IDH1/2 but not DNMT3A mutations are suitable targets for minimal residual disease monitoring in acute myeloid leukemia patients: a study by the Acute Leukemia French Association.

Acute myeloid leukemia (AML) is a heterogeneous disease. Even within the same NPM1-mutated genetic subgroup, some patients harbor additional mutations in FLT3, IDH1/2, DNMT3A or TET2. Recent studies have shown the prognostic significance of minimal residual disease (MRD) in AML but it remains to be determined which molecular markers are the most suitable for MRD monitoring. Recent advances in next-generation sequencing (NGS) have provided the opportunity to use multiple molecular markers. In this study, we used NGS technology to assess MRD in 31 AML patients enrolled in the ALFA-0701 trial and harboring NPM1 mutations associated to IDH1/2 or DNMT3A mutations. NPM1 mutation-based MRD monitoring was performed by RTqPCR. IDH1/2 and DNMT3A mutations were quantified by NGS using an Ion Torrent Proton instrument with high coverage (2 million reads per sample). The monitoringof IDH1/2 mutations showed that these mutations were reliable MRD markers that allowed the prediction of relapse in the majority of patients. Moreover, IDH1/2 mutation status predicted relapse or disease evolution in 100% of cases if we included the patient who developed myelodysplastic syndrome. In contrast, DNMT3A mutations were not correlated to the disease status, as we found that a preleukemic clone with DNMT3A mutation persisted in 40% of the patients who were in complete remission, reflecting the persistence of clonal hematopoiesis.


INTRODUCTION
Acute myeloid leukemia (AML) is a highly heterogeneous malignancy, especially in terms of the molecular and phenotypic characteristics. Heterogeneity is also observed within the same genetic subgroup of AML tumors. For example, within nucleophosmin 1 (NPM1) mutant subgroup, some patients have concomitant mutations in fms-like-tyrosine kinase 3 (FLT3), isocitrate dehydrogenase (IDH) 1 and 2, DNA methyltransferase 3A (DNMT3A) or ten-elven translocation 2 (TET2) genes. A multi-hit model of leukemogenesis, in which class I mutations confer the proliferation or survival advantages of blast cells and class II mutations block myeloid differentiation, has been observed in most cases of AML [1]. Recently, other studies have also reported the epigenetic effects of class III mutations on AML [2,3]. However, the exact roles of each alteration in leukemogenesis and the mechanisms of disease progression remain largely unkown, especially with respect to recent data on molecular intraclonal heterogeneity. Recent studies have shown that minimal residual disease (MRD) in AML patients, during or after treatment, has prognostic value [4][5][6][7][8][9]. However, there are many questions regarding the clinical assessment of MRD in AML patients. First, which of the potential molecular and/or cellular markers should be assessed? Second, what type of biological sample should be analyzed? Third, where should the sensitivity threshold be set, and what are the relevant time-points to consider for MRD assessment? One study found that IDH1/2 gene mutations persisted in patients who were in complete remission (CR), although other molecular markers were not analyzed at the time of AML diagnosis [10]. These mutations may be attributed to a preleukemic clone that acquires additional mutations promoting proliferation and differentiation block, which eventually results in leukemia. This preleukemic clone may be able to survive initial chemotherapy treatments. DNMT3A mutations, which occur in 20% of de novo AML cases, lead to abnormal DNA methylation patterns, which is likely to alter the expression of various target genes [11]. The prognostic impact of DNMT3A mutations seems to be unfavorable [12,13], but their applicability MRD monitoring remains unclear [14].
Recent technological advances in next-generation sequencing (NGS) have provided new opportunities for MRD monitoring in AML patients and the possibility to simultaneously analyze multiple biomarkers and to detect subclonal populations.
In this study, we used NGS technology to monitor MRD using IDH1/2 and DNMT3A mutations in a cohort of NPM1 mutated AML patients. Our objective was to evaluate the suitability of IDH1/2 and DNMT3A mutations as a target for MRD detection by NGS and to compare the data with NPM1 mutation-based MRD assessed by RTqPCR

