Abstract

   Background: About half of adults with acute myeloid leukemia with
   normal cytogenetics (CN-AML) have NPM1 mutations. There is controversy
   regarding their prognosis and best therapy. Methods: We studied 150
   subjects with these features using targeted regional sequencing.
   Prognostic stratification was carried out based on risk factors, and we
   assessed the effects of two post-remission strategies with and without
   transplant across risk cohorts. Results: In multi-variable analyses, a
   positive MRD test after the second consolidation cycle (HR = 6.00; 95%
   CI [3.31, 10.85]; p < 0.001), DNMT3A mutations (HR = 3.01 [1.57, 5.78];
   p < 0.001), FLT3-ITD mutation with high variant allele frequency (HR =
   4.40 [1.89, 10.24]; p < 0.001) and DDX11 mutations (HR = 4.38 [2.38,
   8.04]; p < 0.001) were independently correlated with higher cumulative
   incidence of relapse (CIR) and worse leukemia-free survival (LFS) (HR =
   5.49 [3.01, 10.04]; p < 0.001; HR = 2.99 [1.60, 5.62]; p < 0.001; HR =
   4.20 [1.87, 9.40]; p < 0.001; and HR = 4.22, 95% CI [1.99, 8.95], p <
   0.001). Subjects with ≥1 high-risk co-variate who received a transplant
   had a lower CIR and better LFS, whereas others did not. Conclusions: We
   identified co-variates associated with CIR and LFS in subjects of
   NPM1-mutated CN-AML.

   Keywords: acute myeloid leukemia, normal cytogenetics, NPM1 mutation,
   risk stratification

1. Introduction

   Mutations in nucleophosmin 1 (NPM1), FMS-like tyrosine kinase 3 (FLT3)
   and CCAAT/enhancer-binding protein α (CEBPA) contribute to risk
   stratification of cytogenetically normal acute myeloid leukemia
   (CN-AML) [[54]1,[55]2,[56]3]. NPM1 mutation is reported in 25–35% of
   adults with AML and an even higher percentage of those with normal
   cytogenetics, 45–64% [[57]4]. AML with NPM1 mutation was first
   recognized as a distinct entity in the 2008 World Health Organization
   (WHO) classification system and carried forth in subsequent editions
   [[58]5].

   The distinguishing feature of NPM1 mutation is aberrant cytoplasmic
   localization [[59]6]. The NPM1 mutation appears to be a late event in
   leukemia development and is usually associated with co-mutations such
   as DNA methyltransferase 3 alpha (DNMT3A), Fms-like tyrosine kinase 3
   internal tandem duplication (FLT3-ITD) and isocitrate dehydrogenase
   (NADP(+)1/2) (IDH1/2), which also affect prognosis [[60]2,[61]3].

   People with NPM1/NRAS proto-oncogene GTPase (NRAS), NPM1/RAD21 cohesin
   complex component (RAD21), or NPM1/FLT3-tyrosine kinase point mutation
   (FLT3-TKD) genotypes have relatively good outcomes, whereas those with
   NPM1/WT1 transcription factor (WT1)-mutated or
   NPM1/FLT3-ITD/DNMT3A-mutated genotypes have adverse outcomes
   [[62]3,[63]7,[64]8,[65]9].

   These complexities make it important to consider co-mutations in
   subjects with NPM1 mutations to predict outcomes. Our aim was to
   utilize high-depth targeted regional sequencing (TRS) technology to
   detect more myeloid-cancer-related mutations, combined with clinical
   data, to estimate the prognosis of adults with CN-AML harboring NPM1
   mutations.

2. Materials and Methods

2.1. Subjects and Treatment Protocols

   A total of 1004 consecutive newly diagnosed, untreated persons with AML
   were studied from January 2012 to December 2019 at Peking University
   People’s Hospital (China). AML was diagnosed using the WHO 2022
   criteria [[66]5]. Cytogenetic studies were carried out using standard
   techniques. Targeted regional sequencing was carried out on a cohort of
   150 consecutive subjects with NPM1-mutated CN-AML who presented the
   following characteristics: (1) age > 15 years; and (2) achieved a
   histological complete remission with or without normalization of blood
   cell concentrations ([67]Figure 1).

Figure 1.

