NPM1-mutated acute myeloid leukemia: from bench to bedside

The nucleophosmin (NPM1) gene encodes for a ubiquitous multifunctional shuttling protein¹  with predominant nucleolar localization. NPM1 is the most commonly mutated gene in adult acute myeloid leukemia (AML; ∼30% of cases).²  The most distinguishing feature of NPM1 mutants is their aberrant cytoplasmic localization,³  which led to the discovery of NPM1 mutations by immunohistochemistry (IHC), prior to the next-generation sequencing (NGS) era.²  It was a long journey before NPM1-mutated AML was recognized as a distinct entity in the 2017 World Health Organization (WHO) classification of hematopoietic neoplasms.^4,5  Together with double-CEBPA–mutated AML, NPM1-mutated AML is 1 of the 2 WHO leukemia entities defined by a single-gene mutation. Here, we focus on new insights about NPM1 wild-type (NPM1wt) protein and NPM1-mutated AML.

NPM1wt is a nucleolar chaperone protein, active as an oligomer (pentamer/decamer). The structure and putative functions of NPM1wt have been reviewed extensively^6-9  and are summarized in Figure 1. Here, we focus on newly described nucleolar functions of NPM1wt.

The nucleolus, a membrane-less organelle, is formed through a “liquid-liquid” phase separation (LLPS) process^10-12  that causes the segregation of NPM1wt and other nucleolar components from the surrounding nucleoplasm. This process is similar to how oil and water separate from each other when mixed. LLPS is facilitated by the interaction between pentameric NPM1wt, proteins containing R-motifs (ie, multivalent arginine-rich linear motifs [sharing features with nucleolar localization signals (NoLSs)] and nascent ribosomal RNA [rRNA]).^10,13  These heterotypic interactions are critical for nucleolar phase separation and for maintaining the directionality of ribosome biogenesis through the different subcellular compartments (ie, from the phase-separated nucleolus to the nonphase-separated nucleoplasm).¹³  Examples include the interactions of NPM1wt with p14Arf R-motifs¹⁴  or with SURF6 and rRNA.¹³ 

The nucleolus is made-up of a fibrillar core, surrounded by the dense fibrillar component and the granular component (GC). The multilayered structure of the nucleolus is determined by LLPS between nucleolar proteins, primarily NPM1wt and fibrillarin that segregate in the GC and dense fibrillar component, respectively¹¹  (Figure 1B). Within the GC phase, NPM1wt serves as a scaffolding protein for nascent rRNA¹³  and helps prevent irreversible aggregation of cytoplasmic misfolded proteins migrating to the nucleolus following stress.¹⁵ 

NPM1wt also participates in rRNA 2′-O-methylation¹⁶  (Figure 1B). Specifically, NPM1wt binds directly to several C/D box small nucleolar RNAs¹⁶  and to the methyltransferase fibrillarin. The fibrillarin–NPM1–small nucleolar RNA complex actively methylates rRNA, and NPM1wt loss results in altered rRNA 2′-O-methylation and deregulated translation.¹⁶  Accordingly, germline loss-of-function mutations in the NPM1 acidic region may contribute to the pathogenesis of dyskeratosis congenita,¹⁶  a well-known ribosomopathy. However, the role of altered translation in NPM1-mutated AML remains to be established.

The nucleus-cytoplasm shuttling of NPM1wt is regulated by specific signal motifs³  (Figures 1A and 2A). A bipartite nuclear localization signal (NLS), recognized by importin α/β,¹⁷  drives NPM1wt from the cytoplasm to the nucleoplasm. NPM1wt then localizes to the nucleolus through an aromatic NoLS, a 3-helix structure located at the C terminus.^18,19  NoLS contains 2 tryptophans (W288 and W290) that stabilize the hydrophobic core of triple helix, enabling its proper folding. This is critical for localizing NPM1wt to the nucleolus¹⁸  through interaction with G-quadruplexes of ribosomal DNA.²⁰ 

