Purpose:

Acute myeloid leukemias (AML) are clonal diseases that develop from leukemic stem cells (LSC) that carry an independent prognostic impact on the initial response to induction chemotherapy, demonstrating the clinical relevance of LSC abundance in AML. In 2018, the European LeukemiaNet published recommendations for the detection of measurable residual disease (Bulk MRD) and suggested the exploration of LSC MRD and the use of multiparametric displays.

Experimental Design:

We evaluated the performance of unsupervised clustering for the post-induction assessment of bulk and LSC MRD in 155 patients with AML who received intensive conventional chemotherapy treatment.

Results:

The median overall survival (OS) for Bulk+ MRD patients was 16.7 months and was not reached for negative patients (HR, 3.82; P < 0.0001). The median OS of LSC+ MRD patients was 25.0 months and not reached for negative patients (HR, 2.84; P = 0.001). Interestingly, 1-year (y) and 3-y OS were 60% and 39% in Bulk+, 91% and 52% in Bulk-LSC+ and 92% and 88% in Bulk-LSC−.

Conclusions:

In this study, we confirm the prognostic impact of post-induction multiparametric flow cytometry Bulk MRD in patients with AML. Focusing on LSCs, we identified a group of patients with negative Bulk MRD but positive LSC MRD (25.8% of our cohort) with an intermediate prognosis, demonstrating the interest of MRD analysis focusing on leukemic chemoresistant subpopulations.

Translational Relevance

Detection of measurable residual disease (MRD) is increasingly important for therapeutic tailoring of acute myeloid leukemias (AML). Several methods have been developed for its assessment, including multiparametric flow cytometry. To improve the sensitivity of this technique, we focus on the follow-up of chemoresistant leukemic subpopulations: leukemic stem cells (LSC).

Impressively, an unsupervised multivariate analysis of the data allows to calculate the MRD of all patients with AML with an available diagnostic immunophenotype. Moreover, our results show the prognostic impact of LSC MRD, while highlighting its complementary contribution to the value of bulk MRD.

These results demonstrate the usefulness of combining MRD approaches, using the phenotypic characteristics of different subpopulations of leukemic cells, approaches that could be modulated according to the patients’ treatments.

Acute myeloid leukemias (AML) are clonal diseases that develop from self-renewing leukemic stem cells (LSC). LSCs, enriched within the CD34+CD38− subpopulation (1), represent a tiny minority of all leukemic cells at diagnosis (2). LSCs are more resistant than the overall leukemic population to intensive chemotherapy (i.e., anthracyclines and cytarabine; refs. 3–5). Furthermore, LSCs levels have been shown to carry an independent and strong prognostic impact on both the initial response to induction chemotherapy and overall survival (OS), demonstrating the clinical relevance of stem cell abundance in AML (6, 7). This has been reinforced by studies showing that relapsing patients have an enriched population of LSCs (8) with an increase in LSC-related gene signatures (9).

The detection of measurable residual disease (MRD) is becoming increasingly important in therapeutic adaptation. Several methods have been developed for its evaluation, including multiparametric flow cytometry (MFC). It has been widely demonstrated that MFC is a mandatory diagnostic tool and is very efficient for the follow-up of any type of AML. The "different from normal (DfN)" and "leukemia-associated immunophenotype (LAIP)" analysis strategies are complementary (10). Unfortunately, exhaustive identification of LAIPs may require a large number of marker combinations, which could significantly increase the cost of the method and complicate both analysis and comparison with reference bone marrow (RBM) counterparts. In 2018, the European LeukemiaNet (ELN) has published recommendations, including a combined approach incorporating "DfN" and "LAIP" (10). Simultaneously, the ELN expert panel mentioned that the use of new software, based on unsupervised analysis, should be considered to reduce the subjectivity of interpretation although only few studies addressed this issue (11). Among these methods, the unsupervised FlowSOM methodology seems to be the most suitable for routine work and especially for the analysis of MRD in AML. It combines short analysis time, high sensitivity (12) and complies with ELN recommendations by combining "DfN" and "LAIP" approaches.

In its latest paper (13), ELN recommends an initial MFC MRD after two induction cycles. Among the many articles on which these recommendations are based, few have investigated the early response after a single cycle and even fewer have done so with an LSC analysis. We therefore decided to conduct this study in such a way as to provide a new perspective.

In this work, we evaluated the performance of unsupervised clustering for the post-induction assessment of bulk and LSC MRD in 155 patients with AML who received intensive conventional chemotherapy treatment.

Patients

Between January 1, 2017, and December 31, 2020, 1,044 patients with AML (>18 years, excluding promyelocytic AML) were included in the DATAML registry (authorization #915285). Immunophenotyping at diagnosis was available for 701 patients from Toulouse University Hospital (median age, 62 years) of whom 336 received a standard intensive chemotherapy regimen combining 3 days of daunorubicin or 5 days of idarubicin and 7 days of cytarabine. Treatments have been described in detail elsewhere (14) and in some situations, a third drug may be added such as midostaurin, lomustine, etc. Finally, 155 patients underwent bone marrow (BM) MRD assessment by MFC after hematologic recovery from induction chemotherapy, between day 35 and day 42 (Supplementary Table S1 and Supplementary Fig. S1). AML patient samples were obtained after written informed consent in accordance with the Declaration of Helsinki. BM samples were stored in the HIMIP collection (BB-0033–00060). In accordance with French law, the HIMIP collection was declared to the Ministry of Higher Education and Research (DC 2008–307 collection 1) and obtained a transfer agreement (AC 2008–129) after approval by the Comité de Protection des Personnes Sud-Ouest et Outremer II (institutional review board).

