Acute myeloid leukemia (AML) is a heterogeneous pathology in terms of its cytogenetic and molecular alterations, which are used for prognostic stratification and as therapeutic targets.1-3 Some studies have shown the negative impact of a high allelic burden at diagnosis regarding the mutations of some genes (EZH2, SRSF2, TP53) on the evolution of AML.4-6 The most studied gene is TP53; different variant allele frequency (VAF) thresholds (i.e., 10% or 40%) at diagnosis could have an impact on patients’ outcomes.7,8 Although the mutational burden, according to VAF measurements, has been associated with the prognosis of patients, this parameter is not well established for risk stratification. In this study, we analyzed the impact of the mutational burdens of gene variants detected with a myeloid panel via next-generation sequencing in a cohort of AML patients included in a large epidemiological registry of the “Programa Español para el Tratamiento de las Hemopatías Malignas” (PETHEMA) (ClinicalTrials.gov Identifier: NCT02607059), focusing on overall survival (OS).
This was a non-interventional, systematic, retrospective chart review of data from patients enrolled in the PETHEMA registry, which included patients diagnosed with AML, regardless of the treatment administered. This study was conducted in a cohort of 3,018 adult patients with AML who were diagnosed between 2003 and 2021 and underwent testing with a next-generation sequencing panel; these patients were diagnosed in 108 centers belonging to PETHEMA cooperative group. The study was approved by a formally constituted review board. The samples were obtained at diagnosis, during refractoriness, and at relapse; the comprehensive mutational profile of this cohort was published previously.3 The patients were assigned to therapeutic groups based on the front-line approach: intensive chemotherapy, non-intensive chemotherapy such as hypomethylating agents, or low-dose cytarabine schemes; patients who received venetoclax-based schedules were excluded because of the low number of such patients. The mutational profiles were determined in seven Spanish PETHEMA reference laboratories, which were instructed to use next-generation sequencing to assess the mutational status of genes that define diagnosis and prognosis as well as guide treatment options (ASXL1, BCOR, CEBPA, EZH2, FLT3, IDH1, IDH2, NPM1, RUNX1, SF3B1, SRSF2, STAG 2, U2AF1, ZRSR2, and TP53). Moreover, there was a recommendation to study other genes for which there is proven evidence of their relevance in AML pathogenesis (ABL1, BRAF, CALR, CBL, CSF3R, DNMT3A, ETV6, GATA2, HRAS, JAK2, KIT, KRAS, MPL, NRAS, PTPN11, SETBP1, TET2, and WT1). The next-generation sequencing methods were harmonized and periodically validated across centers.3,9 Using the single-nucleotide polymorphism database (NCBI, dbSNP150), variants with a VAF less than 0.01 in the general population were discarded. Other databases used to search the filtered variants were the Catalogue of Somatic Mutations in Cancer (COSMIC) and VarSome.
All the statistical analyses were performed using SPSS version 22 (IBM, Armonk, NY, USA) and Stata InterCooled for Windows version 16 (StataCorp LLC, College Station, TX, USA); statistical significance was considered at a P value less than or equal to 0.05. A χ2 test was used to assess the associations between categorical variables, and a median test, Student t test, and analysis of variance were performed to compare differences in the median and mean values of continuous variables. The analysis was performed using VAF as a continuous variable (for those genes without mutations, the value of the variable was 0). The VAF was expressed as a percentage of one. The prognostic impacts of the mutational burdens of gene variants were analyzed with respect to the type of leukemia treatment received. Cox proportional hazard models were used to assess the association of variables (clinical data and mutational load) with the patients’ leukemia-free survival (LFS) and OS. For multivariate analyses, we adjusted for patients’ age (continuous variable) and VAF of gene mutations (1% increments). Mixed regression models combine fixed and random effects to analyze correlated data. In this study, we used mixed-effects machine-learning regression to account for patients’ heterogeneity by treating patients as random factors and assess the impact of VAF on survival, considering gene mutations, death, and relapse as fixed factors; this approach allowed for efficient analysis of multiple gene mutations per patient. The receiver operating characteristic curve was constructed under the nonparametric assumption, and analysis was performed to identify the cutoff score that would assist in distinguishing between live and dead patients for each gene.
