Polymorphisms involving gain or loss of CpG sites are significantly enriched in trait-associated SNPs

Abstract

Some single nucleotide polymorphisms (SNPs) influence the existence of CpG sites, the basis of DNA modification such as methylation and hydroxymethylation. These polymorphisms can lead to gain or loss of CpG sites and were defined as CpG site related SNPs (cgSNPs) in this study. The cgSNPs change DNA sequence and might potentially affect DNA modification such as methylation. However, the functional consequence of cgSNPs is poorly understood. We observed that a considerable proportion (23.0%) of common variants were cgSNPs in human genome. Mutations involving loss of CpG sites were associated with reduced levels of methylation (~20.2%) using The Cancer Genome Atlas (TCGA) data. Using public databases (SCAN and seeQTL) of expression quantitative trait loci (eQTLs), we found that the cgSNPs were significantly enriched in eQTLs via logistic regression and simulation test. Furthermore, we observed that cgSNPs were more likely to be trait-associated loci especially cancers using a catalog of published genome-wide association studies (GWAS) recorded by National Human Genome Research Institute (NHGRI). Our results indicated that cgSNP might be meaningful as annotation either in SNP functional prediction or in screening for trait-associated SNPs.

Keywords: CpG site; DNA methylation; cancer; epigenetic; single nucleotide polymorphism (SNP).

Figure 1. The methylation levels at CpG sites, which existed in normal tissue sample and lost in paired tumor sample due to mutations (cgMut), were higher in normal than tumor tissue. While no differences of methylation levels of nearby CpG sites were observed

The differences of methylation levels between normal and paired tumor tissue sample (TCGA ID = 3518) in cgMut related and nearby CpG sites (classified into 6 bins including ~ ± 10bp, ~ ± 50bp, ~ ± 100bp, ~ ± 500bp, ~ ± 1kb and ~ ± 2kb away from cgMut) were presented as box plot of the 50^th percentile (P50, Median) and range of difference of methylation levels. The top and bottom of the box represent the 75^th and 25^th percentile. The whiskers indicate the 10^th and 90^th percentile. An example was given in ~ ± 2kb bin.

Figure 2. cgSNPts are significantly enriched in eQTLs

The theoretical distribution of the counts of cgSNPts in 300 draws (each draw containing 500 SNPs, which is matched to the 500 eQTLs) is shown in histogram. The observed count of cgSNPts in the 500 eQTLs was 303 (shown as a solid circle). According to the observed count of cgSNPts and the distribution of the counts of cgSNPts in 300 matched draws, the enrichment P-value was 0.04.

Figure 3. Trait-associated loci (all the traits were pooled) were significantly associated with cgSNPts

We classified the trait-associated loci into 8 bins according to the reported P-values and did analyses for each bin. The numbers of trait-associated loci in 8 bins were shown in slash bar graphs. The solid circles indicated the ORs of cgSNPts from logistic regressions. The whiskers represented 95% confidence intervals (CI) of ORs. A ‘*’ was marked if the logistic regression of enrichment test achieved statistical significance in each bin (P < 0.05).

Figure 4. cgSNPts are significantly enriched in trait-associated loci with the reported P-values ranging from E-5~ to E-11~, while no enrichment was observed with P-values less than E-14

The theoretical distributions of the count of cgSNPts in 300 draws (each draw containing 500 SNPs, which is matched to the 500 trait-associated loci) are shown in the histograms. The observed counts of cgSNPts in the 500 trait associated-loci are shown as solid circles. The left graph shown that cgSNPts were significantly enriched in trait-associated loci with P-values ranging from E-5~ to E-11~ (count = 317/500, P = 0.023), while the right graph indicated that cgSNPts were not enriched in trait-associated SNPs with P-values less than E-14 (count = 306/500, P = 0.31).

Figure 5. The comparisons of the enrichment effect sizes between cgSNPts in cancer associated loci and non-cancer associated loci

The OR and 95% CI were revealed by logistic regressions in 8 bins (according to the reported P-values). The effect sizes of the associations between cancer associated loci and cgSNPts were marked with solid circles. The effect sizes of the associations between non-cancer trait-associated loci and cgSNPts were marked with solid squares.

Figure 6. cgSNPts are significantly enriched in cancer associated loci

The theoretical distributions of the number of cgSNPts in 300 draws (each draw containing 500 SNPs, which is matched to the 500 cancer associated loci) are shown in the histogram. The observed count of cgSNPts in the 500 cancer associated loci is 345 (shown as a solid circle). According to the observed count of cgSNPts and the distribution of the counts of cgSNPts in 300 matched draws, the enrichment P-value was less than 0.003.