- Aberrant DNA methylation, especially promoter hypermethylation, plays a key role in cancer development and progression. The document analyzes DNA methylation patterns in esophageal tissues using Illumina arrays.
- It finds location-specific heterogeneity between normal, Barrett's esophagus (BE), and esophageal adenocarcinoma (EAC) tissues. Methylation increases with chronological age and shows drift over time in BE, with thresholds linked to cancer formation.
- In EAC, promoter hypermethylation of "drift islands" containing many variably methylated CpG sites is linked to gene silencing, including of tumor suppressor genes. Spatial patterns of methylation drift differentiate pre-cancerous
TEST BANK For Radiologic Science for Technologists, 12th Edition by Stewart C...
CISNET Multiscale Meeting Presentation: Drift CpGs in Normal and Progressed Esophagus
1. Spatial and temporal epigenetic pattern
gradients differentiate normal and
progressed tissues in esophagus
Sean Maden
Data Analyst Assistant
Grady Lab
Clinical Research, Fred Hutch
2. Epigenetics Overview
• Epigenetics - a host of biological features and mechanisms that
affect gene expression independent of sequence modifications:
• Genome-wide CpG methylation
• Chromatin remodeling
• Nucleosome positioning
• lnc- and microRNA activity, etc.
• Trend of the field: Work towards integrative and multiscale
analysis incorporating CpG methylation and other genomic,
molecular and clinical data
3. Aberrant Methylation in Cancer
• Promoter CpG methylation (or
“hypermethylation”) can
reduce or knock out a gene’s
expression.
• In cancer: repressed tumor
suppressors, upreg.
oncogenes..
• Methylation linked to changes
in cell potency during normal
development, which may be
reversible
• Cancer Stem Cell (CSC) model
• Clonal grown pattern
• Etc.
Image modified from:
Toh et al. “Epigenetics in cancer stem cells” (2017) Molecular Cancer 16:29.
4. CpG Methylation in Esophageal Cancers
• Extensive integrative analysis
showed location-specific
molecular heterogeneity
• CpG Island Methylator
Phenotype (CIMP) in lower-
esophagus (where BE and
EAC form)
• Molecular evidence
distinguishing EAC from ESCC
and non-CIN GEAs
Image modified from:
The cancer Genome Atlas Research Network. “Integrated genomic characterization of oesophageal carcinoma” (2017) Nature 541, 169-175.
5. Measuring Methylation: Infinium Array Technology
Images from: Illumina Infinium Array documentation, Georg Leubeck
• Illumina Infinium HM450 BeadChip array
• Assays genome-wide CpG methylation, with enriched mapping of CpG island, gene
promoters and bodies, and enhancer regions
• Uses Infinium assay with bisulphite-converted DNA
• Methylation is read as either:
Beta-value (M/[M+U]) or M-value (logit2-Beta val)
bj =
Im, j
(Im, j + Iu, j )
; M -value = logit2 (bj ) = log2
bj
1- bj
æ
è
çç
ö
ø
÷÷
6. Chronological and Biological Age
• Chronological age: age of patient
• Biological age: age of tissue
• Correlated with chronological age
• Potentially stratifies patients on risk than chronological age
• Ways of calculating tissue age
• Elastic net, lasso penalized regression modeling
• Correlating on Chron. Age and comparing to normal tissue
7. Methylation Drift in Aging
• Methylomic drift: age-dependent, progressive methylation change distinct
from normal tissue
• Most drift CpGs start as hypomethylated, and become methylated with age in
premalignant and diseased tissue
• Biological basis in DNMT chemistry (accumulation and propagation of errors, de novo
methylation)
• Consistent with a non-uniform maintenance (eg. hotspots or mosaicism in tissues and
cells) and stochastic process
• Potentially linked to mitotic age
• Utility: determine BE dwell time based on population-based Bayesian analyses
8. Drift in Nondysplastic BE
• Curtius et al 2016 described 67 drifting
CpGs in BE
• Methylation data from longitudinal BE,
matched BE-SQ, and sporadic BE used to
inform a Bayesian model for estimation of
BE dwell time
• Next: explore methylation patterns at
these 67 CpGs and other correlated
CpGs to better understand drift
dynamics and how to differentiate
progressed and diseased tissues
Curtius et al. “A Molecular Clock Infers Heterogeneous Tissue Age Among Patients with Barrett’s Esophagus”
Image courtesy of Georg Leubeck
9. Longitudinal BE Sample Methylation At Drift CpGs
Most CpGs
increase in
methylation with
age, some
dependence of
increase on
starting point.
Heterogeneity in
drift rate may be
an independent
risk factor for
progression to
EAC
BETRNet
longitudinal
samples
(youngest and
oldest available),
methylation
shown at each
drift CpG
individually
10. Methylation correlates with Age at Drift CpGs
• Cross-sectional BE samples: population-wise positive corr. with chron. age
Methylation (Beta-value) Age (years)
12. Correlating Drift Across the Methylome
• CpG islands are concentrated regions of CG dinucleotides
• 40% of islands normally hypomethylated
• ~60% of gene promoters overlap CpG Islands
• 67 drift CpGs map to 51 unique islands, with 50 in island regions
Images courtesy of Georg Leubeck
Island-assoc.