Among the 3,018 samples analyzed (Figure 1A), 2,464 were from patients at first AML diagnosis (81.6%), and the remaining 554 samples were from 473 patients at relapse/ refractory episodes. The most frequently mutated gene was DNMT3A (24.3%), followed by NPM1 (22.5%), TET2 (21.2%), and RUNX1 (18.8%).
In the ‘diagnosis group’ (2,464 patients), the median age at first AML diagnosis was 67 years (range, 18-98). Patients received front-line intensive chemotherapy schemes (55.6%), hypomethylating treatment with a single agent (27.1%) or low-dose cytarabine-based treatments (14.9%). In patients who received intensive chemotherapy schedules, 70.3% achieved complete remission and 36.2% underwent allogeneic hematopoietic stem cell transplantation. The risk group according to the European LeukemiaNet (ELN) 2017 classification was favorable in 15.0% of cases, intermediate in 34.0%, and adverse in 51.1%. OS and LFS analyses were performed among 2,464 patients at initial diagnosis; the median OS (1,381 patients) was 12.6 months (95% confidence interval [95% CI]: 11.4-13.7 months) and the median LFS (1,137 patients) was 10.1 months (95% CI: 9.3-10.9 months). The complete response rate in the ‘diagnosis group’ was 49.1% (487/991 patients).
Figure 1.Study design and Kaplan-Meier curve of overall survival depending on the cutoff of variant allele frequencies of some genes. (A) Study design. (B) Kaplan-Meier curve of overall survival, depending on the cutoff of the variant allele frequencies (VAF) of ASXL1, JAK2, RUNX1, SRSF2, TET2 and TP53. The entire diagnostic cohort is represented. Blue represents patients with VAF below the cutoff and red represents patients with VAF above the cutoff. CR: complete remission; Allo-HSCT: allogeneic hematopoietic stem-cell transplantation; Auto-HCST: autologous hematopoietic stem-cell transplantation; ELN: European LeukemiaNet; LDAC: low-dose cytarabine; OS: overall survival.
The patient’s age, leukocyte count, and low mutational loads for some genes, such as ASXL1, FLT3, RUNX1 or TP53, or high mutational loads for DNMT3A or NPM1 were associated with achieving a complete response. In the multivariate logistic regression model, a higher age of the patient (odds ratio [OR]=0.935, 95% CI: 0.913-0.958, P<0.001) and higher mutational load for the SRSF2 gene (OR=0.978, 95% CI: 0.967-0.990, P<0.001) were associated with a lower probability of achieving a complete response. However, a higher mutational load for NPM1 (OR=1.025, 95% CI: 1.007-1.043, P<0.001) was associated with a greater chance of achieving a complete response.
To avoid negative cases impacting the analyses of the VAF effect on OS, we carried out a mixed-effects machine learning regression (Online Supplementary Table S3). We observed that increased allelic loads for ASXL1 (OR=1.317, 95% CI: 0.084-2.550, P=0.036), FLT3 (OR=1.382, 95% CI: 0.148-2.615, P=0.028), JAK2 (OR=1.400, 95% CI: 0.167-2.633, P=0.026), RUNX1 (OR=2.215, 95% CI: 0.982-3.448, P<0.001), SRSF2 (OR=3.263, 95% CI: 2.030-4.496, P<0.001), TET2 (OR=2.662, 95% CI: 1.429-3.896, P<0.001), TP53 (OR=4.712, 95% CI: 3.479-5.946, P<0.012), and U2AF1 (OR=1.270, 95% CI: 0.036-2.503, P=0.044) were associated with an adverse prognosis for OS; however, an increase in NPM1 burden conferred a good prognosis (OR= -2.417, 95% CI:-3.651 to -1.184, P<0.001). The results were obtained in comparison with those for the ABL1 mutation load; any differences observed when compared with some previous results were associated with the comparator gene, but SRSF2, TP53, and NPM1 were consistent in all analyses. This model for LFS was not significant.