RESULTS
Of the 31 NPM1 mutated AML patients, 8 patients harbored an IDH1 mutation, 9 an IDH2 mutation and 15 DNMT3A mutation. Sequencing data showed sufficient sequencing depth with a median of 2,012,459 reads for IDH1/2 (range: 102,657 to 5,160,118 reads) and a median of 966,298 reads for DNMT3A (range: 565,152 to 2,700,349 reads). This coverage allowed the detection of MRD with a sensitivity of approximately 0.001%. Despite such an extensive coverage, a median of 520 reads were positive for mutations in the negative controls, reflecting cross-contamination due to the multiple steps involved in the preparation of the gene libraries (i.e., in the intra-run steps, including preliminary PCR, barcode purity, and adaptors/barcodes ligation, and the inter-run steps, including OT2 and clonal amplification). Thus, the detection limit was 0.07% for IDH1/2 mutation analysis (0.002 -0.097, p < 0.01, Fisher's exact test) and 0.11% for DNMT3A mutation analysis (0.001-0.426, p < 0,01, Fisher's exact test).
MRD level was evaluated at the following time points: post induction (MRD1), post first consolidation course (MRD2) and post second consolidation courses (MRD3). The median clinical followup of the cohort was 673 days (range: 131-2637 days). NPM1 mutation and IDH1/2 mutation MRD levels and NPM1 mutation and DNMT3A mutation MRD levels were highly correlated (r = 0,68183, p < 0,0001; r = 0,55514, p < 0,0001, respectively). Of the 17 IDH1/2 mutationpositive patients, we found concordant MRD results between IDH1/2 and NPM1 mutation levels by RTqPCR in 13 cases. Four patients who relapsed were positive for both IDH1/2 and NPM1 mutations, and 9 patients who remained in complete remission were negative for both NPM1 and IDH1/2 mutations. In the 4 remaining patients, we observed a discrepancy between the NPM1 and IDH1/2 mutation levels. These patients presented with one or more MRD time-points with undetectable NPM1 mutation, whereas the IDH1/2 mutation levels ranged from 0.5% to 47% ( Table 1). Three of the patients relapsed after 504, 395 and 158 days, and all of these patients harbored similar NPM1 mutation levels at diagnosis ( Figure 1). The patient who did not relapse developed a NPM1negative myelodysplastic syndrome. In the 15 DNMT3A mutant, we found concordant MRD results between DNMT3A mutation rates and NPM1 mutation levels in 9 cases. Eight out of these 9 patients were positive for both NPM1 and DNMT3A mutations and relapsed, while the remaining case, who remained in persistent CR, was negative for both NPM1 and DNMT3A mutations. In the 6 discordant cases, NPM1 mutations were undetectable in most MRD time-points, whereas DNMT3A mutation levels ranged from 5% to 45% during MRD follow-up. All of these patients remained in first complete remission after a median follow-up of 4 years ( Table 2). DNMT3A mutations were detected at different time-points during follow-up (i.e., post-induction, post consolidation 1, and later), whereas other markers studied were undetectable ( Figure 2). These results are consistent with clonal heterogeneity in AML, particularly on the molecular level. Samples from 3 of these patients were further examined to investigate the origin of the discrepancy using cell sorting analysis.
In the 3 patients for which cell subpopulations were available, DNMT3A mutations were found in the whole peripheral blood and bone marrow collected at which time-point, but not in the DNA extracted from a skin biopsy. These findings suggest that DNMT3A mutations were somatically acquired ( Figure 3). The percentage of mutations in which gene in these patients were comparable in all the bone marrow cell subpopulations analyzed. In complete remission, all of the following cell subpopulations collected from peripheral blood (i.e., CD 56+ NK cells, CD 19+ B lymphocytes, CD 14+ monocytes, CD 66+ granulocytes, CD 34+ CD 45 -low blasts), and (i.e., CD 34+ CD 38-CD 123+ and aldehyde dehydrogenase (ALDH) intermediate leukemic stem cells (LSCs) or CD 34+ CD 38-CD 123+ and ALDH high hematopoietic stem cells (HSCs)) isolated from bone marrow harbored DNMT3A mutations but none of the molecular abnormalities (i.e., NPM1, IDH1/2 and FLT3-ITD mutations) identified at AML diagnosis. Interestingly, DNMT3A mutations were not found in CD 3+ T lymphocytes of these patients (Table 3).

