Genome-Wide DNA Methylation Analysis Reveals Biological and Clinical Insights In Relapsed Childhood Acute Lymphoblastic Leukemia: A Report From The COG ALL Target Project

Although survival of children with B-cell acute lymphoblastic leukemia (B-ALL) has improved substantially over time, 15% to 20% of patients will relapse, and most of those who experience a bone marrow relapse will die. A better understanding of genetic and epigenetic aberrations in relapsed ALL will facilitate new strategies for risk stratification and targeted therapy. In this collaborative study with the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) project, we performed high resolution genome-wide DNA methylation profiling using the HELP (HpaII tiny fragment Enrichment by Ligation-mediated PCR) array on a total of 178 (110 diagnosis, 68 relapse) leukemia samples from 111 patients with childhood B-ALL enrolled on the Children’s Oncology Group (COG) clinical trials who experienced relapsed, and 12 normal preB samples isolated from the bone marrows of 12 healthy individuals. The HELP array covers 117,521 CpG sites, annotated to ∼22,000 gene promoters. For eight diagnosis/relapse pairs, base-pair resolution DNA methylation using the eRRBS (enhanced Reduced Representation Bisulfite Sequencing) method was also performed on Illumina HiSeq2000.

The median relapse time for the 111 patients was 21.8 months (range 2.1 to 56.2). Unsupervised clustering analysis using the HELP data revealed seven clusters: one cluster contained only the 12 normal preB samples; four clusters were enriched with MLLr, ETV6/RUNX1, Trisomy 4+10, and TCF3/PBX1 samples, respectively. The sixth cluster was not enriched for specific cytogenetic cases, but interestingly, all cases in this cluster were NCI High Risk (age>10 years or WBC>=50,000; p<0.0001, Fisher’s Exact test) while the seventh cluster has a mixture of other cases.

Supervised analysis of HELP profiles between paired relapse/diagnosis samples (n=67) revealed a markedly aberrant DNA methylation signature (1011 probesets, 888 genes, FDR<0.01 and methylation difference dx >25%, paired t-test), with 70% of the genes hyper- and 30% hypo-methylated in relapse samples. Using a Bayesian predictor and leave-one-out cross validation, this methylation signature could predict a sample as diagnosis or relapse with 95.3% accuracy. When comparing early (<36 months; n=50) versus late relapses (>=36 months; n=18), we detected a profound hypermethylation signature in early relapse (96.6% of the 610 probesets, 544 genes, FDR<0.01, dx >25%).

Finally, we identified 1800 probesets (1658 genes) as differentially methylated within all cytogenetic subtypes described above compared to the normal preB samples (Dunnett’s test with normal preB as reference, FDR<0.01, dx>25%). Again the majority (70%) of those genes were hypermethylated in relapse as compared to diagnostic and normal preB.

The base-pair resolution and more comprehensive eRRBS methylation analysis for the eight pairs of samples identified 39,679 CpG sites as differentially methylated (dx >25%, FDR<0.01), with 78.2% CpG sites hyper- and 21.2% hypo-methylated in relapse samples. Remarkably, the hypermethylated CpGs are primarily in promoter regions (50%, defined as +/-1kb to TSS), followed by intergenic (26%), then intragenic (14%), and exonic (10%) regions. In contrast, the hypomethylated CpGs are mainly in intragenic (48%), followed by intergenic (31%), exonic (14%) and promoter (7%) regions. The hypermethylated CpGs were mainly in CpG islands (86%) or CpG shores (10%), while hypomethylated CpGs were not (CpG islands: 8%, CpG shores: 27%). We further identified 3040 differentially methylated regions (DMRs) with a median size 426 bp. 78.4% of those DMRs were hyper- (1362 gene promoters) and 21.6% hypo-methylated (98 promoters) in relapse compared to diagnostic samples. Gene set enrichment and Ingenuity pathway analysis showed epigenetically disrupted pathways that are highly involved in cell signaling, and embryonic and organismal development.

Taken together, our genome-wide high resolution DNA methylation analysis on a large cohort of relapsed childhood B-ALL from the COG trial identified unique methylation signatures that correlated with relapse and with specific genetic subsets. Those methylation signatures featured prevailing promoter hypermethylation and to a lesser extent, intrageneic hypomethylation. Epigenetically dysregulated gene networks in those relapse samples involved cell signaling, and embryonic and organismal development.