Cytometry

MFC analysis of leukemic cells at diagnosis and post-induction was performed on fresh BM samples collected on EDTA-K. Two antibody combinations were used. One combination, for leukemic bulk analysis, associated CD34, CD13, CD38, CD7, CD33, CD56, CD117, HLA-DR, and CD45 as recommended by the ELN (13) and experimented by Tettero and colleagues (15) The second combination comprised CD34, CD38, CD133, CD135, CD45, CD45RA for CD34+ subpopulation analysis (Supplementary Table S2). Samples were processed on Navios instruments (Beckman-Coulter, Miami, FL). In parallel, 10 RBM samples were obtained and analyzed by MFC on the same instrument with the same antibody combinations to construct unsupervised immunophenotypic model patterns as reported (16).

FlowSOM process

For each antibody combination, compensations were checked and fluorescence intensities of each marker were normalized for each file. All CD45+ BM cells of RBM were used to construct the model pattern used for bulk MRD and only CD34+ BM cells were used to construct the model pattern for LSC MRD. The normalized RBM files were then merged and processed to obtain two 64-nodes FlowSOM minimal spanning tree (MST; ref. 16): a first one containing all CD45+ BM cells and a second one focused on CD34+ progenitors cells. For each patient, the merged RBM file was then processed together with the normalized diagnosis and post-induction follow-up samples in the FlowSOM module (Bioconductor version 3.3.2 with flowsom and flocore packages) integrated with the Kaluza analysis software (Beckman Coulter) as reported (16). The MST (nodes) of the 3 files (RBM, diagnosis and follow-up) were redistributed in an unsupervised manner, according to the set of immunophenotypic characteristics of the merged samples.

Specific linked gates were defined to track each unique node concomitantly in the RBM, diagnosis and follow-up files. When this tracking gate was moved from one node to another, the respective statistics of the mean fluorescence intensity for each marker and the number of cells in the considered node were displayed for the RBM, diagnosis and follow-up BM. This allowed the detection of MRD nodes with the "DfN" and "LAIP" approach in an unsupervised manner. Specifically, for the search for LSCs, nodes of interest were selected from CD34+CD38− cells included in the CD34+ population after excluding CD34+CD38+ cells. MRD was considered positive if the percentage of cells in a given node of the follow-up sample was at least twice that of the same node in RBM (11) and if the sum of those nodes was greater than 0.1% of the analyzed cells, as recommended by the ELN (10). For the tube based on the LSC search, because no recommendation exists, we chose a patient-dependent threshold representing twice the value found in the RBM files.

Molecular analysis

The presence of FLT3-ITD was tested as described (17). Electrophoregram peaks were quantified using GeneMarker 2.2 (SoftGenetics, State College, PA). Extended DNA resequencing was performed using Illumina NextSeq500 and Haloplex HS (Agilent, Santa Clara, CA) targeted on the complete coding regions of 52 genes: ASXL1, ASXL2, ATM, BCOR, BCORL1, CBL, CCND2, CEBPA, CSF3R, CUX1, DDX41, DHX15, DNMT3A, EP300, ETV6, EZH2, FLT3, GATA1, GATA2, IDH1, IDH2, JAK2, KDM5A, KDM6A, KIT, KMT2D, KRAS, MGA, MPL, MYC, NF1, NPM1, NRAS, PHF6, PIGA, PPM1D, PRPF8, PTPN11, RAD21, RUNX1, SETBP1, SF3B1, SMC1A, SMC3, SRSF2, STAG2, TET2, TP53, U2AF1, WT1, ZBTB7A, and ZRSR2. Data were processed through two algorithms from GATK (https://software.broadinstitute.org/gatk), HaplotypeCaller (scaling accurate genetic variant discovery) to tens of thousands of samples (18, 19). The mean depth was 2,190 reads. Identified variants were curated manually and named according to the rules of the Human Genome Variation Society (hgvs.org). Molecular data are stored in the European Nucleotide Archive (https://www.ebi.ac.uk/ena/).

Statistical analysis

Complete remission, relapse, leukemia-free survival (LFS) and OS were defined according to standard ELN 2017 criteria (20).

The follow-up of patients in complete remission were censored at the time of last contact. Risk groups for prognosis were evaluated for OS and LFS by univariate analysis (log-rank test) and by a multivariate model of Cox regression. All calculations were performed using STATA version 13 software (STATA Corp., College Station, TX), all graphs were drawn using Graph Pad Prism software (San Diego, CA).

Comparisons were performed using a Mann–Whitney test for continuous variables and Fisher exact test for categorical variables with GraphPad Prism. Statistical test results are graphically expressed: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Data Availability

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Unsupervised analysis of Bulk and LSC post-induction MRD

In order to evaluate post-induction MRD levels in the 155 patients with AML treated with intensive chemotherapy, we applied an unsupervised clustering method (FlowSOM, Supplementary Fig. S2A) to RBM samples of healthy donors and cells from diagnosis and follow-up. Using this method, we highlighted a small number of nodes, specific to leukemic bulk, at diagnosis with the myeloid panel (CD34, CD13, CD38, CD7, CD33, CD56, CD117, HLADR, and CD45) and specific to CD34+CD38− LSCs with the LSC panel (CD34, CD38, CD133, CD135, CD45, CD45RA). After applying the node tree to the RBM and to the follow-up point, we could measure the node size and define the unique sensitivity threshold of the technique for each patient (on the RBM) and the MRD value (on the follow-up point).

