Subtyping Burkitt Lymphoma by DNA Methylation

Abstract

Burkitt lymphoma (BL) is an aggressive germinal center B-cell-derived malignancy. Historically, sporadic, endemic, and immunodeficiency-associated variants were distinguished, which differ in the frequency of Epstein-Barr virus (EBV) positivity. Aiming to identify subgroups based on DNA methylation patterns, we here profiled 96 BL cases, 17 BL cell lines, and six EBV-transformed lymphoblastoid cell lines using Illumina BeadChip arrays. DNA methylation analyses clustered the cases into four subgroups: two containing mostly EBV-positive cases (BL-mC1, BL-mC2) and two containing mostly EBV-negative cases (BL-mC3, BL-mC4). The subgroups BL-mC1/2, enriched for EBV-positive cases, showed increased DNA methylation, epigenetic age, and, in part, proliferation history compared to BL-mC3/4. CpGs hypermethylated in EBV-positive BLs were enriched for polycomb repressive complex 2 marks, while the CpGs hypomethylated in EBV-negative BLs were linked to, for example, B-cell receptor signaling. EBV-associated hypermethylation affected regulatory regions of genes frequently mutated in BL (e.g., CCND3, TP53) and impacted superenhancers. This finding suggests that hypermethylation may compensate for the lower mutational burden of pathogenic drivers in EBV-positive BLs. Though minor, significant differences were also observed between EBV-positive endemic and sporadic cases (e.g., at the SOX11 and RUNX1 loci). Our findings suggest that EBV status, rather than epidemiological variants, drives the DNA methylation-based subgrouping of BL.

Keywords: Africa; Burkitt lymphoma; DNA methylation; Epstein–Barr virus; immunodeficiency.

FIGURE 1

DNA methylation‐based subgrouping of Burkitt lymphoma (BL) cases. (A) UMAP visualization (25 neighbors) of 96 BL cases based on the 309 078 CpGs representing the global DNA methylation landscape. (B) UMAP visualization (15 neighbors) of 9313 CpGs (standard deviation (SD) > 0.25) colored and shaped according to the number of clusters (k = 2, k = 4) determined by the combination of CpGs filtering using SD and k‐means clustering. (C) Heatmap depicting DNA methylation levels of the 9313 CpGs across the 96 BL samples. Columns represent individual samples grouped into four optimal clusters as determined by SD and k‐means clustering. Rows represent CpGs, further categorized into three modules (BL‐M1‐3) using k‐means clustering. Sample features are annotated at the top of the heatmap, including the four clusters (SD, k‐means), epidemiological variants, EBV status, age at diagnosis, epigenetic age based on Horvath clock, proliferation history determined with epiCMIT, B‐cell fraction calculated from DNA methylation data, purity score received from the InfiniumPurify package. CpGs are annotated using chromatin states defined in germinal center B cells (gcBCs) and the Kulis modules, which are grouped according to the patterns (I–IV) in the paper [17] (for details see Supporting Information). (D) Bar plot showing the distribution of epidemiological variants (sporadic, endemic, and immunodeficiency‐associated BL) and EBV status across the four identified clusters. (E) Box plots illustrating key biological and epigenetic features of the clusters, including biological age (age at diagnosis), epigenetic age calculated using the Horvath clock, proliferation history determined using the epiCMIT package, and median DNA methylation levels based on 309 078 CpGs and 9313 CpGs. n/a: not applicable. Statistical comparisons are summarized in Tables S10 and S11.

FIGURE 2

Comparative DNA methylation profiling of EBV‐negative and EBV‐positive Burkitt lymphoma (BL). (A) Box plots comparing EBV‐negative and EBV‐positive BL for biological age (age at diagnosis), epigenetic age calculated using the Horvath clock, proliferation history (epiCMIT), and median DNA methylation levels across 309 078 CpGs. (B) Heatmap depicting DNA methylation levels of 11 839 CpGs found significantly differentially methylated between EBV‐negative and EBV‐positive BL cases (adjusted p < 0.01, |Δβ| > 0.3, corrected for fixation technique and array). Sample features are annotated at the top of the heatmap, including the four clusters (SD, k‐means), epidemiological variants, EBV status, age at diagnosis, epigenetic age based on Horvath clock, proliferation history determined with epiCMIT, B‐cell fraction calculated from DNA methylation data, purity score received from the InfiniumPurify package. Columns represent samples, rows depict CpGs. CpGs are annotated using chromatin states defined in germinal center B cells (gcBCs) and the Kulis modules, which are grouped according to the patterns (I–IV) in the paper [17]. (C) Transcription factor enrichment analysis (based on ENCDOE and ChEA) on the genes associated with the 11 839 DMCs. (D) Bar plot displaying the distribution of the 11 839 DMCs within chromatin states defined in gcBCs. (E) Enrichment analysis of genes using a subset (4416 CpGs) of the 11 839 CpGs (non‐PP2 signature), not associated with binding sites of SUZ12/EZH2 and not located within poised‐promoter regions. Enrichment analysis was performed for ENCODE and ChEA transcription factors, as well as Wiki pathways. The y‐axis displays the top 10 most significant gene ontology terms. n/a: not applicable; PP2: poised promoter and polycomb repressive complex 2.

FIGURE 3

Heatmap of DNA methylation levels in regulatory regions of genes recurrently mutated in Burkitt lymphoma (BL). Heatmap displaying significant differentially methylated CpGs of EBV‐positive BL compared to EBV‐negative BL (adjusted p < 0.01, |Δβ| > 0.2) within regulatory regions (promoter, enhancer) for recurrently mutated genes in BL. Median DNA methylation for the groups EBV‐positive sBL (n = 6) and eBL (n = 19), as well as EBV‐negative sBL (n = 33), were calculated. In addition, median gene expression for germinal center B cells (gcBCs, n = 5) and EBV‐negative sBLs (n = 21) are included as annotation bars on the right side, showing that four genes (CACNA1G, CSMD3, GPC5, and SYT14) were not expressed in either gcBCs or sBLs. Genes are ordered according to the mutational frequency in BLs based on the findings by López et al. [4], with the gene with the highest frequency placed at the top. Gene annotations reflect multiple mutational frequency parameters: Higher frequencies based on EBV status and geographic origin [8, 9] and patient age. Additional annotations indicate replication timing in the GM12878 cell line (ENCODE Repli‐seq data), chromatin states defined in gcBCs, and superenhancers the selected CpGs are located in [36].