Multimodal epigenetic microarray classifies AML, MDS and controls using peripheral blood

Abstract Objectives: Aberrant DNA methylation (5mC) and hydroxymethylation (5hmC) patterns are hallmarks of acute myeloid leukemia (AML) and myelodysplastic syndromes (MDS), yet epigenetic markers remain absent from diagnostic workflows. We combined direct chemoenzymatic 5mC and 5hmC labeling with microarray technology for enhanced epigenetic profiling of AML and MDS, by simultaneously detecting differentially methylated and hydroxymethylated genomic regions between cohorts. Within a large multicenter trial funded by the Horizon Cancer Mission (NCT05735704), this study focuses on myeloid conditions, aiming to develop a simple, peripheral blood based tool for AML and MDS detection and monitoring using epigenetic markers. Methods Peripheral blood samples from newly diagnosed AML patients, MDS patients, and controls were collected across four clinical centers. Genomic DNA was extracted, then 5hmC and 5mC were simultaneously labeled by chemoenzymatic reaction to produce dual-labeled DNA with distinct fluorophores. Labeled DNA was hybridized to custom-designed microarrays targeting 13,000 genomic loci susceptible to epigenetic modulation in AML. Following hybridization, microarrays were scanned for optical detection of fluorescent signals. We performed data analysis using logistic regression and developed a multi-channel approach that combines both epigenetic modifications into an integrated classification model. Classification performance was assessed through multiple leave-one-out cross-validations. Results Blood samples from 65 AML (male/female 51/49%; median age 65 [range 18-90]), 54 MDS (50/50%; 74 [19-90]), and 62 control (31/69%; 66 [32-87]) participants were analyzed. According to ELN-2022 risk stratification of evaluable cases, 18% were classified as favorable risk, 36% as intermediate and 36% as adverse. The majority of MDS patients had low risk disease according to IPSS-R (69%), with the remainder comprising intermediate and high-risk patients and those with missing risk data; overall, 37% of MDS patients required therapeutic intervention. Analysis of AML vs controls achieved 98.3% AUC with 96.9% sensitivity and 93.5% specificity using 200 biomarkers. Multimodal 5mC/5hmC analysis outperformed single-marker analysis. Similar enhanced multimodal performance was achieved for MDS vs controls using 100 biomarkers and yielded 96.4% AUC with 90.7% sensitivity and 95.2% specificity. Notably, for AML versus MDS discrimination, 5hmC analysis achieved the best performance with an AUC of 98.2%, correctly classifying 90.8% of AML cases and 96.3% of MDS cases. Analysis of genomic loci enriched in differential markers relative to the full array revealed 78 enriched genes (>2x); HIC1, a known tumor suppressor hypermethylated in various cancers was enriched in all 3 comparisons. While HIC1's 16 markers had predominantly normal 5mC and 5hmC signals across cohorts on the full array, the single differential marker from AML vs Control analysis was hypermethylated in AML. Protocadherin gene family, involved in cell adhesion and linked to cancer development when dysregulated, had 483 markers on full array, 86% with normal methylation in AML compared to controls. Protocadherin markers within AML vs Control differential analysis were mostly hypermethylated (85%) in AML samples. Conclusions Our results demonstrate successful distinction between all three cohorts by analyzing epigenetic markers from peripheral blood using a novel methodology. Enrichment analysis revealed biologically relevant targets including HIC1 and Protocadherin silencing in AML. This work presents the first integrated analysis combining 5mC and 5hmC data into a unified classification approach with enhanced performance compared to single-marker analysis. Notably, 5hmC patterns showed superior discrimination between AML and MDS, potentially reflecting distinct active demethylation dynamics between more indolent states and overt acute myeloid leukemia. Our approach enables improved clinical decision-making regarding invasive bone marrow procedures and provides reliable monitoring for MDS disease progression. This first-in-class epigenetic microarray offers a low-cost, non-invasive tool with significant potential for enhancing diagnostic precision in myeloid malignancies.