Methylation profiles were calculated as described previously6 and are available from the HEP database/browser at www.epigenome.org . Kruskall-Wallis tests were used to determine differential methylation between tissues (T-DMRs), measuring the proportion of uncorrected p-values that were smaller 0.001 for all CpGs. As this test is insensitive to samples that were only measured in a single sample such as sperm and placenta, the obtained number of T-DMRs is unlikely to be overstated due to putative aberrant methylation within these samples. Some T-DMRs were experimentally validated by sequencing independent DNA samples. Equality between two groups (age and sex) was performed using Wilcoxon tests.
For the analysis of co-methylation, median methylation values were used over all technical replicates to minimize any skewing effect because of possible outliers. In addition, we excluded all CpGs where the methylation values derived from the forward and reverse reads of the same amplicon differed by more than 10%. Based on this criterion, 38% of CpGs were excluded from the analysis. As only one DNA strand was analysed following bisulfite conversion, no assessment of hemimethylation was possible in this case. Methylation changes were calculated based on the absolute methylation differences between CpG pairs of identical samples. To minimize a bias introduced by the amplicon selection, the analysis was performed using both, individual CpGs (window size 20,000bp) and CpGs of the same amplicons. Co-methylation of CpGs was described as a function of similar methylation levels over distance (in bp).
For scatter plots, equal amounts of measurements were binned and ranked by numerical order of the X-axis values, representing means of X- and Y- data. For box plots and histograms, data were binned according to the intervals indicated on the X-axis containing different numbers of measurements.
For the analysis of co-methylation, median methylation values were used over all technical replicates to minimize any skewing effect because of possible outliers. In addition, we excluded all CpGs where the methylation values derived from the forward and reverse reads of the same amplicon differed by more than 10%. Based on this criterion, 38% of CpGs were excluded from the analysis. As only one DNA strand was analysed following bisulfite conversion, no assessment of hemimethylation was possible in this case. Methylation changes were calculated based on the absolute methylation differences between CpG pairs of identical samples. To minimize a bias introduced by the amplicon selection, the analysis was performed using both, individual CpGs (window size 20,000bp) and CpGs of the same amplicons. Co-methylation of CpGs was described as a function of similar methylation levels over distance (in bp).
For scatter plots, equal amounts of measurements were binned and ranked by numerical order of the X-axis values, representing means of X- and Y- data. For box plots and histograms, data were binned according to the intervals indicated on the X-axis containing different numbers of measurements.