The nuclear export of NPM1wt is mediated by the interaction of 2 N-terminal nuclear export signals (NESs) with the nuclear exporter XPO1 (also named CRM1).⁷  Binding of XPO1 to some of its cargo occurs within the nucleolus.²¹  We hypothesize that localization of NPM1wt to the GC, the external component of the nucleolus,¹¹  may facilitate the access of XPO1 to this area and its interaction with NPM1wt. However, NPM1wt is not exported efficiently, because the N-terminal NESs have low affinity for XPO1.¹⁷  We believe that nuclear export may also be weak because it is opposed by multiple forces that anchor NPM1wt to the nucleolus, including binding of NoLS to G-quadruplexes and interactions of NPM1wt with nucleolar proteins and rRNA.¹⁰  Ultimately, the import and anchoring signals exceed the export signals^3,17  so that, at steady-state, NPM1wt resides in the nucleolus, whereas only a minimal fraction (although functionally relevant) shuttles between cell compartments.

Nucleolar localization of NPM1wt depends upon its oligomerization status, because pentamers usually reside in the nucleolus, whereas monomers are nucleoplasmic.²²  Phosphorylation of S48, S88, T95, and S125 favors the permanence of NPM1wt in a monomeric state, promoting its nucleoplasmic localization.²³  Phosphorylation of T199, T219, T234, and T237 by a cell cycle–dependent kinase during mitosis hinders the interaction between NPM1wt and rRNA.²⁴  We hypothesize that altering this interaction results in loss of NPM1wt phase separation, facilitating its distribution through the cell during mitosis. Acetylation of K229, K230, K257, and K267 seems to favor its nucleoplasmic localization and binding to RNA polymerase II,²⁵  likely facilitating DNA transcription. Finally, sumoylation and ubiquitination at specific sites may influence NPM1 localization and stability.¹⁹ 

The main features of NPM1 mutations are summarized in Table 1. NPM1 mutations are always heterozygous and primarily restricted to exon 12 (<1% involving other exons).⁷  They probably arise from replication errors primed by an illegitimate terminal deoxynucleotidyl transferase activity.²⁶ NPM1 mutations usually consist of 4-bp insertions that cause a frameshift in the last few C-terminal amino acids of the protein, leading to loss of W288 and W290 (or W290 alone) and generation of a new C-terminal NES²⁷  (Figures 1A and 2B). Both changes are required for cytoplasmic localization of NPM1 mutants.^3,27 

Loss of tryptophan(s) causes unfolding of the C-terminal triple helix,¹⁸  impairing NPM1’s ability to reside within the nucleolus²⁰  (Figure 2B). The new C-terminal NES²⁷  reinforces the export activity of the N-terminal NESs,²⁸  increasing the probability of NPM1 mutants wandering in the nucleoplasm to be exported to the cytoplasm. Importantly, XPO1 recognizes the C-terminal NES more efficiently than the N-terminal NESs, likely as a result of unfolding of the C-terminal triple helix.¹⁷  Eventually, the export exceeds the import and anchoring signals, and NPM1 mutants are delocalized to the cytoplasm of AML cells (Figure 2B).

NPM1 mutants act dominantly over NPM1wt, causing its cytoplasmic delocalization through the formation of NPM1 mutant/NPM1wt heterodimers.^3,27  However, the exact proportion of NPM1wt localized to the cytoplasm of NPM1-mutated cells has never been quantified, and its biological relevance remains unclear.

All NPM1 mutations, including those rarely occurring outside of exon 12 (ie, exon 5,²⁹  exon 9,³  and exon 11³ ) generate cytoplasmic leukemic mutants (Figure 3A). This is also a feature of rare NPM1 fusion proteins, including NPM1-HAUS1,³⁰  and those generated by novel translocations [ie, t(5;10)(q35;q23) NPM1-RPP30, t(5;18)(q34;q12) NPM1-SETBP1, and t(5;6)(q35;q23) NPM1-CCDC28A].³¹  These fusion proteins lose the NPM1wt C-terminus (including W288 and W290) and acquire a C-terminal NES that is either newly created or already present in the fusion partner.