   [68]Figure 1
   [69]Open in a new tab

   Subject recruitment and cohort assignment. AML, acute myeloid leukemia;
   CN, cytogenetically normal; CR, complete response.

   The protocols for induction and consolidation therapies were reported
   [[70]10,[71]11,[72]12]. After the second consolidation cycle, the
   physicians assessed the risks and benefits of the patient receiving an
   allotransplant versus continuing with consolidation chemotherapy.
   Allogeneic hematopoietic stem cell transplantation (allo-HSCT) was
   carried out as reported [[73]10,[74]11,[75]12].

2.2. Response Definitions

   Histological complete response (CR) was defined as <5 percent bone
   marrow blasts. Complete hematological recovery was defined as blood
   neutrophils > 1.0 × 10^9/L, platelet concentration > 100 × 10^9/L
   without platelet transfusions, and no RBC transfusions. Relapse was
   defined as >5 percent bone marrow blasts, blasts in the blood or
   extra-medullary leukemia.

2.3. Measurable Residual Disease Testing

   Measurable residual disease (MRD) testing was performed using
   multi-parameter flow cytometry (MPFC) with a panel of eight antibody
   combinations that recognize CD7, CD11b, CD13, CD14, CD16, CD19, CD33,
   CD34, CD38, CD41, CD45, CD56, CD61, CD64, CD71, CD117, CD123, and
   HLA-DR. After each chemotherapy cycle, bone marrow samples were tested
   for MRD based on the leukemia-associated immune phenotype (LAIP)
   identified at diagnosis. An MRD test result ≥ 0.01 percent was scored
   as positive.

2.4. High-Depth Targeted Regional Sequencing and Analyses

   High-depth targeted region sequencing (TRS) analysis was performed on
   bone marrow samples from subjects, focusing on 236 genes
   ([76]Supplement Table S1) known or considered to be potential mutation
   hotspots in hematological cancers. DNA was extracted using DNAzol^®
   kits (Invitrogen, Carlsbad, CA, USA) using the manufacturers’
   instructions. Sequencing details were reported [[77]12]. The following
   criteria were used to filter raw variant results: (1) average effective
   sequencing depth on target per sample ≥ 1000×; (2) mapping quality ≥
   30; (3) base quality ≥ 30; and (4) variant allele frequency (VAF) ≥ 1%
   for SNVs and small indels. Burrows–Wheeler alignment (BWA version
   0.7.12) was used to align the trimmed reads. The MarkDuplicates tool
   from Picard was used to mark the PCR duplicates. Indel Realigner and
   Base Recalibrator from Genome Analysis Toolkit (GATK version 3.8) were
   applied for re-alignment and re-calibration of the BWA data. Variant
   calling, including SNVs and small indels, was carried out in Mutect2.
   ANNOVAR software (V2020) was used to annotate all the variants
   including 1000 G projects, COSMIC, SIFT and PolyPhen. Typical and
   atypical NPM1 mutations (type A/B/D) were detected by real-time
   quantitative polymerase chain reaction (RT-qPCR) and first-generation
   sequencing [[78]13]. FLT3-ITD mutations were detected by capillary
   electrophoresis fragment analyses.

2.5. Endpoints and Statistical Methods

   The last follow-up was 31 December 2020. The primary endpoint was
   cumulative incidence of relapse (CIR) measured from CR. The secondary
   endpoint was leukemia-free survival (LFS) defined as the survival
   period with continuous CR from CR1. Subjects receiving a transplant
   were censored at the time of transplantation when evaluating
   co-variates for CIR and LFS.

   Categorical variables were expressed in percentages and analyzed by
   Pearson’s Chi-square test. Normally distributed continuous variables
   were described as mean (standard deviation) values and analyzed by
   independent-sample t-test. Non-normally distributed continuous
   variables were described as median (range) values and analyzed by the
   Mann–Whitney U test. Homogeneity of variances between groups was
   assessed using the Levene test. CIR was calculated using competing
   risks with non-relapse mortality (NRM), and the Fine–Gray test was used
   for univariate analysis. The LFS rate was calculated using the
   Kaplan–Meier method with the log-rank test. Competing risks model and
   multi-variate Cox regression analyses were performed for the
   co-variates selected by the univariate analyses with a criterion of p <
   0.10. The variance inflation factor (VIF) was estimated to check for
   multi-collinearity amongst co-variates included in the multi-variate
   analyses. Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment
   analysis for the mutational genes was carried out by using the R
   ClusterProfiler package. Analyses were performed using SPSS software
   version 25.0 (Chicago, IL, USA) and R software version 4.1.0.
   ([79]http://www.Rproject.org). p < 0.05 was considered significant.

