AML typical mutations (CEBPA, FLT3, NPM1) identify a high-risk chronic myelomonocytic leukemia independent of CPSS molecular

Approximately 90% of individuals diagnosed with chronic myelomonocytic leukemia (CMML) harbor at least 1 somatic mutation affecting various pathways. Mutations in TET2 (∼60%), SRSF2 (∼50%), ASXL1 (∼40%), and the oncogenic RAS pathway (∼30%) are the most frequent,^1-3 and ASXL1, NRAS, RUNX1, and SETBP1 mutations are associated with poor prognosis.^4,5 The TP53 mutation is infrequent in CMML but has also been associated with adverse prognosis.⁵ 

Patients with CMML may also harbor (albeit infrequently) mutations commonly found in acute myeloid leukemia (AML), such as CEBPA, FLT3, IDH1/2, and NPM1. These mutations are well-known in the context of AML, both for their prognostic implications and, in some cases, as therapeutic targets. However, data on the clinical significance of these AML-associated mutations and their potential impact on patient with CMML management are scarce,^6-8 likely due in part to their relatively rare occurrence. CEBPA and FLT3 mutations have been reported in only 4% to 8%^9,10 and <5%,^9,11,12 respectively, of patients with CMML. IDH mutations have been described in <5%^13-16 of patients with CMML and are frequently considered mutually exclusive with TET2 mutations.^13,NPM1 mutations have been detected in ∼3% to 5% of patients with CMML.^6,7,17-23 Although current knowledge of the frequency and prognostic impact of NPM1 mutations in CMML is limited to findings from small series and case reports,²³ they are generally associated with a rapidly evolving clinical course and indicate the need for intensive AML treatment protocols when feasible, including allogeneic hematopoietic stem cell transplantation (alloHSCT).^8,24-32 The status of these AML-associated mutations is not currently included in CMML prognostic models (CMML Prognostic Scoring System molecular [CPSS-Mol], the Groupe Français des Myélodysplasies, and the Mayo Molecular Model^4,33,34), although they may have prognostic implications, in that they can identify higher-risk patients. In addition, they may have therapeutic implications, particularly regarding IDH1/2^34,35 and FLT3 inhibitors.^11,36-39 

To elucidate the clinical and biological characteristics and prognostic impact of CEBPA, FLT3, IDH1/2, and NPM1 mutations in CMML, we have retrospectively analyzed data on the, to our knowledge, largest cohort to date including patients with these mutations, with the aim of suggesting personalized treatment of these patients based on the mutational status of these AML-associated genes.

This was a retrospective analysis of 127 patients with CMML. A total of 83 patients had been diagnosed with CMML at our hospital from 2015 to 2022 and an additional 44 patients had been diagnosed at other institutions affiliated with the Spanish Group of Myelodysplastic Syndromes. The patients from our center were included if they had available targeted next-generation sequencing (NGS) data regardless of their mutational status, whereas the patients selected from other institutions were known to harbor mutations in at least 1 AML-associated gene. Thus, the present cohort was enriched with patients with CMML harboring AML-like mutations. Patient diagnoses were made on the basis of the 2022 criteria established by the World Health Organization and the International Consensus Classification in 2022.^40-42 Bone marrow samples were collected both at the time of CMML diagnosis and, whenever feasible, at progression to AML. Cytogenetic analysis involved the examination of G-banded metaphase cells.

Genetic analyses were performed on all patients at the time of CMML diagnosis. Targeted NGS was performed using the Ion AmpliSeq AML Research Panel (Thermo Fisher Scientific), Oncomine Myeloid Research Assay panel (Thermo Fisher Scientific), or the Sophia DDM Myeloid Solution panel (Sophia Genetics), as previously described.^43-45,FLT3-internal tandem duplication (ITD), FLT3-tyrosine kinase domain (TKD) (835/836 variants), IDH1, IDH2, and NPM1 mutations were also assessed with conventional polymerase chain reaction–based techniques, as previously described.^46,47 The FLT3-ITD allelic ratio was assessed using polymerase chain reaction DNA fragment analysis (3500xL Genetic Analyzer; Applied Biosystems, Thermo Fisher Scientific), as previously described.^46,47 We acknowledge the limited capacity of NGS techniques to detect some CEBPA variants.