Nuclear export of NPM1 mutants is also fine-tuned by the strength of the C-terminal NES, depending on the loss of W288 and W290 (insertion of a weaker NES) or W290 only (insertion of a stronger NES)³²  (Figure 3B). Rarely (eg, in exon 5 NPM1 mutations), the new NES may be strong enough to cause nuclear export of leukemic mutants without losing any of the C-terminal tryptophans (“super-NES”). Thus, all NPM1 mutations act to maximize the export of NPM1 mutants, pointing to cytoplasmic delocalization as being critical for leukemogenesis.^3,32 

Clonal hematopoiesis represents the expansion of a clonal population of hematopoietic precursors carrying ≥1 somatic mutation, without evidence of hematologic malignancy.^33-35  Unlike mutations of genes involved in chromatin remodeling (eg, DNMT3A, TET2, ASXL1) and RNA splicing (eg, SF3B1, SRSF2), NPM1 mutations have not been detected in individuals with clonal hematopoiesis.³⁵  Rather they act as “gatekeepers” for AML.³⁵  A widely accepted model of clonal evolution, also supported at the single-cell level,³⁶  sees NPM1-mutated AML developing from preexisting clonal hematopoiesis (eg, DNMT3A or IDH2 driven)^33,34,37-39  (Figure 4A). Accordingly, 1 patient with IDH2-mutated clonal hematopoiesis developed AML shortly after the detection of NPM1 mutation.⁴⁰  Moreover, AML patients carrying both NPM1 and DNMT3A mutations at diagnosis, who respond optimally to standard therapy, show persistence of DNMT3A mutations, despite achieving the eradication of NPM1 clone (as assessed by quantitation of NPM1-mutant transcripts) and experiencing long-term complete remission (CR) and cure.^41-43  However, the return to the status of clonal hematopoiesis may predispose these patients to the development of a second hematologic neoplasm. Indeed, although most relapses in NPM1-mutated AML are due to the recurrence of the original NPM1-mutated clone, ∼5% to 10% are characterized by the absence of NPM1 mutations.⁴⁴  These cases are thought to develop a second AML that is favored by the persistence of clonal hematopoiesis following the eradication of the original NPM1-mutated clone.^45,46  This is supported by the long latency between CR and relapse with NPM1 wild-type AML^44-46  (giving enough time for the second genetic hit to occur) and the different mutational patterns detected at diagnosis and relapse. NPM1-mutated AML may also relapse with NPM1 mutation switch (eg, from type D to A),^47,48  likely representing another example of second AML arising from persistent clonal hematopoiesis. Rarely, clonal hematopoiesis may predispose to develop 2 hematological neoplasms, as we reported in a patient with angioimmunoblastic T-cell lymphoma and NPM1-mutated AML⁴⁹  (Figure 4B). Thus, NPM1-mutated AML patients in CR with persistent clonal hematopoiesis should undergo long-term surveillance for the development of a second malignancy. Allogeneic hematopoietic stem cell transplantation (allo-HSCT) is the only therapy that has demonstrated the capability to eradicate clonal hematopoiesis.⁵⁰ 

NPM1 and other mutations, especially those affecting FLT3²  and DNMT3A,⁵¹  frequently co-occur in AML patients, implying molecular synergisms promoting AML development.^52,53  Cooperation mechanisms have been investigated extensively in mouse models. Npm1/Dnmt3a double-mutated mice develop AML but only after a long latency and following transplants,³⁸  suggesting that other mutations are necessary for AML development. Conversely, Npm1/Flt3–internal tandem duplication (ITD) double-mutated mice generate a fully penetrant and short-latency AML^54,55  that is due, at least in part, to GATA1 epigenetic deregulation.⁵⁶ Npm1/Flt3–tyrosine kinase domain (TKD) double-mutated mice also develop AML.⁵⁷  In this model, the Npm1 mutant relocates mutated Flt3 from the cell surface to the endoplasmic reticulum, leading to Stat5 activation.⁵⁷ Npm1/Flt3-ITD/Dnmt3a triple-mutated mice die of a particularly aggressive AML that is chemoresistant, likely due to impaired nucleosome remodeling.⁵⁸  Accordingly, patients with this AML genotype show a very poor prognosis.^53,59  These mutations synergize to drive overexpression of hepatic leukemia factor, which leads to increased self-renewal and the aberrant immunophenotype (GPR56^high/CD34^low) of triple-mutated cells.^60,61 

About 20% of NPM1-mutated AMLs carry NRAS or KRAS mutations.⁵³ Npm1/Nras double-mutated mice develop AML with granulocytic differentiation and a less aggressive behavior than Npm1/Flt3-ITD AML, which instead shows monocytic differentiation and short latency.^62,63  Similarly, patients with NPM1/NRAS double-mutated AML exhibit a relatively good prognosis.⁵³  All of the above mouse models may serve as a platform for testing novel treatment strategies.

