Strategies for identifying NUP98 rearrangements in adult myeloid neoplasms

Abstract

Nucleoporin 98 rearrangements (NUP98r) are recurrent in myeloid neoplasms and are subtype-defining for acute myeloid leukemia (AML) in the World Health Organization Classification 5th edition (WHO5) and the International Consensus Classification (ICC). Identification of NUP98r is essential given the frequency of treatment resistance and possibility of sensitivity to targeted therapies. However, NUP98r is often cryptic on karyotype and has over 40 described partners. Therefore, it is underdiagnosed in the absence of dedicated testing that is not always routine practice, e.g., RNA-based next generation sequencing (NGS), NUP98 break-apart fluorescence in situ hybridization, or real-time-quantitative polymerase chain reaction for specific NUP98 fusions. Historically, AML with NUP98r has received the most attention in pediatric AML, where its incidence is highest, but has been increasingly characterized in adult AML. By contrast, the incidence and behavior of NUP98 fusions in myelodysplastic syndromes (MDS) is less understood and based predominantly on case reports. In this study, we describe our adult institutional experience with a clinically validated anchored multiplex PCR RNA-based targeted NGS assay, explore strategies for rational use of specific testing for NUP98r including a proof-of-principle based on WT1 and FLT3- ITD mutational status, and integrate our results with a review of the literature. In total, we identified 3 MDS and 15 AML patients with NUP98r as the genetic driver, including two novel fusion partners (FGF14 and LAMC3), thus highlighting the utility of NGS testing to detect NUP98 fusions. Recognition of NUP98r in myeloid neoplasms is crucial for accurate diagnosis and prognosis, with significant implications for therapy or enrollment in clinical trials.

Introduction

Acute myeloid leukemia (AML) with nucleoporin 98 rearrangement (NUP98r) comprises one of several genetically defined AML subtypes that have been newly incorporated into both the 2022 International Consensus Classification (ICC) and the 5^th edition of the World Health Organization Classification (WHO5), where a diagnosis of AML can be made with a blast count under 20%.^1,2 AML with NUP98r has historically been associated with adverse clinical outcomes and chemotherapy resistance although recent preclinical models have raised the possibility of rational targeted therapy with menin inhibitors.³ Thus, routine identification of NUP98 rearrangement is important for clinical care of patients with AML and future improvement of risk-adapted therapy. However, the entity is underdiagnosed by many clinical practices since the rearrangements are frequently cryptic on conventional karyotype due to the location of NUP98 at 11p15.4 near the terminal end of the short arm of chromosome 11. The NUP98 gene, which encodes a component of the nuclear pore complex, rearranges with over 40 unique fusion partners, all involving the N-terminal end of NUP98 and notable for partner-specific enrichments for monocytic, myelomonocytic, megakaryoblastic, or erythroid differentiation.⁴ Accordingly, reliable detection across the entire spectrum of fusion partners requires complex testing modalities such as whole transcriptome RNA-seq, targeted RNA-based next generation sequencing (NGS) fusion assays with coverage of NUP98 rearrangements, optical genome mapping (OGM), whole genome sequencing (WGS), or fluorescence in situ hybridization (FISH) break-apart probes for NUP98. Of these, NUP98 FISH is cheapest and fastest but does not identify the specific NUP98 partner and would not detect other cryptic rearrangements that are potential drivers in the absence of NUP98r. By contrast, WGS is the most comprehensive and has started to become adopted clinically, with accurate risk categorization for AML and myelodysplastic syndromes (MDS) including reliable detection of NUP98r.^5,6

In recent years, RNA-based NGS fusion assays have enabled estimates of the prevalence of NUP98r in AML, ranging from 7.2-8.0% of pediatric AML and 2.5-5.0% of adult AML.^6-12 NGS has also revealed distinct co-mutational patterns, including enrichment of FLT3-ITD and WT1 variants, particularly in AML with NUP98::NSD1. NUP98r has been less studied in myeloid neoplasms outside of AML. Although presumed to be exceedingly rare, their frequency may be underappreciated.^13-15 Mouse models of NUP98::NSD1 have generated conflicting data, with one study showing almost universal transformation to AML, compared to other studies indicating a weak leukemogenic potential alone, but increased when combined with FLT3-ITD.^16-18 The presence of a chronic or pre-leukemic phase of NUP98r AML may be clinically relevant for the possibility of earlier detection and intervention.

Here, we report morphological, clinical, and molecular findings at our institution of adults with NUP98r AML or MDS, and explore features in our data and in public datasets which could prompt specific testing for NUP98r, potentially providing the basis for cost-effective strategies in clinical practices that do not employ screening for NUP98r. In particular, a myelomonocytic morphology and immunophenotype, WT1 mutations in MDS, and concurrent FLT3-ITD and WT1 mutations in AML, in the absence of another subtype defining genetic aberration (e.g., NPM1), highly enrich for myeloid neoplasms with NUP98::NSD1 or occasionally other NUP98 rearrangement partners.

