1. Chromatin state dynamics in nine human cell types elucidate regulators and disease-associated SNPs
Jason Ernst
Joint work with Pouya Kheradpour, Luke Ward
Brad Bernstein and Manolis Kellis

2. Challenge: interpreting disease-associated variants
CATGACTG
CATGCCTG
Epigenomics
Disease variants
GWAS studies implicate thousands of non-coding loci associated with disease
Challenges towards interpreting disease variants:
Find ‘true’ causative SNP among many candidates in LD
Determining type of function: especially outside protein-coding
Reveal relevant cell type of activity
Link to upstream regulators and downstream targets
This talk: chromatin tools to address these challenges

3. Challenge of data integration in many marks/cells
Construct antibodies pull down chromatin  ChIP-seq tracks
Epigenomic information retains genome ‘state’ in differentiation and development
Histone tail modifications
Two types:
DNA methyl.
Histone marks
Dozens of chromatin tracks
Understand their function
Reveal their combinations
Annotate systematically
Common chromatin states
Explicitly model combinations
Unsupervised approach, probabilistic model
DNA packaged into chromatin around histone proteins

4. From ‘chromatin marks’ to ‘chromatin states’
Learn de novo significant combinations of chromatin marks
Reveal functional elements, even without looking at sequence
Use for genome annotation
Use for studying regulation dynamics in different cell types
Promoter states
Transcribed states
Active Intergenic
Repressed

5. ENCODE: Study nine marks in nine human cell lines
9 human cell types
81 Chromatin Tracks (2^81 combinations)
H3K4me1
H3K4me2
H3K4me3
H3K27ac
H3K9ac
H3K27me3
H4K20me1
H3K36me3
CTCF
+WCE
+RNA
HUVEC
Umbilical vein endothelial
NHEK
Keratinocytes
GM12878
Lymphoblastoid
K562
Myelogenous leukemia
HepG2
Liver carcinoma
NHLF
Normal human lung fibroblast
HMEC
Mammary epithelial cell
HSMM
Skeletal muscle myoblasts H1
Embryonic x
15 chromatin states
(for each cell type)

6. Chromatin states dynamics across nine cell types
Key points to make:
Chromatin states enabled us to study the dynamic nature of chromatin across many cell types.
By distinguishing 15 different types of chromatin states, we could summarize all significant combinations of 81 different chromatin tracks and 2.4 billion reads in just nine chromatin annotation tracks, one for each cell type.
For example, the same gene (WLS), is ‘poised’ in embryonic stem cells (ES), repressed in three other cell types (K562, blood, and liver), and active in the other five cell types.
This allows us to now define ‘vectors’ of activity for each region of the genome, based on the chromatin annotation in the nine cell types.
Single annotation track for each cell type
Summarize cell-type activity at a glance
Can study 9-cell activity pattern across

7. Introducing multi-cell activity profiles
Gene
expression
Chromatin
States
Active TF motif
enrichment
TF regulator
expression
Dip-aligned
motif biases
HUVEC
NHEK
GM12878
K562
HepG2
NHLF
HMEC
HSMM H1
TF On
TF Off
Motif aligned
Flat profile ON
OFF
Active enhancer
Repressed
Motif enrichment
Motif depletion

8. Linking Distal Regulatory Elements to Genes
Which gene(s) is this active enhancer in HMEC likely regulating? ? ?
HMEC state
IRF6 expression
C1orf107 expression
-1.6
4.2
3.7
0.9
-1.7
-0.7
0.1
0.4
0.3
-0.1
0.5
-1.3
0.0
1.2
-1.1
H3K27ac signal
Compute correlations between gene expression levels and enhancer associated histone modification signals

9. Linking Distal Regulatory Elements to Genes
Which gene(s) is this active enhancer in HMEC likely regulating?
0.6
Random gene expression
-1.1
-1.0
-0.5
0.5
-0.8
4.0
HMEC state
IRF6 expression
-0.7
H3K27ac signal
-1.7
-1.6
Combine intensity signal from all marks:
Train logistic regression classifier to discriminate real from random correlations, conditioned on state, TSS dist, cell type
-1.7
0.9
-1.6
-1.6
4.2
3.7
Random
Real
Compare correlations between enhancer and gene expression between real and randomized data

10. Enhancer-gene links supported by eQTL-gene links
eQTL study
Validation rationale:
Expression Quantitative Trait Loci (eQTLs) provide independent SNP-to-gene links
Do they agree with activity-based links?
15kb
Individuals
Indiv. 1
-0.5 A
Indiv. 2
-1.5 A
Indiv. 3
-1.8 A
Example: Lymphoblastoid (GM) cells study
Expression/genotype across 60 individuals (Montgomery et al, Nature 2010)
120 eQTLs are eligible for enhancer-gene linking based on our datasets
51 actually linked (43%) using predictions  4-fold enrichment (10% exp. by chance)
Indiv. 4
3.1 C
Indiv. 5
1.1 A
Indiv. 6
-1.8 A
Indiv. 7
-1.4 A C
Indiv. 8
3.2 C
Indiv. 9
4.4 … … …
Independent validation of links.
Relevance to disease datasets.
Expression level of gene
Sequence variant at distal position