Categorical variables were expressed as numbers and percentages, whereas continuous variables were presented as median (range). Group differences were assessed with the Fisher exact test, χ²test, Mann-Whitney U test, or Student t test, as appropriate. Overall survival (OS) was defined as the duration from CMML diagnosis to death from any cause or the last follow-up. OS was computed using the Kaplan-Meier method and compared using the log-rank test. A multivariate analysis was performed using Cox proportional hazards regression analysis. The median follow-up was determined using the Kaplan-Meier estimate of potential follow-up.⁴⁸ The cumulative incidence of progression (CIP) to AML was estimated with a competing risk approach, wherein death from any cause without disease progression served as a competing event. The cumulative incidence of relapse (CIR) was estimated with a competing risk approach, wherein death without relapse served as a competing event. The impact of quantitative covariates was assessed using the Fine-Gray regression model.⁴⁹ Patients with no events were censored at the time of the last follow-up. All statistical analyses were performed with IBM SPSS Statistics, Version 23.0 (IBM Corp, Armonk, NY) or R, version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria). Significance was set at P <.05.

The prognosis of patients was assessed using the CPSS and CPPS-Mol.^4,50 Treatment response to hypomethylating agents was categorized into 5 groups based on the criteria established by Savona et al⁵¹: no response, complete response, hematologic response, partial response, and clinical benefit.

We used a validation series from different centers affiliated with the Spanish Group of Myelodysplastic Syndromes. The validation series consists of 168 patients with CMML with available NGS data (supplemental Tables 1 and 2).

The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the ethics committee/institutional review board of Hospital Clinic Barcelona (2021-1230; on 20 July 2022). This study was approved by all the participating centers. All participants provided informed consent for inclusion in the study.

CEBPA mutations were detected in 7 patients, with 5 occurring in the basic leucine zipper (bZIP) region. Among these, 4 were bZIP in-frame mutations, all of which were point mutations, with no bZIP in-frame insertions or deletions observed. FLT3 mutations were detected in 8 patients: 2 had FLT3-ITD; 5 had FLT3-TKD, 2 of whom harbored 2 different variants; and 1 had both FLT3-ITD and FLT3-TKD. A total of 6 FLT3-TKD1 variants and 2 FLT3-TKD2 variants were detected. All these variants were distinct from typical D835/I836 mutations except for one. IDH1 mutations were detected in 12 patients and IDH2 in 26. All the IDH2 mutations were the c.419G>A p.(Arg140Gln) variant. NPM1 mutations were observed in 11 patients (Table 1; Figure 1; supplemental Figure 1).

Patients harboring CEBPA, FLT3, or NPM1 mutations had unfavorable characteristics and a poorer prognosis than those without these mutations (Table 2). Specifically, in comparison with CEBPA wild-type patients, those with CEBPA mutations had a higher bone marrow blast count (P = .002); were more frequently categorized as CMML-2 according to the 2022 International Consensus Classification/World Health Organization classification (P = .001); had elevated risk according to CPSS-Mol (P = .06); more frequently received modifying treatments (P = .04), such as chemotherapy (P = .01) and alloHSCT (P = .003); and had a shorter OS when censored at the time of alloHSCT (P = .002) (Table 2). When we examined the OS of 4 patients with bZIP in-frame CEBPA mutations vs the 3 patients without bZIP in-frame CEBPA mutations, we found no significant differences (21.8 months [95% confidence interval (CI), 8.2 to not achieved(NA)] vs 22.2 months [95% CI, 10 to NA]; P = .446; supplemental Figure 2).