N_Shore
OpenSea
S_Shore
13. Functional Units of Drift: Promoter Island Drift
Images courtesy of Georg Leubeck
Drift is highly correlated among
CpGs at 1,317 Islands that drift
(contain at least 5 drift CpGs) in
NDBE
14. Transition of drift patterns from BE to EAC
Image courtesy of Georg Leubeck
• Subtypes in BE and EAC visible in density plots of drift CpG
methylation.
• Going from low to high average methylation, there are
distinct unimodal-to-bimodal density pattern transitions in
BE and EAC
• Could be leveraged as a discrete criterion for drift
subtyping
15. Transition of drift patterns from BE to EAC
Image courtesy of Georg Leubeck
• Methylation density pattern
evolution suggests a
threshold effect:
1. Gradual drift changes
accumulate, with
maintenance of low-
level mode
2. Changes become
selective and drive
cancer formation
unimodal low (L), bimodal intermediate (I), and bimodal high (H).
16. Functional Consequences of Drift
• Evidence of functional ramifications of drift at promoter drift islands with increased methylation
• Agreement/validation with two datasets: TCGA (N=87 EACs,, methylation arrays and RNAseq)
and Krause et al (N=51 EACs, Queensland Australia, methylation arrays and Agilent expr. arrays).
• Hypermethylation at Island/promoter drift CpGs results in silencing of disease-assoc. genes
• 20/35 sig. repressed genes in both datasets,
• CHFR (checkpoint with forkhead associated and ring finger domains) repressed in both
datasets, and enriched in EAC comparison to GEA in TCGA Nature paper.
• In network analysis of repressed genes, 3-fold enrichment of DNA binding transcription factor
activity term, and the Krueppel-assoc. box domain zinc finger (KRAB, ZNF) TFs.
17. Conclusions
• Spatial distribution of CpGs contributes to an understanding of their functional
importance. CpG methylation can be correlated at functional units such as islands
• Progressive age-dependent methylation changes in BE and EAC show a threshold
effect and affect
• In EAC, functional consequences of promoter drift island hypermethylation on
gene expression
• Functional consequences of drift in BE necessary to better understand
mechanisms of progression. We have a hypothesis in drift patterns and dwell time.
Now need more expression data
18. Thanks! This work was supported by:
• Dr.’s Bill Grady, Ming Yu, members of Grady Lab, and the Genomics
Core at Fred Hutch
• Collaborators at Case Western and UW, including Dr.’s Joe Willis,
Andrew Kaz, and Amitabh Chak
• Dr.’s Georg Leubeck, Bill Hazelton, and Kit Curtius
• BETRNet and MEMO consortium facilitators, doctors, and patients
• Organizers and facilitators at CISNET 2017
GRADY LAB
Original analyses run in R with modules
from Bioconductor and CRAN
19. Supplemental Slide 1
• Methylation-based subtypes identified
in EAC identified using most variant
CpG probes in model-based clustering.
• EAC subtype signature apparent in BE,
suggesting they arise early in the
progression sequence
• No apparent clinical basis for subtypes
(ie. correlation with age, gender,
smoking, survival).
• Molecular features distinguish the EAC
subtypes: ERBB2 expr./CN gain; mRNA
and microRNA expr.; response of cell
cultures to drug treatments
BETRNet Cohort
20. Supplemental Slide 2
HM = high methylator; IM = intermediate methylator; LM = low methylator; MM = minimal methylator
22. Supplemental Slide 4: Towards developing screening
panels - Validation
Color = Val1
Sens.
Krause et al. 2016TCGA EACs
Editor's Notes
Thanks to Dr. Georg Leubeck inviting me to present at CISNET. My name is Sean, I am a Data Analyst Asst at Dr. Bill Grady’s lab in Clinical Research at Fred Hutch. We focus on epigenetics of esophageal and colorectal cancers, and a primary part of my work is handling qc pipelines for arrays run at the Hutch, managing data storage and sharing, and encouraging collaborations by sharing code and collaborating on experimental design with other labs.
Note the interconnectedness of certain features. microRNA loci can be regulated by methylation, CpG methylation near histones can affect enhancer binding, while methylation or lack thereof at histones themselves can mark a region as active or inactive (H3K27ac, H3K4me/me3) etc. We are perpetually interested in collaborations!
Image is regulation of key cancer stem cell signaling pathways by epigenetic mechanisms. Methylation-induced decreased expr of DKK1 enhances wnt/beta-catenin signaling; Hedgehog signaling pathway can be activated by Shh promoter hypomethylation coupled with increased HDAC1 expression. Degregulation of these pathways enables CSCs to acquire self-renewal abilities ands drug resistance properties.
