1. Epigenetics System Biology Workshop: Introduction
Irina Shchukina
11/27/2018
Title

2. Outline A very short intro into gene expression regulation
Some NGS technologies available to study regulation
Overview of a standard pipeline for ChIP-seq data processing
Next: practical session with ChIP-seq data

3. Chromatin organization
Heterochromatin – packed, unreachable DNA (6-8) Euchromatin – generally more active regions (1–5)

4. Nucleosomes
Nucleosomes are made of 8 histone proteins (2x H2A, H2B, H3, and H4). +H1 – linker histone
~1.65 loops of DNA include 147 nucleotides

5. Nucleosomes
Nucleosomes are made of 8 histone proteins (2x H2A, H2B, H3, and H4). +H1 – linker histone
~1.65 loops of DNA include 147 nucleotides
histones tails stick outside and can be recognized
chemical modifications of histones influence DNA accessibility
histone modifications are dynamic: they can be added, erased, and recognized

6. Nucleosomes
nucleosomes are made of 8 histone proteins (2x H2A, H2B, H3, and H4). +H1 – linker histone
~1.65 loops of DNA include 147 nucleotides
histones tails stick outside and can be recognized
chemical modifications of histones influence DNA accessibility
histone modifications can be read, erased, and recognized
Examples: H3K4me1, H3K4me3, H3K27ac

7. Transcription factors
Proteins that binds to DNA and controls transcription of DNA into RNA

8. Transcription factors
Proteins that binds to DNA and controls transcription of DNA into RNA
A lot of interactions with other genome regions and corresponding histone modifications are happening! Enhancer may be located very far from the gene it regulates.

9. Role of histone modifications
H3K27ac – distinguishes active enhancers from poised
H3K27me3 – repression of transcription; has only one methyltransferase EZH2, which is a part of PRC2
H3K4me1 – enhancer mark
H3K4me3 – promoters, active transcription
H3K36me3 – gene body, active transcription

10. Chromatin immunoprecipitation: ChIP-seq

11. Chromatin immunoprecipitation: ChIP-seq
Need a good antibody for clean data
Input sample – no specific IP, control for background noise level, required for normalization and peak calling
Can be used for both transcription factors and histone modifications
Different proteins → different types of data → different processing:
TFs bind narrow region of DNA (5-30bp)
Some histone modifications are very broad: H3K36me3 may cover entire body of actively transcribed gene
(Will see some examples in the next part)

12. Ultra-low-input (ULI) ChIP-seq
Standard ChIP-seq protocol requires 1-5 million cells (or even more): serious limitation for multiple studies (human samples, rare populations, etc.)
Ultra-low-input ChIP-seq allows you to go as low as 100,000 cells per sample.
The price is data quality and consistency:
higher noise
variable signal to noise ratio between samples within one prep
hard to process (as you will see in practice session)
Major difference in protocol: no crosslinking step.
Often Mnase digestion is used instead of sonication to decrease level of noise and keep DNA-protein complexes intact without crosslinking.

13. Publicly available data
ENCODE:
Raw and processed data available.
People actually care about this stuff and generated tons of data
Blueprint:

14. Standard ChIP-seq pipeline
Raw data
Alignment
Peak calling
Interpretation
Heavy computational work
Machine-time- and memory-consuming part
GTAC run their standard pipeline for you
Following slides have many keywords, details, names of tools and links, etc – they are for googling

15. Raw sequencing data: FASTQ files
4 lines per read
Line ID
Line 2: actual sequence
Line 3: + id or any description
Line 4: encoded quality of each nucleotide
Raw data
Alignment
Peak calling
Interpretation

16. Alignment
Finds location in genome for each read Tools to use: bowtie or BWA (shouldn’t make much difference)
Raw data
Alignment
Peak calling
Interpretation

17. Alignment Finds location in genome for each read
Tools to use: bowtie or BWA (shouldn’t make much difference)
NB: use consistent reference genome versions! hg18 → hg19 (=GRCh37) → hg38 (=GRCh38) mm8 → mm9 → mm10
If you want compare positional data:
Realign one of your dataset
Liftover tool from UCSC ( bin/hgLiftOver)
Raw data
Alignment
Peak calling
Interpretation

