Association of Gene Expression Patterns in Bone Marrow Cells with Likelihood of Treatment Free Remission after TKI Discontinuation

Approximately half of chronic myeloid leukemia (CML) patients achieving deep molecular remission on tyrosine kinase inhibitors (TKIs) can successfully achieve treatment-free remission (TFR) after stopping TKI therapy. To gain insight into mechanisms determining success or failure of treatment discontinuation, and develop potential predictors of TFR, we studied gene expression in bone marrow (BM) samples from 38 CML patients in whom Imatinib was stopped following achievement of deep molecular response. Samples were obtained prior to treatment discontinuation. RNA-Seq was performed to evaluate differential gene expression in BM mononuclear cells (MNC) from relapsed patients (n=14) versus those that maintained TFR (n=24). Prediction analysis of microarrays identified a 181-gene classifier that predicted relapse with 85.7% sensitivity and 95.8% specificity. In addition, random forests analysis also identified a 54-gene classifier that predicted relapse with 78.6% sensitivity, and 100% specificity. To identify genes associated with relapse, we analyzed gene expression data with time to relapse using the Cox proportional hazards model. Cox-fitted principal component analysis identified a set of 9 genes that correctly predicted 93% of TFR patients, whereas penalized Cox regression identified a set of 10 genes (7 overlapping) that correctly predicted 81.8% of TFR patients. These results indicate that gene expression analysis could potentially identify patients with low risk for relapse. Gene set enrichment analysis indicated that BM cells from patients that relapsed demonstrated significant upregulation of TNFα, inflammation and TGFβ related gene signatures, whereas MNC from patients maintaining TFR were enriched for T cell immunity, antigen response, and adipogenesis related genes. To further study mechanisms underlying relapse versus TFR, we analyzed gene expression in CD34+ stem/progenitor cells and CD45-GlyA-CD31- mesenchymal cells purified from BM samples. CD34+ cells from relapsed patients were enriched for genes related to transcription, protein translation and cell cycle, and from TFR patients for genes related to T cell immune response and antigen presentation. Mesenchymal cells from relapsed patients were enriched for genes related to TNFα and STAT5 signaling, and from TFR patients for adipogenesis gene sets. Importantly, TNFα and TGFβ signaling have been related to retention of resistant, quiescent CML LSC after TKI treatment, whereas T cell activity and adipogenesis have been related to reduced LSC retention and growth. In conclusion, our results support further evaluation of gene expression in BM cells to identify patients at low risk for relapse after TKI discontinuation, and indicate a potential role for immune and inflammatory mechanisms in determining continued remission versus relapse.