Methods

Nucleic acid extracted from blood, bone marrow (BM), or extramedullary disease sites was tested by one or more of several NGS assays: 1) a clinically validated targeted RNA assay (Heme Fusion Assay [HFA]; Integrated DNA Technologies) designed principally to detect fusions through anchored multiplex polymerase chain reaction (PCR) (AMP)¹⁹ and performed on clinical samples as part of patient care from 2017-2024 (N=381; HFA clinical cohort) and on research samples for this study (N=7) at the Center for Integrated Diagnostics at Massachusetts General Hospital; 2) a clinically validated targeted DNA panel (Rapid Heme Panel [RHP*] version 3)²⁰ based on NEBNextDirect (New England BioLabs) to detect single nucleotide variants, small indels, and copy number alterations, and performed on clinical samples as part of patient care from 2019-2024 (N=21209; RHP cohort) at the Center for Advanced Molecular Diagnostics at Brigham and Women’s Hospital; and 3) total RNA sequencing performed on research samples for this study (N=2) at the Dana-Farber Cancer Institute molecular biology core facility. The research RNA-based NGS testing was performed on nucleic acid extracted from archived cytogenetic pellets. Data were processed by default clinical pipelines (HFA, RHP) or by a custom pipeline (total RNA-seq) using adapter trimming by BBDuk, alignment to hg19 by bwa-mem, and manual analysis of bam files. NUP98 break-apart FISH (Empire Genomics; 11p15.4) was performed on 100 interphase nuclei. Overall survival (determined from the date of first diagnosis to death from any cause) was assessed using the Kaplan-Meier method. Public RNA sequencing FASTQ files were downloaded from the Sequence Read Archive (www.ncbi.nlm.nih.gov/sra) for 2 MDS datasets (SRP149374, SRP418365) and aligned to hg19 by bwa-mem.^21,22 Alignments were analyzed by: i) searching for select fusions via grep (restricted to alignments to partner gene regions) for exon-exon junctional sequences (30 nucleotides consisting of 15 from each exon) and their reverse complements across all possible exon combinations producing the fusion or its reciprocal as previously described,²³ followed by manual confirmation of hits; ii) outlier isoform analysis for aberrant expression of isoforms in select genes (KMT2A, UBTF) as previously described,²⁴; and iii) custom variant detection based on pileup data across padded coding sequence of the WT1 gene. Annotations (mutations, fusions, cytogenetics, diagnoses) for the IPSS-M MDS cohort and the Leucegene AML cohort were retrieved from previously published data.³⁰ The study was conducted in accordance with the Declaration of Helsinki and with the approval of the institutional review boards at the Dana-Farber Cancer Institute and Massachusetts General Brigham.

Results

NUP98 rearrangements are effectively detected by anchored multiplex polymerase chain reaction-based targeted RNA sequencing, revealing novel partners and a potential enrichment in high-risk myelodysplastic syndromes

NUP98r was identified through the targeted RNA sequencing HFA assay in 18 patients overall (Table 1 and Online Supplementary Figure S1), with diagnoses of AML (N=14), MDS (N=3), or B/myeloid mixed phenotype acute leukemia (MPAL) (N=1) at the time of initial NUP98r detection. Eight different partner genes were observed, comprising 6 established (DDX10, HOXD13, KDM5A, NSD1, PRRX2, TNRC18) and 2 novel (FGF14, LAMC3) partners (Online Supplementary Table S1). The novel partners both contained domains that form a coiled coil structure, similar to many other non-HOX NUP98 partners.³⁰ NSD1 was the most frequent partner (11/18 patients; 61.1%), with single occurrences of other partners. Fusions involving 6/8 (75%) partners (FGF14, KDM5A, LAMC3, NSD1, PRRX2, TNRC18) from 15/17 (88.2%) evaluated patients were cytogenetically cryptic on conventional karyotyping (Table 1), highlighting the utility of RNA-based NGS for comprehensive detection of NUP98r. Only one case (MPAL_1) had NUP98 FISH performed during initial clinical workup. Nine NUP98r cases (50%) had a normal karyotype. Ten of the 15 (66.7%) AML/MPAL cases showed myelomonocytic or monocytic differentiation by morphology and flow cytometry (Table 1, Online Supplementary Figure S2), including 7/9 (77.8%) with NUP98::NSD1 and 3/6 (50%) with non-NSD1 partners (TNRC18, DDX10, PRRX2). No other subtype-defining alteration was identified by any testing modality in any of the NUP98r cases, including those with novel fusion partners. KMT2A-PTD was also absent from NUP98r cases, akin to mutual exclusivity reported in pediatric AML; this alteration in AML has been associated with aberrant HOXB expression, similar to AML with NUP98::NSD1, mutated NPM1, DEK::NUP214, and UBTF-TD, although it is not subtype-defining.^6,31,32

Table 1.Summary of patient characteristics.

