NPM1-mutated AML: how many diseases?

In this issue of Blood, Othman et al report on the genetic landscape of 1098 intensively treated patients with acute myeloid leukemia (AML) and harboring a nucleophosmin 1 (NPM1) mutation. This data set, the largest to date for this entity, provides important clarifications on the prognostic value of recurrent comutations in NPM1-mutated AML.¹ 

NPM1 mutations are found in ∼30% of AMLs and define an entity with a distinct transcriptome and phenotype.² NPM1-mutated AMLs are globally chemosensitive, with complete remission (CR) rates nearing 90% but with heterogeneity in relapse rates. Across all ages, ∼50% of patients with an NPM1 mutation who are intensively treated will relapse. Identifying those patients with an NPM1-mutated AML who would benefit from allogeneic hematopoietic cell transplantation (allo-HCT) in CR1 has thus long been a matter of investigation.

In compiling FLT3-ITD and measurable residual disease (MRD) data of patients with the NPM1 mutation from 2 large National Cancer Research Institute trials of intensive chemotherapy (both conducted before the availability of midostaurine), Othman et al recently reported that peripheral blood NPM1 MRD assessment after 2 cycles of therapy captured the prognostic impact of FLT3-ITDs. MRD positivity is therefore sufficient to guide allo-HCT indications in this population, irrespective of FLT3-ITD status.³ 

In their current study, Othman et al inspect the prognostic role of comutations beyond FLT3-ITDs through targeted sequencing of a 36-gene panel in 1098 patients with the NPM1 mutation from the previously noted trials, including 644 patients with MRD data. The size of this data set first allowed the authors to confirm the presumed order of mutation acquisition in NPM1-mutated AML, with preleukemic hits such as DNMT3A mutations, and secondary mutations, mostly in signaling genes. Bulk sequencing precluded formal resolution of clonal architecture at the individual level. It could not be addressed whether different clonal architectures lead to different phenotypes and outcomes. For instance, this is relevant for DNMT3A, where R882 and frameshift mutations have been suggested to be secondary or preleukemic hits, respectively.⁴ Othman et al found the patterns of comutations to be nonrandom, and, although the number of recurrently mutated genes was too small to perform a bona fide clustering analysis, 2 archetypes of NPM1-mutated AMLs emerged from their analysis. The first is enriched in DNMT3A, WT1, and FLT3-ITD mutations, whereas the second harbors other signaling mutations such as FLT3-TKD or genes in the RAS pathway. Orthogonal approaches conducted using the large HARMONY database, so far reported in abstract form, produced partially overlapping results.⁵ 

Previous work has shown that patients with the “triple mutation” of NPM1, FLT3-ITD, and DNMT3A had worse prognosis.⁶ Othman et al confirm this finding and extend it to patients harboring WT1 instead of DNMT3A combined with NPM1 and FLT3-ITD mutations. In fact, patients with the NPM1 mutation who harbor mutations in any 2 of DNMT3A, WT1, or FLT3-ITD, despite high initial CR rates, had higher MRD and shorter survival. Not only was this 3-gene classifier validated in an independent cohort, but the authors could also show that this poorer prognosis was not abrogated by allo-HCT, thereby identifying such “higher risk” patients with NPM1-mutated AML as having an unmet medical need (see figure). It is currently unknown whether addition of menin inhibitors can improve patient outcomes, but this simple 3-gene classifier should be considered when stratifying those patients with NPM1-mutated AML who are randomized to future trials testing this emerging drug class. Interestingly, NPM1/DNMT3/FLT3-ITD triple-mutated AMLs were shown to have a distinct biology characterized by high leukemia stem cell (LSC) frequency.⁷ It remains unknown whether such LSC expansion is also present in other higher risk NPM1-mutated AMLs. However, these findings, along with previous work,⁸ suggest that LSC frequency may prove to be the most biologically relevant, if not most robust, diagnostic biomarker of high-risk NPM1-mutated AML. Further work is needed to determine whether this higher risk will remain true with up-to-date treatment strategies, including validated FLT3 inhibitors, maintenance therapy, and monitoring of molecular relapse. It will also be of interest to identify whether this classification is applicable to patients treated less intensively. Future studies will also need to compare the prognostic accuracy of this classifier, with genetic scores integrating variant allele frequency data and/or relying on machine learning, or to strategies integrating flow cytometry or gene expression to assess leukemia stemness.^5,8,9 

One could wonder about the importance of these findings given the prominent role of NPM1 MRD for risk stratification. First, upfront identification of higher risk patients is relevant to tailor induction regimens to specific patient populations. Othman et al describe how higher risk patients clearly, if not specifically, benefited from a regimen based on fludarabine + cytosine arabinoside + granulocyte colony-stimulating factor + idarubicin, as well as from addition of gemtuzumab ozogamycin. They also show that, in contrast to FLT3-ITD considered in isolation, MRD positivity does not fully capture the higher risk of relapse and mortality associated with these genotypes. Patients with these genotypes remain at a weakly but significantly higher risk even after achieving MRD negativity, perhaps because of relapse with wild-type NPM1. Future studies will need to address the potential benefits of additional MRD modalities (eg, flow or next-generation sequencing) on top of NPM1 transcript for these patients.

A final, notable observation by Othman et al involves the differences in presentation, comutation profile, and outcome between the different NPM1 mutations. Rare non-ABD NPM1 mutations were found to be associated with significantly shorter overall survival, independent of comutations and clinical confounders. Both frequent (types A, B, and D) and rare non-ABD NPM1 mutations are thought to converge functionally on the generation of a strong nuclear export signal recruiting the exportin XPO1/CRM1, not only for nuclear export to the cytoplasm but also for creating chromatin condensates with gene-regulating potential.¹⁰ Following on previous contradictory reports with smaller cohorts, these findings raise the so-far speculative possibility that different NPM1 mutations may lead to subtle functional differences with variable oncogenicity and chemoresistance. The fact that the less-recurrent (non-ABD) mutations are also those with potentially higher chemoresistance is a testimony to the fact that different selection pressures shape the clonal architecture of AML before and after exposure to therapy.

A valuable resource for the AML community, the meticulous study by Othman et al leaves us with more new questions to address than definitive answers to old questions. Perhaps its greatest value is to remind us that important clinical information and stimulating biological hypotheses can be garnered from large-scale bulk genomic studies in AML, even in the “single-cell” era.

Conflict-of-interest disclosure: R.I. declares no competing financial interests.