To facilitate the application of results in clinical practice, we attempted to determine a cutoff for each gene to define changes in OS; different optimal cutoff points were obtained, namely ASXL1 (VAF 0.475), JAK2 (VAF 0.038), RUNX1 (VAF 0.043), SRSF2 (VAF 0.028), TET2 (VAF 0.030), and TP53 (VAF 0.024) for some. This confirmed statistically significant differences, with a better OS associated with a low VAF for all genes (ASXL1: low VAF vs. high VAF, 15.84 vs. 13.51 months, P=0.025; JAK2: 15.87 vs. 10.10 months, P<0.001; SRSF2: 16.16 vs. 12.49 months, P<0.001; TET2: 17.02 vs. 10.69 months, P<0.001; TP53: 17.21 vs. 6.95 months, P<0.001), with the exception of RUNX1 (15.41 vs. 16.03 months, P=0.789), for which the results were not statistically significant.
We also evaluated the impact of 1% increases in the mutational load on the risk of death (OS) and relapse (LFS) in the group of patients treated with intensive regimens (N=467 patients with a complete data set) (Table 1, Figure 1B). In a multivariate analysis, we observed a worse OS in older patients (hazard ratio [HR]=1.04, P<0.001) or patients with a higher leukocyte count (HR=1.04, P<0.001); in addition, we observed that higher VAF for BRAF (HR=1.04, P=0.009), EZH2 (HR=1.03, P=0.005), KRAS (HR=1.05, P<0.001), SRSF2 (HR=1.02, P=0.006), TP53 (HR=1.02, P<0.001), and U2AF1 (HR=1.02, P=0.009) were associated with a worse OS, and a higher VAF for IDH1 was associated with a better OS (HR=0.98, P=0.03). Regarding LFS (N=466 patients with a complete data set) (Table 2), in the multivariate analysis, we observed a worse LFS with higher VAF for ASXL1 (HR=1.02, P=0.016) and CALR (HR=1.02, P=0.033), and a better LFS with a higher VAF for IDH2 (HR=0.98, P=0.033). EZH2 is a transcriptional regulation gene, and U2AF1 is a splicing factor gene; both are related to dysplasia and are included in the adverse-risk category in the ELN2022 classification. An association between a higher EZH2 clonal burden and a worse LFS has been reported previously;5 however, to our knowledge, the relationship between a high U2AF1 VAF and worse outcome has not been reported before. To our knowledge, no study has shown that patients with a high CALR VAF have a worse OS or LFS; this could be related to acute leukemias secondary to chronic myeloproliferative neoplasms, which have a worse evolution than that of de novo AML.
Table 1.Overall survival: multivariate analyses of factors at diagnosis in patients in each treatment group.
Table 2.Leukemia-free survival: multivariate analyses of factors at diagnosis in patients in each treatment group.
In the ‘low-dose cytarabine group’, regarding OS (N=158 patients with a complete data set), a higher age (HR=1.06, P=0.002), higher leukocyte count (HR=1.01, P<0.001), and higher VAF for BRAF (HR=1.10, P=0.008), CBL (HR=1.07, P=0.016), DNMT3A (HR=1.01, P=0.015), and TP53 (HR=1.01, P<0.001) were associated with poor outcomes. In the ‘hypomethylating agent group’, regarding OS (N=227 patients with a complete data set), higher VAF of CBL (HR=1.01, P=0.03) and TP53 (HR=1.01, P<0.001) were identified as poor risk factors for OS, as well as a higher blast count in bone marrow (HR=1.01, P=0.011). In patients receiving low-dose cytarabine, splicing factors were not detected as having an impact on OS; in patients who received hypomethylating treatment, epigenetic factors were not detected as having a prognostic impact. These differences have not been previously described and could be related to the type of treatment received; adding venetoclax to low-dose cytarabine may mitigate the poor prognosis of splicing mutations, or hypomethylating agents may eliminate the prognostic impact of genes involved in epigenetic pathways.10
Our results are consistent with already known results, with a negative impact of the TP53 VAF on OS in the global cohort and in each one of the three treatment subgroups. Previously, Short et al. established a VAF threshold of 0.40, showing a better OS in patients with a low TP53 VAF treated with a cytarabine-based regimen;7 other studies have shown similar results although it is difficult to establish a threshold.5,8,11-13
In summary, our results show that mutation allele burden of certain signaling (FLT3, JAK2), transcription factor (RUNX1), epigenetic (ASXL1, TET2), and splicing (SRSF2, U2AF1) genes, in addition to TP53, worsen OS survival in AML patients. We also determined a specific prognostic cutoff for each of those genes. More studies are needed to confirm our results and further establish the prognostic or predictive value of the allele burden in AML patients.