NPM1 mutations are always heterozygous,²  and complete loss of NPM1wt is embryonic lethal,⁶⁴  implying that 1 copy of NPM1wt is necessary for cell survival. Npm1 heterozygous knock-out mice develop a myelodysplastic-like disorder that is due to uncontrolled centrosome duplication and aneuploidy,⁶⁴  demonstrating that NPM1wt is an haploinsufficient tumor suppressor. However, depletion of NPM1wt in NPM1-mutated AML cells is unlikely to drive leukemia through this mechanism. Indeed, NPM1-mutated AML is closely associated with normal karyotype,⁷  likely because both NPM1wt and NPM1 mutants can properly regulate centrosome duplication.

NPM1 mutants induce cytoplasmic delocalization of several nuclear proteins involved in apoptosis, DNA repair and differentiation, including ARF, HEXIM1, FBW7, MIZ1, APE1, and caspases 6 and 8 (see also Falini et al³  and Heath et al⁹ ). Recently, altered localization of the transcription factors CTCF⁶⁵  and PU.1⁶⁶  has been also reported (Figure 5). Overall, the degree to which the function of these proteins is disrupted by the altered localization remains to be determined because they have been mostly studied in vitro or in nonhematopoietic cells,³  and none of these NPM1 partners was confirmed to be delocalized to the cytoplasm of primary NPM1-mutated AML cells.^3,67 

Physiologically, NPM1wt binds to a small nuclear pool of the transcriptional regulator BRD4, exerting an inhibitory effect on its transcriptional activity.⁶⁸  NPM1 mutants alter this equilibrium, because a significant proportion of NPM1 is delocalized into the cytoplasm without BRD4, which is then free to drive the transcription of its target genes,⁶⁸  especially BCL2 and MYC. How this mechanism contributes to AML development remains unclear.

HOXA and HOXB gene clusters are highly expressed in normal adult hematopoietic stem and progenitor cells (HSPCs) but are physiologically silenced in mature blood cells, suggesting a role in self-renewal.⁶⁹  Moreover, ectopic HOX expression in normal HSPCs leads to leukemic transformation.⁶⁹  We provided the first evidence that NPM1-mutated AML exhibits higher HOX levels (specifically HOXA and HOXB)⁷⁰  as compared to NPM1wt AML. However, the HOXA and HOXB expression signature of NPM1-mutated AML is nearly identical to that of normal HSPCs, suggesting persistent expression rather than upregulation of HOX genes⁷¹  and pointing to common mechanisms regulating HOX expression in normal HSPCs and NPM1-mutated AML.⁷¹ 

The histone modifier MLL1 contributes by regulating HOX expression in NPM1-mutated AML, and this function depends on the interaction between MLL1 and its cofactor menin.^72,73  More recently, we showed that, in NPM1-mutated AML cells, HOX expression is directly dependent on NPM1 mutants.⁷⁴  Accordingly, nuclear relocalization or targeted degradation of mutated NPM1 led to immediate loss of HOX expression, followed by differentiation and growth arrest.⁷⁴  Thus, NPM1 mutants act in a gain-of-function manner upstream of HOX to maintain the undifferentiated state of leukemic cells.⁷⁴ 

The link between NPM1 mutants and HOX expression remains unclear. NPM1 could directly bind and displace putative factors from the nucleus that are required for proper differentiation and physiological HOX downregulation (eg, the myeloid transcription factor PU.1⁶⁶  or the architectural protein CTCF,⁶⁵  although its role in HOX regulation in the context of NPM1-mutated AML has recently been questioned⁷⁵ ). In this model, loss of NPM1 mutants from the cytoplasm would restore the normal activity of these factors. Alternatively, HOX expression could be maintained by recruiting NPM1 mutants to HOX loci through their interaction with chromatin-bound XPO1,⁷⁶  and cytoplasmic delocalization of NPM1 would merely represent an epiphenomenon. We propose a model in which NPM1 mutants promote leukemia by acting at the chromatin level and by delocalizing NPM1-interacting partners to the cytoplasm, “killing 2 birds with 1 stone” (Figure 5).