In comparison with FLT3 wild-type patients, those with FLT3 mutations were younger (P = .04); had higher leukocyte (P = .004), monocyte (P = .01), and bone marrow blast counts (P = .003) and lower hemoglobin levels (P = .002); more commonly had the myeloproliferative (MP) CMML subtype (P = .006); were more frequently categorized as CMML-2 (P = .002); had elevated risk according to CPSS (P = .009) and CPSS-Mol (P = .013); had greater transfusion dependence (P < .001); more frequently received modifying treatments (P = .04), such as chemotherapy (P < .001) and alloHSCT (P = .002); had a shorter OS when censored at the time of alloHSCT (P < .001); and had a higher 2-year CIP to AML (P < .001; Table 2). We did not detect significant differences in clinical and biological characteristics or outcomes between patients with FLT3-ITD and FLT3-TKD (data not shown). However, the small sample size of patients with CMML with FLT3 mutations precluded a robust analysis.

In comparison with NPM1 wild-type patients, those with NPM1 mutations had higher leukocyte (P = .002), monocyte (P = .002), and bone marrow blast counts (P < .001) and lower hemoglobin levels indicating anemia (P < .001); more frequently had the MP-CMML subtype (P = .008); were more frequently categorized as CMML-2 (P < .001); had elevated risk according to CPSS (P < .001); had greater transfusion dependence (P < .001); more frequently received modifying treatments (P = .008), such as chemotherapy (P < .001) and alloHSCT (P = .005); had a shorter time to treatment (0.8 vs 3 months; P = .01); had a shorter OS when censored at the time of alloHSCT (9 vs 35.2 months; P < .001); and had a higher 2-year CIP to AML (36.4% vs 15%; P = .01; Table 2). Patients with NPM1 mutations who received intensive 7+3–type chemotherapy had longer OS than those who did not receive it (28.7 vs 5.6 months; P < .001; supplemental Figure 3).

There were no significant differences in patient characteristics and outcomes between patients with mutations in IDH1 or IDH2 and those with wild-type IDH (Table 3).

A total of 21 patients had mutations in CEBPA, FLT3, and/or NPM1 (mutCFN), whereas the remaining 93 patients had the wild-type form of all 3 genes (wtCFN). More than 1 of these mutations was harbored by 17 patients. Of the 83 patients from our institution, 7 (8%) had mutCFN, whereas 14 patients from the other institutions had mutCFN. mutCFN patients had more unfavorable characteristics and poorer prognosis than wtCFN patients (Table 4). In comparison with wtCFN patients, mutCFN patients were younger (63 vs 71 years; P = .03); had higher leukocyte (19.6 × 10⁹/L vs 7.6 × 10⁹/L; P = .003), monocyte (2.9 × 10⁹/L vs 1.7 × 10⁹/L; P = .006), and bone marrow blast count (12% vs 4%; P < .001); and had lower hemoglobin levels (8.6 vs 11.4 g/dL; P < .001) and platelet count (98 × 10⁹/L vs 117 × 10⁹/L; P = .04). They were also more frequently associated with the MP-CMML subtype (61.9% vs 29%; P = .01); were more frequently categorized as CMML-2 (81% vs 14.1%; P < .001); and had elevated risk according to CPSS (75% vs 29.8%; P < .001) and CPSS-Mol (88.2% vs 44%; P = .001).

mutCFN patients had greater transfusion dependence (66.7% vs 21.5%; P < .001) and more frequently received modifying treatment (81% vs 36.6%; P < .001), including hypomethylating agents (61.9% vs 31.2%; P = .01), chemotherapy (57.1% vs 6.5%; P < .001), and alloHSCT (57.1% vs 11.8%; P < .001). They also had a shorter time to treatment (0.9 vs 4.5 months; P = .002), shorter OS (22.2 vs 46.8 months; P = .01), and a higher 2-year CIP to AML (39.8% vs 13.7%; P < .001) (Table 4; Figure 2). mutCFN patients who received chemotherapy (n = 12) showed a trend toward longer OS than those who did not (n = 9; 25 vs 9 months; P = .067; supplemental Figure 4). Twelve mutCFN patients received alloHSCT (supplemental Table 3) and attained longer OS than those who did not (n = 9; 28 vs 12 months; P = .05; Figure 3). Median OS after alloHSCT for these 12 patients was 19.6 months (95% CI, 5-72), the 2-year CIR was 50% (95% CI, 18-75) (supplemental Figure 5), and the 2-year nonrelapse mortality rate was 25% (95% CI, 5-52). The 11 wtCFN patients with CMML who underwent alloHSCT had a post-alloHSCT median OS of 91 months (95% CI, 12 to NA; supplemental Figure 6).