18. Alignment output: BAM/SAM files
Detailed format description: Set of utilities: SAMtools
Raw data
Alignment
Peak calling
Interpretation

19. Visualization
Genome browsers (JBR, IGV, UCSC) work with bigwig (=bw) format BAM to bigwig conversion: deepTools suite, bamCoverage tool
Visualization
Raw data
Alignment
Peak calling
Interpretation

20. Visualization
Genome browsers (JBR, IGV, UCSC) work with bigwig (=bw) format
BAM to bigwig conversion: deepTools suite, bamCoverage tool
You can normalize data before visualization:
Sequencing depth (bamCoverage 1x normalization)
Input (bamCompare)
Visualization
Raw data
Alignment
Peak calling
Interpretation

21. Visualization Different ChIP-seq data actually looks different!
Raw data
Alignment
Peak calling
Interpretation
ChIP–seq: advantages and challenges of a maturing technology. Peter Park. Nat. reviews

22. Peak calling
The most sensitive step. Choosing appropriate tool is very important!
Golden standard tools:
MACS (narrow peaks: TFs and some histone modifications)
SICER (broad histone modifications)
Data quality significantly affects results. ULI-ChIP-seq data requires special treatment: SPAN.
More on your practical session.
Raw data
Alignment
Peak calling
Interpretation

23. Peak calling output: BED
BED format describes genomic intervals (not specific for ChIP-seq). 3 required fields: chromosome, start, end. Other depend on the tool used Detailed description: Toolkit: bedtools Can be visualized in genome browsers together with bigwig files.
Raw data
Alignment
Peak calling
Interpretation

24. Quality control: ENCODE
ENCODE established a set of formal data quality standards and metrics:
Raw data
QC!
Alignment
QC!
Peak calling
QC!
Interpretation

25. Quality control Other things to look at:
Raw data: FASTQC – report per one FASTQ file, MultiQC – summarizes multiple outputs into single report
Aligner usually produces a report. Look for % of aligned reads, % of multimappers, etc. Can be processed with MultiQC.
Peak calling: no ultimate QC metric, need to look for multiple scores.
General advise:
Visualize your data! Check genes that you know are bound/active/repressed/…
Comparing to existing dataset may be a good sanity check
Raw data
QC!
Alignment
QC!
Peak calling
QC!
Interpretation

26. Interpretation: binding motifs
Particularly useful for transcription factors – can be a part of QC routine for TFs with known binding motif.
Major tool: the MEME suite ( MEME-ChIP – specialized version for ChIP-seq.
Not too many sequences (e.g. 1,000 – 2,000), not too wide (~ bp) – otherwise may take forever to calculate
Input is .fa file with actual nucleotide sequences. May be generated using bedtools: bedtools getfasta -fi GRCm38.genome.fa –fo peaks.seq.fa -bed peaks.bed
Submit your jobs and be patient
Raw data
Alignment
Peak calling
Interpretation

27. Interpretation: annotation
Match peaks and genes. Possible approached:
Most common way: find gene that is the closest to the peak. Why you may be losing your favorite gene from resulting list?
Assign all genes located within selected range from a peak
Mostly for TF: consider only peaks located near TSS (e.g. [-10kb, +3kb] around TSS)
Raw data
Alignment
Peak calling
Interpretation

28. Interpretation: pathway enrichment
GREAT:
Choose your genome version
Upload your BED file
May adjust association rules
Submit
Result includes:
Enrichment analysis against multiple pathway databases
Matched peaks and genes
Some positional analysis (e.g. distribution of distances from TSS)

29. Interpretation: differential analysis
Not established analysis. No golden tools. Multiple biological replicates are strongly recommended!
Good review: A comprehensive comparison of tools for differential ChIP-seq analysis. Steinhauser, Kurzawa, Eils and Herrmann

30. Next: practical session
Thank you!
Next: practical session
Raw data
Alignment
Peak calling
Interpretation