NPM1-mutated AML with multilineage dysplasia^77,78  (Figure 6, top panels) should be distinguished from AML with myelodysplasia-related changes (AML-MRC). According to the 2017 WHO classfication of hematopoietic tumors,⁵  when NPM1 mutations and multilineage dysplasia coexist, the genetic abnormality supersedes morphology, and the case is diagnosed as NPM1-mutated AML.⁵  Conversely, AML-MRC should be diagnosed if typical cytogenetic alterations and/or a previous history of myelodysplasia are documented, even in the presence of NPM1 mutations.⁵ 

Unlike AML with recurrent genetic abnormalities, such as t(8;21), inv(16), t(16;16) or PML/RARα, ≥20% bone marrow (BM) blasts are required to diagnose NPM1-mutated AML.⁵ NPM1-mutated myeloid neoplasms showing ≤20% BM blasts are very rare and have been reported as myelodysplastic syndrome (MDS) or chronic myelomonocytic leukemia (CMML).^79-81  However, NPM1-mutated MDS usually showed normal karyotype, CD34 negativity, and good response to chemotherapy, thus resembling AML.^80,81  Similarly, NPM1-mutated CMML cases^81-83  carried normal karyotype, evolved rapidly to AML,⁸²  especially those with higher NPM1 mutation allelic burden,⁸³  and responded better to chemotherapy than to hypomethylating agents.⁸¹  Moreover, in our experience, BM biopsies from NPM1-mutated MDS and CMML often show clusters of NPM1 cytoplasmic-positive blasts, suggestive of early AML. Whether all of the above neoplasms should be classified as NPM1-mutated AML, without regard to blast cell count,⁸⁴  should be a topic for the next revision of the WHO classification.

Immunohistochemical detection of cytoplasmic NPM1 (a surrogate for NPM1 mutations⁸⁵ ) may help in defining myeloid sarcoma⁸⁶  (Figure 6, middle and bottom panels), which is common in NPM1-mutated AML⁸⁷  and difficult to diagnose by molecular assays because of the frequent limited availability of pathological tissue. IHC also predicts NPM1 mutations outside of exon 12 that may be missed with conventional molecular techniques (Figure 3A).

According to the 2017 European LeukemiaNet (ELN) recommendations,⁸⁸ NPM1 mutations convey a relatively favorable prognosis only if FLT3-ITD is absent or shows a low allelic ratio (<0.5, FLT3-ITD^low). Three recent studies based on chemotherapy alone^89-91  did not confirm the favorable prognosis of NPM1-mutated/FLT3-ITD^low AML, whereas, in the RATIFY trial, this genotype showed 73% overall survival at 5 years following chemotherapy plus the FLT3 inhibitor.⁹²  Thus, risk stratification according to FLT3-ITD allelic ratio requires further validation depending on the treatment setting (eg, FLT3 inhibitors or allo-HSCT). The favorable impact of the NPM1-mutated/FLT3-ITD⁻ genotype appears to be less evident in elderly patients,⁸⁹  especially those older than 65 years.⁹³  In general, adverse-risk cytogenetic abnormalities confer a poor prognosis to patients with the favorable NPM1-mutated/FLT3-ITD⁻ genotype.⁹⁴  However, these cases should be classified as AML-MRC, according to the 2017 WHO classification of hematopoietic tumors.⁵ 

NPM1-mutated AML with FLT3-ITD high allelic ratio (>0.5, FLT3-ITD^high) and is assigned to the intermediate-risk group.⁸⁸  Recently, it has been proposed that the 2017 ELN classification should be refined by moving all FLT3-ITD^high patients, regardless of their NPM1 mutation status, from the intermediate-risk category to the adverse-risk category⁹⁵ ; however, further studies are required to support this proposal. Finally, the prognostic impact of NPM1 mutant allele burden^96-98  at diagnosis remains controversial.

