ],
      "name": "Mark Gerstein"
    },
    {
      "affiliations": [
        "Yale School of Medicine, Department of Pathology, New Haven, CT, USA",
        "Yale School of Medicine, Department of Medical Oncology, New Haven, CT, USA"
      ],
      "name": "David L Rimm"
    },
    {
      "affiliations": [
        "Yale School of Medicine, Department of Pathology, New Haven, CT, USA",
        "Yale University, Department of Molecular Biophysics and Biochemistry, New Haven, CT, USA",
        "NEC Laboratories America, Princeton, NJ, USA"
      ],
      "name": "Jonathan Warrell"
    }
  ],
  "title": "515 Tumor agnostic spatially defined immune cell gene signatures predict immunotherapy outcomes",
  "uid": "1dd48017-81a4-54f5-a904-a5ba0f142a61"
}