The majority of NUP98 fusions were detected through clinical testing during the course of patient care, with clinically reported NUP98r in samples from 14 patients across the HFA clinical cohort, including 11/257 (4.3%) of all newly diagnosed AML patients treated at one institution (Massachusetts General Hospital) between 2017-2024 (Online Supplementary Figure S1). The other 3 clinically identified NUP98r patients accounted for 2/46 (4.3%) MDS and 1/3 (33.3%) MPAL from the HFA cohort; however, these test populations were subject to selection bias, particularly given the lack of clear guidelines on which MDS cases should be tested by an RNA fusion assay. Most MDS cases that were tested had high-risk features and clinical concern for AML, particularly elevated blast counts (28/46, 60.9%) and/or high to very high International Prognostic Scoring System (IP-SS)-Molecular scores (29/46, 63.0%) (Online Supplementary Table S2). No cases tested clinically by HFA from patients diagnosed with MDS/myeloproliferative neoplasm (MPN) overlap (0/24), MPN (0/49), or another myeloid neoplasm (0/5) demonstrated a NUP98r.

WT1 mutations are recurrently observed with and without FLT3-ITD in acute myeloid leukemia with NUP98::NSD1 and occur in myeloid neoplasms with other NUP98 rearrangements

We investigated co-mutational profiles of the 14 NUP98r cases from the HFA cohort, as characterized by targeted DNA-based NGS testing (RHP). Consistent with prior studies, the most common co-occurring mutations at diagnosis were: i) FLT3-ITD in 6/14 cases (42.9%; all harboring NUP98::NSD1) with variant allele frequencies (VAF) ranging from <1% to 25%; and ii) WT1 in 6/14 cases (42.9%; 5 with NUP98::NSD1 and 1 with NUP98::LAMC3) with VAF ranging from 5.5% to 47.2%, where 4/14 (28.6%) had multiple (2-4) WT1 mutations (Figure 1, Online Supplementary Table S3). All 6 cases with WT1 mutations harbored one or more frameshift variants for a total of 12 frameshifts (vs. 1 nonsense) including frameshift insertions in all cases. Concurrent FLT3-ITD and WT1 mutations were seen in 3/14 cases (21.4%). Mutations in RUNX1, MYC, TET2, KRAS, and PTPN11 were also each seen more than once (Figure 1). Myelodysplasia-related (MR) gene mutations were not identified, except in RUNX1, which is considered an MR gene by the ICC but not by the WHO5. FLT3-ITDs were detected exclusively in AML cases.^16,33 By contrast, 1/2 (50%) MDS cases demonstrated multiple WT1 mutations up to 40% VAF. WT1 mutations also developed after the initial diagnosis in 3 cases (AML_5, AML_6, AML_7) with fusions to TNRC18, NSD1, and DDX10, again with one or more frameshifts in every case. Thus overall, 9/14 (64.3%) cases presented with or developed a WT1 mutation at various stages of their clonal hierarchies.

Figure 1.Co-mutation plot of the Heme Fusion Assay and Rapid Heme Panel (RHP) cohorts at initial diagnostic bone marrow biopsy. Pathogenic mutations identified in each NUP98r case are represented by blue boxes. If multiple mutations occur in the same gene, the number of concurrent mutations is indicated within the blue box. RHP: Rapid Heme panel. ^*RHP cohort. ^†Novel fusion partner.

Mutational status of WT1 identifies candidate cases for selective testing for NUP98r in myelodysplastic syndromes

We explored whether WT1 mutations detected by up-front DNA-based NGS testing of MDS, in the absence of potential subtype-defining molecular or cytogenetic features determined from routine workups, might provide a rational strategy for initiation of HFA testing with the goal of detecting rare NUP98r cases. As a proof of principle, we interrogated DNA-based NGS testing across all MDS cases of the HFA cohort (N=45 with DNA NGS), revealing WT1 mutations in 4/45 (8.9%) cases: an NPM1-mutated case by RHP, a MECOM-rearranged (MECOMr) case by cy-togenetic studies, a case with KMT2A-PTD by RHP, and the NUP98r case described earlier that was cryptic on karyotype and characterized subsequently by HFA. Under a hypothetical tiered approach to MDS evaluation, the latter 2 cases would be candidates for dedicated NUP98r testing, whereas the first 2 cases were already characterized by their initial workups. However, NUP98r testing of KMT2A-PTD cases may have limited yield, given the mutual exclusivity of KMT2A-PTD with NUP98r (and most other molecular subtypes) reported in pediatric AML and the similar absence of KMT2A-PTD from the NUP98r HFA cohort. Therefore, the hypothetical yield of the proposed tiered strategy would be 1/1 (100%) if KMT2A-PTD cases were deliberately excluded from further testing or 1/2 (50%) otherwise or if a clinical practice did not include detection of KMT2A-PTD as part of their initial workup. Of note, one NUP98r MDS case from the HFA cohort would have been missed by this strategy, since that case harbored only a DNMT3A mutation, which is widely mutated across myeloid neoplasms and thus not amenable to a molecular strategy for rationed testing.