The median size of bulk AML nodes was 20.6% of white blood cells at diagnosis [interquartile range (IQR), 11.9%–45.7%; Supplementary Fig. S2B]. The sensitivity threshold calculated in RBM was 0.08% (IQR, 0.05%–0.19%; Supplementary Fig. S2B). The levels of bulk AML MRD were very heterogenous with a median of 0.10% (IQR, 0.02%–0.48%; Supplementary Fig. S2B). Post-induction fold reduction of bulk leukemic cells was 206 (IQR, 43–1,197; Supplementary Fig. S2C), i.e., 2.31 log reduction, with 58 patients (37%) having a positive MRD above the ELN-defined threshold of 0.1% (10).

Stem cell nodes were specifically defined on the CD34+CD38− subpopulation using the LSC panel. Accordingly, LSC nodes comprised 69.4% of total CD34+CD38− leukemic cells at diagnosis (IQR, 48.3%–85.3%; Supplementary Fig. S2D) versus 1.3% in RBM samples (IQR, 0.5%–2.7%), validating the LSC clustering strategy. To note, we were able to identify CD34+CD38− LSC populations DfN in all AML even those with low CD34 expression (CD34 < 20%; Supplementary Fig. S3). The rate of post-induction residual LSCs was highly heterogeneous with a median of 2.42% of CD34+CD38− cells (IQR, 0.43%–9.91%) and a fold reduction of 27 (IQR, 6–157; Supplementary Fig. S2C), i.e., 1.43 log reduction. After induction, 85 patients (55%) still had detectable LSC above the technical threshold.

Prognostic impact of post-induction MRD

The median follow-up of patients was 26.2 months. The median OS for patients with a positive bulk MRD was 15.2 months and was not reached for negative patients (HR, 3.61; P < 0.0001; Fig. 1A). One-year (y) and 3-y OS were respectively 57% and 38% in bulk MRD-positive patients and 92% and 74% in negative patients. The median OS of patients with a positive LSC MRD was 25.0 months and not reached for negative patients (HR, 3.05; P = 0.0001; Fig. 1B). One-y and 3-y OS were respectively 73% and 46% in LSC MRD-positive patients and 85% and 77% in negative patients. In multivariate analysis, a positive bulk MRD was significantly associated with a worse OS [HR, 2.64; 95% confidence interval (CI), 1.44–4.84; P = 0.002; Supplementary Table S3]. AML de novo status retained a significant impact on OS (HR, 0.44; P = 0.011).

Figure 1.

Estimates of survival end points. A, Kaplan–Meier curves for OS according to the result of bulk MRD (follow-up nodes < 0.1%, blue line; ≥ 0.1%, red line). B, Kaplan–Meier curves for OS according to the result of LSC MRD (follow-up nodes < twice the threshold set by normal samples, blue line; ≥ twice the threshold set by normal samples, red line). C, Kaplan–Meier curves for LFS according to the result of bulk MRD (follow-up nodes < 0.1%, blue line; ≥ 0.1%, red line). D, Kaplan–Meier curves for LFS according to the result of LSC MRD. A–D, Log-rank tests. HR, hazard ratio; 95% CI, 95% confidence interval.

Figure 1.

Estimates of survival end points. A, Kaplan–Meier curves for OS according to the result of bulk MRD (follow-up nodes < 0.1%, blue line; ≥ 0.1%, red line). B, Kaplan–Meier curves for OS according to the result of LSC MRD (follow-up nodes < twice the threshold set by normal samples, blue line; ≥ twice the threshold set by normal samples, red line). C, Kaplan–Meier curves for LFS according to the result of bulk MRD (follow-up nodes < 0.1%, blue line; ≥ 0.1%, red line). D, Kaplan–Meier curves for LFS according to the result of LSC MRD. A–D, Log-rank tests. HR, hazard ratio; 95% CI, 95% confidence interval.

Close modal

The median LFS for bulk MRD-positive patients was 7.2 months and was not reached for bulk MRD-negative patients (HR, 2.80; P < 0.0001; Fig. 1C). One-y and 3-y LFS were respectively 42% and 16% in bulk MRD-positive patients and 71% and 50% in negative patients. Median LFS was respectively 12 months and not reached for LSC MRD-positive and negative patients (HR, 2.49; P = 0.0002; Fig. 1D). One-y and 3-y LFS were respectively 51% and 24% in LSC MRD-positive patients and 74% and 55% in negative patients. In multivariate analysis, positive bulk MRD (HR, 2.91; 95% CI, 1.66–5.10; P < 0.001) and positive LSC MRD (HR, 1.9; 95% CI, 1.09–3.30; P = 0.023) were both independently associated with a shorter LFS (Supplementary Table S4).

Combined analysis of Bulk and LSC MRD

To better describe the prognostic impact of AML MRD, bulk and LSC MRD data were associated to define 3 groups of patients based to their early response to chemotherapy: (i) chemosensitive patients with both negative Bulk and LSC MRD (Bulk-LSC−; n = 57, 37%), (ii) patients with AML with negative bulk MRD but still positive LSC MRD (Bulk-LSC+; n = 40, 26%), and (iii) chemoresistant patients with positive bulk MRD (Bulk+; n = 58, 37%). LSC phenotypic profile of those 3 groups were highly similar (Supplementary Fig. S4).

Median OS were 15.2 months in Bulk+, 38.4 months in Bulk-LSC+ and not reached in Bulk-LSC− (Fig. 2A). One-y and 3-y OS were 57% and 38% in Bulk+, 90% and 57% in Bulk-LSC+ and 95% and 86% in Bulk-LSC−. In multivariate analysis, Bulk+ (HR, 6.09; 95% CI, 2.45–15.1; P < 0.001) was associated with a worse OS (Supplementary Table S5). Median LFS were 7.1 months in Bulk+, 13.2 months in Bulk-LSC+ and not reached in Bulk-LSC− (Fig. 2B). In multivariate analysis, Bulk+ (HR, 5.73; 95% CI, 2.84–11.5; P < 0.001) and Bulk-LSC+ (HR, 2.78; 95% CI, 1.37–5.65; P = 0.005) were associated with a shorter LFS (Supplementary Table S6).

