A Bayesian deconvolution strategy for immunoprecipitation-based DNA methylome analysis

Abstract

DNA methylation is an indispensible epigenetic modification required for regulating the expression of mammalian genomes. Immunoprecipitation-based methods for DNA methylome analysis are rapidly shifting the bottleneck in this field from data generation to data analysis, necessitating the development of better analytical tools. In particular, an inability to estimate absolute methylation levels remains a major analytical difficulty associated with immunoprecipitation-based DNA methylation profiling. To address this issue, we developed a cross-platform algorithm-Bayesian tool for methylation analysis (Batman)-for analyzing methylated DNA immunoprecipitation (MeDIP) profiles generated using oligonucleotide arrays (MeDIP-chip) or next-generation sequencing (MeDIP-seq). We developed the latter approach to provide a high-resolution whole-genome DNA methylation profile (DNA methylome) of a mammalian genome. Strong correlation of our data, obtained using mature human spermatozoa, with those obtained using bisulfite sequencing suggest that combining MeDIP-seq or MeDIP-chip with Batman provides a robust, quantitative and cost-effective functional genomic strategy for elucidating the function of DNA methylation.

Figure 1

Calibration of the Batman model against MeDIP-chip data (a) Estimated CpG coupling factors for a MeDIP-chip experiment as a function of the distance between a CpG dinucleotide and a microarray probe. (b) Plot of array signal against total CpG coupling factor, showing a linear regression fit to the low-CpG portion, as used in the Batman calibration step. This plot shows all data from one array on chromosome 6.

Figure 1

Calibration of the Batman model against MeDIP-chip data (a) Estimated CpG coupling factors for a MeDIP-chip experiment as a function of the distance between a CpG dinucleotide and a microarray probe. (b) Plot of array signal against total CpG coupling factor, showing a linear regression fit to the low-CpG portion, as used in the Batman calibration step. This plot shows all data from one array on chromosome 6.

Figure 2

Comparison of Batman-analyzed MeDIP-chip data with bisulfite-PCR sequencing data from the Human Epigenome Project (a) Plot of MeDIP-chip data against CpG coupling factor, with points colored by methylation values from the HEP bisulfite-sequencing data. All probes that did not overlap at least one CpG annotated in HEP were excluded. (b) Comparisons of MeDIP-chip data with HEP using a range of processing strategies: LOESS-normalized log₂-ratios in a 100bp window centered around a 50mer probe that overlaps a HEP amplicon (top left), simple averaging of the LOESS-normalized log₂-ratios for all probes within a 500bp window (top right), averaging of the LOESS-normalized log₂-ratios for all probes within a 500bp window and then dividing by the observed/expected CpG density (bottom left), Batman analyzed (bottom right). This analysis was derived from 1481 MeDIP-chip probes that overlapped 667 bisulfite-PCR amplicons from the HEP. HEP methylation values for all CpGs that overlapped any given 100bp MeDIP-chip window were averaged. Furthermore, to reduce noise in the HEP dataset, all 100 bp windows were required to have at least 2 HEP scores (i.e. data from the top and bottom bisulfite-PCR strands for windows containing a single CpG site, or from at least 2 different CpG sites) that differed by <50%. The purple – yellow (0 – 30) color bar on the right of each figure shows the total CpG coupling factor for each probe (c) Comparison of Batman-quantified MeDIP data with bisulfite data from HEP. Points show the mean Batman output for regions with a given HEP methylation level. Error bars show 95% bootstrap credible intervals.

Figure 2

Comparison of Batman-analyzed MeDIP-chip data with bisulfite-PCR sequencing data from the Human Epigenome Project (a) Plot of MeDIP-chip data against CpG coupling factor, with points colored by methylation values from the HEP bisulfite-sequencing data. All probes that did not overlap at least one CpG annotated in HEP were excluded. (b) Comparisons of MeDIP-chip data with HEP using a range of processing strategies: LOESS-normalized log₂-ratios in a 100bp window centered around a 50mer probe that overlaps a HEP amplicon (top left), simple averaging of the LOESS-normalized log₂-ratios for all probes within a 500bp window (top right), averaging of the LOESS-normalized log₂-ratios for all probes within a 500bp window and then dividing by the observed/expected CpG density (bottom left), Batman analyzed (bottom right). This analysis was derived from 1481 MeDIP-chip probes that overlapped 667 bisulfite-PCR amplicons from the HEP. HEP methylation values for all CpGs that overlapped any given 100bp MeDIP-chip window were averaged. Furthermore, to reduce noise in the HEP dataset, all 100 bp windows were required to have at least 2 HEP scores (i.e. data from the top and bottom bisulfite-PCR strands for windows containing a single CpG site, or from at least 2 different CpG sites) that differed by <50%. The purple – yellow (0 – 30) color bar on the right of each figure shows the total CpG coupling factor for each probe (c) Comparison of Batman-quantified MeDIP data with bisulfite data from HEP. Points show the mean Batman output for regions with a given HEP methylation level. Error bars show 95% bootstrap credible intervals.