The above WT1 mutational rate (8.9%; 4/45) and hypothetical HFA testing rate (2.2%; 1/45 after also excluding KMT2A-PTD) reflected a high-risk MDS cohort. As a broader estimate of testing rate using the public IPSS-M dataset, 37/2591 (1.4%) of its MDS (WHO4) cases harbored pathogenic WT1 mutations with an enrichment in higher risk subtypes (MDS-EB2: 17/438 [3.9%]; MDS-EB1: 10/464 [2.2%]) and subsequent contribution to criteria for a proposed AML-like group of MDS.^25,34 Of these 37 WT1-mutated MDS cases, 14 showed key alterations considered mutually exclusive with NUP98r: mutated NPM1 (N=6), KMT2A-PTD (N=4), biallelic TP53 mutations (N=1), biallelic DDX41 mutations (N=1), t(6;9)(p23;q34)/DEK::NUP214 (N=1), and t(3;21) (q26;q22)/RUNX1::MECOM (N=1). After their exclusion, hypothetical HFA testing might then apply to 23/2591 (0.9%) MDS cases. These 23 cases demonstrated either single missense variants (N=8 cases), single nonsense variants (N=3), single splice variants (N=2), or purely frameshifts (8 insertions only, 1 deletion only, 1 both), thus an alternative minimalistic strategy might test only the 9/2591 (0.3%) harboring an insertion frameshift. The IPSS-M cohort, however, did not have NUP98r status (or RNA sequencing data) to measure testing yield.

We attempted to further validate the WT1-based strategy by applying it to the RHP cohort (Figure 2A). Screening yielded 17 adult patients with MDS harboring WT1 mutations, of which 7 were found to have an alteration considered mutually exclusive with NUP98r by either RHP or kayotype: NPM1 mutation (N=5), MECOM-r (N=1), or TP53 multi-hit (N=1) (Figure 2A). Cases from 6 additional patients harbored KMT2A-PTD, which we decided not to test further for the reasons discussed earlier. Of the remaining 4 patients, 2 had already undergone testing within the HFA clinical cohort, with one positive for NUP98r and one negative for any fusions by HFA. The final 2 patients underwent retrospective HFA testing for this research study, with one positive for NUP98r and one failing sequencing quality control metrics. Thus, the overall HFA yield for NUP98r within WT1-mutated MDS without a key driver (including KMT2A-PTD) was 2/3 (66.7%) in this limited dataset.

We also reanalyzed public RNA sequencing data from two adult MDS cohorts, resulting in detection of NUP98r in 4 more MDS cases involving 2 different partner genes (3 cases with NUP98::NSD1, 1 with NUP98::HOXA9), for cohort frequencies of 2/215 (0.9% in SRP418365) and 2/109 (1.8% in SRP149374) (Online Supplementary Table S4). No evidence was found for NUP98 fusions involving 22 other partner genes that have been reported previously in MDS, CMML, or AML across predominantly adult studies (Tables 2, 3) (DDX10, EMX1, FGF14, HHEX, HMGB3, HOXA11, HOXA13, HOXC13, HOXD12, HOXD13, KAT7, KDM5A, LNP1, NSD3, PHF23, PRRX1, PRRX2, PSIP1, RAP1GDS1, TLX1, TNRC18, TOP1). We lacked knowledge of WT1 mutational status on the DNA level to fully evaluate the WT1-based strategy. As a proxy, we instead screened the RNA-seq data for expressed WT1 loss-of-function (LOF) mutations (frameshift, nonsense, or splice site) while adopting an approach prioritizing sensitivity in order to partially offset inherent limitations posed by RNA, including variably low WT1 expression, nonsense mediated decay of mutant RNA transcripts, splicing mutations that may not appear within mature RNA transcripts (e.g., by conferring exon skipping rather than intron retention), and shallow sequencing coverage. Our analysis detected expressed WT1 LOF mutations in ¾ (75%) NUP98r cases (3/3 NUP98::NSD1; 0/1 NUP98::HOXA9) and 15/324 (4.6%) MDS cases overall, with individual cohort frequencies of 5/215 (2.3% in SRP418365) and 10/109 (9.2% in SRP149374) (Online Supplementary Table S5). The higher frequency of the latter cohort was hypothesized to be a consequence of CD34⁺ enrichment and was associated with higher expressed VAF; however, the possibility of a component of false positives also existed. Outlier isoform analysis demonstrated 2 KMT2A-PTD and one UBTF-PTD within non-NUP98r cases expressing WT1 LOF mutations. After excluding KMT2A-PTD cases, the hypothetical yield of the WT1-based strategy was 1/3 (33.3% in SRP418365) and 2/10 (20% in SRP149374) but potentially could be greater, given the lack of annotations (e.g., cytogenetics) and the possibility of additional findings upon standard workups (e.g., MECOMr). Of note, all 3 NUP98::NSD1 cases exhibited WT1 frameshift insertions, similar to cases in our local cohort. Thus, the alternative strategy of using WT1 frameshift insertions might largely maintain sensitivity while increasing specificity, with a hypothetical yield across both public cohorts of 3/4 (75%) for NUP98r, where the non-NUP98r case harbored UBTF-TD and would also benefit from subtyping.

