Immune and Myeloid Differentiation Shape Transcriptome to Discriminate Evolutionary Stages of Cellular Transformation

Background: Myeloid mutations retain ability to elicit variable innate immunity response. T-cell “autoimmunity” restricts myeloid clonal expansion suggesting that “clonal evolution” requires genomic and transcriptomic modifications to “enhance cellular fitness”. Gene expression (GE) could assist exome sequencing in monitoring clonal evolution and to understand how transformed myeloid cells behave under immune selective pressure. In this study, our aims are: (a) to describe transcriptomic immune and inflammatory modifications in patients (pt) harboring clonal hematopoiesis (CH), myelodysplastic syndrome (MDS) and Acute Myelogenous Leukemia (AML) (b) to evaluate global immune/inflammatory GE training dataset ability to distinguish CH from Myelodysplastic syndrome (MDS), and Acute Myelogenous leukemia (AML).

Methods: After IRB approval, 650 pt [150, 120, 180 and 200 normal, CH, MDS and AML cases, respectively] were included for analysis. Non-parametric ANOVA was used to detect differential Major Histocompatibility Complex (MHC)-1 A, B C, IL1b, interleukin 6 (IL-6), IL8, CD274 (PDL1) GE expression within individual groups. Next, Bayesian statistics reprioritized transcriptomic data from 85 to 20 annotated genes (Fig.1) including MHC-1 and 2, inflammation [i.e., TNFa, IL-6, IL-8, among others], Interferon (INF) signaling, differentiation [i.e., CD34, CD38, CD33, among others]. Individual gene subgroups were selected for class prediction based on sensitivity and specificity, and final receiver operator curve (ROC) procedure.

Results: In CH vs MDS vs AML, HLA-A, B and C GE discriminated CH from MDS and AML [p=<0.0001 and p=<0.0001, each MCH-1 gene]; as did GE for IL-6 [p=0.006 and p=<0.0001, for MDS and AML]; and IL-8 [p=0.01 and p=<0.0001, for MDS and AML]. Given that progressive myeloid differentiation characterizes MDS and AML pathogenesis, we integrated myeloid maturation GE [i.e., CD34, CD33, CD117] into our immune/inflammatory GE training (1/3 of sample) and validation prediction to identify CH vs MDS vs AML. Our approach allowed accurate discrimination between CHIP vs MDS (AUC=74.5%; CI 0.66-0.82), CHIP vs AML (AUC=80%; CI 0.71-0.89), MDS vs AML (AUC=90.8%; CI 0.85-0.95) (Fig. 2). Interestingly, when considering GE for MHC 1 and 2, MDS was accurately discriminated from AML (AUC= 76.1%; CI 0.6-0.8) and “normal polyclonal hematopoiesis” from MDS (AUC=66.7%; CI 0.5-0.7).

Conclusions: Our annotated GE pathways inform evolutionary transcriptomic modifications, which improve accuracy in detecting CH, MDS and AML. It is likely that cells “reprogram” transcriptome under selective immune pressure allowing fitness, and likely expansion. External reproducibility of our data, and integration within CH risk score and Molecular- international prognostic score system (M-IPSS) for MDS may significantly improve predictive accuracy. Additionally, it could also inform transformative events leading to hemopoietic malignancy conversion to investigate therapeutic interventions.