Figure 2

Comparison of Batman-analyzed MeDIP-chip data with bisulfite-PCR sequencing data from the Human Epigenome Project (a) Plot of MeDIP-chip data against CpG coupling factor, with points colored by methylation values from the HEP bisulfite-sequencing data. All probes that did not overlap at least one CpG annotated in HEP were excluded. (b) Comparisons of MeDIP-chip data with HEP using a range of processing strategies: LOESS-normalized log₂-ratios in a 100bp window centered around a 50mer probe that overlaps a HEP amplicon (top left), simple averaging of the LOESS-normalized log₂-ratios for all probes within a 500bp window (top right), averaging of the LOESS-normalized log₂-ratios for all probes within a 500bp window and then dividing by the observed/expected CpG density (bottom left), Batman analyzed (bottom right). This analysis was derived from 1481 MeDIP-chip probes that overlapped 667 bisulfite-PCR amplicons from the HEP. HEP methylation values for all CpGs that overlapped any given 100bp MeDIP-chip window were averaged. Furthermore, to reduce noise in the HEP dataset, all 100 bp windows were required to have at least 2 HEP scores (i.e. data from the top and bottom bisulfite-PCR strands for windows containing a single CpG site, or from at least 2 different CpG sites) that differed by <50%. The purple – yellow (0 – 30) color bar on the right of each figure shows the total CpG coupling factor for each probe (c) Comparison of Batman-quantified MeDIP data with bisulfite data from HEP. Points show the mean Batman output for regions with a given HEP methylation level. Error bars show 95% bootstrap credible intervals.

Figure 3

Mapping quality and genomic coverage of the MeDIP-seq data (a) Histogram showing the fractions of high-quality paired-end read mappings in 50kb windows across the genome. (b) Fraction of methylated regions (>60% methylation) which are not covered by reads in our MeDIP-seq dataset. As with all the MeDIP-seq analyses, the reads are extended to a length of 500bp.

Figure 3

Mapping quality and genomic coverage of the MeDIP-seq data (a) Histogram showing the fractions of high-quality paired-end read mappings in 50kb windows across the genome. (b) Fraction of methylated regions (>60% methylation) which are not covered by reads in our MeDIP-seq dataset. As with all the MeDIP-seq analyses, the reads are extended to a length of 500bp.

Figure 4

Comparison of Batman-analyzed MeDIP-seq data with bisulfite-PCR sequencing data from the Human Epigenome Project (a) MeDIP-seq read depth (i.e. the number of confidently placed reads overlapping a given point in the genome) for points overlapping HEP amplicons, plotted against total CpG coupling factor. Points are colored according to sperm DNA methylation (yellow – blue represents 0 – 100% methylation), as measured by in HEP. (b) MeDIP-seq versus sperm bisulfite-PCR sequencing data from the Human Epigenome Project (HEP). 100 bp MeDIP-seq tiles are plotted against 1,322 overlapping HEP bisulfite-PCR amplicons. As in Figure 2b, HEP methylation values for all CpGs that overlapped any given 100bp MeDIP-seq tile were averaged, and all 100 bp windows were required to have at least 2 HEP scores (i.e. either data from the top and bottom strand for a single CpG site, or at least 2 CpG sites) that differed by <50%. The purple – yellow (0 -30) color bar on the right of each figure shows the total CpG coupling factor for each 100 bp tile. The same data stratified by CpG density is displayed in Supplementary Figure 4 online.

Figure 4

Comparison of Batman-analyzed MeDIP-seq data with bisulfite-PCR sequencing data from the Human Epigenome Project (a) MeDIP-seq read depth (i.e. the number of confidently placed reads overlapping a given point in the genome) for points overlapping HEP amplicons, plotted against total CpG coupling factor. Points are colored according to sperm DNA methylation (yellow – blue represents 0 – 100% methylation), as measured by in HEP. (b) MeDIP-seq versus sperm bisulfite-PCR sequencing data from the Human Epigenome Project (HEP). 100 bp MeDIP-seq tiles are plotted against 1,322 overlapping HEP bisulfite-PCR amplicons. As in Figure 2b, HEP methylation values for all CpGs that overlapped any given 100bp MeDIP-seq tile were averaged, and all 100 bp windows were required to have at least 2 HEP scores (i.e. either data from the top and bottom strand for a single CpG site, or at least 2 CpG sites) that differed by <50%. The purple – yellow (0 -30) color bar on the right of each figure shows the total CpG coupling factor for each 100 bp tile. The same data stratified by CpG density is displayed in Supplementary Figure 4 online.

Figure 5

Genomic coverage and web display of the MeDIP-seq data (a) Genomic coverage of MeDIP-seq (measured as fraction of CpGs). Genomic features are from the Ensembl genome database (release 45). The first ten bars are mutually exclusive, i.e. repeats are not included when considering subsequent features. Numbers in parentheses indicate the total number of CpGs within the human genome in that category. Promoters are defined as 2kb regions centered on annotated transcriptional start sites, and Reg. features represent non-promoter Regulatory Features in Ensembl. The colors represent the range of DNA methylation levels (b) MeDIP-seq data integrated into Ensembl along with MeDIP-chip data of the same sperm DNA sample, and sperm bisulfite-PCR data from the Human Epigenome Project. The yellow – green – blue color gradient represents 0 – 100% methylation.

Figure 5

Genomic coverage and web display of the MeDIP-seq data (a) Genomic coverage of MeDIP-seq (measured as fraction of CpGs). Genomic features are from the Ensembl genome database (release 45). The first ten bars are mutually exclusive, i.e. repeats are not included when considering subsequent features. Numbers in parentheses indicate the total number of CpGs within the human genome in that category. Promoters are defined as 2kb regions centered on annotated transcriptional start sites, and Reg. features represent non-promoter Regulatory Features in Ensembl. The colors represent the range of DNA methylation levels (b) MeDIP-seq data integrated into Ensembl along with MeDIP-chip data of the same sperm DNA sample, and sperm bisulfite-PCR data from the Human Epigenome Project. The yellow – green – blue color gradient represents 0 – 100% methylation.