Mutational status of WT1/FLT3-ITD identifies candidate cases for selective testing for NUP98r in acute myeloid leukemia

In AML, RNA-based NGS testing was a part of routine work-ups in some but not all our local institutional practices during the study period. As access to testing expands, it is likely that clinical practices will increasingly adopt either universal upfront RNA-based NGS or a tiered approach with reflex testing of all AML cases that do not have a characterized subtype after routine workup. However, in a setting of limited resources with goals of maximizing positive predictive value, we explored the utility of WT1/FLT3-ITD dual mutations in the absence of subtype-defining genetic features (by RHP or cytogenetics) as another potential rational strategy in AML for initiation of HFA testing. Screening of the RHP cohort identified 41 adult patients with AML harboring both a WT1 mutation and FLT3-ITD, of which 31 were found to have a genetic abnormality considered mutually exclusive with NUP98r, including NPM1 (N=15), KMT2A-PTD (N=7), PML::RARA (N=6), DEK::NUP214 (N=1), MECOMr (N=1), and CEBPA bZIP domain mutation (N=1) (Figure 2B), of which all but KMT2A-PTD further enabled AML classification by the ICC or the WHO5. Of the remaining patients, 3 had al ready undergone testing within the HFA clinical cohort, with one positive for NUP98::NSD1 and 2 negative for fusions. The final 7 patients included 5 with available material for retrospective HFA testing for this research study, yielding detection of NUP98r in 4/5 (80%) cases (Online Supplementary Figure S1). One of the 4 patients with successful confirmation of NUP98r at the AML stage also had NUP98r detected at an earlier MDS stage through the WT1 screen. This case (MDS_3) progressed from MDS-MLD to MDS-IB2 together with rising peripheral blasts and relapsed quickly after transplant as AML, with emergence and outgrowth of FLT3-ITD across the serial samples (Figure 3A, B). FISH analysis performed for this study supported the early clonal nature of the NUP98r at the initial MDS timepoint (Figure 3C). Thus, the overall HFA yield for NUP98r within FLT3-ITD⁺/WT1⁺ AML without a mutually exclusive molecular alteration by RHP or karyotype was 5/8 (62.5%). To further characterize the 3 cases which remained unresolved after HFA, 2 had available material for total RNA-sequencing, revealing UBTF-TD in both.

Figure 2.Mutually exclusive genetic alterations in myelodysplastic syndromes with WT1 mutations and acute myeloid leukemia with WT1 and FLT3-ITD mutations. (A) Adult myelodysplastic syndrome (MDS) cases with WT1 mutation in the Rapid Heme Panel database (N=17), labeled with detected genetic driver. NOS: no genetic driver detected. *1 case sample with no known genetic driver failed QC metrics on Heme Fusion Assay. (B) Adult acute myeloid leukemia (AML) cases with WT1 mutation and FLT3-ITD in the Rapid Heme Panel database (N=41), labeled with detected genetic driver. AML-MR: myelodysplasia-related, by the International Consensus Classification (ICC); NOS: no genetic driver detected. Of note, 2 of the 3 AML-NOS cases subsequently underwent total RNA-sequencing, and both were found to harbor UBTF-T D.

Table 2.Acute myeloid Leukemia Literature review, aduLt onLy.

Table 3.MDS and MDS/MPN Literature review, adult and pediatric.

