Long-read epigenomic diagnosis and prognosis of Acute Myeloid Leukemia

Abstract

Acute Myeloid Leukemia (AML) is an aggressive cancer with dismal outcomes, vast subtype heterogeneity, and suboptimal risk stratification. In this study, we harmonized DNA methylation data from 3,314 patients across 11 cohorts to develop the Acute Leukemia Methylome Atlas (ALMA) of diagnostic relevance that predicted 27 WHO 2022 acute leukemia subtypes with an overall accuracy of 96.3% in discovery and 90.1% in validation cohorts. Specifically, for AML, we also developed AML Epigenomic Risk, a prognostic classifier of overall survival (OS) (HR=4.40; 95% CI=3.45-5.61; P<0.0001), and a targeted 38CpG AML signature using a stepwise EWAS-CoxPH-LASSO model predictive of OS (HR=3.84; 95% CI=3.01-4.91; P<0.0001). Finally, we developed a specimen-to-result protocol for simultaneous whole-genome and epigenome sequencing that accurately predicted diagnoses and prognoses from twelve prospectively collected patient samples using long-read sequencing. Our study unveils a new paradigm in acute leukemia management by leveraging DNA methylation for diagnostic and prognostic applications.

Figure 1. The Acute Leukemia Methylome Atlas Study.

A) Overall design of computational models. B) Acute Leukemia Methylome Atlas (ALMA) defined by dimension reduction algorithm PaCMAP in the overall population. Each data point is a patient sample deriving from bone marrow or peripheral blood at diagnosis, relapse, or remission. As an unsupervised model, only 5mC values of 331,556 CpGs were used to define the clusters. C) Clinical implementation of long-read nanopore sequencing pipeline. AML, acute myeloid leukemia; ALL, acute lymphoblastic leukemia; MDS, myelodysplastic syndrome; MPAL, mixed phenotype acute leukemia; APL, acute promyelocytic leukemia; PaCMAP, pairwise controlled manifold approximation; WHO, World Health Organization; PB, peripheral blood; BM, bone marrow; EDTA, Ethylenediaminetetraacetic acid; HMW gDNA, high molecular weight genomic DNA.

Figure 2. ALMA Subtype classification according to WHO 2022 subtypes.

Sankey diagram displaying discovery cohort comparison between WHO 2022 diagnosis (left) and ALMA Subtype (right) for samples with A)or without B) WHO 2022 clinical annotation available. The width of the bands indicates the number of patient samples in each category. Comparison applied to validation cohort describing samples with C) or without D) WHO 2022 annotation available. E)Receiver operating characteristic (ROC) curves for the discovery cohort (n=2471) and F) validation cohort (bottom, n=200). The curves depict the true positive rate against the false positive rate for model predictions. Specific per-class AUC results are available in the electronic notebook. G)Table summarizing classification performance metrics: accuracy, macro F1, weighted F1, and Cohen’s Kappa for the training and test cohorts, indicating the overall predictive performance.

Figure 3. Patient outcomes and multivariate analyses of AML Epigenomic Risk groups.

Patient outcomes by AML Epigenomic Risk groups in A) discovery cohort and B) validation cohort with EFS (top) and OS (bottom) of AML Epigenomic Risk^high (orange) and AML Epigenomic Risk^low (blue). Multivariate analysis adjusting for other confounding variables of OS in C) discovery cohort and D) validation cohort and EFS in E) discovery cohort and F) validation cohort. MRD 1, minimal residual disease at end of first induction; FLT3 ITD, FMS-like tyrosine kinase-3 internal tandem duplication; CI, confidence interval; Ref., reference.

Figure 4. Development and testing of 38CpG AML signature.

A) Stepwise workflow for generating the 38CpG-AMLsignature model, including data preprocessing, batch correction, survival analysis, and penalized Cox PH modeling, resulting in a concise 5mC-based prognostic risk score for AML. B) Manhattan plot showing EWAS results with the significance of CpG probes across the genome, highlighting 200 CpGs that remained significant after p<10e-5 threshold. C) Stability analysis of 200 selected CpGs, showing the frequency of non-zero coefficients across 1000 model loops, which led to 38 CpGs being selected. D) Distribution of 38CpG-HazardScores in the discovery (left) and test (right) cohorts, dividing patients into high and low risk based on a 50% cutoff. E) Kaplan-Meier survival curves for EFS (top) and OS (bottom) in the discovery cohort (left, n=946) and F) validation cohort (right, n=200), comparing 38CpG-AMLsignature^high (orange) and 38CpG-AMLsignature^low (blue) groups.

Figure 5. Forest plots of 38CpG AML signature against known confounding factors.

A) Discovery cohort by OS. B) Discovery cohort by EFS. C) Validation cohort by OS. D) Validation cohort by EFS.

Figure 6. Specimen-to-result prospective testing using rapid nanopore long-read sequencing.

A) Nanopore test cohort samples (n=12) mapped onto Acute Leukemia Methylome Atlas (ALMA) based on the combined methylation values of 331,556 CpGs. Each point represents a patient sample colored by ALMA Subtype. Samples with similar epigenomes cluster together, forming the island patterns seen through the scatterplot. Patient UF1829 was misplaced in two dimensions, but correctly placed in five dimensions. B) Sankey plot comparing clinical diagnosis information provided at sample collection (left) with ALMA Subtype classifier predictions (right) for the nanopore test cohort. The flow lines illustrate the correspondence between initial diagnostic classifications and the predicted ALMA subtypes. NUP98-fusion, Nucleoporin 98 fusion; PB, Peripheral Blood; BM, Bone Marrow; MDS, Myelodysplastic Syndrome; S/P: Status Post; Flow, Flow cytometry; TKD, Tyrosine Kinase Domain; CNS, Central Nervous System; BMBx, Bone Marrow Biopsy; NPM1, Nucleophosmin 1.