Although the 2017 ELN classification is of practical value, genomic classification of AML has shown that the prognosis of NPM1-mutated AML may depend on a variety of accompanying mutations other than FLT3.⁵³  For example, patients with NPM1/N-RAS, NPM1/RAD21,⁵³  or NPM1/FLT3-TKD⁹⁹  genotypes seem to have a relatively good prognosis. Conversely, cases with NPM1/WT1 double-mutated⁹⁵  or NPM1/FLT3-ITD/DNMT3A triple-mutated⁵⁹  genotypes show a particularly adverse outcome. A search for DNMT3A mutations has been recommended in a recent proposal to update the 2017 ELN classification.⁹⁵  A limitation to the wide application of genomic classification⁵³  is that NPM1-mutated AML is fragmented into many small molecular subsets whose prognostic impact may be difficult to study in clinical trials. Measurable residual disease (MRD) assessment in NPM1-mutated AML patients may help to overcome this limitation.

NPM1 mutations are ideal targets for MRD monitoring because they are frequent, stable at relapse, and not found in subjects with clonal hematopoiesis. We established a real-time quantitative polymerase chain reaction (RQ-PCR) assay to measure NPM1-mutant transcripts to predict hematological relapse and long-term survival.¹⁰⁰  RQ-PCR remains the standard method for MRD monitoring in NPM1-mutated AML, although digital droplet PCR^101,102  and NGS^103,104  are emerging as alternative options. Digital droplet PCR has the advantage of a high sensitivity, and NGS can detect all types of NPM1 mutants, including those not recognized by the currently used kits for RQ-PCR. At present, both techniques are more time consuming and less cost effective than RQ-PCR, but we envision that they will become more widely used in the future.

MRD negativity measured by RQ-PCR in peripheral blood (PB)⁵⁰  or BM¹⁰⁵  and deep reduction in NPM1 mutant transcripts (>3 log in BM,¹⁰⁶  to <1% in BM,¹⁰⁷  or >4 log in PB¹⁰⁸ ) at multiple time points after chemotherapy predict a low risk for leukemia relapse and good survival. Notably, MRD assessment in PB after 2 chemotherapy cycles appears to be even more accurate than genetic-based prognostication in predicting relapse risk.⁵⁰ 

Patients in first CR who remain MRD positive after consolidation therapy benefit from allo-HSCT.^108-110  However, this benefit may be only partial, because pretransplant MRD positivity predicts a poor outcome.^111,112  Future studies should assess whether converting pretransplant NPM1 MRD positivity to negativity is feasible and clinically meaningful. In our experience, the risk of posttransplant leukemia relapse is reduced by using myeloablative HLA-haploidentical transplantation with regulatory and conventional T-cell–adoptive therapy.¹¹³ 

In general, any confirmed increase in NPM1 transcripts in BM and/or PB reliably predicts hematological relapse (usually occurring within 3 months).^105,114,115  According to ELN recommendations,¹¹⁵  patients undergoing standard treatment should be tested by RQ-PCR at least at diagnosis, after 2 cycles of therapy, and at the end of treatment. MRD should be measured every 3 months for the first 2 years of follow-up. Then, timing of MRD monitoring should be personalized according to relapse risk. These recommendations are also applicable for the follow-up of patients undergoing allo-HSCT.

Patients in molecular relapse could benefit from preemptive therapy, including 5-azacitidine,¹¹⁶  venetoclax-based regimens,¹¹⁷  or immunotherapy.¹¹⁸  However, the value of preemptive treatment remains unclear.

Fit adult (≤60 years) patients with NPM1-mutated AML are treated with intensive chemotherapy. NPM1-mutated AML cells usually express high levels of CD33 that is treatable with gemtuzumab ozogamicin. Adding this agent to chemotherapy improved event-free survival and overall survival in 1 study,¹¹⁹  but it resulted in lower relapse incidence with no survival advantage in another study.¹²⁰  In addition to chemotherapy, NPM1-mutated AML patients should receive inhibitors of FLT3, if this gene is mutated.^92,121  Allo-HSCT is usually recommended for NPM1-mutated AML with FLT3-ITD^high.^88,122  Conversely, it is generally accepted that patients with NPM1-mutated AML without FLT3-ITD or with FLT3-ITD^low should not undergo allo-HSCT.^92,123-125  It remains to be investigated whether the small subset of patients with low transplant–related risk, HLA-identical donor,¹⁰⁹  and suboptimal reduction in NPM1 MRD after 2 chemotherapy cycles benefit from allo-HSCT in first CR.