Figure 2.

Estimates of survival end points for combined analysis of MRD. A, Kaplan–Meier curves for OS according to the combined result of bulk and LSC MRD (bulk follow-up nodes < 0.1% and LSC follow-up nodes < twice the threshold set by normal samples, blue line; < 0.1% and ≥ twice the threshold set by normal samples, pink line; bulk MRD ≥ 0.1% regardless of the result of LSC MRD, red line). B, Kaplan–Meier curves for LFS according to the combined result of Bulk and LSC MRD. A and B, log-rank tests.

Figure 2.

Estimates of survival end points for combined analysis of MRD. A, Kaplan–Meier curves for OS according to the combined result of bulk and LSC MRD (bulk follow-up nodes < 0.1% and LSC follow-up nodes < twice the threshold set by normal samples, blue line; < 0.1% and ≥ twice the threshold set by normal samples, pink line; bulk MRD ≥ 0.1% regardless of the result of LSC MRD, red line). B, Kaplan–Meier curves for LFS according to the combined result of Bulk and LSC MRD. A and B, log-rank tests.

Close modal

Of the 155 patients, 67 (43.2%) were transplanted, including 24, 19 and 24 with Bulk-LSC−, Bulk-LSC+ and Bulk+ MRD, corresponding to 42.1%, 47.5% and 41.4% of each group, respectively (Supplementary Table S1). In a landmark analysis at 6 months from diagnosis, when 142 (91.6%) patients were still alive, the OS of transplant recipients were similar in Bulk-LSC− and Bulk-LSC+ MRD patients (Supplementary Fig. S5A and S5B); in contrast, Bulk+ MRD transplant recipients had significantly longer OS (1y OS 79.2%) than that of non-transplanted Bulk+ MRD patients (57.2%, P = 0.011; Supplementary Fig. S5C).

As there was no significant interaction in our multivariate analysis, between Bulk and LSC MRD and age, cytogenetics, NPM1 mutation or FLT3-ITD for each end point, we concluded that the prognostic significance of MRD was independent of these pretherapy prognostic parameters.

Characteristics of patients with chemoresistant and chemosensitive AML

Although the overall immunophenotypic profiles of AML blasts, depending on MRD status, were similar (Supplementary Fig. S6A), some specific markers at diagnostic showed significant, even if moderate, differences (Supplementary Fig. S6B). In our cohort, markers such as CD36 and CD11b (which are part of the markers we study at diagnosis but which we do not use in our follow-up method) have a significantly higher expression in Bulk+ MRD patients, as well as CD16 and CD56. Conversely, CD13 expression was significantly decreased in this group of patients.

To explore cytogenetic and molecular differences between chemoresistant (Bulk+) and chemosensitive (Bulk-LSC−) patients, we focused on 72 patients of the cohort (47%) for whom genetic and next-generation sequencing mutational data on 46 genes were available. The distribution of abnormalities in these patients was representative of what is known in AML (Fig. 3A). Abnormalities were grouped into 7 functional classes defined by The Cancer Genome Atlas (TCGA; ref. 21). As expected, NPM1 mutations were positively correlated to the chemosensitive Bulk-LSC− group (OR, 3.3; 95% CI, 1.2–9.5; Fig. 3B).

Figure 3.

Mutational and genetic abnormalities. A, Distribution of recurrent genetic abnormalities in 72 patients of the cohort related to the results of combined bulk and LSC MRD. B, Forest plot according to the 7 functional classes described by TCGA and associated with combined MRD status. OR not significant, gray; OR significant, colored according to MRD group.

Figure 3.

Mutational and genetic abnormalities. A, Distribution of recurrent genetic abnormalities in 72 patients of the cohort related to the results of combined bulk and LSC MRD. B, Forest plot according to the 7 functional classes described by TCGA and associated with combined MRD status. OR not significant, gray; OR significant, colored according to MRD group.

Close modal

MRD in specific AML subgroups

Although our cohort size was relatively small, we wanted to study the impact of Bulk and LSC MRD in specific subgroups of AML. Indeed, the recent increase in therapeutic choices offers new options to clinicians and early MRD could be an aid to therapeutic decisions.

FLT3-ITD, IDH1/2 mutations, NPM1 mutation associated with wild-type FLT3 and adverse cytogenetics were detected in 38, 40, 28, and 36 patients, respectively. In these groups, 3-y OS and 3-y LFS ranged between 21% to 57% and 0% to 11% in Bulk+ and between 75% to 91% and 56% to 75% in Bulk-LSC−, respectively (Fig. 4; Table 1). The number of patients did not allow us to interpret the prognostic impact of Bulk-LSC+ MRD. Finally, to explore those clinically relevant AML groups, we compared the phenotypic characteristics of LSC. It appeared that LSC from NPM1 m AMLs were low expressers of CD45RA, CD135, and CD133 compared with LSC from AML with adverse cytogenetics (Supplementary Fig. S7).

Figure 4.

Kaplan–Meier representation of estimates of survival endpoints for combined analysis of MRD in patients with mutated FLT3-ITD (A and B) or IDH1/2 (C and D) or NPM1 with FLT3wt (E and F) or adverse cytogenetic (G and H). OS (A, C, E, G) and LFS (B, D, F, H), log-rank tests.

Figure 4.

Kaplan–Meier representation of estimates of survival endpoints for combined analysis of MRD in patients with mutated FLT3-ITD (A and B) or IDH1/2 (C and D) or NPM1 with FLT3wt (E and F) or adverse cytogenetic (G and H). OS (A, C, E, G) and LFS (B, D, F, H), log-rank tests.

