{
  "abstract": "Background Robust, non-invasive biomarkers for Immune checkpoint blockade (ICB) response and toxicity are an important unmet clinical need. Peripheral blood—easily accessible and rich in immune information—offers a compelling window into the tumor immune microenvironment (TIME). We present three complementary studies leveraging blood-based single-cell and proteomic profiling with machine learning (ML) to develop predictive models for TIME composition, ICB efficacy, and irAE risk in patients with head and neck squamous cell carcinoma (HNSCC) and melanoma.Methods In the first study, 1 2 we analyzed matched single-cell RNA sequencing (scRNA-seq) data from peripheral blood mononuclear cells (PBMCs) and tumors in HNSCC to develop TIMEP, a machine learning framework that infers tumor immune cell composition from PBMC profiles. In the second [Chang et al, 2025], we performed a meta-analysis of 1,232 HNSCC patients across 11 cohorts to evaluate tumor and circulating B cells as biomarkers of ICB response. In the third,3 we assessed baseline CyTOF (>40 markers) and Olink serum proteomics (>2,900 proteins) in four melanoma cohorts to identify predictors of severe irAEs using machine learning.Results TIMEP successfully infers the infiltration levels of 12 major immune cell types in the tumor and predicts 17–47% of immune gene expression patterns from blood. Exhausted CD8+ T cell signatures and a derived memory B cell:regulatory T cell (Treg) ratio are strong predictors of ICB response.In the HNSCC meta-analysis, both tumor and importantly, circulating B cell levels are robust predictors of ICB response, surpassing 20 established biomarkers including tertiary lymphoid structure signatures. Mechanistically, elevated B cell abundance associates with an immune activated ‘hot’ tumor microenvironment.In melanoma, elevated baseline activated CD8+ central memory T cells and pre/post-treatment serum CXCL9 changes predicted severe treatment toxicity. A baseline bivariate model generalizes well across multiple cohorts (mean AUC > 0.78). Notably, this biomarker decouples toxicity from response, offering a valuable tool for clinical decision-making.Conclusions These studies collectively demonstrate the power of blood-based immune profiling—via scRNA-seq, CyTOF, and proteomics—combined with AI, to non-invasively infer TIME states, predict ICB efficacy and toxicity ( figure 1). Our findings lay a solid basis for further prospective testing going forward.References Yingying Cao#, Tian-Gen Chang#, Fiorella Schischlik#, Kun Wang, Sanju Sinha, Sridhar Hannenhalli, Peng Jiang, Eytan Ruppin*. Inferring characteristics of the tumor immune microenvironment of patients with HNSCC from single-cell transcriptomics of peripheral blood. Cancer Research Communications 2024;4(9):2335–2348.Tian-Gen Chang#, Aris Spathis#, Alejandro A. Schäffer, Niki Gavrielatou, Fengshen Kuo, Dongya Jia, Sumit Mukherjee, Cem Sievers, Panagiota Economopoulou, Mirian Anastasiou, Myrto Moutafi, Lipika R Pal, Joris Vos, Andrew Sangho Lee, Stanley Lam, Karena Zhao, Peng Jiang, Clint T. Allen, Periklis Foukas, Georgia Gomatou, Grégoire Altan-Bonnet, Luc GT Morris*, Amanda Psyrri*, Eytan Ruppin*. Tumor and blood B cell abundance outperforms established immune checkpoint blockade response prediction signatures in head and neck cancer. Annals of Oncology 2025;36(3):309–320.Tian-Gen Chang#*, Magdalena Kovacsovics-Bankowski#, Fernando Pablo Canale, Julia Martinez Gomez, Lucia Boffelli, Amos Stemmer, Chi-Ping Day, Mitchell Paul Levesque, Lukas Flatz, Nicolas Gonzalo Nuñez*, Eytan Ruppin*, Siwen Hu-Lieskovan*. Pre-existing peripheral CD8+ central memory T cell activation predicts severe toxicity and limited benefit from immune checkpoint blockade in melanoma. (under review)Abstract 483 Figure 1",
  "authors": [
    {
      "affiliations": [
        "National Cancer Institute, National Institutes of Health, Bethesda, MD, USA"
      ],
      "name": "Tiangen Chang"
    },
    {
      "affiliations": [
        "National Cancer Institute, National Institutes of Health, Bethesda, MD, USA"
      ],
      "name": "Eytan Ruppin"
    }
  ],
  "title": "483 New biomarkers predicting immunotherapy response and toxicity from the blood",
  "uid": "304bd14f-a812-591f-9ba3-becfe7f5ff95"
}