Identification of a key role of widespread epigenetic drift in Barrett's esophagus and esophageal adenocarcinoma

Abstract

Background: Recent studies have identified age-related changes in DNA methylation patterns in normal and cancer tissues in a process that is called epigenetic drift. However, the evolving patterns, functional consequences, and dynamics of epigenetic drift during carcinogenesis remain largely unexplored. Here we analyze the evolution of epigenetic drift patterns during progression from normal squamous esophagus tissue to Barrett's esophagus (BE) to esophageal adenocarcinoma (EAC) using 173 tissue samples from 100 (nonfamilial) BE patients, along with publically available datasets including The Cancer Genome Atlas (TCGA).

Results: Our analysis reveals extensive methylomic drift between normal squamous esophagus and BE tissues in nonprogressed BE patients, with differential drift affecting 4024 (24%) of 16,984 normally hypomethylated cytosine-guanine dinucleotides (CpGs) occurring in CpG islands. The majority (63%) of islands that include drift CpGs are associated with gene promoter regions. Island CpGs that drift have stronger pairwise correlations than static islands, reflecting collective drift consistent with processive DNA methylation maintenance. Individual BE tissues are extremely heterogeneous in their distribution of methylomic drift and encompass unimodal low-drift to bimodal high-drift patterns, reflective of differences in BE tissue age. Further analysis of longitudinally collected biopsy samples from 20 BE patients confirm the time-dependent evolution of these drift patterns. Drift patterns in EAC are similar to those in BE, but frequently exhibit enhanced bimodality and advanced mode drift. To better understand the observed drift patterns, we developed a multicellular stochastic model at the CpG island level. Importantly, we find that nonlinear feedback in the model between mean island methylation and CpG methylation rates is able to explain the widely heterogeneous collective drift patterns. Using matched gene expression and DNA methylation data in EAC from TCGA and other publically available data, we also find that advanced methylomic drift is correlated with significant transcriptional repression of ~ 200 genes in important regulatory and developmental pathways, including several checkpoint and tumor suppressor-like genes.

Conclusions: Taken together, our findings suggest that epigenetic drift evolution acts to significantly reduce the expression of developmental genes that may alter tissue characteristics and improve functional adaptation during BE to EAC progression.

Keywords: Barrett’s esophagus (BE); DNA methylation; Endogenous retroviruses (ERVs); Epigenetic drift; Esophageal adenocarcinoma (EAC); Neoplastic progression; Tissue age; Transcriptional repression in cancer.

Fig. 1

Proportion of CpGs and CpG islands that drift differentially in Barrett’s esophagus vs. normal squamous (NS) esophagus among over 146 k hypomethylated probes in NS tissue

Fig. 2

Representative karyographs of four autosomes. Chromosomes 2 and 12 exhibit typical methylomic drift patterns while chromosomes 17 and 19 exhibit high-density methylomic drift. Top track: chromosome banding. Middle track: array-based CpG island positions. Bottom track: positions of CpG islands that undergo methylomic drift in 64 BE samples (mean levels color-coded)

Fig. 3

CpG island-level methylation heatmap (β values) of 1317 drift CpG islands (rows) and 10 NS and 64 non-dysplastic BE samples (columns) ordered by their respective means. See text for details

Fig. 4

Pairwise correlations between island-CpGs and other CpGs designated as island, shore, and shelf, associated with the same island, as a function of genomic distance at a resolution of 10 bp. “Static” (nondrifting) CpG islands (black), drift-associated islands (red). Shaded area represents the approximate boundary location between shores and shelves

Fig. 5

a Typical drift patterns for BE and EAC samples by type of β value distribution. Shown are the methylation distributions for 11,425 island-associated drift CpGs (minimum of 5 drift CpGs per island). The three drift groups are based on unimodal low drift (group L), bimodal intermediate drift (group I), and bimodal high (group H). b Simulated methylation densities (arbitrary time scale) for an island-like region of 50 CpGs and 1000 cells mimicking the array-based measurements of epigenetic drift in panel a. Details provided in text

Fig. 6

CpG methylation transition rates based on paired longitudinal biopsies (collected at least 3–4 years apart) from the Cleveland Clinic (CC) (black) and Case Western (CW) (red). Forward (increasing) methylation transition rates represent the annual rate of CpG probes advancing past a threshold of β = 0.2, while retarding transition rates represent the fraction of CpG probes transitioning from high to low methylation (below β = 0.2)

Fig. 7

See “Methods”