Close modal
Table 1.

OS and LFS in clinically relevant AML subgroups according to MRD status.

1-y OS3-y OSP1-y LFS3-y LFSP
AML subgroupMRD statusPatients n (%)Median OS (months)%%vs. othersMedian LFS (months)%%vs. others
FLT3-ITD Bulk-LSC− 15 (39) NR 87 87 0.0077 NR 79 63 0.0073 
 Bulk-LSC+ 9 (24) NR 89 78 0.44 20.2 56 42 0.99 
 Bulk+ 14 (37) 10.9 35 21 0.0001 3.6 11 11 0.0002 
IDH1/2m Bulk-LSC− 18 (45) NR 100 91 0.012 32.8 75 56 0.006 
 Bulk-LSC+ 10 (25) NR 80 60 0.27 12.3 60 40 0.78 
 Bulk+ 12 (30) NR 67 57 0.075 6.2 30 10 0.001 
NPM1m FLT3wt Bulk-LSC− 13 (46) NR 92 81 0.21 NR 67 67 0.046 
 Bulk-LSC+ 7 (25) NR 100 86 0.34 NR 71 57 0.66 
 Bulk+ 8 (29) 12.4 56 44 0.015 4.4 14 <0.0001 
Adverse cytogenetics Bulk-LSC− 6 (17) NR 100 75 0.041 NR 100 75 0.032 
 Bulk-LSC+ 9 (25) NR 89 56 0.52 16.8 63 25 0.59 
 Bulk+ 21 (58) 10.5 43 25 0.02 10 45 0.12 
1-y OS3-y OSP1-y LFS3-y LFSP
AML subgroupMRD statusPatients n (%)Median OS (months)%%vs. othersMedian LFS (months)%%vs. others
FLT3-ITD Bulk-LSC− 15 (39) NR 87 87 0.0077 NR 79 63 0.0073 
 Bulk-LSC+ 9 (24) NR 89 78 0.44 20.2 56 42 0.99 
 Bulk+ 14 (37) 10.9 35 21 0.0001 3.6 11 11 0.0002 
IDH1/2m Bulk-LSC− 18 (45) NR 100 91 0.012 32.8 75 56 0.006 
 Bulk-LSC+ 10 (25) NR 80 60 0.27 12.3 60 40 0.78 
 Bulk+ 12 (30) NR 67 57 0.075 6.2 30 10 0.001 
NPM1m FLT3wt Bulk-LSC− 13 (46) NR 92 81 0.21 NR 67 67 0.046 
 Bulk-LSC+ 7 (25) NR 100 86 0.34 NR 71 57 0.66 
 Bulk+ 8 (29) 12.4 56 44 0.015 4.4 14 <0.0001 
Adverse cytogenetics Bulk-LSC− 6 (17) NR 100 75 0.041 NR 100 75 0.032 
 Bulk-LSC+ 9 (25) NR 89 56 0.52 16.8 63 25 0.59 
 Bulk+ 21 (58) 10.5 43 25 0.02 10 45 0.12 

Abbreviation: NR, not reached.

In this study, we confirm the prognostic impact of post-induction MFC Bulk MRD in patients with AML treated with intensive chemotherapy, complementing previous studies (22–27). We wanted to evaluate MRD in this early therapeutic point, on one hand to homogenize our results and on the other hand because the different publications on the subject were performed after two cycles of chemotherapy or without a specific follow-up point (24, 28). Of note, analysis of MRD panels with the unsupervised multivariate clustering FlowSOM method identified LAIP/DfN clusters in 100% of patients with AML in this cohort. This artificial intelligence-based strategy greatly facilitated the homogenization of the results (data normalization phase, follow-up over time with the reference diagnosis, computerized assistance for the choice of LAIP/DfN). The speed of the analysis is also a determining factor for its applicability in an MFC platform. The power of clustering relies largely on the choice of markers. For example, we showed that CD13 and CD56 were differentially expressed between AML with negative and positive post-induction MRD. Although these markers are part of the classical LAIP-DfN profiles and thus influenced our ability to detect residual cells, they seemed to be related to the intrinsic chemoresistance of leukemic cells. Indeed, CD56 expression was associated with shorter survival of patients with AML and a low probability of achieving complete remission (29). CD56 plays a critical role in the regulation of cell survival, drug resistance, and self-renewal of AML blasts in connection with activation of the MAPK or glycolysis signaling pathway (30).

Nevertheless, a sizable group of Bulk- MRD patients still relapsed with only 69% LFS after 1 year. One reason for this disappointing result could be that the threshold for MRD positivity was set too high (0.1% as recommended by ELN; ref. 10). Unfortunately, this threshold is mainly dictated by the choice of flow cytometry markers and normal hematopoietic subpopulations present in the BM. To circumvent this problem, we focused on LSCs, a well-known chemoresistant subpopulation (31, 32). Using this strategy, we isolated a group of patients with negative Bulk MRD but positive LSC MRD (25.8% of this cohort). As expected, these patients had intermediate survival between Bulk+ MRD and Bulk-LSC− MRD patients. Similarly, our study of immunophenotypic markers in AML highlighted that CD36 expression correlated with a poorer response to chemotherapy. This membrane fatty acid receptor is of great interest in AML. CD36 is indeed involved in lipid metabolism and chemoresistance of LSCs. Indeed, according to Ye and colleagues (33), CD36+ LSCs secrete inflammatory cytokines that enable lipolysis of adipocytes. The release of fatty acids and their recovery by CD36 allows their use by LSCs as an additional energy source that is normally scarce in the marrow.

