Introduction: In acute myeloid leukemia (AML), the prognostic relevance of minimal residual disease (MRD) detection has increased within the past decade and currently, clinical trials are evaluating the value of MRD-guided therapy. Immunotherapy represents an exciting prospect to eliminate residual chemorefractory leukemic cells. Development and implementation of these therapies rely on effective and highly sensitive methods to detect and characterize target antigen expression on residual cells.
Methods: To test the feasibility of the recently proposed viSNE (visualization of t-distributed stochastic neighbor embedding) algorithm to detect and characterize rare, aberrant populations we analyzed patient samples acquired within 3 months prior to hematological relapse, which were MRD-negative by traditional, manual gating. 400.000 events were acquired using an 8 color, 10 parameter panel on a Navios® flow cytometer (Beckman Coulter, Brea, CA, USA). For the bioinformatic analysis, the patient's sample and a healthy donor sample measured on the same machine were combined digitally and viSNE clustering was applied to the resulting dataset. As previously suggested, MRD positivity was defined as the presence of a distinct cluster of >100 cells which consisted of >90% patient cells. Then, expression intensity of the target antigens CD33 and CD123 was compared to primary diagnosis and relapse.
Results: From our database, we identified 12 patients with comprehensive MRD flow assessment available within 3 months prior to hematological relapse who were in CR by cytomorphology at the time of MRD assessment. 9 of these patients were MRD-positive by traditional LAIP-gating (>0.1%). This first analysis therefor focused on the remaining 3 patients with MRD-levels at 0.01%, 0.03% and 0.07% at 21, 42 and 37 days prior to hematological relapse. Using viSNE, distinct clusters of 5058, 225 and 1043 events consisting of 95.5%, 92.6% and 93.8% patient cells were identified, consistent with MRD positivity as defined above. In all three cases, the cells identified by the viSNE algorithm showed the immunophenotype seen at relapse, while there were immunophenotypical shifts compared to primary diagnosis, possibly explaining false-negative MRD-assessment by traditional gating. Specifically, there was a decrease in CD34-expression in one case, and an increase in CD33-expression in 2 cases. Concerning the target antigens CD33 and CD123, both antigens were present on residual cells, showing stable (CD123) expression or an increased (CD33) expression compared to primary diagnosis. Target antigen expression was identical compared to the immunophenotype seen at relapse.
Conclusion: viSNE clustering detected minor, aberrant populations in 3 patients with subsequent hematological relapse, which were not detected by traditional LAIP-gating strategy. viSNE might therefor aid in the analysis of high-dimensional flow cytometry data to detect small, aberrant populations. Target antigen expression could then be determined on these residual cells indicating feasibility of this approach to characterize these rare events for the development of personalized immunotherapeutic strategies as well as response assessment during such therapies. Further validation is ongoing.
Citation Format: Thomas Köhnke, Sandra Rechkemmer, Veit Leonhard Bücklein, Wolfgang Hiddemann, Marion Subklewe. Automated target antigen characterization on residual leukemic cells by flow cytometry. [abstract]. In: Proceedings of the CRI-CIMT-EATI-AACR Inaugural International Cancer Immunotherapy Conference: Translating Science into Survival; September 16-19, 2015; New York, NY. Philadelphia (PA): AACR; Cancer Immunol Res 2016;4(1 Suppl):Abstract nr B002.
- ©2016 American Association for Cancer Research.