On the Origins of AML Relapse

While induction into remission is effective in the majority of acute myeloid leukemia (AML) patients, disease recurrence is common, especially among the elderly. Understanding the origins of AML relapse would permit better treatments targeting the specific cells that survive chemotherapy. While some evidence suggests that AML relapse can originate either from a minor or a major clone that is already present at diagnosis, the exact origins of AML relapse are still obscure. In the current study we aimed at identifying the origins of AML by identifying genetic variants that appear at relapse, and to then track these variants back into specific cell populations present at diagnosis. We hypothesized that relapse might have multiple origins: from the major blast population, from rare leukemia imitating cells (LIC) as detected using xenografting, or from preleukemic stem cells (preL-HSCs).

Methods: The bulk diagnosis and relapse samples of peripheral blood from eleven AML patients were analyzed, first by whole genome sequencing (50X coverage) to identify somatic mutations and genetic variants which were specifically present at relapse (relapse variants-RVs). The presence RVs was then reassessed in phenotypically defined sub-fractions sorted from the diagnosis samples, at a sensitivity of 1 in 1000 by digital PCR. The following sub-populations were genotyped: 1) isolated CD33+ blasts (the major population) 2) phenotypically defined leukemic and preleukemic stem cells 3) functionally defined leukemia initiating cells (LICs) harvested from xenografts (an average of 30 xenografts were generated from each diagnosis and relapse sample).

Results: LICs, but not the dominant blast population from diagnosis carried the RVs in 3 of 11 cases. In these patients CD33-CD34+CD45RA+ immature cells from diagnosis also carried the RVs. In a second subset of 3 of 11 AML samples, relapse originated from a minor clone present within the CD33+ leukemic blasts; these samples did not produce exnografts. Other samples (2/11) exhibited relapse samples that arose from a combined origin (both LICs, and CD33+ blasts, or from the major clone (1/11). In two cases we could not identify the origins of relapse. As our initial results suggested that the cells responsible for AML relapse can come from distinct origins within the diagnosis sample, we next asked whether other functional and phenotypic differences might be present between the patients that have different relapse origins. RNA sequencing analysis of bulk cells from diagnosis demonstrated a remarkable clustering of the global gene expression that correlated with the origin of relapse. Unsupervised hierarchical clustering grouped together the AML samples who relapsed from the LICs, while all other samples were in a very distinct second cluster. The gene expression signature of the samples that relapsed from LICs was consistent with a monocytic phenotypic signature, while the other samples were more progenitor-like. To further expand and validate our findings we used the same unsupervised clustering on the RNA sequencing data of AML samples who relapsed in the TCGA dataset (n=86). Remarkably, the similar two main clusters were generated; comparison by GSEA provided evidence that the gene expression clusters in our study were generated by the same genes as in the TCGA clusters.

Conclusion: Our results provide for the first time evidence that AML can relapse from distinct, predictable and pre-existing origins: AMLs with a monocytic phenotype relapse from chemo-resistant LICs; and AMLs with a progenitor gene expression pattern (yet lacking xenografting capacity) that relapse from CD33+ cells. These results pose a series of predictions as to the success of different therapies. For example, in the former group the major monocytic clone is sensitive to chemotherapy, yet relapse originates from CD33-CD34+CD45RA+ cells and would therefore be predicted to be resistant to Anti-CD33 therapeutics. On the other hand, relapse in the latter group originates from CD33+ cells and these are predicted to be sensitive to Anti-CD33 therapeutics. The results of this study document the complexity in origins of AML relapse and have important implications for the design of future more effective and personalized strategies for preventing AML relapse.