AUTHORS
Norberto Kzan, MD, Luiza Viza Fonseca, MD2, Laura Ribeiro Ribeiro de Matos, MD3, Francisco Marchi, PhD4, Pedro V. B. Orsini, MD5, Luiz Ferreira Cortes6, Paulo Vidal Campregher, MD, PhD7, Fabio Pires S Santos, MD, PhD8
AFFILIATIONS
1 Einstein Hospital Israelita, São Paulo, São Paulo, Brazil
2 Einstein Hospital Israelita, São Paulo, São Paulo, Brazil
3 Einstein Hospital Israelita, São Paulo, São Paulo, Brazil
4 Alma Genomics, Gainesville, Florida, United States
5 Alma Genomics, Gainesville, Florida, United States
6 Einstein Hospital Israelita, São Paulo, São Paulo, Brazil, Genesis Genomics, São Paulo, Brazil
7 Einstein Hospital Israelita, São Paulo, São Paulo, Brazil, Genesis Genomics, São Paulo, Brazil
8 Einstein Hospital Israelita, São Paulo, São Paulo, Brazil
ABSTRACT
Background: Aging is associated with an increase in DNA methylation at Polycomb repressive complex 2 (PRC2) target regions and is one of the explanations behind epigenetic clocks that estimate age based on methylation status of selected CpGs. More recently, causal epigenetic clocks have been developed that separate adaptive (AdaptAge) from damaging (DamAge) methylation programs. It is currently not known how the altered epigenetic landscape of AML affects epigenetic aging measures, and whether these measures reflect chronological aging or the molecular biology of leukemia. We evaluated PRC2-target and causal epigenetic-aging metrics in a large multicohort dataset of AML patients.
Methods: We analyzed the publicly available ALMA compilation of methylation datasets, which harmonized several public datasets into a batch-corrected matrix of 331,556 CpGs shared across 450K and EPIC platforms. We calculated PRC2-target methylation (PRC2m), PRC2-AgeIndex, and AdaptAge and DamAge as raw scores and epigenetic-age acceleration measures. The core analysis included 449 patients with WHO 2022 subtype assignment and a complete 77-gene mutational panel, while secondary, mutation-independent analyses included 1,449 patients. Associations were evaluated using quantile regression adjusted for cohort, with age modelled using restricted cubic splines, and Benjamini-Hochberg correction for multiple testing. Subtype discrimination was assessed using multinomial and binary regression with bootstrap optimism correction. Overall survival was analyzed using Cox models
adjusted for age, ELN 2022 genetic risk, and cohort.
Results: PRC2m correlated strongly with AdaptAge (Spearman rho=0.88; N=449), whereas PRC2m and PRC2-AgeIndex were nearly collinear (rho approximately 0.999), indicating a shared PRC2-related axis. All four primary metrics differed across WHO 2022 subtypes (global p<0.0001). CEBPA-mutated AML showed higher PRC2m, PRC2-AgeIndex, and AdaptAge and lower DamAge, whereas core-binding-factor AML demonstrated the opposite profile. DNMT3A mutations (n=87) were associated with lower PRC2m and AdaptAge and higher DamAge; in contrast, IDH1/IDH2 mutations were associated with lower DamAge. Raw scores showed no detectable association with chronological age. Adding individual metrics to an age-plus-cohort model improved WHO 2022 subtype discrimination, most strongly for AdaptAge (likelihood-ratio p=2×10⁻⁷; delta-Nagelkerke R²=0.056). Adjusted binary models achieved AUCs of 0.84 for predicting core-binding-factor AML using AdaptAge and 0.80 for KMT2A-rearranged AML using DamAge. In the overall survival cohort (N=399; 221 deaths), no epigenetic aging metric added prognostic information in adults or children after covariate adjustment. In an exploratory analysis restricted to NPM1-mutated AML (N=119; 64 deaths), PRC2m, PRC2-AgeIndex, and AdaptAge were associated with survival (likelihood-ratio p approximately 0.002), increasing the optimism-corrected C-index from 0.675 to 0.707–0.715.
Conclusions: Epigenetic aging-derived methylation metrics in AML appear to capture molecular state more strongly than chronological age. They reproduce distinct DNMT3A- and IDH-associated methylation patterns, distinguish WHO 2022 subtypes, and provide modest incremental information on subtype classification. A potential prognostic association was seen in NPM1-mutated AML that should be validated in independent cohorts.
DISCLOSURES
Norberto Kzan, MD: No relevant conflicts of interest to declare.