Older fit patients (>60 years) with NPM1-mutated AML treated with intensive chemotherapy experience 15% to 20% long-term survival.^126,127  However, these results may be ameliorated using venetoclax combined with hypomethylating agents.¹²⁸  Eligible patients with NPM1-mutated AML also benefit from postremission reduced-intensity conditioning¹²⁹  or nonmyeloablative¹³⁰  allo-HSCT. Older unfit patients with NPM1-mutated AML rarely show long-term response to single hypomethylating agents.¹³¹  Combining venetoclax with hypomethylating agents or low-dose cytarabine is emerging as the standard of care in this setting, inducing CR in ∼90% of cases.^132,133 

Given the good CR rates, the National Cancer Research Institute recommends¹³⁴  that, during the COVID-19 pandemic, venetoclax-based regimens should be considered as an alternative to induction chemotherapy in NPM1-mutated AML patients, particularly those aged > 50 years with comorbidities.¹²⁸  Stringent molecular monitoring for NPM1 mutant transcripts is mandatory if such a strategy is adopted. This approach could help to bridge patients through the pandemic with the objective to deliver definitive therapy, including allo-HSCT, later on.

Because of the poor prognosis,⁸¹  we tend to treat patients with NPM1-mutated MDS and CMML following the same recommendations for NPM1-mutated AML; however, this issue remains controversial.

Nuclear export of NPM1 mutants is XPO1 dependent.²⁷  Thus, XPO1 inhibitors can redirect NPM1 mutants to the nucleus.^66,74,135  However, XPO1 inhibitors are not NPM1 specific, because they also inhibit the nuclear export of other NES-containing proteins (eg, TP53 and p21).^136,137  Consequently, their antileukemic activity may occur through more complex mechanisms. The selective XPO1 inhibitor selinexor shows low efficacy in AML patients,¹³⁸  probably because it can be administered only once or twice a week due to its toxicity.¹³⁸  This schedule does not ensure the stable XPO1 inhibition required to induce differentiation and growth arrest of NPM1-mutated AML cells¹³⁹  (L.B., unpublished data, December 2019). Because of its lower blood-brain barrier penetration, the new XPO1 inhibitor eltanexor¹⁴⁰  is better tolerated than selinexor and is currently administered up to 5 times per week in phase 1/2 studies (NCT02649790), holding promise for therapeutic success.

A functional HOX network is necessary to maintain the undifferentiated state of NPM1-mutated AML.^62,72,74  Therefore, targeting HOX expression may prove effective in this AML entity. The histone methyltransferase MLL1 and its cofactor menin induce the trimethylation of lysine 4 on histone 3, a histone mark that correlates with active transcription of HOX genes and their cofactor MEIS1 (HOX/MEIS). Reducing HOX/MEIS expression by inhibiting menin-MLL interaction is feasible and effective in NPM1-mutated AML.⁷²  Recently, 2 novel menin-MLL inhibitors (MI3454 and VTP-50469) exhibited striking antileukemic activity against NPM1-mutated AML in mice.^73,141  Intriguingly, both inhibitors preferentially downregulated the HOX cofactor MEIS1 without impairing the expression levels of HOX genes. We believe that loss of MEIS1 may interfere with the function of multiple HOX transcription factors, leading to differentiation and growth arrest.¹⁴²  Menin-MLL inhibition prevented the transformation of Npm1-mutated mouse hematopoietic progenitors into leukemic cells, implying that it could also be effective in Npm1-mutated preleukemia.⁷³  However, we believe that menin-MLL inhibition is more likely to benefit patients with frank NPM1-mutated AML, because NPM1 mutations do not drive preleukemic clonal hematopoiesis.

Because pharmacologic XPO1 inhibition also potently downregulates HOX expression in NPM1-mutated AML,⁷⁴  we envision synchronized inhibition of menin-MLL and XPO1 as an appealing therapeutic option to block HOX/MEIS expression in NPM1-mutated AML.

