{
  "abstract": "Background The clinical adoption of bulk RNA sequencing (RNA-seq) for tumor transcriptome analysis is hampered by the complex cellular composition of tumors, which confounds the gene expression of malignant cells (MCs) with transcripts from the tumor microenvironment (TME). Accurate compartment-specific gene expression profiling is crucial for identifying reliable biomarkers for cancer diagnostics and personalized treatment. Helenus, a machine learning (ML) tool developed to deconvolve bulk RNA-seq data into gene expression profiles for MCs and TME cells, was used to improve patient stratification for immune checkpoint inhibition (ICI).Methods Helenus was trained on ~200 million simulated transcriptomes reflecting diverse tumor types and purities. Helenus demonstrated improved accuracy in separating gene expression profiles of MCs and TMEs in validation tests over existing methods. Helenus was applied to both published 1-3 and two new cohorts (n=45 and n=136) of non-small cell lung cancer (NSCLC) patients, to uncover consistent predictors of ICI response in this combined lung adenocarcinoma (LUAD) cohort (n=404). Functional gene expression signatures were developed internally, as described previously.4 ssGSEA scores were used to determine if a gene signature was differentially enriched in tumor samples. Survival was analyzed based on biomarker expression level (low, medium, high) before and after Helenus’ application using Kaplan-Meier estimates (logrank test P-values). Cox regression models provided hazard ratios (HRs) to assess associations between gene expression and outcomes.Results Helenus unveiled previously obscured immunotherapeutic predictive biomarkers by refining the TME-specific gene expression profiles. Helenus-reconstructed PD-L1 expression in the TME significantly stratified progression-free survival (PFS) in LUAD patients undergoing ICI therapy (HR=0.304, P=0.002), in contrast to IHC (immunohistochemistry) and uncorrected RNA-seq, which did not differentiate responders by PD-L1 expression ( figure 1). Across cohorts, Helenus-enhanced TME expression amplified the significance and reproducibility of multiple response-associated gene signatures, particularly those linked to cytokine-mediated recruitment of T and B cells and formation of tertiary lymphoid structures (TLS). In the combined LUAD cohort, the TLS-related signature predicted PFS more effectively after Helenus than uncorrected RNA-seq (HR=0.507, P=4e-5), unmasking immune mechanisms associated with effective ICI response (figure 2).Conclusions Helenus uncovers compartment-specific expression signals from bulk RNA-seq data, enabling reliable detection of shared biological mechanisms underlying ICI response. By purifying TME expression, Helenus not only improves stratification based on PD-L1 but also reveals consistent and clinically relevant immune signatures across cohorts. This approach holds significant potential for refining biomarker-driven patient selection in immuno-oncology.References Ravi A, Hellmann MD, Arniella MB, Holton M, Freeman SS, Naranbhai V, et al. Genomic and transcriptomic analysis of checkpoint blockade response in advanced non-small cell lung cancer. Nat Genet. Nature Publishing Group; 2023;55:807–19.Jung H, Kim HS, Kim JY, Sun J-M, Ahn JS, Ahn M-J, et al. DNA methylation loss promotes immune evasion of tumours with high mutation and copy number load. Nat Commun. Nature Publishing Group; 2019;10:4278.Kang J, Lee JH, Cha H, An J, Kwon J, Lee S, et al. Systematic dissection of tumor-normal single-cell ecosystems across a thousand tumors of 30 cancer types. Nat Commun. Nature Publishing Group; 2024;15:4067.Bagaev A, Kotlov N, Nomie K, Svekolkin V, Gafurov A, Isaeva O, et al. Conserved pan-cancer microenvironment subtypes predict response to immunotherapy. Cancer Cell. 2021;39:845-865.e7.Ethics Approval The study was conducted according to the Declaration of Helsinki and was approved by the Institutional Review Board of Saint Luke’s Hospital of Kansas City (SLHS IRB-22-051) and Inova Schar Cancer Institute (U23-03-5013).Abstract 154 Figure 1Progression-free survival (PFS) of ICI-treated lung adenocarcinoma (LUAD) patients separated by PD-L1. Only samples from the combined cohort with paired IHC samples were analyzed hereAbstract 154 Figure 2PFS of ICI-treated LUAD patients (combined cohort) separated by TLS-associated cell recruitment signature",
  "authors": [
    {
      "affiliations": [
        "BostonGene Corporation, Waltham, MA, USA"
      ],
      "name": "Natalia Khotkina"
    },
    {
      "affiliations": [
        "BostonGene Corporation, Waltham, MA, USA"
      ],
      "name": "Boris Shpak"
    },
    {
      "affiliations": [
        "BostonGene Corporation, Waltham, MA, USA"
      ],
      "name": "Maria Savchenko"
    },
    {
      "affiliations": [
        "BostonGene Corporation, Waltham, MA, USA"
      ],
      "name": "Aleksandr Serdiukov"
    },
    {
      "affiliations": [
        "BostonGene Corporation, Waltham, MA, USA"
      ],
      "name": "Linda Balabanian"
    },
    {
      "affiliations": [
        "BostonGene Corporation, Waltham, MA, USA"
      ],
      "name": "Sheila T Yong"
    },
    {
      "affiliations": [
        "Inova Schar Cancer Institute, Fairfax, VA, USA"
      ],
      "name": "Janakiraman Subramanian"
    },
    {
      "affiliations": [
        "Saint Luke’s Cancer Institute, Kansas City, MO, USA"
      ],
      "name": "Dhruv Bansal"
    },
    {
      "affiliations": [
        "BostonGene Corporation, Waltham, MA, USA"
      ],
      "name": "Aleksander Bagaev"
    }
  ],
  "title": "154 Machine-learning tool Helenus enhances gene expression profiling to improve biomarker discovery in lung adenocarcinoma cohorts and advance the precision of immunotherapy",
  "uid": "75686164-41ce-58cb-9ab0-74a41d6cc3f7"
}