2.6. Trial Registration

   This trial is registered at Clinicaltrials.gov ([80]NCT01455272,
   [81]NCT02185261) and in chictr.org (ChiCTR-OCH-10000940).

3. Results

3.1. Subject Baseline Co-Variates and CIR or LFS

   Subject co-variates are displayed in [82]Table 1. For post-remission
   therapy, 99 did not received a transplant and 51 received a transplant.
   The transplant cohort was younger, with medians of 33 years
   (Interquartile Range, IQR, 18 years) versus 53 years (IQR, 19 years; p
   < 0.001). Other co-variates were similar. Five-year CIR and LFS in the
   no-transplant cohort were 60% and 36%, compared to 28% and 60% (HR
   0.44, 95% CI [0.24, 0.82], p = 0.01; HR 0.59, 95% CI [0.35, 1.01], p =
   0.05) in the transplant cohort. ([83]Supplement Figure S1).

Table 1.

   Subject co-variates and post-remission therapy.
   No Transplant (N = 99) Transplant (N = 51) p-Value
   Gender (male/female) 48/51 30/21 0.30
   Age (median, range) 53 (16–76) 33 (17–61) <0.001
   WBC (×10^9/L) (median, range) 29.8 (0.3–224.4) 35.1 (0.9–266.0) 0.69
   Hemoglobin (g/L) (mean, SD) 87 (22) 88 (25) 0.86
   Platelets (×10^9/L) (median, range) 68 (6–806) 59 (10–535) 0.61
   Bone marrow blasts (%) (median, range) 72 (26–99) 70 (21–96) 0.31
   CR after first induction (n, %) 78 (79%) 44 (86%) 0.28
   MRD-positive (n, %) 50 (51%) 23 (45%) 0.61
   [84]Open in a new tab

   Abbreviations: CR, complete remission; MRD, measurable residual
   disease.

3.2. Genomic Analyses and Mutation Topography

   One subject had a mis-sense and two in-frame insertion mutations. Eight
   had >1 type of NPM1 mutation referred to multi-hit ([85]Figure 2). A
   total of 771 other mutations (not all 236 genes in the panel were found
   to have mutations) were detected, with a median of five (Range: 1–12)
   per subject, and >5 mutations were detected in 56 subjects. Mis-sense
   mutations were the most common (N = 564) followed by frame-shift
   insertions (N = 182) and in-frame insertions (N = 64). Mutations in ≥10
   subjects were divided into eight subgroups: (1) signaling pathway
   (FLT3, FLT3-ITD, PTPN11, NARS, KRAS, LILRB3, MACF1 and NF1); (2) DNA
   methylation (DNMT3A, TET2, IDH1 and IDH2); (3) histone methylation
   (KMT2A and KMT2D); (4) cohesion complex (DDX11 and RAD21); (5)
   chromatin modifiers (SRCAP); (6) transcription factors (GATA2); (7)
   tumor suppressors (WT1); (8) and others (CCDC168, PCLO, DNAH2).

Figure 2.

   [86]Figure 2
   [87]Open in a new tab

   Genomic landscape. Genes mutated in ≥10 subjects are shown. Each column
   indicates data of one sample; each row represents a gene. Mutated genes
   were color-coded for mis-sense (green), non-sense (blue), in-frame ins
   (red), splice-site (orange), frameshift del (pink), in-frame del
   (yellow), frame-shift ins (purple) and multi-hit (black). The top bar
   indicates mutation load (mutation/Mb DNA), and the right bar indicates
   mutation frequency.

   DNMT3A (N = 74) was the most frequently mutated gene followed by
   FLT3-ITD (N = 52, 34.67%), FLT3 (FLT3 mutation types other than
   FLT3-ITD, N = 51, 34%), and IDH2 (N = 44, 29.33%). Median variable
   allele frequency (VAF) was 0.43 (Range, 0.01–1.00). The VAFs of
   mutations in the DNA methylation pathways and histone methylation were
   significantly higher compared with those of mutations in the activated
   signaling pathways (0.43 [0.01, 0.92] versus 0.17 [0.01, 0.78]; p <
   0.001; 0.44 [0.01, 0.60] versus 0.17 [0.01, 0.78], p < 0.01;
   [88]Supplement Figure S2).

