Background: The heterogeneous and adaptive nature of AML-associated genomic and proteomic landscape may account for disease relapse and poor prognosis, as therapy-associated selective pressure drives the emergence and expansion of AML clones with features different from those detected at diagnosis. The evolving focus is on single-cell analytical tools to fully capture the pathobiological heterogeneity of AML. We leveraged CyTOF to interrogate the proteomic heterogeneity in patients with R/R AML, which could potentially permit the design of rational combinatorial therapeutic approaches targeting vulnerabilities in these cells..

Methods: To dissect AML heterogeneity and its contribution to treatment failure, we designed a 51-parameter CyTOF panel and interrogated cellular hierarchy, major immune phenotypes, anti-apoptotic molecules, signaling pathways, exhaustion markers and attractive targets for CAR T-cell therapy in R/R AML (n:13).

Results: First, we generated two-dimensional t-SNE maps and observed that the leukemic bone marrow compartment harbored immature (CD34+CD38- and CD34+CD38+) and mature leukemic blasts (CD33+CD34-) and major immune subsets. Constitutively active signaling pathways characterized by high levels of p-4EBP1, p-MEK1/2, p-S6 and p-AKT, marked immature and mature leukemia cells and comparative analysis revealed that monocytic blasts harbored more active signaling networks. The proportions of these subpopulations varied significantly across patients. We initially assessed the distribution of anti-apoptotic molecules across these leukemia compartments. Strikingly, Bcl-2 levels were considerably high within less-differentiated leukemic cell compartments and CD68 expressing leukemic blasts with monocytic differentiation had significantly lower levels of Bcl-2. This suggests that differentiated leukemic cells could preferentially survive under selection pressure of Bcl-2 inhibitors. On the contrary, we observed a trend towards higher Mcl-1 levels in differentiated leukemia cells. These findings provide a rationale for combining therapeutic modalities to target different leukemia subpopulations. Indeed, Bcl-2 and Mcl-1 inhibitors (Venetoclax and AZD5991) resulted in highly synergistic effects in AML PDX models. Hence, this analysis supports the hypothesis that Mcl-1 overexpression is a resistance factor to Bcl-2 inhibition, usually understood as developing in the same cell.

Next, we assessed expression patterns of putative CAR T-cell targets expressed on leukemic cells and identified significant variegated expression patterns in R/R AML samples. CD123 demonstrated patchy distribution across immature and differentiated leukemic blasts while CD33 expression was the main characteristic of differentiated leukemic blasts. Of note, the immature leukemia compartment demonstrated variable levels of CD33 and complete lack of CD33 on CD34+ leukemic cells was observed in a subgroup of patients. CLL-1 was uniformly expressed across all leukemia compartments, but was not ubiquitously expressed in all patient samples, revealing substantial interpatient heterogeneity in R/R AML and highlighting the concept of targeting at least two antigens concomitantly.

Lastly, we sought to undertake high-dimensional assessment of immune compartments to identify major immune phenotypes in heavily treated R/R AML patients and discover the link between immune phenotypes and AML-associated traits. In line with leukemia cell heterogeneity, we found a significant degree of variation in immune cell composition among patients. Inhibitory molecules PD-1 and TIGIT were significantly expressed on CD4 and CD8 T-cells respectively, providing a rationale for use of combinatorial immunotherapeutic approach for the treatment of AML.

Conclusion: Single-cell profiling of R/R AML using CyTOF reveals significantly heterogeneous expression patterns of molecules targeted by BH3 mimetics (Bcl-2 and Mcl-1), CAR T-cells, and other antibody-based immunotherapeutic therapies. This approach provides a rationale to develop combinatorial therapeutic approaches targeting distinct leukemia sub-populations with discrete expression patterns of established and novel putative targets. An example is the combined targeting of Bcl-2 and Mcl-1, which are differentially expressed in early and more differentiated leukemia subpopulations.


Carter:AstraZeneca: Research Funding; Syndax: Research Funding; Amgen: Research Funding; Ascentage: Research Funding. Andreeff:Centre for Drug Research & Development; Cancer UK; NCI-CTEP; German Research Council; Leukemia Lymphoma Foundation (LLS); NCI-RDCRN (Rare Disease Clin Network); CLL Founcdation; BioLineRx; SentiBio; Aptose Biosciences, Inc: Membership on an entity's Board of Directors or advisory committees; Daiichi-Sankyo; Jazz Pharmaceuticals; Celgene; Amgen; AstraZeneca; 6 Dimensions Capital: Consultancy; Daiichi-Sankyo; Breast Cancer Research Foundation; CPRIT; NIH/NCI; Amgen; AstraZeneca: Research Funding; Amgen: Research Funding.

Author notes


Asterisk with author names denotes non-ASH members.