This first study combining Bulk and LSC MFC MRD and unsupervised analysis strategy has some weaknesses. Indeed, it is a monocentric study with a relatively small cohort size and a short follow-up time. But despite this, we observe survival results comparable with the study of Zeijlemaker and colleagues (28) who carried out their MRD analyses after two cycles of chemotherapy. In addition, our study allows interesting observations on groups of patients classified by genetic or molecular abnormalities and suggests an additional prognostic value of post-induction MRD. Indeed, we can observe that in patients with adverse cytogenetics, OS and LFS were significantly different between Bulk-LSC− and Bulk-LSC+ MRD patients. Astonishingly, in the Bulk-LSC− MRD patients relatively small population, no deaths or relapses were reported, whereas Bulk-LSC+ and Bulk+ patients had a poor prognosis. Similarly, in the group of good prognosis NPM1 mutated and FLT3wt, a Bulk+ MRD was predictive of a relapse within 14 months. Of course, all these results will have to be confirmed in a larger cohort of patients and could be extended to more genetics subgroups. Indeed, it would be interesting to study whether LSCs show differences in phenotypic expression according to genetic subgroups.

The design of our LSC panel can certainly be improved in the future. First, we chose to focus on markers such as CD45RA and CD133 that were already familiar in our laboratory. Some markers like CD96 (34), CLL1 (35), TIM3 (36), CD36 (33, 37), CD39 (38), or CALCRL (39) could improve the sensitivity and specificity of this follow-up method, but we could not add them because the calculation of our MRD depends on the markers used for diagnosis. Then the threshold can also be questioned. In the absence of a specific study on LSC MRD by unsupervised multivariate method, we chose to keep the validated threshold in Bulk MRD, i.e., a threshold representing twice the value found in the same RBM node (11).

One of the major advantages of MFC remains the speed of the technique. FlowSOM's unsupervised clustering analysis significantly improves the speed of analysis and the rendering of results. In a period where OS and relapse-free survival is increasing thanks to therapeutic improvements, it becomes essential to have a quick and reliable answer on the effectiveness of a treatment. We believe that early assessment MRD in MFC may have a role to play in future clinical trials, as it does in other hematologic diseases, and even more with unsupervised analysis.

In the future era of personalized medicine, we can envision an important place for MRD by MFC combined with unsupervised clustering software and a chemoresistant cell monitoring strategy as done here with the LSC. The use of new markers for cohorts of patients treated with new therapies will certainly require the adaptation of CMF MRD analysis tools and will need many additional studies.

S. Bertoli reports personal fees from AbbVie, Astellas, BMS-Celgene, Novartis, Pfizer, and Jazz Pharmaceuticals outside the submitted work. S. Tavitian reports other support from Novartis, Servier, and Pfizer outside the submitted work. F. Huguet reports personal fees from Novartis, Incyte, Pfizer, and Servier outside the submitted work. P. Bories reports honoraria from Kite-Gilead and AbbVie on topics unrelated to the present manuscript. C. Récher reports grants, personal fees, and nonfinancial support from AbbVie, Astellas, BMS, and Jazz Pharmaceuticals; grants from Amgen, Iqvia, and MaatPharma; personal fees and nonfinancial support from Novartis and Servier; and personal fees from Takeda outside the submitted work. F. Vergez reports grants from Pierre Fabre and nonfinancial support from Sysmex outside the submitted work. No disclosures were reported by the other authors.

A. Canali: Conceptualization, data curation, software, formal analysis, validation, investigation, writing–original draft, writing–review and editing. I. Vergnolle: Formal analysis, validation, investigation. S. Bertoli: Investigation. L. Largeaud: Investigation. M.-L. Nicolau: Investigation. J.-B. Rieu: Investigation. S. Tavitian: Investigation. F. Huguet: Investigation. M. Picard: Investigation. P. Bories: Investigation. J.P. Vial: Software, investigation. N. Lechevalier: Software. M.C. Béné: Software, writing–review and editing. I. Luquet: Investigation. V. Mansat-De Mas: Investigation. E. Delabesse: Investigation. C. Récher: Conceptualization, investigation, writing–review and editing. F. Vergez: Conceptualization, data curation, software, formal analysis, validation, investigation, writing–original draft, writing–review and editing.

We would like to thank the technicians of the hematology flow cytometry laboratory at the Toulouse Oncopole University Cancer Institute for processing all the samples used for this work.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