   We identified 10 pairs of co-occurring genes and 5 pairs of mutually
   exclusive mutations, which were statistically significant. RAD21, NRAS,
   and PTPN11 were co-mutations, whereas IDH2 and TET2 mutations, and
   PTPN11 and FLT3-ITD mutations were mutually exclusive mutations
   ([89]Supplement Figure S3A). Using KEGG pathway enrichment analysis, we
   found that co-mutations in CN-AML with NPM1 mutations were related to
   cancer and metabolism ([90]Supplement Table S2, Supplement Figure S3B).
   Associations between mutations and subject-related co-variates at
   diagnosis are shown in [91]Table 2. DNMT3A mutations were associated
   with increased age and high WBC and platelet concentrations at
   diagnosis. FLT3-ITD mutations were associated with a higher percentage
   of bone marrow blasts.

Table 2.

   Associations of mutations with subject-related variables.
   Variable DNMT3A FLT3-ITD IDH2 PTPN11
   WT
   (N = 76) MUT
   (N = 74) WT
   (N = 98) MUT
   (N = 52) WT
   (N = 106) MUT
   (N = 44) WT
   (N = 121) MUT
   (N = 29)
   Gender, male/female 35/41 43/31 48/50 30/22 53/53 25/19 62/59 16/13
   Age (median, range) 41 (17–73) 50 (22–76) * 49 (17–76) 47 (16–68) 48
   (16–76) 48 (17–65) 47 (16–76) 48 (17–67)
   WBC (×10^9/L) (median, range) 21 (0–228) 37 (2–226) * 29 (0–266) 38
   (1–185) 32 (1–266) 25 (0–192) 31 (0–266) 30 (3–224)
   Hemoglobin (g/L) (mean, SD) 85 (24) 91(23) 85 (23) 92 (24) 86 (24) 91
   (22) 88 (24) 86 (19)
   Platelets (×10^9/L) (median, range) 48 (6–267) 76(11–806) * 69 (6–535)
   56(10–806) 59 (6–806) 78 (7–535) 64 (6–806) 73 (9–211)
   Bone marrow blasts (%) (median, range) 70 (23–96) 72 (21–99) 65 (21–99)
   76 (27–96) * 70 (21–94) 73 (31–99) 71 (24–99) 57 (21–91)
   [92]Open in a new tab

   WT, wild type; MUT, mutation; SD, standard deviation. * p < 0.05 when
   compared with wild type.

3.3. Prognostic Co-Variates

   Clinical variables and gene mutations occurring in ≥10 patients were
   included in the univariate analyses and six variables with p < 0.10
   were eligible for subsequent analyses ([93]Supplement Table S3).
   Considering the interaction between MRD test results after the first
   and second consolidation cycles and referring to previous reports
   [[94]14], we used a backward elimination process and ultimately
   included the MRD test results after the second consolidation cycle as a
   co-variate.

   Eventually, four co-variates were significantly correlated with CIR in
   multi-variable analyses: (1) positive MRD test after the second
   consolidation cycle (HR = 6.00 [3.31, 10.85]; p < 0.001), (2) DNMT3A
   mutation (HR = 3.01 [1.57, 5.78], p < 0.001); (3) FLT3-ITD mutation
   with high VAF (≥0.5) (HR = 4.40 [1.89, 10.24]; p < 0.001) and (4)
   DEAD/H-box helicase 11 (DDX11) mutation (HR = 4.38 [2.38, 8.04]; p <
   0.001) and LFS (HR = 5.49 [3.01, 10.04]; p < 0.001; HR = 2.99 [1.60,
   5.62]; p < 0.001; HR = 4.20 [1.87, 9.40]; p < 0.001; and HR = 4.22, 95%
   CI [1.99, 8.95], p < 0.001). ([95]Table 3). The VIFs of the co-variates
   included in the multi-variate analyses were below five, suggesting no
   significant multi-collinearity among the co-variates.

Table 3.

