Analysis of shotgun bisulfite sequencing of cancer samples

Transcription

1 Analysis of shotgun bisulfite sequencing of cancer samples Kasper Daniel Hansen Postdoc with Rafael Irizarry Johns Hopkins Bloomberg School of Public Health Brixen, July 1st, 2011

2 The basis of phenotypic variation: species

3 The basis of phenotypic variation: tissues

4 Epigenetics Heritable changes in phenotype that are not caused by changes in DNA.

5 DNA In humans: methylation occurs at CpG dinucleotides (28.2M) GC CG me GC CG me me GC CG me me GC CG me GC GC GC CG CG CG CpGs are depleted genomewide. CpGs tend to cluster together (clusters are termed CpG Islands), these clusters are enriched in or near promoters. is associated with openness of the DNA. Hypermethylation (high) is associated with gene silencing Hypomethylation (low) is associated with active genes is inherited (at least in cell division).

6 Measuring DNA methylation PCR does not preserve methylation information. Hybridization is not affected by methylation. Pretreatment Enzyme digestion Affinity enrichment Sodium bisulphite Analytical step Locus-specific analysis Gel-based analysis Array-based analysis NGS-based analysis HpaII-PCR MeDIP-PCR MethyLight EpiTYPER Pyrosequencing Southern blot RLGS MS-AP-PCR AIMS Sanger BS MSP MS-SNuPE COBRA DMH MCAM HELP MethylScope CHARM MMASS MeDIP mdip mcip MIRA BiMP GoldenGate Infinium Methyl seq MCA seq HELP seq MSCC MeDIP seq MIRA seq RRBS BC seq BSPP WGSBS AIMS, amplification of inter-methylated sites; BC seq, bisulphite conversion followed by capture and sequencing; BiMP, bisulphite Illumina methylation arrays: GoldenGate (early 2007, 1.5k CpGs), 27k (late 2007), 450k (2011) Laird, Nature Reviews Genetics (2010)

7 Bisulfite treatment The gold standard for measuring DNA methylation at single CpGs is bisulfite treatment followed by Sanger or Pyro sequencing Bisulfite treatment converts unmethylated Cs to Us (= T) CG CG CG CG CG CG CG me me me Bisulfite treatment TG CG CG CG TG TG TG Can be used genome-wide, but requires the same sequencing effort as whole genome DNA sequencing (= expensive).

8 Cancer and DNA methylation DNA methylation in cancer was the first epigenetic modification discovered in cancer (~25 years ago). Focus (at least lately) in the literature have been on hyper methylation of CpG islands in promoters (tumor supprs) hypo methylation of select repeat elements although global hypomethylation hypo methylation of selected genes (typically oncogenes) have also been described. terminology Hyper: goes up, Hypo: goes down DMR: differentially methylated region

9 CpG Islands shores M a Brain (normal) Liver (normal) Spleen (normal) b M Brain (normal Liver (normal) Spleen (normal) Colon (normal) Colon (tumor) CpG density Genes Shore PRTFDC1 Island CpG density Genes 0.10 Shore Island HS3ST Chr. 10 Location Chr. 16 Location Many changes are not in CpG islands, but in regions bordering CpG islands; termed CpG Island shores. Irizarry et al. Nature Genetics (2009)

10 Acknowledgements Increased methylation variation in epigenetic domains across cancer types Kasper Daniel Hansen 1,2,10, Winston Timp 2 4,10, Héctor Corrada Bravo 2,5,10, Sarven Sabunciyan 2,6,10, Benjamin Langmead 1,2,10, Oliver G McDonald 2,7, Bo Wen 2,3, Hao Wu 8, Yun Liu 2,3, Dinh Diep 9, Eirikur Briem 2,3, Kun Zhang 9, Rafael A Irizarry 1,2 & Andrew P Feinberg 2,3 Nature Genetics, Advance Online

11 Increased methylation variation across all cancers 0.20 (% %$ (( %) (, %+ %# ) %' * %" ($ %, (# $% % %* $ " %% ( () (+ (* -%+ (" -(' -(# -%) -% -%' -" -%( -(, -+ -) -(( -* -%# -(* -(" -' -* $, -( -(+ -$ -" -$, -$% -# %( -( -($ -%$ -%% -%" -%* -% -(% -$ -%, -() -+ -# -%, -+ -%, -# -* -% -) -% -) -) -" -( -" -+ -* -$ 0.10 Lung Cancer 0.30 Increased variation between normals and cancers, for the same regions across all 5 cancer types (lung, colon, breast, thyroid and Wilms) regions in 290 samples <43=+ -(( -%% -$ -* -( -() -) -%$ -%, -%+ -% -%# -" -# -(, -%* -' -($ -(* -%" -(% -%) -%' -(" -%( (" %' " % # ( %$ ' * (% %* (* %+ ) %# %( %% $ (( %" (, %, () ($ %) "& "% "$ "# The same regions that distingush cancers from normals, distinguishes normal tissue types. " ()*+,-)./()0123*0456 '" Lung Normal %

12 Design 3 colon cancers and their matched normal mucosa 2 adenomas ABI SOLiD, 50bp reads ~5x coverage on CpGs We traded coverage for biological replicates.

