Large-scale methylation domains mark a functional subset of neuronally expressed genes

Abstract

DNA methylation is essential for embryonic and neuronal differentiation, but the function of most genomic DNA methylation marks is poorly understood. Generally the human genome is highly methylated (>70%) except for CpG islands and gene promoters. However, it was recently shown that the IMR90 human fetal lung fibroblast cells have large regions of the genome with partially methylated domains (PMDs, <70% average methylation), in contrast to the rest of the genome which is in highly methylated domains (HMDs, >70% average methylation). Using bisulfite conversion followed by high-throughput sequencing (MethylC-seq), we discovered that human SH-SY5Y neuronal cells also contain PMDs. We developed a novel hidden Markov model (HMM) to computationally map the genomic locations of PMDs in both cell types and found that autosomal PMDs can be >9 Mb in length and cover 41% of the IMR90 genome and 19% of the SH-SY5Y genome. Genomic regions marked by cell line specific PMDs contain genes that are expressed in a tissue-specific manner, with PMDs being a mark of repressed transcription. Genes contained within N-HMDs (neuronal HMDs, defined as a PMD in IMR90 but HMD in SH-SY5Y) were significantly enriched for calcium signaling, synaptic transmission, and neuron differentiation functions. Autism candidate genes were enriched within PMDs and the largest PMD observed in SH-SY5Y cells marked a 10 Mb cluster of cadherin genes with strong genetic association to autism. Our results suggest that these large-scale methylation domain maps could be relevant to interpreting and directing future investigations into the elusive etiology of autism.

Figure 1.

Global DNA methylation levels within the cell lines and methylation domain types. (A) The distribution of average percent methylation across 20-kb windows is shown for each of the three cell lines. (B) Average percent methylation within the four methylation domain types (N-HMD, L-HMD, B-PMD, or B-HMD) defined using the HMM. The H1 cell line is included for comparison. Error bars show standard deviations. (C) Average percent methylation of promoter CpG islands in human cerebral cortex and the SH-SY5Y and IMR90 cell lines. N-HMD, neuronal HMD; L-HMD, lung HMD; B-PMD, PMD in both cell lines; B-HMD, HMD in both cell lines.

Figure 2.

Examples of B-PMDs, N-HMDs, and L-HMDs in the three cell lines. Data were loaded as tracks in the UCSC Human Genome Browser. Hidden Markov model PMD annotations for the IMR90 and SH-SY5Y cells are shown as black bars. Percent methylation tracks display MethylC-seq data for individual CpG sites. Red = 80%–100% DNA methylation, green = 60%–80%, blue = 1%–60%, black = 0%, and white = no sequence coverage. PMDs in IMR90 cells appear blue and green. Due to lower sequence coverage, PMDs in SH-SY5Y cells appear black. (A) B-PMDs mark a cluster of 11 olfactory receptor genes and part of an HLA gene cluster. B-PMDs are shaded in gray. (B) N-HMDs mark CNTNAP2, a neurexin and autism candidate gene (Alarcon et al. 2008), and DPP6, a gene highly expressed in brain and critical for membrane excitability (Kim et al. 2008). N-HMDs are shaded in green. (C) L-HMDs mark CFTR, CAV1, and CAV2, all genes with important functions in lung (Mehta 2005; Gosens et al. 2008). L-HMDs are shaded in pink. OR, olfactory receptor; HLA, human leukocyte antigen; CNTNAP2, Contactin-associated protein-like 2; DPP6, dipeptidyl-peptidase 6; CAV1&2, caveolins 1 and 2; CFTR, cystic fibrosis transmembrane conductance regulator.

Figure 3.

Cell type specific expression of transcripts contained within PMDs. Gray circles are normalized expression values for all microarray probes, averaged over triplicates in both cell lines. Red triangles represent probes to genes within the specified domain type. Light green lines mark the 75th percentile for expression values. Percentages in each quadrant show the percentage of probes in that PMD type for that quadrant.

Figure 4.

The characteristics of genes in PMDs. (A) Venn diagram showing overlap of PMDs between SH-SY5Y and IMR90 cells. The numbers of genes in each domain type are shown. (B–D) GO biological process classifications for genes in each domain type using DAVID. Bonferroni correction was applied to the P-values. Redundant GO terms containing similar lists of genes were excluded and a full list can be found in Supplemental Table 1. Because of the large number of olfactory receptor genes in B-PMDs, the “sensory perception of smell” category had a Bonferroni P-value of 2.9 × 10⁻²⁷⁷.

Figure 5.

B-PMD at the 5p14.1 locus implicated in autism. Diagram of the largest PMD in SH-SY5Y cells that does not include centromeric or telomeric sequence (9.7 Mb). The domain includes four cadherin (CDH) genes. Both the rs4307059 and rs10038113 SNPs have been implicated in autism spectrum disorder (Wang et al. 2009). Notably, rs7727656 was associated with hippocampal atrophy (Potkin et al. 2009).