   Multivariate analyses of CIR and LFS.
   Variables Multivariate Analysis of CIR Multivariate Analysis of LFS
   HR (95% CI) p-Value HR (95% CI) p-Value
   MRD test (+/−) 6.00 (3.31–10.85) <0.001 5.49 (3.01–10.04) <0.001
   DNMT3A (Mutated vs. WT) 3.01 (1.57–5.78) <0.001 2.99 (1.60–5.62) <0.001
   FLT3-ITD (Mutated with VAF ≥ 0.5 vs. WT or mutated with VAF < 0.5) 4.40
   (1.89–10.24) <0.001 4.20 (1.87–9.40) <0.001
   DDX11 (Mutated vs. WT) 4.38 (2.38–8.04) <0.001 4.22 (1.99–8.95) <0.001
   [96]Open in a new tab

   MRD, measurable residual disease; WT, wild type; VAF, variant allele
   frequency; CIR, cumulative incidence of relapse; HR, hazard ratio; CI,
   confidence interval; LFS, leukemia-free survival.

   For the newly discovered DDX11 mutations, we found that, compared to
   wild-type subjects, those with the DDX11 mutations had a significantly
   higher CIR and a notably shorter LFS ([97]Supplement Figure S4).

3.4. Impact of Transplants

   We assigned subjects with ≥1 adverse risk co-variates to a high-risk
   cohort (N = 104) and others to a low-risk cohort (N = 46) with
   corresponding 5-year CIRs of 73% [58, 88%] and 37% [16, 57%]; (HR =
   3.85 [1.98, 7.50]; p < 0.001) and 5-year LFSs of 24% [14, 43%] and 58%
   [41, 84%] (HR = 3.77 [1.90, 7.48]; p < 0.001) ([98]Figure 3).

Figure 3.

   [99]Figure 3
   [100]Open in a new tab

   Cumulative incidence of relapse (A) and leukemia-free survival (B) in
   the high-risk and low-risk cohort.

   Transplant recipients in the high-risk cohort (N = 37) had a lower
   5-year CIR, 33% [16, 49%] versus 74% [60, 89%] (HR = 0.37 [0.19, 0.74];
   p < 0.01), and a better 5-year LFS, 58% [43, 78%] versus 23% [13, 41%]
   (HR = 0.45 [0.25, 0.83]; p = 0.01), compared with those in the
   no-transplant cohort (N = 67). In the low-risk cohort, 14 subjects
   received a transplant and 32 did not receive a transplant. Five-year
   CIRs were 17% [0, 40%] versus 32% [11, 54%] (HR = 0.63 [0.13, 3.05]; p
   = 0.57), and five-year LFSs were 66% [44, 100%] versus 62% [44, 88%]
   (HR = 1.22 [0.37, 4.06]; p = 0.75) ([101]Figure 4).

Figure 4.

   [102]Figure 4
   [103]Open in a new tab

   Cumulative incidence of relapse and leukemia-free survival in the
   transplant and no-transplant subjects of high-risk cohort (A,B) and
   low-risk cohort (C,D).

4. Discussion

   We found that in subjects with CN-AML and NPM1 mutation, those with ≥1
   high-risk co-variate, including positive MRD test after the second
   consolidation cycle, DNMT3A mutation, FLT3-ITD mutation with high VAF
   and DDX11 mutation, had a higher CIR and worse LFS compared to subjects
   without these co-variates. We also found a lower CIR and better LFS in
   subjects in the high-risk cohort who received a transplant, which was
   not observed in the low-risk cohort.

   Compared to the recent study published by Yao et al. [[104]15], our
   research revealed that the mutation landscape of NPM1 mutations in
   CN-AML subjects was similar to their findings. We also identified
   DNMT3A and FLT3-ITD mutations as poor prognostic factors. However,
   their study did not emphasize that the subjects were cytogenetically
   normal AML patients. Furthermore, we discovered a novel DDX11 mutation
   that adversely affects prognosis. In addition, we considered the impact
   of MRD as a clinical factor on prognosis.