1.
Sarry
J-E
,
Murphy
K
,
Perry
R
,
Sanchez
PV
,
Secreto
A
,
Keefer
C
, et al
.
Human acute myelogenous leukemia stem cells are rare and heterogeneous when assayed in NOD/SCID/IL2Rγc-deficient mice
.
J Clin Invest
2011
;
121
:
384
95
.
2.
Dick
JE
.
Stem cell concepts renew cancer research
.
Blood
2008
;
112
:
4793
807
.
3.
Costello
RT
,
Mallet
F
,
Gaugler
B
,
Sainty
D
,
Arnoulet
C
,
Gastaut
JA
, et al
.
Human acute myeloid leukemia CD34+/CD38− progenitor cells have decreased sensitivity to chemotherapy and Fas-induced apoptosis, reduced immunogenicity, and impaired dendritic cell transformation capacities
.
Cancer Res
2000
;
60
:
4403
11
.
4.
Ishikawa
F
,
Yoshida
S
,
Saito
Y
,
Hijikata
A
,
Kitamura
H
,
Tanaka
S
, et al
.
Chemotherapy-resistant human AML stem cells home to and engraft within the bone-marrow endosteal region
.
Nat Biotechnol
2007
;
25
:
1315
21
.
5.
Thomas
D
,
Powell
JA
,
Vergez
F
,
Segal
DH
,
Nguyen
N-YN
,
Baker
A
, et al
.
Targeting acute myeloid leukemia by dual inhibition of PI3K signaling and Cdk9-mediated Mcl-1 transcription
.
Blood
2013
;
122
:
738
48
.
6.
Vergez
F
,
Green
AS
,
Tamburini
J
,
Sarry
J-E
,
Gaillard
B
,
Cornillet-Lefebvre
P
, et al
.
High levels of CD34+CD38low/-CD123+ blasts are predictive of an adverse outcome in acute myeloid leukemia: a Groupe Ouest-Est des Leucemies Aigues et Maladies du Sang (GOELAMS) study
.
Haematologica
2011
;
96
:
1792
8
.
7.
Vergez
F
,
Nicolau-Travers
M-L
,
Bertoli
S
,
Rieu
J-B
,
Tavitian
S
,
Bories
P
, et al
.
CD34+CD38−CD123+ leukemic stem cell frequency predicts outcome in older acute myeloid leukemia patients treated by intensive chemotherapy but not hypomethylating agents
.
Cancers
2020
;
12
:
1174
.
8.
Ho
GJ
,
Liew
SM
,
Ng
CJ
,
Shunmugam
HR
,
Glasziou
P
.
Development of a search strategy for an evidence based retrieval service
.
PLoS One
2016
;
11
:
e0167170
.
9.
Hackl
H
,
Steinleitner
K
,
Lind
K
,
Hofer
S
,
Tosic
N
,
Pavlovic
S
, et al
.
A gene expression profile associated with relapse of cytogenetically normal acute myeloid leukemia is enriched for leukemia stem cell genes
.
Leuk Lymphoma
2015
;
56
:
1126
8
.
10.
Schuurhuis
GJ
,
Heuser
M
,
Freeman
S
,
Béné
M-C
,
Buccisano
F
,
Cloos
J
, et al
.
Minimal/measurable residual disease in AML: a consensus document from the European LeukemiaNet MRD Working Party
.
Blood
2018
;
131
:
1275
91
.
11.
Vial
JP
,
Lechevalier
N
,
Lacombe
F
,
Dumas
P-Y
,
Bidet
A
,
Leguay
T
, et al
.
Unsupervised flow cytometry analysis allows for an accurate identification of minimal residual disease assessment in acute myeloid leukemia
.
Cancers
2021
;
13
:
629
.
12.
Van Gassen
S
,
Callebaut
B
,
Van Helden
MJ
,
Lambrecht
BN
,
Demeester
P
,
Dhaene
T
, et al
.
FlowSOM: Using self-organizing maps for visualization and interpretation of cytometry data: FlowSOM
.
Cytometry
2015
;
87
:
636
45
.
13.
Heuser
M
,
Freeman
SD
,
Ossenkoppele
GJ
,
Buccisano
F
,
Hourigan
CS
,
Ngai
LL
, et al
.
2021 Update on MRD in acute myeloid leukemia: a consensus document from the European LeukemiaNet MRD Working Party
.
Blood
2021
;
138
:
2753
67
.
14.
Bertoli
S
,
Tavitian
S
,
Huynh
A
,
Borel
C
,
Guenounou
S
,
Luquet
I
, et al
.
Improved outcome for AML patients over the years 2000–2014
.
Blood Cancer J
2017
;
7
:
635
.
15.
Tettero
JM
,
Freeman
S
,
Buecklein
V
,
Venditti
A
,
Maurillo
L
,
Kern
W
, et al
.
Technical aspects of flow cytometry-based measurable residual disease quantification in acute myeloid leukemia: experience of the European LeukemiaNet MRD Working Party
.
HemaSphere
2022
;
6
:
e676
.
16.
Lacombe
F
,
Lechevalier
N
,
Vial
JP
,
Béné
MC
.
An R-derived FlowSOM process to analyze unsupervised clustering of normal and malignant human bone marrow classical flow cytometry data
.
Cytometry A
2019
;
95
:
1191
7
.
17.
Largeaud
L
,
Cornillet-Lefebvre
P
,
Hamel
JF
,
Dumas
P-Y
,
Prade
N
,
Dufrechou
S
, et al
.
Lomustine is beneficial to older AML with ELN2017 adverse risk profile and intermediate karyotype: a FILO study
.
Leukemia
2021
;
35
:
1291
300
.
18.
Poplin
R
, et al
.
Scaling accurate genetic variant discovery to tens of thousands of samples
.
bioRxiv
2017
Available from: http://biorxiv.org/lookup/doi/10.1101/201178.
19.
McKenna
A
,
Hanna
M
,
Banks
E
,
Sivachenko
A
,
Cibulskis
K
,
Kernytsky
A
, et al
.
The Genome Analysis Toolkit: A MapReduce framework for analyzing next-generation DNA sequencing data
.
Genome Res
2010
;
20
:
1297
303
.
20.
Döhner
H
,
Estey
E
,
Grimwade
D
,
Amadori
S
,
Appelbaum
FR
,
Büchner
T
, et al
.
Diagnosis and management of AML in adults: 2017 ELN recommendations from an international expert panel
