cfDNA Methylation-Based Pan-Hematologic Prognostic Classification

Background: The prognosis in hematologic (heme) malignancies remains challenging due to the complex and diverse biology of these diseases and invasive procedures. Although risk stratification/prognostic methods (e.g., cytogenetics, mutations, clinical parameters) exist for single heme indications, there is a lack of prognostic tools that simultaneously address the biological heterogeneity across multiple hematologic neoplasms. This gap underscores the need for a pan-heme prognostic classifier. Here, we developed a comprehensive prognostic classifier based on targeted methylation sequencing of cfDNA in peripheral blood.

Methods: Our pan-heme classifier was trained using 621 heme participants (32 Hodgkin lymphoma, 265 Non-Hodgkin Lymphomas [NHL], 128 myeloproliferative neoplasms, 72 plasma cell neoplasms, 124 monoclonal gammopathy of undetermined significance [MGUS]) and evaluated with a subset of 320 heme patients (18 Hodgkin lymphomas, 188 NHL, 70 myeloproliferative neoplasms, 44 plasma cell neoplasms) from the Circulating Cell-free Genome Atlas substudy 2 using a six-fold nested cross-validation. The top K principal components (PCs) of methylation beta values with age, sex, body mass index, smoking status, stage, and heme subtype were used to fit a Cox model. Personalized risk scores were generated for each sample by exponentiating the linear predictor. The C-index was calculated using risk scores and survival outcomes. Participants were stratified into high- and low-risk groups based on the risk scores of corresponding indications, with scores above 66.7th percentile categorized as high risk and below this threshold categorized as low risk. The risk classification was assessed by Kaplan-Meier curves and log-rank tests. To better understand methylation biology driving the classification performance, we conducted differential methylation analysis between high-risk (N=213) and low-risk (N=107) groups, using beta-binomial regression with an arcsine link function to model the region-level count data. Hypothesis testing was performed by Wald-test for each region. The differentially methylated regions (DMRs) were considered significant if q-value was less than 0.05 and the absolute value of beta-binomial model delta coefficients was greater than 0.1. The closest genes were annotated to significant DMRs, and pathway over-representation analysis was conducted using the annotated genes from significant hypermethylated DMRs or hypomethylated DMRs.

Results: Methylation features are significant predictors of survival outcome even after adjusting for clinical covariates. The pan-heme risk stratification achieved an overall C-index of 0.70 (95% CI: 0.62-0.78), indicating good discrimination. The survival of the high- and low-risk groups was significantly different (log-rank p-value<0.0001). Noteworthy results were observed for specific indications: Myelodysplastic syndromes (MDS; N=16) showed a robust discriminatory power with a C-index of 0.86 (95% CI: 0.67-1.00). Hodgkin lymphoma (N=18) achieved a C-index of 0.91 (95% CI: 0.76-1.00). Diffuse large B-cell lymphoma (DLBCL; N=33) yielded a C-index of 0.71 (95% CI: 0.52-0.89). Follicular lymphoma (N=50) reached a C-index of 0.63 (95% CI: 0.39-0.86). Plasma cell neoplasm (N=44) showed a C-index of 0.65 (95% CI: 0.37-0.90). Moreover, the differential methylation analysis identified 4,110 DMRs. Overlapped features in significant DMRs compared to background were enriched in genomic regions, including CpG islands, CpG shelves, and CpG shores. Pathway enrichment analysis of hypermethylated DMRs in high-risk groups revealed significant associations with calcium, cAMP, MAPK, Ras, Rap1, Wnt, and hippo signaling pathways.

Conclusions: We developed a novel, minimally-invasive, pan-heme prognostic classifier utilizing cfDNA blood-based targeted methylation sequencing technology. This cfDNA-based approach demonstrated potential for risk stratification across a diverse range of heme malignancies. Further, an approach utilizing a simple, minimally-invasive blood draw instead of a combination of multiple testing modalities, offers a pragmatic approach to disease prognosis across large heterogeneous groups of hematologic malignancies.