Footnotes
- Received July 21, 2024
- Accepted February 14, 2025
Correspondence
Disclosures
No conflicts of interest to disclose.
Contributions
Funding
This work was funded by Instituto de Salud Carlos III (ISCIII), Spain, and co-funded by the European Union: PMP22/00069, PI22/01088, PI19/01518.
References
- Döhner H, Wei AH, Appelbaum FR. Diagnosis and management of AML in adults: 2022 recommendations from an international expert panel on behalf of the ELN. Blood. 2022; 140(12):1345-1377. Google Scholar
- Sargas C, Ayala R, Larráyoz MJ. Comparison of the 2022 and 2017 European LeukemiaNet risk classifications in a reallife cohort of the PETHEMA group. Blood Cancer J. 2023; 13(1):77. Google Scholar
- Sargas C, Ayala R, Larráyoz MJ. Molecular landscape and validation of new genomic classification in 2668 adult AML patients: real life data from the PETHEMA registry. Cancers (Basel). 2023; 15(2):438. Google Scholar
- Goel S, Hall J, Pradhan K. High prevalence and allele burden-independent prognostic importance of p53 mutations in an inner-city MDS/AML cohort. Leukemia. 2016; 30(8):1793-1795. Google Scholar
- Hamilton BK, Rybicki L, Hirsch C. Mutation clonal burden and allogeneic hematopoietic cell transplantation outcomes in acute myeloid leukemia and myelodysplastic syndromes. Bone Marrow Transplant. 2019; 54(8):1281-1286. Google Scholar
- Wu M, Guo ZW, Huang GN, Ye YB. Features and impacts on the prognosis of gene mutations in patients with acute myeloid leukemia. Neoplasma. 2021; 68(5):1072-1078. Google Scholar
- Short NJ, Montalban-Bravo G, Hwang H. Prognostic and therapeutic impacts of mutant TP53 variant allelic frequency in newly diagnosed acute myeloid leukemia. Blood Adv. 2020; 4(22):5681-5689. Google Scholar
- Zhao D, Eladl E, Zarif M. Molecular characterization of AML-MRC reveals TP53 mutation as an adverse prognostic factor irrespective of MRC-defining criteria, TP53 allelic state, or TP53 variant allele frequency. Cancer Med. 2023; 12(6):6511-6522. Google Scholar
- Sargas C, Ayala R, Chillón MC. Networking for advanced molecular diagnosis in acute myeloid leukemia patients is possible: the PETHEMA NGS-AML project. Haematologica. 2020; 106(12):3079-3089. Google Scholar
- Senapati J, Urrutia S, Loghavi S. Venetoclax abrogates the prognostic impact of splicing factor gene mutations in newly diagnosed acute myeloid leukemia. Blood. 2023; 142(19):1647-1657. Google Scholar
- Sasaki K, Kanagal-Shamanna R, Montalban-Bravo G. Impact of the variant allele frequency of ASXL1, DNMT3A, JAK2, TET2, TP53, and NPM1 on the outcomes of patients with newly diagnosed acute myeloid leukemia. Cancer. 2020; 126(4):765-774. Google Scholar
- Kim K, Maiti A, Loghavi S. Outcomes of TP53-mutant acute myeloid leukemia with decitabine and venetoclax. Cancer. 2021; 127(20):3772-3781. Google Scholar
- Grob T, Al Hinai ASA, Sanders MA. Molecular characterization of mutant TP53 acute myeloid leukemia and high-risk myelodysplastic syndrome. Blood. 2022; 139(15):2347-2354. Google Scholar
Data Supplements
Figures & Tables
Article Information
This work is licensed under a Creative Commons Attribution 4.0 International License.