Clinical impact of the genomic landscape and leukemogenic trajectories in non-intensively treated elderly acute myeloid leukemia patients

Abstract

To characterize the genomic landscape and leukemogenic pathways of older, newly diagnosed, non-intensively treated patients with AML and to study the clinical implications, comprehensive genetics analyses were performed including targeted DNA sequencing of 263 genes in 604 patients treated in a prospective Phase III clinical trial. Leukemic trajectories were delineated using oncogenetic tree modeling and hierarchical clustering, and prognostic groups were derived from multivariable Cox regression models. Clonal hematopoiesis-related genes (ASXL1, TET2, SRSF2, DNMT3A) were most frequently mutated. The oncogenetic modeling algorithm produced a tree with five branches with ASXL1, DDX41, DNMT3A, TET2, and TP53 emanating from the root suggesting leukemia-initiating events which gave rise to further subbranches with distinct subclones. Unsupervised clustering mirrored the genetic groups identified by the tree model. Multivariable analysis identified FLT3 internal tandem duplications (ITD), SRSF2, and TP53 mutations as poor prognostic factors, while DDX41 mutations exerted an exceptionally favorable effect. Subsequent backwards elimination based on the Akaike information criterion delineated three genetic risk groups: DDX41 mutations (favorable-risk), DDX41^wildtype/FLT3-ITD^neg/TP53^wildtype (intermediate-risk), and FLT3-ITD or TP53 mutations (high-risk). Our data identified distinct trajectories of leukemia development in older AML patients and provide a basis for a clinically meaningful genetic outcome stratification for patients receiving less intensive therapies.

Fig. 1. Mutational and cytogenetic landscape of older patients with acute myeloid leukemia.

Mutational (A) and cytogenetic (B) profile, as well as distribution of AML according to the International Consensus Classification (C) in 604 older patients with newly diagnosed AML. A Genes with mutations present in ≥4% of AML. B Cytogenetic abnormalities present in ≥4% of AML; abnormalities were determined by conventional chromosome analysis, fluorescence in-situ hybridization, and EPIC-array analysis. Frequencies given in percent.

Fig. 2. Oncogenetic tree model using a modeling algorithm by Szabo.

In an oncogenetic tree model, the root represents a state of disease before occurrence of mutations. Each node represents a gene mutation and each branch represents a distinct biologic clone thus illustrating the different clones and temporal sequence of acquisition of mutations.

Fig. 3. Unsupervised hierarchical clustering using mutational and cytogenetic data.

Hierarchical Dirichlet Processes were employed to build the cluster plot. Mutations and cytogenetic data that was present ≥1% of AML were used. The distribution of ELN risk strata and ICC entities across the newly identified classes is indicated by different colors.

Fig. 4. Prognostic value of current AML classifications and of proposed new genetic risk categories for older AML patients.

Overall survival by European LeukemiaNet (ELN) 2017 (A) and ELN 2022 (B) genetic risk classification, by ICC categories (C) and proposed risk categories for older AML patients who are not eligible for intensive chemotherapy derived from multivariable Cox models of the current study (D). A, B Both risk classifications did not provide clinically meaningful separation of the survival curves. In cases previously stratified according to the 2017 ELN stratification, the 2022 ELN stratification entailed a change of strata in 14% (n = 82) of the patients, with re-classification to a more adverse-risk category in 13% (n = 75) and to a more favorable in 1% (n = 6) of the cases. D After applying a backwards elimination algorithm on the multivariable Cox model, a reduced prognostic model yielded genetic factors with a significant impact on OS: DDX41, FLT3-ITD, and TP53. The results of the reduced model were visualized using predicted survival probabilities for all combinations of the resulting mutations while clinical variables are fixed at the median or mode. This led to a stratification into three risk categories: DDX41^mut as favorable, DDX41^wt/TP53^wt/FLT3-ITD^neg as intermediate, and TP53^mut or FLT3-ITD^pos as adverse.

Fig. 5. Impact of clinical, mutational, and cytogenetic features on overall survival.

Forest plot displaying hazard ratios based on results from Cox regression analysis using clinical variables [age, sex, ECOG performance status ( ≥ 2 vs. <2), white blood cell counts (log₁₀-transformed)], treatment, and all mutations with a frequency of ≥4%; HR hazard ratio, CI confidence interval.