Establishing Lymphoma Type-Specific Cytokine Signatures Using Tissue-Based RNA or Peripheral Blood Cell-Free RNA (cfRNA)

• Published in: Blood 2024-11, Vol.144 (Supplement 1), p.4367-4367
• Main Authors: Albitar, Maher, Zhang, Hong, Agersborg, Sally, Charifa, Ahmad, Pecora, Andrew L, Leslie, Lori A., Feldman, Tatyana, Ip, Andrew, Goy, Andre
• Format: Article
• Language: English
• ISSN: 0006-4971; 1528-0020
• DOI: 10.1182/blood-2024-203834

Introduction: Cytokines and chemokines play important roles in lymphoma growth and response to therapy. They are also relevant for the clinical symptoms and various manifestations of the disease. Evaluating RNA levels of large numbers of cytokines/chemokines and their receptors in tissue is now possible using next generation sequencing (NGS). However, it is not known if the tumor microenvironment is reflected in peripheral blood cell-free RNA (cfRNA). Using NGS, we evaluated the RNA levels of 36 cytokines/chemokines and their receptors in tissue samples from patients with various types of lymphoid neoplasms and used machine learning to establish signatures that distinguish between various types of lymphoma. We also explored if these signatures are adequately reflected when peripheral blood cfRNA is tested instead of tissue RNA. Methods: RNA was extracted from tissue samples with confirmed chronic lymphocytic leukemia (CLL) (N=184), diffuse large B-cell lymphoma (DLBCL) (N=287), mantle cell lymphoma (MCL) (N=74), T-cell lymphoma (N=276), and bone marrow samples with low level B-cell lymphoid neoplasm not otherwise classified (N=750). Peripheral blood cfRNA was extracted from 19 patients with DLBCL, 23 with mantle cell lymphoma, 39 with T-cell lymphoma, 16 with CLL, and 361 with low level B-cell lymphoid neoplasm not otherwise classified. Cellular RNA and cfRNA were sequenced using a 1500-gene panel. The expression levels of 36 cytokines/chemokines were used in this analysis. Using Bayesian statistics, we first evaluated and ranked the sensitivity and specificity of each biomarker individually with 10-fold cross validation using leave-one-out. Then, two-thirds of the tissue samples were used for training and building models and one-third was used for testing these models. Each model was then used to test if cfRNA samples showed the same results obtained from tissue samples. Results: For differentiating MCL from DLBCL, random forest used a signature of 16 biomarkers to distinguish between the two diseases (AUC: 0.928, CI: 0.855-1.00). The top 16 biomarkers are: TGFBR2, IL21R, TGFBI, CXCR4, TNFAIP3, IL8, TGFB3, IL2, TNFRSF17, CTLA4, TNFRSF4, TNFRSF6B, IL2RA, TNFRSF10B, IL3RA, and CXXC4. The same random forest algorithm using these 16 biomarkers as measured in cfRNA was also able to reliably distinguish MCL from DLBCL (AUC of 0.789, CI: 0.652-0.927). Using the same approach to distinguish between CLL and DLBCL, only 6 biomarkers (TGFBR2, IL8, IL21R, CCL2, TNFRS