Figure 4

Cumulative distributions of the empirical P value for the observed number of pairs of CpGs 8 to 10 bp (A) and 12 to 14 bp (B) apart. The empirical P value (likelihood of observing as many or more CpG pairs at 8 to 10 bp/12 to 14 bp by chance) was determined for each CGI by comparison with 1,000 observations obtained by shuffling the CGI sequence while maintaining the dinucleotide frequencies. Here, we show the cumulative distributions of this P value among the CGIs in each of the three CGI categories (DMRs: DMR CGIs, other oocyte-methylated CGIs, and unmethylated CGIs). For a particular P value X, the cumulative distribution value Y equals the proportion of CGIs in the respective category with P < X. A vertical line marks X = 0.05, the P value threshold that we considered significant. At this threshold, 43%, 37% and 19% of unmethylated, DMR and other methylated CGIs, respectively, were significantly enriched with CpG pairs at 8 to 10 bp (A), versus 39%, 2% and 19% for CpG pairs at 12 to 14 bp. Overall, CpG pairs at 8 to 10 bp thus are enriched in DMR CGIs relative to CpG pairs at 12 to 14 bp, while no such distance-specific enrichment is observed for unmethylated or other methylated CGIs. This pattern is robust, that is, holds true for a wide range of P value thresholds.