Figure 3.MDS_3 progression to acute myeloid leukemia. (A) Summary of aspirate blast count and mutations present in bone marrow (BM) biopsies for this patient at the time of myelodysplastic syndrome with multilineage dysplasia (MDS-MLD), MDS with excess blasts/increased blasts 2 (MDS-IB2), and acute myeloid leukemia (AML) diagnoses. (B) Representative histology of BM cores from MDS-IB2 and final AML biopsies (HE, 60x magnification). (C) Fluorescence in situ hybridization (FISH) testing on the MDS-IB2 biopsy using NUP98 (11p15.4) break-apart probe; two representative interphase nuclei each showing one intact NUP98 signal and one split 3’ green signal (t) and 5’ red signal (c), indicating a rearrangement. HFA: Heme Fusion Assay.

Examination of the well-characterized Leucegene AML cohort (n=452) demonstrated similar findings (Online Supplementary Table S6). Out of 17 AML cases positive for both FLT3-ITD and WT1 mutations, 15 harbored a genetic alteration considered mutually exclusive with NUP98r, again with NPM1 mutations (N=8) and KMT2A-PTD (N=5) as the most common, along with PML::RARA (N=1) and classic biallelic CEBPA mutations (N=1). Of the 2 remaining cases, one harbored NUP98::NSD1 while the other could be considered AML-MR. Thus, the hypothetical yield of a WT1/FLT3-ITD strategy would be 1/2 (50%) if KMT2A-PTD status is determined up front. The Leucegene cohort also contained 16 additional AML cases with WT1 mutations but lacking FLT3-ITD, of which 11 harbored a genetic alteration considered mutually exclusive with NUP98r, including PML::RARA (N=6), NPM1 mutations (N=2), classic biallelic CEPBA mutations (N=1), RUNX1::RUNX1T1 (N=1), and KM-T2A::AFDN (N=1). After their exclusion, 5 cases remained, with 2 harboring NUP98::NSD1, 2 potential AML-MR, and one AML-NOS. Thus, the hypothetical yield of a WT1-based strategy regardless of FLT3-ITD status would be 3/7 (42.9%) if KMT2A-PTD cases are excluded.

NUP98r myeloid neoplasms have an aggressive clinical course with poor outcomes even after stem cell transplantation

In our two cohorts, 11 of the 14 AML patients plus the one patient with MPAL were initially treated with induction chemotherapy (daunorubicin plus cytarabine [7+3] or vincristine, doxorubicin, methotrexate, plus cytarabine). Three of 14 patients with AML and all 3 MDS patients received hypomethylating agent (HMA)-based therapy with decitabine and venetoclax or with decitabine alone. One patient with AML died two days after starting 7+3 induction chemotherapy, and 2 patients failed to achieve remission, while the other 15 proceeded to hematopoietic stem cell transplant (SCT) in first complete remission (CR1). Post transplant relapse was seen in 60.0% (9/15) of patients transplanted in CR1, including MDS_3 who relapsed with AML (Figure 3). Of the remaining patients, 5 are in remission at 26 days, 42 days, 9.4 months, 19.6 months, and 21.3 months after SCT, and one has achieved sustained remission (98 months) after a second SCT (Table 1). In the HFA cohort, 8 patients died with a mean overall survival (OS) of 14 months. The median OS of the RHP cohort was 12 months (Figure 4).

Discussion

NUP98r is a rare genetic finding that is AML-defining in new classification systems but prone to under-detection without dedicated or complex testing. It portends a poor prognosis and likely requires dedicated therapeutic approaches. Here, utilizing a clinically validated targeted RNA sequencing approach, we studied the frequency of NUP98r in myeloid neoplasms in adult patients at two large academic centers and found 18 cases overall, including 11/257 (4.3%) of all newly diagnosed AML patients treated at one institution. In doing so, we also detected NUP98r in patients with MDS, uncovered novel NUP98 fusion partners, and identified frequent co-mutations which could be leveraged to prompt dedicated testing for NUP98r.

