Molecular subclusters across the continuum of myelodysplastic neoplasms and acute myeloid leukaemia define distinct clinical entities

Abstract Introduction The distinction between myelodysplastic neoplasms (MDS) and acute myeloid leukaemia (AML) is becoming increasingly blurred particularly at the arbitrary 20% blast cut-off at which AML is traditionally diagnosed. There is a move towards lowering this cut-off to 10% blasts allowing patients within the 10-20% range to access both AML and MDS therapies. There is however a clinical need for more precise methods to classify patients on the continuum between these diseases to inform treatment decisions. Using clustering techniques applied to targeted sequencing data derived from a large unselected population-based cohort we investigated whether molecular subtypes could be identified across AML and MDS which could better direct therapeutic interventions and improve outcomes. Methods Patients diagnosed with AML and MDS between 2004-2012 from a catchment population of ∼4 million (14 centres) were sequenced with a targeted 293-gene panel. A total of 1085 patients with complete genetic and cytogenetic data were available for analysis. Patients with previously well-defined good risk translocations were excluded from further analysis leaving a final cohort of 1007 patients (AML=445, MDS=561). A total of 69 genetic features (both mutational and structural) were analysed, which were either detected in at least 15 of the 1000 analysed patients or of established significance in myeloid neoplasm. To identify genetic subgroups, the data were modelled as a finite mixture of Bernoulli distributions, providing a data-driven probabilistic interpretation of group membership strength. The number of identifiable clusters was selected by using the Akaike Information Criterion (AIC) likelihood penalization method. Genetic subgroups were then correlated with overall survival, disease progression and treatment specific responses. All patients were followed up for mortality until 01 August 2023. Results A total of 970 patients (96%) harboured at least one of the 69 genetic features. Eight distinct clusters were identified and were named after the most discriminatory abnormality within each cluster. All clusters contained both AML and MDS cases however five clusters showed a predominance of MDS cases (EZH2, SF3B1, SRSF2, TP53, U2AF1₁57) while AML cases predominated in the remaining 3 (BCOR, KMT2A-PTD, NPM1). The genomic profile within each cluster was strikingly similar across both AML and MDS. With respect to overall survival (OS) within the total cohort, the SF3B1 cluster showed the best OS while TP53 was by far the worst. These effects persisted when analysing the MDS and AML cohorts separately. The SF3B1 group had the best OS within MDS, but also in AML patients where the better outcome was equivalent to both the NPM1 (p=1. 0) and U2AF1₁57 groups (p=0. 4). The TP53 group predicted the worst outcome across both diseases though the BCOR group was comparably poor in MDS, particularly after 2 years (TP53 vs BCOR OS; 1 year 15% (CI 9-26) vs 48% (CI 32-72), 2 years 5% (CI 2-14) vs 9% (CI 2-33). Within each cluster group, OS was not significantly different between MDS and AML patients for the BCOR (p=0. 4), KMT2APTD (p=0. 2), NPM1 (p=1. 0) and U2AF1₁57 (p=0. 4) groups and approached significance for EZH2 (p=0. 07). For other groups MDS patients had a significantly better OS and blast percentage was the main discriminating factor with respect to outcome. Those with 5% blasts had the best outcome while above this threshold OS was broadly comparable. The risk of progression to AML was highest for those MDS cases within the AML predominant clusters (BCOR HR 16. 8; CI 6. 3-44. 5; NPM1 HR 10. 6; CI 4. 1-27. 9) and TP53 cluster (HR 15. 1; CI 6. 5-35. 1). A total of 269 patients received intensive chemotherapy as a first line treatment. Those patients within the NPM1, SF3B1 and U2AF1₁57 had the best outcome following intensive therapy though these patients were significantly younger than other groups combined (p0. 0001). In contrast patients within the TP53 and SRSF2 groups had a very poor outcome with all treatment interventions. Conclusion By analysing MDS and AML cases on a continuum, distinct molecular subgroups can be identified with diverse outcomes and responses to treatment in many cases irrespective of blast percentage. While some of these groups are expected from previous studies, other groups are not currently recognised as distinct entities. Validation of the impact of these groups in the era of novel therapies is however needed.