The multivariate analysis identified mutCFN as an adverse prognostic factor for OS independent of age and CPSS-Mol (hazard ratio, 2.452; 95% CI, 1.237-4.86; P = .01; Table 5).

In the validation cohort, mutCFN patients (n = 11) had a higher 2-year CIP to AML (51.1% vs 17.2%; P < .01; supplemental Figure 7A), a shorter median OS when censored at the time of alloHSCT (22.8 vs 44 months; P < .01; supplemental Figure 7B), and a trend toward shorter OS (22.8 vs 47 months; P = .076; supplemental Figure 7C). Given the rarity of these mutations in CMML, and to obtain more robust results, we conducted an analysis of the combined series, resulting in a total of 282 patients. mutCFN patients (n = 32) had a shorter median OS (22.8 vs 46.9 months; P < .001) and a higher 2-year CIP to AML (46.2% vs 15.7; P < .001; supplemental Figure 8). The multivariate analysis identified mutCFN as an adverse prognostic factor for OS independent of age and CPSS-Mol (hazard ratio, 1.851; P = .017; supplemental Table 4). Interestingly, mutCFN patients also had a mutational profile distinct from that of wtCFN patients (Figure 4). They had a higher incidence of DNMT3A mutations (n = 9 [45%] vs 6 [6.5%]; P < .001) and a lower incidence of TET2 mutations (n = 4 [20%] vs 55 [59.1%]; P = .002). They also showed a trend toward a lower incidence of SRSF2 mutations (n = 3 [15.8%] vs 30 [35.7%]; P = .1).

Given that most mutCFN CMML cases were classified as CMML-2 and exhibited similarities to AML, we compared mutCFN CMML cases with both CMML-2 and M4/M5 AML. We did not find statistically significant differences in the main clinical or prognostic characteristics (supplemental Results; supplemental Tables 5-8).

Five mutCFN patients with CMML had available paired samples taken at CMML diagnosis and at progression to AML. At the time of progression to AML, all of the AML-associated mutations present at CMML diagnosis were retained. In all patients with complete NGS at CMML diagnosis and transformation, new mutations emerged at the onset of AML. These mutations were observed in genes associated with signaling (FLT3, n = 1; NRAS, n = 1), transcription (RUNX1, n = 1), and splicing (STAG2, n = 1) pathways (supplemental Figure 9).

CEBPA, FLT3, IDH1/2, and NPM1 mutations are commonly found in AML but are rare in CMML, and their status is not included in current CMML prognostic models. In fact, there is little information available on their impact on patient prognosis or therapeutic strategy in CMML. In the present study, we have found that patients with CMML with at least 1 mutation in CEBPA, FLT3, and/or NPM1 have a genetic profile quite similar to that of patients with AML. Moreover, these patients had a dismal prognosis independently of their CPSS-Mol classification, with shorter OS and a higher CIP to AML, underscoring the unmet need for more AML-type clinical approaches and novel treatments for these patients.

Previous studies have found that CEBPA mutations are linked to an immature myeloid blast phenotype. The specific frequency of CEBPA mutations in CMML ranges from 4% to 8%.^9,10 Studies in AML have reported that high rates of complete remission and long overall and relapse-free survival were attained only by patients with bZIP in-frame insertion-deletion (inDel) CEBPA mutations.^52,53 We detected CEBPA mutations in 7 of our patients with CMML. These patients had higher-risk characteristics and poor outcome. Interestingly, 4 patients harbored bZIP in-frame mutations according to Taube et al’s⁵² definition, none of which were the most frequently associated with mutated CEBPA-AML, the bZIP in-frame InDel mutation. Recently, a study showed that the beneficial effect of CEBPA mutation in AML is restricted to CEBPA bZIP in-frame InDel mutation.⁵³ We did not find any of these mutations in our study. The negative impact of CEBPA mutations in non-AML myeloid neoplasms has been scarcely investigated, but some studies identify these mutations as indicators of poor prognosis.^21,54,55 This highlights the need for further investigation into specific implications for patient outcome and treatment response.