In our review of the literature, less than 200 adult NUP98r with AML (Table 2) and far fewer adult NUP98r cases with other myeloid diagnoses (14 MDS, 4 CMML) (Table 3) have been described to date. The frequency of NUP98r in MDS is difficult to estimate precisely, given the lack of large comprehensive studies (Table 3). The most applicable study tested 101 consecutive adult MDS patients at a single institution by OGM, resulting in detection of NUP98r in 2/101 (2.0%) cases (1 NUP98::NSD1, 1 NUP98::PRRX2). Similarly, our reanalysis of public RNA-sequencing data from 2 adult MDS cohorts revealed NUP98r in 2/215 (0.9%) and 2/109 (1.8%) patients (3 NUP98::NSD1, 1 NUP98::HOXA9). Our study of the HFA clinical cohort revealed NUP98r at a slightly greater incidence in 2/46 (4.3%) adult MDS patients (1 NUP98::NSD1, 1 NUP98::FGF14). However, this cohort was subject to non-universal testing patterns and enriched for high-risk MDS. We also identified another high-risk MDS case with NUP98::NSD1 through our dedicated strategies. Although relatively small, these studies suggest that this genetic aberration may be more common in MDS than previously thought, particularly in high-risk patients. Of note, an older study testing only for NUP98::NSD1 by RT-PCR detected no cases out of 193 MDS patients.³⁵ Finally, since NUP98r is AML-defining in both the ICC (if >10% blasts) and the WHO5 (if >5% BM / >2% blood blasts) when with increased blasts, the 2/3 of the NUP98r MDS reported in the literature with at least 5% blasts would now be diagnosed as AML. Therefore, screening of MDS cases will also be important to identify cases that are actually AML, if they meet the blast criteria and have NUP98r.

Only 2 NUP98r cases in our cohort were recognized by karyotype (11.2% of 17 evaluable) (Table 1), highlighting the need for testing beyond conventional cytogenetics. Indeed, the most common NUP98r gene partners in adult and pediatric AML (NSD1 and KDM5A) are well-known to produce karyotypically cryptic fusions. Moreover, a substantial proportion of uncommon NUP98r gene partners may similarly generate cryptic fusions according to recent comprehensive studies of adult AML enabled by RNA-based NGS. The largest such study reported 4 uncommon partners that were always cryptic by karyotype (HMGB3, KMT2A, PSIP1, and TNRC18), 4 that were never karyotypically cryptic (HOXA9, TOP1, DDX10, and HHEX), and one that was variably cryptic (PRRX2).¹⁰ In our study, uncommon fusions involving both novel partners (FGF14 and LAMC3) as well as TNRC18 were karyotypically undetectable, versus 2 (DDX10 and HOXD13) that were characterizable. By contrast, in a study of pediatric AML with NUP98r, uncommon partners were mostly detectable by conventional karyotype G-banding.⁷

Figure 4.Overall survival curve of the Heme Fusion Assay clinical cohort and the Rapid Heme Panel cohort. Mean overall survival was 14 months and 12 months, respectively.

Until universal screening for NUP98r becomes widely adopted as part of routine workups, strategies for rational test utilization are critical to ensure accurate detection of this entity. These strategies may be particularly beneficial for MDS, where guidelines for RNA-based NGS testing are lacking. We propose a tiered approach in the absence of universal screening and show here that it is possible to identify a subset of cases with high likelihood of NUP98r based on results from standard molecular testing. Specifically, MDS with WT1 mutations and AML with FLT3-ITD and WT1 co-mutations are enriched for NUP98r and thus represent candidates for follow-up dedicated testing in the absence of AML subtype defining alterations and KMT2A-PTD. These cases alternatively could harbor UBTF-TD, another highrisk alteration that is more common in pediatric MDS/AML but also occurs rarely in adults, including the 2 AML-NOS cases in our FLT3-ITD⁺/WT1⁺ AML cohort and a WT1⁺ MDS case from the public RNA sequencing data; of note, most UBTF-TD should eventually be detectable during up-front testing by adding UBTF exon 13 to DNA-based panels. Further development of strategies to detect NUP98r may be warranted to leverage other known features, such as its association with FAB subtypes M4 and M5 (e.g., 10/15 cases in our NUP98r cohort) or the high frequency of a normal karyotype (9/17 evaluable cases in our cohort).

Although identification of NUP98r cases is critical for appropriate diagnosis and prognosis of AML, the optimal approach to NUP98r testing must balance cost and turnaround time with sensitivity. The most economical and fastest testing option is NUP98 FISH, with a proposed reimbursement in the United States of $145.28 per test (CPT code 88368) and a turnaround time as short as 1-2 days but longer if run in batches/infrequently. However, since the incidence of NUP98r cases is less than 5% of cases of adult AML, the overall cost of universal testing for all AML patients would be quite high relative to the very low pre-test probability – the cost to the healthcare system is effectively greater than 20 times the individual FISH cost, or more than $2,905.60 per each NUP98r case detected. In MDS, where NUP98r is rarer (potentially 2% of cases), universal testing would be even more costly. Selective testing, such as through WT1 or FLT3-ITD/WT1 strategies, is, therefore, a much better fit for NUP98 FISH. Larger studies will be needed to better characterize yield and to further develop and optimize strategies.

