Unsupervised Flow Cytometry Analysis: Application to Minimal Residual Disease Detection in a Cohort of 40 Acute Myeloblastic Leukemia Patients with Molecular Markers

Minimal/measurable residual disease (MRD) assessment in the course of acute myeloblastic leukemia (AML) monitoring is a topic that currently reaches undisputable interest. Multiparameter flow cytometry (MFC) appears as a valuable, rapid tool to assess response to therapy. Within the European LeukemiaNet (ELN) a task force has promoted the use of a backbone panel of monoclonal antibodies likely to cover both the most common leukemia associated immunophenotyping patterns (LAIP) and different from normal (DfN) immunophenotypes (Schuurhuis et al., Blood 2018). Classical gating strategies of MRD assessment however rely on supervised/operator-dependent approaches. The development of mass-cytometry has led to the emergence of new unsupervised analysis software. Among them, the FlowSOM (Flow Self Organizing Maps) R-based solution has been recognized as the most efficient, applicable to both mass and classical flow cytometry.

Combination of the R-based FlowSOM definition of self-organized minimal spanning trees (MST) and the Kaluza® classical flow cytometry analysis software allows for a combined exploration of a normal bone marrow (NBM) matrix (derived from 19 normal BM samples)together with individual diagnosis and follow-up (FU) samples. This strategy was applied to 40 AML patients treated with a conventional 3+7 regimen for whom at least post induction FU was available, together with targets allowing for molecular MRD assessment (NPM1, RUNX1, CBFB-MYHorBCR-ABL). According to previously published panels (GEIL, Cytometry part B, 2018), two tubes were tested in MFC and results were compared to molecular MRD results for 96 FU time points. All patients could be evaluated in MFC at all time points. Comparison of molecular and MFC data showed an overall good concordance (80.2%), especially for FU1 time points (87.5%) with a sensitivity of 0.89 and a specificity of 0.75. Yet, outcome after induction appeared to be more correlated with MFC-negative MRD. This work also allowed to demonstrate the particular ability of the FlowSOM algorithm to disclose subtle chemoresistant subclones during FU. Moreover, by systematic comparison of normal bone marrow and FU points, MRD could be asserted without ambiguity and regenerating BM identified as such.

Of note, this work could be performed on retrospective data, based on the fact that instruments were harmonized as reported (Lacombe et al. Leukemia, 2016) and identical panels were used all along. This makes this strategy easily transferrable to a network of laboratories using consensus settings and panels, as proposed by the European LeukemiaNet (Schuurhuis et al., Blood 2018).

All in all, this study shows that unsupervised MFC MRD assessment appears as an accurate tool to appreciate response to chemotherapy in AML, and the FlowSOM + Kaluza method, combined with a systematic diagnostic FU / NBM multicomparison approach, increases the resolving power in the monitoring of MRD compared to the classic LAIP/different from normal approaches as implemented to date.