   DNMT3A, a protein responsible for DNA methylation, is mutated in
   approximately 20% of all AML cases. It commonly co-mutates with NPM1 in
   60% of cases, and in about 30% of cases, both NPM1 and FLT3 mutations
   are present [[105]16,[106]17,[107]18,[108]19,[109]20]. Compared to
   wild-type DNMT3A subjects, those with DNMT3A mutations tend to be
   older, have higher WBC concentration, and more frequently have normal
   karyotype at diagnosis [[110]16,[111]18,[112]19,[113]21,[114]22].
   Although several studies reported that DNMT3A mutations were
   independently associated with poor outcomes in AML, especially in
   CN-AML [[115]18,[116]21,[117]23,[118]24], the impact of DNMT3A
   mutations in clinical decision-making remains debatable
   [[119]19,[120]20,[121]25]. In our study, DNMT3A mutations were the most
   frequently co-occurring event in NPM1-mutated CN-AML subjects (49.33%).
   These subjects were older and had a higher WBC concentration, which was
   consistent with previous studies
   [[122]16,[123]17,[124]18,[125]19,[126]21,[127]22]. We further
   identified DNMT3A mutations as one of the poor prognostic factors, and
   subjects harboring DNMT3A mutations were classified into high-risk
   cohort, for whom allo-HSCT could improve survival.

   FLT3-ITD has consistently been associated with high WBC and BM blast
   concentration, an increased risk for relapse, and inferior survival
   [[128]26,[129]27,[130]28]. In the 2022 ELN guidelines [[131]29], the
   FLT3-ITD allelic ratio is no longer considered in risk classification.
   This change reflects the impact of FLT3 inhibitor use [[132]30] on the
   natural course of AML with FLT3-ITD mutations and the increasing
   importance of MRD in guiding treatment decisions [[133]29]. In our
   study, subjects with FLT3-ITD mutations with high VAF were classified
   into the high-risk cohort and may benefit from allo-HSCT; this may be
   related to our data, where among the 19 subjects with high VAF of
   FLT3-ITD mutations, 12 were found to have concurrent DNMT3A mutations.
   Several studies have reported NPM1/FLT3-ITD/DNMT3A, the most common
   triple mutation pattern in NPM1-mutated patients, with extremely poor
   prognosis [[134]9,[135]15,[136]16,[137]17,[138]18], and allo-HSCT has
   been shown to significantly improve the survival of this subgroup
   [[139]15].

   We further identified that DDX11 mutations had prognostic significance
   in our study. DDX11 encodes an iron–sulfur cluster DNA helicase
   required for development, mutation and over-expression in cancers
   [[140]31]. DDX11 has been shown to be associated with the progression
   of multiple cancers, including melanoma [[141]32], lung adenocarcinoma
   [[142]33], and hepatocellular carcinoma [[143]34]. In multivariate
   analyses conducted in a CIBMTR study, it was revealed that DDX11
   mutations affect survival by increasing the risk of both relapse and
   transplant-related mortality in patients with myelodysplastic syndrome
   [[144]35]. Encouragingly, this study further validates the recent
   findings published by us, which indicated that CN-AML patients with
   DDX11 mutations have a significantly higher CIR and poorer LFS
   [[145]36], and DDX11 dysfunctions were linked to AML via promoting cell
   proliferation [[146]37]. Other than that, DDX11 mutations have not been
   reported in NPM1-mutated CN-AML, and the mutation frequency was 8% (N =
   12) in our study; nine of these subjects relapsed.

   In addition to baseline genetic characterization, the 2022 ELN
   guidelines also emphasize the importance of response to initial therapy
   and early assessment of MRD in individual risk assignment [[147]29]. In
   clinical practice, MRD can serve as an important biomarker for
   prognosis, prediction, monitoring, and efficacy evaluation [[148]38].
   However, MRD-negative patients may still relapse in some cases, while
   some MRD-positive patients may remain in CR. Additionally, MRD
   positivity might be more common in genetically adverse AML cases
   compared to genetically favorable ones, suggesting that MRD is
   influenced to some extent by genetic factors [[149]39]. Therefore,
   integrating MRD monitoring results with additional genetic data during
   the treatment process for risk stratification may further enhance
   prognostic prediction.

   Our study has important limitations. First, it was retrospective study.
   Second, the subjects received diverse therapies and were not randomly
   assigned to receive a transplant. Third, our censoring of transplant
   recipients was imperfect. Fourth, there were few subjects with DDX11
   gene mutations. Given these limitations, our conclusions need
   validation. Our data, if validated, could assist physicians in
   predicting the prognostic risk stratification for subjects with CN-AML
   and NPM1 mutations, and could potentially be incorporated into clinical
   decision-making.

Acknowledgments