On the surface, RNA sequencing appears to be more costly, with a proposed reimbursement in the United States of $2,919.60 per test for a targeted RNA sequencing panel (CPT code 81455, 2025 Clinical Diagnostics Laboratory fee schedule) and a longer turnaround time (at least 4-7 days) than FISH. However, universal RNA sequencing allows for essentially 100% detection of NUP98r fusions, identification of gene partners, and appropriate disease subclassification. In addition, RNA sequencing approaches capture not only NUP98r cases but a wide spectrum of clinically important alterations that are critical for diagnosis, prognosis, and treatment of AML, some of which may also be cryptic by metaphase karyotype. Similar results may be obtained by WGS-based or whole transcriptome-based methods. On the other hand, a tiered approach with only karyotype and a DNA panel upfront would have an initial turnaround time of 3-5 days. Based on the results, reflex testing for RNA sequencing or FISH could be added. This strategy increases pre-test probability and decreases costs compared to universal testing, but results in longer turnaround times and lower sensitivity of NUP98r detection. Importantly, all testing algorithms are institution-specific, influenced by the availability of individual tests, testing schedules, and local logistics. Thus, testing decisions are ideally managed/supervised by pathology, as algorithmic testing in hematopathology has previously been shown to improve cost-effectiveness.³⁶

NUP98r has consistently been associated with worse outcomes in studies of both pediatric and adult AML.^7,11,37-40 In our study, we observed high relapse rates even after SCT in CR1 (60% of patients). Therefore, there is a need to identify NUP98r at diagnosis and to develop more effective treatment strategies. In pre-clinical models, NUP98r AML has demonstrated sensitivity to Menin inhibition, with eviction of both NUP98 fusion proteins and KMT2A (MLL1) from chromatin at a critical set of pro-leukemic genes.3 Given the recent approval of Menin inhibitors for AML with KMT2A rearrangement and their active development for NPM1-mutated AML, there are several phase I clinical trials (e.g., clinicaltrials.gov NCT05326516 and NCT05453903) that also recruit patients with NUP98r AML.^41-44 Recent PDX mouse models of NUP98r have also indicated that the combination of a Menin inhibitor with a CDK4/6 inhibitor (palbociclib) or a FLT3 inhibitor (gilteritinib) has a synergistic anti-leukemic effect.⁴⁵ In addition, several alternative treatments may be promising for NUP98r AML. One example is venetoclax, a BCL-2 inhibitor, which may be effective against AML with HOXA/B gene overexpression.^46,47 Another example is dasatinib, an inhibitor of ABL and SRC family kinases, which has synergistic effects on cells with NUP98::NSD1 and FLT3-ITD.⁴⁸

In conclusion, our results indicate that AML with NUP98r cases are usually cytogenetically cryptic and can be missed with conventional molecular testing, such as karyotype testing, FISH for common translocations, and myeloid-directed NGS panels looking at DNA mutations. Targeted RNA sequencing with anchored multiplex PCR or hybrid capture enrichment, whole transcriptome sequencing or other genome wide technologies, such as optical genome mapping, should be considered to detect NUP98r alterations.

Our high-yield tiered approach could be used to perform dedicated testing in the subset of AML and MDS that are enriched for NUP98r, which we, like others, demonstrated to be associated with poor prognosis. In fact, NUP98r should be specifically investigated in MDS as well, since it could lead to a change in diagnosis to AML and since the ability to detect NUP98r prior to leukemic transformation may allow for earlier intervention.

Footnotes

• Received April 22, 2025
• Accepted August 6, 2025

Correspondence

^#HKT and VN contributed equally as senior authors

V. Nardi vnardi@mgb.org

H.K. Tsai hktsai@mgb.org

Disclosures

The authors have no conflicts of interest related to this research. RCL reports consulting for Qiagen, Bluebird Bio, Vertex Pharmaceuticals, Verve Therapeutics, Geron Corporation, Takeda Pharmaceuticals and Jazz Pharmaceuticals. ATF reports consulting for Servier, Brystol Myers Squibb, Astellas, Amgen, Kura, Syndax, AstraZeneca, Daiichi Sankyo, Prelude, AbbVie, Schrödinger, Takeda, Rigel, Gilead, Genentech, Autolus, Genmab, and has received clinical trial support from AbbVie, Servier, Kura, and Brystol Myers Squibb. MRL has received research funding to the institution from Novartis, AbbVie, and is part of the advisory board and provides consulting for Pfizer, Novartis, Kite and Jazz. VN reports consulting for Predicta Biosciences. All other authors have no disclosures.

Contributions

LDY, HKT, and VN designed the study, collected the data, performed analysis and interpreted the data, and wrote, reviewed, and revised the manuscript. All authors critically reviewed and edited the manuscript for publication.

Funding

This study was supported by the Vickery-Colvin grant #VicColv 48 to LDY and VN from Massachusetts General Hospital Department of Pathology, Boston, MA, USA.