The FLT3 mutation has been reported in <5%^35,56 of patients with CMML, whereas it was detected in 8 patients in our study. Although the FLT3 mutation does not necessarily predict transformation to AML, it may help determine the optimal therapeutic strategy, which could include FLT3 inhibitors, such as midostaurin⁵⁷ and sorafenib,⁴³ that have been approved for AML.^11,36-39 Interestingly, 2 of our patients with FLT3 mutations received sorafenib following a post-alloHSCT relapse (1 molecular and 1 morphological relapse) and attained prolonged responses for 8 months and a year, respectively.⁴³ Previous reports have found an association between FLT3 mutations in CMML and leukocytosis, the MP-CMML subtype, shorter OS,⁵⁶ and progression to AML.^23,58 Consistent with these findings, our patients with FLT3 mutations had higher-risk characteristics and worse outcome. In contrast to findings in AML,^59,60 most FLT3 mutations detected in our study were TKDs, with 6 TKD1 and 2 TKD2 variants, most of which were not p.(Asp835Tyr). This aspect remains unexplored in CMML.

IDH1 and IDH2 mutations have been reported in <5% of patients with CMML^13-16: IDH1 in only 1% and IDH2 in 5%.^15,35 All the IDH2 mutations in our patients involved the IDH2R140 hot spot, which is consistent with a previous report.⁶¹ The prognostic impact of IDH mutations in CMML is not clear. One study found that OS did not differ between IDH-mutant and IDH wild-type patients,¹⁴ whereas in another study, patients with IDH1 mutations had a shorter OS than those with IDH2 mutations.¹⁵ We found no significant differences in outcomes between patients with and without IDH1/2 mutations. Nevertheless, patients with CMML harboring IDH mutations may well be candidates for targeted therapy with IDH inhibitors, such as ivosidenib and enasidenib.³⁸ One of our patients received off-label treatment with the IDH2 inhibitor enasidenib for molecular relapse after alloHSCT, and maintained a long-lasting complete response for 5 years and a half, remaining in molecular remission to this day.⁴³ 

NPM1 mutations have been observed in 3% to 5% of patients with CMML,^6,7,17-23 whereas we detected them in 11 of our patients. Previous studies have shown that patients with CMML with NPM1 mutations have a high risk of progressing to AML, leading some authors to consider them as “early-stage” AML given their shared clinical and molecular characteristics.^7,19,20 Moreover, there is currently a debate on whether AML with NPM1 mutation should be classified as AML.^29,41,62 Patients with CMML with NPM1 mutations were found to have lower hemoglobin levels, an elevated leukocyte count, increased percentages of bone marrow monocytes and blasts, a heightened likelihood of progression to AML, and poorer OS compared with those with wild-type NPM1.^6,7,18 Remarkably, patients with CMML with NPM1 mutations who progressed to AML did not exhibit the favorable prognosis typically associated with de novo AML with NPM1 mutations.^{7,19,21,23,42} In line with these previous studies,^6,7 our patients with CMML with NPM1 mutations had more adverse clinical characteristics and worse outcome than those with wild-type NPM1. Management of patients with CMML with these AML-associated mutations poses a challenge given the scarcity of studies with large patient cohorts.^7,8,18,23,27 Notably, some results indicate that patients with NPM1 mutations respond well to chemotherapy,^7,8,24 implying that they might derive greater benefits from AML-type chemotherapy than from the standard CMML treatment. In fact, in the present study, OS was longer in patients with NPM1 mutations treated with chemotherapy. However, this may reflect selection bias toward fitter patients, and the rarity of the condition makes it difficult to draw solid conclusions. Additionally, although alloHSCT is currently the only potential curative strategy for CMML, it is not clear whether patients with NPM1-mutated CMML should receive alloHSCT at first complete response.^7,8,18,23,27 In our patient cohort, those who received chemotherapy followed by alloHSCT had a better prognosis than those who did not, although we are aware of the treatment selection bias based on their baseline characteristics. However, we also observed a high CIR after alloHSCT, which highlights both the complexity of treating these patients and the need for further studies with larger numbers of patients.

