Establishing Distinct Cytokine Signatures Differentiating between Acute Myeloid Leukemia, Myelodysplastic Syndrome, and Chip Using Bone Marrow RNA or Cell-Free RNA (cfRNA)

Cytokines are essential for various immune functions and the overall inflammatory response to myeloid neoplasms in bone marrow. They play a major role in bone marrow microenvironment in normal and abnormal hematopoiesis. Cytokines exert their functions by interacting with their receptors, and full evaluation of the cytokines’ roles requires evaluating their receptors as well. Using next generation sequencing (NGS) and machine learning, we measured the expression of 36 cytokines/chemokines and cytokine receptors in the bone marrow (BM) of patients with acute myeloid leukemia (AML), myelodysplastic syndrome (MDS), and clonal hematopoiesis of indeterminate potential (CHIP), and established cytokine signatures that differentiate between these diseases. We also explored if peripheral blood cell-free RNA (cfRNA) reflects BM environment.

Methods: RNA was extracted from the bone marrow (BM) samples of patients with AML (N=515), MDS (N=825), and CHIP (N=915). cfRNA was extracted from the peripheral blood of patients with AML (N=30), MDS (N=184), and CHIP (N=502). BM RNA and cfRNA were sequenced using a 1500-gene targeted RNA next generation sequencing (NGS) panel. More than 80 million reads and a percentage of spliced reads above 20% were required for acceptable evaluation. The expression levels of 36 cytokines/chemokines were used in this analysis. Using Bayesian statistics, each of the 36 biomarkers was ranked based on its sensitivity and specificity of distinguishing between two diagnostic classes with 10-fold cross validation by leave-one-out. Random forest algorithms were developed using two-thirds of the BM samples and top-ranked biomarkers to build signatures that distinguished between two diagnostic classes. One-third of the bone marrow samples were used for testing these algorithms. Each model was then used to test if cfRNA samples showed the same results obtained from BM samples.

Results: In distinguishing between AML and MDS, we first used Bayesian statistics with 10-fold cross validation to rank the studied 36 cytokines/chemokines and receptors. After ranking, random forest showed that a cytokine signature of 20 top-ranked biomarkers can reliably distinguish between BM with MDS from BM with AML (AUC: 0.874, CI: 0.843-.906). The biomarkers in the signature are: TNFRSF10D, TNFAIP3, TNFRSF4, IL3RA, IL8, TGFBR3, CXXC4, IL1RAP, IL7R, IFNG, TNFRSF10B, IL2, TNF, TGFBR2, CXCR4, TNFRSF14, CTLA4, IL12RB2, TGFBI, IL21R. Using the same algorithm and the same biomarkers but as measured using peripheral blood cfRNA, AML was distinguishable from MDS (AUC: 0.706, CI: 0.617-0.795). Using a similar approach, we were able to distinguish between BM with MDS and BM with CHIP using random forest and a cytokine signature of 20 biomarkers (AUC: 0.761, CI: 0.722-0.800). The biomarkers in this signature are TGFBR2, TNFRSF14, CXCR4, IL1RAP, TNFRSF10D, TNFAIP3, IL8, IL12RB2, IL1B, CTLA4, IL7R, IL21R, TNFRSF9, TNF, IL3, TNFRSF17, TGFBR3, TNFRSF4, TNFRSF10B, and IL13RA2, in order. Using the same signature and biomarkers but as measured using peripheral blood cfRNA from patients with MDS or CHIP, we were able to distinguish between the two diseases with AUC of 0.712 (CI: 0.668-0.756). While both signatures share multiple biomarkers (IL8, CTLA4, CXCR4...), the signature distinguishing AML from MDS is uniquely using IL3, IL1B and TNFRSF17 while the signature for distinguishing MDS from CHIP is uniquely using CXX4, IL2, and TGFB1.

Conclusions: There are unique cytokines/chemokines and receptor signatures for each of the AML, MDS, and CHIP. This indicates that these biomarkers are crucial for defining each of these diseases. Furthermore, our data shows that cfRNA is reliable in reflecting bone marrow findings and can be used as an alternative to bone marrow samples for measuring and monitoring these signatures.