TCGA paper largely compared EACs to gastric cancers, finding that gastroesophageal adenocarcinoma with chromosomal instability were similar according to microRNA, mRNA, protein, and chromosome platform data, but that clustering on methylation differentiated the two into clusters. Hypermethylated clusters were enriched for EACs, and samples in the highest cluster showed silencing of MGMT and Checkpoint with forkhead associated and ring finger domain (CHFR) genes.
Primary assay is Infinium 2, which involves single-bp extension at the end of a 50bp probe. A fluorescent marker on the added base is read in a single channel, resulting in a red (unmethylated) or green (methylated) signal. Another Inf 1 assay covers a small subset of about 5% of CpGs, and is a variation on this tech with a single color channel instead of two. 420k CpGs detected after quality control measures.
Curtius et al 2016 describe and validated 67 drift CpGs undergoing age-related drift. These were then used with longitudinal BE samples, paired BE-NSQ samples, and sporadic samples to derive a Bayesian model for calculation of BE dwell time.
This is a selection of 9 BETRNet BE patients and their longitudinally collected samples. Note a general tendency for increased methylation over time. It is clear there is heterogeneity in drift rates between patients, and this my contribute to risk of EAC independent of BE dwell time. Ideally we would want to compare matched SQ tissues with each BE sample.
69 nondysplastic BE samples are taken from among several datasets, including BETRNet and MEMO cohorts through University of Washington. Shown are methylation density and mean methylation at 67 drifting CpGs in BE. Note evidence of a unimodal, low methylation distribution transitioning to a bimodal, high methylation distribution as chron. age increases. These trends are noted later in the context of drift at functional units.
Drift within a patient at cross-sectional samples from throughout their esophagus can vary considerably. That is to say, there is variance in the deviance from normal tissue methylation for samples from that same patient at the same chronological age. This could be important for advancing theories about the significance of BE and how to screen for BE. For instance, if BE advances up the esophagus over time, this could be consistent with patient 1, where old BE is detected at the top, and younger BE is detected at the bottom, and could suggest greater risk of progression if lower BE is younger due to continued damage response. Old BE near the GEJ, by contrast, may suggest BE formed but does not elicit the same inflammatory response. These insights could change whether and what intervention is deemed necessary.
Image shows pairwise correlation of island CpGs and others in island, shelf and shore as a function of distance from island proper. Results shown for islands containing CpGs that are correlated with the former 67 CpGs. 64 NDBE patients samples were used.
Image is density of methylation at 11,425 island-associated drift CpGs (min, 5 drift-CpGs per island). Three drift groups reflect discrete density patterns: unimodal low (L), bimodal intermediate (I), and bimodal high (H). For EACs, TCGA and Krause et al cohorts were used, and 200 repressed genes found from expression array and sequencing data, whose promoters overlapped one of 1240 of drift islands. Checkpoint with forkhead associated and ring finger domain (CHFR) gene a mitotic stress checkpoint gene significantly repressed in both Krause and TCGA cohorts.
Image is density of methylation at 11,425 island-associated drift CpGs (min, 5 drift-CpGs per island). Three drift groups reflect discrete density patterns: unimodal low (L), bimodal intermediate (I), and bimodal high (H). For EACs, TCGA and Krause et al cohorts were used, and 200 repressed genes found from expression array and sequencing data, whose promoters overlapped one of 1240 of drift islands. Checkpoint with forkhead associated and ring finger domain (CHFR) gene a mitotic stress checkpoint gene significantly repressed in both Krause and TCGA cohorts.
Image is density of methylation at 11,425 island-associated drift CpGs (min, 5 drift-CpGs per island). Three drift groups reflect discrete density patterns: unimodal low (L), bimodal intermediate (I), and bimodal high (H). For EACs, TCGA and Krause et al cohorts were used, and 200 repressed genes found from expression array and sequencing data, whose promoters overlapped one of 1240 of drift islands. Checkpoint with forkhead associated and ring finger domain (CHFR) gene a mitotic stress checkpoint gene significantly repressed in both Krause and TCGA cohorts.
A panel search could recapitulate the observed threshold effect in drift. Here is shown a separate but similar workflow used for a related project in EAC. Samples are from TCGA. RPMM groups correspond to hyper and hypo methylated subtypes, roughly related to the EAC-H and EAC-I/L groups from prior slide. Conceivably this approach could be adopted for drift leveraging the threshold effect, where the cutoff is informed by the likelihood a CpG will advance or retard from one mode to the other.
In this example, DMPs are differentially methylated probes from normal vs. tumor comparison in BETRNet EACs. Validation 1 is N=87 TCGA EACs and 2 is N=125 Krause et al EACs for which clusters HM and NHM are defined. In this context, panels have a single set of coordinates for spec/sens due to discrete criteria. Panels are designed from among DMPs with a predefined set of parameters (ie. number of CpGs that must be above a given threshold to classify a sample as HM). Panels can then be compared along four axes (Val1 Sens and Spec, and Val2 Sens and Spec), or two coordinates from each validation, to determine the best overall performing panel.