Ns126:Calendar/NOTES/2015-6-26
Methylation Block in HM450 microarray
Motivation
Dr.Zhang has defined methylation haplotype and methylation blocks based on GWBS data. and we are sure that these regions must have special biological, genomic or evolution functions. Therefore, I want to do the similar analysis in HM450K array. I will compare the regions identified in HM450 with the regions disovered by Dr. Zhang. In our hypothesis, the regions in my analysis would be a sutset of the regions from Dr. Zhang.
Materials and Method
I collected pair-wised samples of 11 cancers from TCGA project.
Method
HDR: High CpG density region. A intuitive method adopted from Takai and Jones's sliding-window algorithm were applied to identify the HDR region within methylation 450K microarray as the following steps.
Set a 100-base window in the beginning of genomic position of the array and shift the window to the right when there are least 2 probes in the windows. Calculate the total probes in extent regions until the last window does not meet the criteria. All the regions in which at least 4 probes were collected and be defined as high CpG density regions and calculate the average Pearson correlation among all the probes in cancer and normal samples, respectively. Please check the code.
Materials and Dataset
TCGA Dataset
PBMC dataset
Download the data with the R code
- GSE53045:Epigenome analysis of smoking in peripheral blood mononuclear cells (PBMC) samples:50 smokers and 61 non-smokers.
- GSE32148:DNA methylation in peripheral blood (PBMC) from individuals with 17 Crohns' disease, 11 ulcerative colitis and 20 normal controls
- GSE35069:Differential DNA Methylation in Purified Human Blood Cells (CD4+,CD8+,CD19+)
- GSE36054:Methylation Profiling of Blood DNA from Healthy Children
- GSE36064:Methylation Profiling of Blood DNA from 78 Healthy Children (peripheral blood leukocytes from healthy children, Harvard Age Study)
- GSE41169:Blood DNA methylation profiles in a Dutch population (62 schizophrenia and 33 normal )
- GSE44132:Antenatal Blood DNA Methylation Profiles (Depression or Bipolar Disorder)
- GSE42861:Differential DNA methylation in the PBMC from 354 Rheumatoid arthritis and 337 normal
iPS and ES dataset
Download the data with the R code
- GSE54115: Aberrant DNA methylation reprogramming during iPS cell generation is dependent on the choice of reprogramming factors
- GSE56851: Increased Risk of Genetic and Epigenetic Instability in Human Embryonic Stem Cells Associated with Specific Culture Conditions
- GSE54769: Tissue- and Aging-specific DNA-Methylation Patterns are erased in Mesenchymal Stromal Cells derived from Induced Pluripotent Stem Cells.
Result
High density CpG regions in 450K array
- HighdensityCpgRegion450k.jpg
Figure 1. High density CpG regions
For different cancer dataset, the number of the high density CpG region (HDR) were slightly different because some probes might totally or majority missed in the dataset and these probes were removed and then to find the HDR. From the HDR analysis, you could find that there is not significant difference for the number of the HDR in different cancer dataset. the number of the HDR in all the 11 dataset ranged from 4858 to 5290. The average length of the HDR is 171bp (IQR=117bp) and the average distance between neighbor CpG site is about 27bp(IQR=13bp). the average number of the CpG site in the HDR is 6 (range from 5 to 50).
Correlation of high density CpG regions in 450K array
- Density.CorrelationDifferenceBetweenCancerNormal.jpg
Figure 2. Density plot for the correlation of high density CpG regions in 450K array between cancer and normals
- CorrelationDifferenceBetweenCancerNormal.jpg
Figure 3. Comparison of the correlation for the high density CpG regions in 450K array between cancer and normals
- It is very interesting result: 1) cancer have two peak 2) cancer have more higher correction regions
Comparing the correlation of HDR, we could find the similarity of the correlation between cancer and normal were interrupted. the cancer tissues had more high correlated HDRs than the normal tissues. What's more, obviously, there is only one peak in the density plot of correlations in the normal tissues while there is an extra peaks in the cancer tissues.
Based on the Figure 3, I think, the region whose average correlation is large than 0.5 (cut-off) can be defined as methylation block. However, the performance of cut-off of 0.4, 0.5, 0.6 were analysis in the following section. The result can be found in the below figure. Although the number of the conservative regions were quite different (left), the relative proportion of the conservative regions were almost same (right). I don't know how to set a unique threshold to define methylation block, In additional, there is no such unique threshold to define LD in population genetics, therefore, I think an arbitrary cut-off should be set here.
