Background: In acute myeloid leukemia (AML), detection of minimal residual disease (MRD) by flow cytometry is an adverse prognostic factor besides pre-treatment risk classifications, including cytogenetic and molecular aberrations. High dimensional multiparameter flow cytometry (MPFC) offers improved sensitivity and specificity, however manual analysis is increasingly challenging. In this study, we explore the value of the recently proposed viSNE algorithm to quantify MRD levels in patients with AML achieving complete remission (CR) after intensive induction chemotherapy.

Methods: Bone marrow samples from patients with AML (excluding patients with acute promyelocytic leukemia) were analyzed by 8-10 MPFC using a NAVIOS flow cytometer (Beckman Coulter, Brea, CA, USA). Only patients achieving a CR or CR with incomplete blood count recovery (CRi) post-induction were included in this analysis. Manual gating of MRD flow data was performed as described previously (Köhnke et al., Leukemia 2014) using a cutoff for MRD positivity of 0.1%. The viSNE algorithm was performed as described previously (Amir et al., Nat. Biotech. 2013) and MRD positivity was defined as the presence of a distinct cluster of >100 cells which consisted of >90% patient cells. Kaplan-Meier estimator and log-rank test as well as Cox's proportional hazards regression model were used to analyze survival data.

Results: Post-induction flow cytometry and clinical data of 38 patients with AML achieving CR (n=34) or CRi (n=4) were available for analysis (median age 53 years; de-novo AML n=32, tAML n=1, sAML n=5). Most patients belonged to the intermediate cytogenetic risk group (MRC favorable n=5, intermediate n=22, adverse n=11).

19/38 patients were MRD positive post-induction by manual gating. 12/19 patients deemed MRD positive relapsed, whereas 3/19 patients deemed MRD negative relapsed. Therefore, MRD positivity by manual gating correlated with reduced relapse free survival (median RFS for MRD positive patients: 7.5 months vs. median not reached for MRD negative patients, log-rank test p=0.017). For overall survival (OS), no significant impact of MRD positivity could be detected so far (p=0.3), however follow-up was short (median follow-up 9.3 months). MRD positivity by manual gating remained an independent risk factor for RFS (HR 4.8, p=0.021) when compared to genetic risk and age.

MRD positivity by viSNE clustering was seen in 19/38 patients. 10/19 patients deemed MRD positive by viSNE relapsed, while 5/19 deemed MRD negative by viSNE experienced a relapse. This resulted in a trend towards shorter RFS for MRD positivity by viSNE (median RFS 9.9 months vs. 19.0 months for MRD negative patients, p=0.185). Among the patients deemed MRD positive by viSNE who did not relapse, i.e. false-positive patients, follow-up was very short (<3.5 months) in 4/9 cases and final judgment whether these patients are truly negative should be withheld. However, differentiation between healthy monocytes and potential MRD cells by viSNE seems to be especially challenging and warrants optimization.

Nevertheless, within the group of patients deemed MRD negative by manual gating, viSNE was able to detect all patients who experienced a subsequent relapse. To maximize sensitivity for MRD detection, we therefore combined the results of both methods and defined MRD positivity as positivity by manual gating and/or viSNE clustering. Using this combined strategy, 28/38 patients were defined as MRD positive and 10/38 as MRD negative. Within the MRD positive group, 15/28 patients relapsed, whereas none of the patients in the MRD negative group relapsed (Figure 1, median RFS for MRD positive patients: 7.7 months, log-rank test p=0.028). Of note, within the MRD negative group only 2 patients (both with intermediate genetic risk) went on to receive a allogeneic stem cell transplantation whereas 8 remained in remission with chemotherapy alone.

Conclusion: In summary, viSNE clustering used in combination with manual gating improves the sensitivity of MRD detection by flow cytometry. Importantly, viSNE is able to detect MRD positivity within the manual MRD negative patient population. This is the first report of MRD detection in AML using this method. However, improvements in the algorithm as well as further studies are needed to validate the prognostic value of viSNE clustering in AML.


No relevant conflicts of interest to declare.

Author notes


Asterisk with author names denotes non-ASH members.