These AML-associated mutations are not currently included in CMML prognostic scores.^4,33,34 However, our findings suggest that including CEBPA, FLT3, and/or NPM1 mutations would help to increase prognostic accuracy despite the infrequency of these mutations in CMML. Our multivariate analysis demonstrated their independent adverse prognostic impact, suggesting that their mutational status can improve the stratification of patients with CMML.

Our study also suggests that patients with CMML with CEBPA, FLT3, and/or NPM1 mutations exhibit a comutational pattern more typical of AML than of CMML. For example, we identified a positive correlation between DNMT3A, FLT3-ITD, and NPM1 in these patients, which is not commonly observed in CMML but is frequently seen in AML.^7,17,63 Conversely, we observed a negative correlation with other typical CMML mutations, such as TET2 and ASXL1.^7,17 

Research involving paired samples has documented clonal evolution from CMML to AML, indicating the emergence of new mutations and an increase in variant allele frequency, including NPM1, FLT3, and SH2B3.²³ In our study, we observed the emergence of FLT3, NRAS, RUNX1, and STAG2.

Our study has several limitations, including its retrospective nature and the relatively small numbers of patients with CMML with CEBPA, FLT3, and/or NPM1 mutations due to the rarity of these alterations in CMML. To address this limitation and increase the number of patients with these mutations under study, our cohort was enriched with these specific mutations, which were subsequently confirmed in a validation cohort. Clearly, prospective studies involving larger cooperative cohorts are warranted. Our understanding of the pathophysiology of CMML is improving, and one of the goals is to move toward a molecular-based classification and risk assessment. Therefore, despite the limitations of our study, our findings provide the basis for progressing toward this goal.

In conclusion, we have found that CEBPA, FLT3, and NPM1 mutations represent an unfavorable molecular prognostic factor in CMML not captured by existing molecularly integrated prognostic models. We therefore propose that considering these mutations in the prognostic stratification of patients with CMML will provide a useful tool for decision-making and patient management. Patients with CMML harboring these mutations exhibit a distinct genetic profile that closely resembles that of patients with AML and is linked to poorer prognosis. These patients may thus derive potential benefits from a more intensive AML-type treatment approach, including intensive chemotherapy and alloHSCT when feasible, as well as certain targeted therapies approved for AML.

The authors thank Ferran Nadeu for his help in preparing the OncoPrint representation.

This study was supported by Contractes Clínic de Recerca Emili Letang-Josep Font 2021 grants from the Hospital Clínic de Barcelona; Instituto de Salud Carlos III through the project “FIS PI22/01660” cofunded by the European Union (M.D.-B. and J.E.); and Convocatòria d’Intensificació Interna per als professionals de l’HCB 2023 (M.D.-B.) and Collaboration Scholarship, financed by the Spanish Ministry of Education (V.T.-M.).

These organizations had no role in the design of the study; the collection, analysis, or interpretation of data; the writing of the manuscript; or the decision to publish the results.

Contributions: All authors were involved directly or indirectly in the care of patients, sample procurement, and collection of clinical data; M.D.-B. designed the study and led the research team; S.C.-D. and M.D.-B. performed the statistical analyses and wrote the manuscript; and all authors reviewed and accepted the final version of the manuscript.

Conflict-of-interest disclosure: F.G. is a speaker for AbbVie. J.E. has received consultancy honoraria from AbbVie, Novartis, Astellas, Jazz Pharmaceuticals, Bristol Myers Squibb–Celgene, Pfizer, and Daiichi Sankyo; and research grants from Novartis and Jazz Pharmaceuticals. M.D.-B. has served as a consultant, adviser, or speaker and received travel grants from Bristol Myers Squibb, AbbVie, Astellas, Jazz Pharmaceuticals, Takeda, and Novartis. The remaining authors declare no competing financial interests.

Correspondence: Marina Díaz-Beyá, 'Department of Hematology, Hospital Clínic de Barcelona, C/Villarroel, 170, Barcelona, 08036, Spain; email: diazbeya@clinic.cat.