- Picture4ConservationFragtmentin450kdataset.png
Figure 4A. Conversation Methylation Block Screening among 11 TCGA Normal dataset.
- ConservationFragtmentin450kdataset.Cancer.jpg
Figure 4B. Conversation Methylation Block Screening among 11 TCGA Cancer dataset.
As the obove figure showed, different cut-off of the definition of methythylation block were almost similar when average correlation is larger than 0.4, 2015-7-10-001-code
- 2421 Common methylation blocks in TCGA normal datset: File:Conservative.methylation.block.0.6.bed.txt
- 155 Shared methylation blocks in 11 normal samples: File:Conservative.methylation.block.C11.0.6.bed.txt
Similarly, we can get the shared high correlated HDRs in cancer samples 2015-7-10-002-code
Here, I want to show that these 2421 regions were no specially selected in 11 cancers. Suppose, these 2421 regions were derived from a same population, then the frequency of each regions occurred in 11 samples would have binomial distribution. therefore, we test each regions one by one and no any one of the regions significantly deviate from the theoretical distribution. In this analysis, P=
Methylation Haplotype based on PBMC dataset
- GSE35069.wb.pmbc.cd4.8.19.jpg
Figure 5. PCA Analysis to GSE35069: Genome-wide DNA methylation profile for PBMC, CD4+, CD8+ and other cell types in human blood.
- GSE42861.PC12.Status.Gender.jpg
Figure 6. PCA Analysis to GSE42861: Differential DNA methylation in the PBMC from 354 Rheumatoid arthritis and 337 normal.It is very interesting that disease status is not the largest difference among the dataset while gender is the most obvious soucre of the variance or difference.
- GSE35069.HDRC.Distribution.Density.jpg
Figure 7. The distribution of the correlations of high density of CpG region in methylation 450K beadchips in Havard Age Project of PBMC Dataset. Normal Samples
- GSE42861.HDRC.Distribution.Density.jpg
Figure 8. The distribution of the correlations of high density of CpG region in methylation 450K beadchips in RA PBMC dataset. Normal Samples
- Shared High HDRC regions by CD4+ T cells, CD8+ T cells, CD56+ NK cells, CD19+ B cells, CD14+ monocytes, neutrophils, and eosinophils cells were attached in File:GSE35069.Share.High.HDRC.0.6.bed.txt. Binomial test showed there is no any region was over-preferred, indicating these regions were conservative in different elements of the whole blood cells. In this analysis, P=0.34, in another way, the probability for each region in each sample was 0.34.
- Shared High HDRC regions by rheumatoid arthritis and normal were attached *File:GSE42861.Share.High.HDRC.0.6.bed.txt
Methylation Haplotype based on iPS and ES dataset
- IPSCs-MSCs.jpg
Figure 9. PCA Analysis to GSE54769: Tissue- and Aging-specific DNA-Methylation Patterns are erased in Mesenchymal Stromal Cells derived from Induced Pluripotent Stem Cells.
- GSE54115.ips.es.hf.jpg
Figure 10. PCA Analysis to GSE54115: Aberrant DNA methylation reprogramming during iPS cell generation is dependent on the choice of reprogramming factors.
- GSE54769.HDRC.Distribution.Density.jpg
Figure 11. The distribution of the HDRs in dataset of GSE54769: Tissue- and Aging-specific DNA-Methylation Patterns are erased in Mesenchymal Stromal Cells derived from Induced Pluripotent Stem Cells.
- GSE54115.ips.hf.hdr.correlation.jpeg
Figure 12. The distribution of the HDRs in HF,Yamanaka iPS and Thomson iPS. There are 604, 381 and 592 high CpG density regions whose average correlations are larger than 0.6 in HF,Yamanaka iPS and Thomson iPS. The analysis is based on GSE54115.
- Shared high HDRC regions for GSE54115 were attached in File:GSE54115.Share.High.HDRC.0.6.bed.txt
- Shared high HDRC regions for GSE54769 were attached in File:GSE54769.Share.High.HDRC.0.6.bed.txt
- HDRC.PBMC.TCGA.Overalpregion.jpg
Figure 12. Shared Methylation block regions between PBMC and TCGA dataset.