Increased Local Disorder of DNA Methylation Forms the Basis of High Intra-Leukemic Epigenetic Heterogeneity and Enhances CLL Evolution

We recently demonstrated that genetic heterogeneity contributes to evolution and clinical outcome in chronic lymphocytic leukemia (CLL) (Landau et al., Cell 2013). As both genetic and epigenetic alterations contribute to the cellular phenotype and its fitness for selection, we hypothesized that heterogeneity in DNA methylation impacts leukemic evolution as well.

We analyzed existing CLL methylation data (Kulis et al., Nat Gen 2012), and found higher heterogeneity in 450K arrays of 139 CLLs compared to 23 normal B cell samples, with a higher proportion of sites having intermediate levels of methylation (13.7% vs 10.7%, P<0.0001). Next, we re-analyzed whole genome bisulfite sequencing data from 2 CLL and 3 normal B cell samples to distinguish between 2 possible explanations for heterogeneity: 1) a mixture of cell subpopulations with ordered but differing methylation states, and 2) increased proportion of cells with ‘noisy’ or disordered methylation states. We found that the bulk of heterogeneity resulted from the latter, as evidenced by a high proportion in CLL of individual sequencing reads with a discordant methylation state in neighboring CpGs.

To validate this observation, we applied reduced representation bisulfite sequencing (RRBS) to generate genome-scale information of methylation at critical sites (e.g., CG islands) of 108 genetically-characterized CLLs. As predicted, we found higher proportions of discordant reads in CLL than either samples from 30 normal B cells or from 35 diverse healthy human tissues (average of 24.7% vs. 17.6% or 16.3%, respectively, P<0.00001). Similarly high proportions of discordant reads were also found in 31 cancer cell lines. Samples with somatic mutations in methylation modulators (e.g., DNMT3A) had increased proportion of discordant reads (P<0.001). Importantly, the presence of methylation ‘noise’ was associated with reduced correlation between promoter methylation and gene silencing (P<0.0001), and with substantial allele specific expression, both measured by RNA-seq of 48 CLLs. These findings suggest that much of DNA methylation heterogeneity in CLL stems from locally disordered methylation, and shed light on how epigenetic alterations in human cancer arise in part from epigenetic drift.

Increased epigenetic ‘noise’ is expected to result in a more plastic evolutionary landscape that facilitates the emergence of fitness-enhancing genetic and epigenetic alterations. Indeed, the increase in discordant reads in CLL compared to normal samples was most prominent in genomic regions important to transcription (promoters, CG islands [P <1*e-14] vs. introns, CGI shelves [P <1*e-2]). In addition, significantly higher levels of discordant reads were found in promoters of genes silenced in CLL, in differentially methylated regions and in genes important to pluripotency potential (e.g., stem cell modules, Q<1*e-14). Finally, samples with increased disordered methylation had higher numbers of subclonal, but not clonal mutations (P<0.001 vs. P=0.26, respectively).

To observe the relationship between genetic and epigenetic evolution, we used RRBS to longitudinally study 14 CLLs with characterized patterns of genetic evolution (median time between samples 3.5 yrs; 6 unevolved, 8 evolved). While unevolved CLLs revealed largely unchanged epigenetic landscapes, evolved CLLs demonstrated decreased methylation over time in promoters enriched for stem cell and MYC targets gene sets (N=347, Q<0.0001), and increased methylation in TP53 targets (N=336, Q<0.00001). Moreover, in evolved CLLs, specific promoters with disordered methylation across all samples (e.g., TERT, WT1), revealed changes over time in methylation proportions corresponding to increases in subclone size inferred from the genetic analysis.

Increased heterogeneity was found to be associated with shorter retreatment-free survival in a cohort of 46 patients treated after sampling (P=0.03, HR of 3.13), in a regression model including established risk indicators (IGHV unmutated status, del(17p), del(11q) and treatment prior to sample).

Our data suggest that heterogeneity in CLL DNA methylation results from stochastic drift that subsequently undergoes positive selection. Thus, we propose that locally disordered DNA methylation may have a similar role to genetic instability, enhancing the ability of cancer cells to search for superior evolutionary trajectories.