11. Coordinated activity reveals activators/repressors
Enhancer activity
Gene activity
Predicted
regulators
Activity signatures for each TF
Key points to make:
Using these correlations in activity enabled us to start piecing together enhancer regulatory networks, which have been previously inaccessible, linking regulators to enhancers and enhancers to target genes.
Putting it all together, we can (a) define 20 distinct profiles of activity (labeled A through T) across the nine cell types, (b) observe the expression patterns of associated genes, showing upward of 0.9 correlation with enhancer activity, (c) discover enriched regulatory motifs revealing candidate regulators, (d) distinguish activators and repressors based on positive or negative correlations between motif enrichment in active regions and expression of the corresponding regulator.
[click-animate] For example, cluster Oct 4 is a predicted activator of enhancers active in embryonic stem (ES) cells. The motif is enriched in ES-specific enhancers (cluster A), and the Oct 4 TF is expressed specifically in the same cell type
[click-animate] similarly, Ets is a predicted activator of cluster G, associated with GM and HUVEC activity but not either one alone. This is important for the next slide, as we predict that a disruption in the Ets1 motif in patients of lupus erythromatosus is responsible for disruption of the corresponding enhancer and disregulation of immunity gene HLA-DRB1 (Human Leukocyte Antigen) in the major histocompatibility locus.
Enhancer networks: Regulator  enhancer  target gene
Ex1: Oct4 predicted activator of embryonic stem (ES) cells
Ex2: Gfi1 repressor of K562/GM cells

12. Causal motifs supported by dips & enhancer assays
Dip evidence of TF binding (nucleosome displacement)
Enhancer activity halved by single-motif disruption
 Motifs bound by TF, contribute to enhancers

13. Revisiting disease- associated variantsxx
Revisiting disease- associated variants
Disease-associated SNPs enriched for enhancers in relevant cell types
E.g. lupus SNP in GM enhancer disrupts Ets1 predicted activator

14. SNPs from GWAS Enrich for Cell Type Specific Strong Enhancer Chromatin States in Biologically Relevant Cell Types
Cell Type
Title
Author/
Journal
# SNPs in Strong enhancers
Total #SNPs
Fold
FDR
K562
Multiple loci influence erythrocyte phenotypes in the CHARGE Consortium.
Ganesh et al
Nat Genet 2009 9 35 17
0.02
HepG2
Biological, clinical and population relevance of 95 loci for blood lipids
Teslovich et al
Nature 2010 13
101 11
GM12878
Genome-wide association study meta-analysis identifies seven new rheumatoid arthritis risk loci
Stahl et al
Nat Genet 2010 7 29 15
0.03
Genome-wide meta-analyses identify three loci associated with primary biliary cirrhosis
Liu et al 4 6 41
Genome-wide association study in a Chinese Han population identifies nine new susceptibility loci for systemic lupus erythematosus.
Han et al 18 21
Six new loci associated with blood low-density lipoprotein cholesterol, high-density lipoprotein cholesterol or triglycerides in humans.
Kathiresan et al
Nat Genet 2008 5 24
Genome-wide association study of hematological and biochemical traits in a Japanese population
Kamatani et al 39 12
A genome-wide meta-analysis identifies 22 loci associated with eight hematological parameters in the HaemGen consortium.
Soranzo et al 28
Meta-analysis of genome-wide association data identifies four new susceptibility loci for colorectal cancer.
Houlston et al 3 66
Genome-wide association study identifies eight loci associated with blood pressure.
Newton-Chen et al 30
0.04
Ernst et al, Nature 2011

15. Ex1: Systemic lupus erythrematosus SNP: Ets-1 motif
SNP in lymphoblastoid GM enhancer state
Disrupts Ets1 motif instance, predicted GM regulator
 Model: Disease SNP abolishes GM-specific enhancer

16. Ets-1 is a predicted activator of GM enhancers
Enhancer activity
Gene activity
Predicted
regulators
Activity signatures for each TF
Ets expression  Ets-1 motif enrichment in enhancers
 Model: Ets-1 disruption would abolish enhancer state

17. Chromatin state dynamics: Contributions summary
Chromatin states capture mark combinations
Reveal promoter/enhancer/insulator/transcribed regions
Chromatin states capture chromatin dynamics
Single annotation track for each cell type
Nine tracks instead of 2^81 combinations
Activity profiles capture correlated changes
Gene expression vs. chromatin: EnhancerGene links
Motifs vs. TF expr vs. chromiatin: Activators/Repressors
Regulatory predictions validated: eQTLs/dips/lucif.
eQTLs: links. Dips: binding. Luciferase assays: motif role
Interpret disease-associated variants
Intergenic SNPs enriched for cell-type specific enhancers
Mechanistic predictions reveal potential drug targets

18. Collaborators and Acknowledgements
MIT compbio group:
Pouya Kheradpour
Lucas Ward
Manolis Kellis
ENCODE consortium
Funding
NHGRI, NIH, NSF, HHMI, Sloan Foundation
MGH Pathology/HHMI:
Tarjei Mikkelsen
Noam Shoresh
Charles B. Epstein
Xiaolan Zhang
Li Wang
Robyn Issner
Michael Coyne
Manching Ku
Timothy Durham
Bradley E. Bernstein