NPM1wt is a nucleolar stress sensor,¹⁴³  and the nucleolus of NPM1-mutated AML cells may be particularly vulnerable to stress because it is NPM1 depleted as a result of haploinsufficiency and cytoplasmic delocalization.⁷  Therefore, inducing nucleolar stress could be therapeutically effective in NPM1-mutated AML. Actinomycin D, which triggers nucleolar stress by inhibiting RNA polymerase I, induces CR in NPM1-mutated AML,¹⁴⁴  although the mechanism of action remains unclear. Other drugs causing nucleolar stress through inhibition of ribosome biogenesis¹⁴⁵  should be investigated in this setting. Blocking NPM1 oligomerization¹⁴⁶  represents another potential therapeutic approach, because pentameric/decameric NPM1wt is required for proper nucleolus formation.

We proposed for the first time that the unique C-terminal sequences of NPM1 mutants represent a potential immunotherapeutic target¹⁴⁷  and that cytoplasmic localization of NPM1 mutant may favor its processing by a class I degradation pathway, leading to HLA presentation.¹⁴⁷  This concept was further expanded demonstrating that specific autologous T-cell responses against NPM1 mutant peptides can be elicited.^148-151  Immune responses were associated with molecular remission in 1 AML patient,¹⁵²  and they may explain the relatively favorable outcome of NPM1-mutated AML.¹⁵³  Recently, the T-cell receptor specific for 1 of these peptides bound to HLA-02*01 was isolated; following its retroviral transduction into CD4⁺ and CD8⁺ T cells, it elicited an antileukemic effect against NPM1-mutated cells.¹⁴⁹  Thus, using engineered T cells against NPM1-mutated peptides bound to HLA represents a promising therapeutic approach.

NPM1 mutations are AML-driving events occurring in the setting of preleukemic clonal hematopoiesis. Evidence that eradication of NPM1-mutated clones results in AML cure^41-43  (although persistent clonal hematopoiesis may require close monitoring) should encourage the development of targeted treatments for this AML entity.

Endogenous tagging of wild-type and mutated NPM1⁷⁴  may help us to better understand the functions of these proteins, and mass spectrometry may assist in unraveling their interactome in the cytoplasm. Clarifying how NPM1 mutants deregulate the HOX program and how HOX transcription factors control their downstream targets may also lead to new therapeutic avenues. Finally, functional genomic studies may highlight the sensitivity of combinatorial mutational events to specific drugs (eg, unexpected in vitro sensitivity to ibrutinib¹⁵⁴  of NPM1-mutated AML cells carrying FLT3 and/or DNMT3A mutations).

Clinically, it should be clarified whether diagnosis of NPM1-mutated AML is justified, irrespective of blast percentage. Standardization of NPM1 MRD measurement, better definition of its impact at different time points, and its inclusion in the design of future clinical trials in young and elderly patients are warranted. Combining venetoclax with other drugs (eg, arsenic¹⁵⁵  or XPO1 inhibitors)¹⁵⁶  should be investigated further in NPM1-mutated AML.

The authors thank Ildo Nicoletti for providing the 3-dimensional reconstruction of the confocal images shown in Figure 3B and Timothy Ley (Washington University, St Louis, MO) for critically reading the manuscript and for providing Figure 4A. The authors apologize to those investigators whose papers could not be cited due to space limitations.

This work was supported by the Associazione Italiana per la Ricerca sul Cancro (AIRC IG 2019 number 23604 and AIRC Start-Up 2019 number 22895), the European Research Council (Advanced Grant 2016 number 740230 and Consolidator Grant 2016 number 725725), and the ARC Foundation for Cancer Research (Leopold Griffuel Prize to B.F.).

This review is dedicated to the memory of David Grimwade, who greatly contributed to establishing the clinical impact of MRD monitoring in NPM1-mutated AML.

Contribution: B.F. coordinated the preparation of the review, and all authors performed experiments related to the review and wrote the manuscript.

Conflict-of-interest disclosure: B.F. holds a patent on NPM1 mutants (number 102004901256449). The remaining authors declare no competing financial interests.

Correspondence: Brunangelo Falini, Institute of Hematology and Centro Ricerche Emato-Oncologiche, University of Perugia, Piazzale Menghini, 9, Perugia 06131, Italy; e-mail: brunangelo.falini@unipg.it.