DISCUSSION
Our data suggest that the use of NGS to monitor MRD based on IDH1/2 mutations is feasible and effective, as this method enabled us to predict relapse in the majority  [15][16][17][18].
Of these targeted inhibitors, 3734 AG-120 is effective at lowering 2-HG levels and restoring cellular differentiation in primary AML cells. This therapy could be more personalized by monitoring IDH1/2 mutation levels by NGS during and after treatment.
In contrast, DNMT3A mutations were not a suitable markers for MRD monitoring because of the persistence of a preleukemic clone carrying DNMT3A mutations in 40% of the patients who were in complete remission after a median follow-up of 1439 days (range: 1154-2637 days) which likely reflects the intraclonal molecular heterogeneity of hematopoiesis. Changes in gene mutation frequency were reported between AML diagnosis and relapse, and the expansion of a subclone initially present at a low frequency at the time of diagnosis has been observed at relapse [19].
Our results of the sorted cell populations confirm the molecular heterogeneity of hematopoietic clones at complete remission Liran et al. [20] also reported the presence of DNMT3A mutations at a high allelic frequency in highly purified HSCs, progenitors and mature blood cell fractions in AML patients in complete remission but did not observe concurrent NPM1 mutations, present in the blast cells at AML diagnosis. DNMT3A mutant HSCs showed a multi-lineage repopulation advantage over the non-mutated HSCs in xenograft experiments, which suggests that these cells were pre-leukemic HSCs [21].
Altogether, these data suggest that DNMT3A mutations could induce "pre-leukemic" abnormal hematopoiesis but remain insufficient for leukemogenesis.
We were not able to perform the mutation screening in samples collected before the diagnosis of AML in our patient population to determine whether DNMT3A mutations may have preexisted in the patients whose samples showed a discrepancy between DNMT3A and NPM1 mutation MRD levels. However, the observed DNMT3A variant allele frequency in the patients who were in complete remission after chemotherapy ranged from 5 to 45% at post-induction and increased during follow-up, unlike other MRD markers.
Although we could not evaluate whether mutated DNMT3A was present in HSCs before AML diagnosis, 3 groups independently reported the emergence of neoplastic blood cell clones with aging [22][23][24]. Jaiswal et al. reported that DNMT3A mutation were the most frequent mutations observed with aging and that patients with DNMT3A mutations had a 10-to 50-fold higher propensity for developing hematologic cancer [24]. Similarly, Genovese et al. reported that the frequency of mutations among individuals older than 65 years was 10% and that the most frequent mutations affected DNMT3A gene [23]. They also reported that DNMT3A mutation was associated with increased risk for developing hematologic cancer that was related to the earlier clone. Our patients were not over 65 years of age and demonstrated an elevated VAF during the CR stage that was higher than the level observed with aging, which suggests that the mutant HSCs were resistant to chemotherapy. How these abnormal hematopoietic clones will be involved in relapse or in the occurrence of new hematological malignancies should be monitored on a long term period.

Patients and samples
This retrospective study included 94 samples from 31 NPM1 mutation-positive patients (23-70 years old; median: 60 yrs) who were newly diagnosed with AML from the Acute Leukemia French Association (ALFA) -0701 trial.

Molecular analysis
NPM1 mutation monitoring by RTqPCR was performed as previously described [25]. IDH1/2 and DNMT3A mutation monitoring was performed by NGS using an Ion Torrent ProtonTM instrument (life technologies). To obtain very high coverage (i.e., approximately 2 million reads per sample), 24 samples were analyzed per run. Bioinformatic analysis was performed as described in our previous work [26].
Samples from three patients in CR who had persistent DNMT3A mutations but no other abnormalities were more extensively investigated. DNMT3A quantification by NGS was performed on the following samples: skin, whole peripheral blood, whole bone marrow and blood subpopulations.

Flow cytometry
The following blood subpopulations were sorted using positive selection immuno-bead kits (Easy Sep Stem Cell): Neutrophils targeted by CD 66b antibodies (Ab), monocytes targeted by CD 14 Ab, T lymphocytes targeted by CD 3 Ab, B lymphocytes targeted by CD 19 Ab and NK cells targeted by CD 56 Ab. The bone marrow subpopulations that were notably enriched in leukemia-initiating cells included CD 34+ , CD 38-, CD 123+ , and ALDH intermediate cells. These cells were sorted using a FACS ARIA Sorp based on the membrane expression levels of CD 34 , CD 38 , and CD 123 and the level of ALDH activity [27].

CONCLUSION
The NGS technique is an effective tool to monitor MRD in AML patients, but choosing the appropriate MRD markers is crucial to avoid results that are not related to the disease status. Altogether, our findings show that DNMT3A mutation does not participate to relapse or leukemia progression during our period of clinical follow up. Screening of leukemia-initiating mutations, such as DNMT3A, NPM1 or IDH1/2 mutations should be performed at diagnosis but only NPM1 and IDH1/2 are robust target for MRD monitoring.