.
Blood
2017
;
129
:
424
47
.
21.
The Cancer Genome Atlas Research Network
.
Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia
.
N Engl J Med
2013
;
368
:
2059
74
.
22.
Ivey
A
,
Hills
RK
,
Simpson
MA
,
Jovanovic
JV
,
Gilkes
A
,
Grech
A
, et al
.
Assessment of minimal residual disease in standard-risk AML
.
N Engl J Med
2016
;
374
:
422
33
.
23.
Köhnke
T
,
Sauter
D
,
Ringel
K
,
Hoster
E
,
Laubender
RP
,
Hubmann
M
, et al
.
Early assessment of minimal residual disease in AML by flow cytometry during aplasia identifies patients at increased risk of relapse
.
Leukemia
2015
;
29
:
377
86
.
24.
Terwijn
M
,
van Putten
WLJ
,
Kelder
A
,
van der Velden
VHJ
,
Brooimans
RA
,
Pabst
T
, et al
.
High prognostic impact of flow cytometric minimal residual disease detection in acute myeloid leukemia: data from the HOVON/SAKK AML 42A study
.
JCO
2013
;
31
:
3889
97
.
25.
Sui
JN
,
Chen
QS
,
Zhang
YX
,
Sheng
Y
,
Wu
J
,
Li
JM
, et al
.
Identifying leukemia-associated immunophenotype-based individualized minimal residual disease in acute myeloid leukemia and its prognostic significance
.
Am J Hematol
2019
;
94
:
528
38
.
26.
Rossi
G
,
Giambra
V
,
Minervini
MM
,
De Waure
C
,
Mancinelli
S
,
Ciavarella
M
, et al
.
Leukemia-associated immunophenotypes subdivided in “categories of specificity” improve the sensitivity of minimal residual disease in predicting relapse in acute myeloid leukemia
.
Cytometry B Clin Cytom
98
:
216
25
(
2019
).
27.
Freeman
SD
,
Virgo
P
,
Couzens
S
,
Grimwade
D
,
Russell
N
,
Hills
RK
, et al
.
Prognostic relevance of treatment response measured by flow cytometric residual disease detection in older patients with acute myeloid leukemia
.
J Clin Oncol
2013
;
31
:
4123
31
.
28.
Zeijlemaker
W
,
Grob
T
,
Meijer
R
,
Hanekamp
D
,
Kelder
A¨L
,
Carbaat-Ham
JC
, et al
.
CD34+CD38− leukemic stem cell frequency to predict outcome in acute myeloid leukemia
.
Leukemia
2019
;
33
:
1102
12
.
29.
Raspadori
D
,
Damiani
D
,
Lenoci
M
,
Rondelli
D
,
Testoni
N
,
Nardi
G
, et al
.
CD56 antigenic expression in acute myeloid leukemia identifies patients with poor clinical prognosis
.
Leukemia
2001
;
15
:
1161
4
.
30.
Sasca
D
,
Yun
H
,
Giotopoulos
G
,
Szybinski
J
,
Evan
T
,
Wilson
NK
, et al
.
Cohesin-dependent regulation of gene expression during differentiation is lost in cohesin-mutated myeloid malignancies
.
Blood
2019
;
134
:
2195
208
.
31.
van Rhenen
A
,
Feller
N
,
Kelder
AL
,
Westra
AH
,
Rombouts
E
,
Zweegman
S
, et al
.
High stem cell frequency in acute myeloid leukemia at diagnosis predicts high minimal residual disease and poor survival
.
Clin Cancer Res
2005
;
11
:
6520
7
.
32.
van Rhenen
A
,
Moshaver
B
,
Kelder
A
,
Feller
N
,
Nieuwint
AWM
,
Zweegman
S
, et al
.
Aberrant marker expression patterns on the CD34+CD38− stem cell compartment in acute myeloid leukemia allows to distinguish the malignant from the normal stem cell compartment both at diagnosis and in remission
.
Leukemia
2007
;
21
:
1700
7
.
33.
Ye
H
,
Adane
B
,
Khan
N
,
Sullivan
T
,
Minhajuddin
M
,
Gasparetto
M
, et al
.
Leukemic stem cells evade chemotherapy by metabolic adaptation to an adipose tissue niche
.
Cell Stem Cell
2016
;
19
:
23
37
.
34.
Hosen
N
,
Park
CY
,
Tatsumi
N
,
Oji
Y
,
Sugiyama
H
,
Gramatzki
M
, et al
.
CD96 is a leukemic stem cell-specific marker in human acute myeloid leukemia
.
Proc Natl Acad Sci USA
2007
;
104
:
11008
13
.
35.
van Rhenen
A
,
van Dongen
GAMS
,
Kelder
A
,
Rombouts
EJ
,
Feller
N
,
Moshaver
B
, et al
.
The novel AML stem cell–associated antigen CLL-1 aids in discrimination between normal and leukemic stem cells
.
Blood
2007
;
110
:
2659
66
.
36.
Han
C
,
Jan
LY
,
Jan
Y-N
.
Enhancer-driven membrane markers for analysis of nonautonomous mechanisms reveal neuron–glia interactions in Drosophila
.
Proc Natl Acad Sci USA
2011
;
108
:
9673
8
.
37.
Farge
T
,
Saland
E
,
de Toni
F
,
Aroua
N
,
Hosseini
M
,
Perry
R
, et al
.
Chemotherapy-resistant human acute myeloid leukemia cells are not enriched for leukemic stem cells but require oxidative metabolism
.
Cancer Discov
2017
;
7
:
716
35
.
38.
Aroua
N
,
Boet
E
,
Ghisi
M
,
Nicolau-Travers
M-L
,
Saland
E
,
Gwilliam
R
, et al
.
Extracellular ATP and CD39 activate cAMP-mediated mitochondrial stress response to promote cytarabine resistance in acute myeloid leukemia
.
Cancer Discov
2020
;
10
:
1544
65
.
39.
Larrue
C
,
Guiraud
N
,
Mouchel
P-L
,
Dubois
M
,
Farge
T
,
Gotanègre
M
, et al
.
Adrenomedullin-CALCRL axis controls relapse-initiating drug tolerant acute myeloid leukemia cells
.
Nat Commun
2021
;
12
:
422
.

Supplementary data