13 Mapping Bisulfite conversion makes the genome into an (appr) 3 letter alphabet, making mapping hard. We could not use existing tricks for unbiased alignment of bisulfite sequencing data: we wrote a custom aligner, Merman. We can map ~20M CpGs uniquely. U U U M U M U M CG Genome Coverage (for this CpG): 8 3 M s and 5 U s (Unmethylated)

14 Global levels of methylation Pair 1 Normal Cancer Pair 2 Normal Cancer Pair 3 Normal Cancer Adenoma 1 Adenoma 2 Bisulfite conversion rates estimated using λ phage spike-in to be %

15 One sample, small region ~ 14kb Smoothing using a binomial model (local likelihood) Adaptive bandwidth ( important)

16 Small region kb normals cancers Islands Genes Shore (hypo) Island (hyper) Shore (hypo) Loss of methylation boundaries in cancer

17 Boundary Shifts (inwards and outwards) kb normals cancers Islands Genes kb Islands Genes

18 Novel hypomethylation kb Islands Genes

19 Capture bisulfite "#$%&'())"*"+," -./ "#$012#(3+ "#$012#( ~40,000 capture regions, ~400,000 CpGs Red: Average of cancers Blue: Average of normals Green: Difference between cancers and normals

20 Capture bisulfite "#$%&'())"*"+," -./ "#$012#(3+ "#$012#( ~40,000 capture regions, ~400,000 CpGs Red: Average of cancers Blue: Average of normals Green: Difference between cancers and normals

21 More capture "#$%&'())"*"+," -./ "#$012#(3+ "#$012#( "#$%&'())"*"+," -./ "#$012#(3+ "#$012#(

22 normals cancers 0.2 Large blocks of hypomethylation 100kb PMD LOCKS LAD Islands Genes Consistent boundaries 100kb PMD LOCKS LAD Islands Genes (Some) coincides with LADs, LOCKs and PMDs. Related to structural conformation of the DNA in the nucleus

23 What predicts hypomethylation in blocks? PMD LOCKS LAD Islands Genes 100kb normals cancers density 2 6 normals cancers density normals, in blocks cancers, in blocks normals, outside blocks cancers, outside blocks change methylation methylation blocks not blocks repeats blocks repeats not blocks not repeats not repeats

24 Blocks are enriched for hyper-variables genes Standardized log expression N T COL11A1 CHI3L1 MMP10 SLCO1B3 KRT6A GPR109B KRT23 CST1 CADPS WISP3 CCL26 STC1 IL13RA2 CHI3L1 MMP7 MMP3 KRT6B PAH ZG16 IL1A TMPRSS3CXCL11 INHBA STC1 CLDN2 MAGEA6 Some of these genes are associated with tumor progression

25 Quality control: M-bias WGBS Fraction of filtered evidence indicating presence of methylation Normal 1 Normal 2 Normal 3 Cancer 1 Cancer 2 Cancer 3 Adenoma 1 Adenoma Offset from beginning of read

26 Quality control: M-bias Capture 907:.516/1+/+5;.-0-8/-<58-6:-/5685:7.564/=0-,-6:-/1+/2-.>?;7.5 &' &# &( @1027;/' B76:-0/A B76:-0/" B76:-0/' " # $ % Based on this, we trim 15bp. This improves the concordance *++,-./+012/ /1+/0-78

27 Conclusions Large blocks of hypomethylation in cancer Global hypomethylation, expression variability LOCKs/LADs Structural changes (boundaries) in small regions Unified framework for shore/islands hypo/hyper methylation With our smoothing technique, 4-5x is good enough Verified by high coverage padlock bs capture biological replicates are very useful Quality assessment (M-bias plots)

28 Advantages of biological replicates p= kb normal cancer p= kb

29 The effect of copy number variation (CNV) Pair 1 difference Pair 2 difference difference Pair 2 log 2 (CNR) Pair 1 log 2 (CNR) log 2 (CNR) Chr 20