Conservation and divergence of DNA methylation in eukaryotes

Abstract

DNA methylation is one of the most important heritable epigenetic modifications of the genome and is involved in the regulation of many cellular processes. Aberrant DNA methylation has been frequently reported to influence gene expression and subsequently cause various human diseases, including cancer. Recent rapid advances in next-generation sequencing technologies have enabled investigators to profile genome methylation patterns at single-base resolution. Remarkably, more than 20 eukaryotic methylomes have been generated thus far, with a majority published since November 2009. Analysis of this vast amount of data has dramatically enriched our knowledge of biological function, conservation and divergence of DNA methylation in eukaryotes. Even so, many specific functions of DNA methylation and their underlying regulatory systems still remain unknown to us. Here, we briefly introduce current approaches for DNA methylation profiling and then systematically review the features of whole genome DNA methylation patterns in eight animals, six plants and five fungi. Our systematic comparison provides new insights into the conservation and divergence of DNA methylation in eukaryotes and their regulation of gene expression. This work aims to summarize the current state of available methylome data and features informatively.

Introduction

DNA methylation, which occurs in the genome of a vast array of bacteria, plants, fungi and animals, has been proven to play an important role in many cellular processes. These include embryonic development, transcription, chromatin structure, X chromosome inactivation, genomic imprinting and chromosome stability.1–6 Consistent with these important roles, a growing number of human diseases are now known to be associated with aberrant DNA methylation, often in CpG islands or promoter regions.7 Consequently, regulation of DNA methylation is a vital part of normal development and when dysfunctional, may activate disease states such as cancer.8–10 For example, hypomethylation of the global genome or hypermethylation of specific tumor suppressor genes may contribute to human carcinogenesis.11,12 Therefore, the precise detection of both the extent and location of DNA methylation in specific genes and the genome in its entirety (often referred as the “methylome”) is urgently needed.13 Recent rapid advances in sequencing technologies have enabled profiling of the genome methylation pattern at single-base resolution. So far more than 20 eukaryotic methylomes have been generated by next-generation sequencing technologies, as summarized in Table 1. Analysis of this vast amount of methylome data has dramatically enhanced our knowledge of the conservation, divergence and biological function of DNA methylation in eukaryotes.14–16 In this paper, we first briefly review the technologies of DNA methylation detection and then systematically compare the features of whole genome DNA methylation in eight animals, six plants and five fungi. This paper provides an overview of DNA methylation features and differences in these organisms, with data primarily generated since November 2009. This study also provides some new insights into the evolution and role of DNA methylation in eukaryotes.

Methods for Genome-Wide DNA Methylation Profiling

In recent years, many methods have been developed to detect the genome-wide cytosine methylation status.17–19 Widely used methods include restriction enzyme digestion of methylated DNA followed by hybridization to high-density oligo-nucleotide arrays or sequencing4,20 and capture of methylated genomic DNA with antibodies that target 5-methylcytosine (methylated DNA immunoprecipitation, MeDIP) or methyl-binding domain (MBD) proteins, followed by array hybridization or sequencing.21–23 These approaches can determine genome-wide methylation levels and patterns but have major limitations in resolution, restriction enzyme bias, difficulty in characterizing genome regions that are rich in repeats and, most importantly, an inability to detect DNA methylation at a single-base resolution.

Currently, the “gold standard” method for the detection of DNA methylation at single-base resolution is an integration of sodium bisulfite (BS) conversion and next-generation sequencing. Sodium bisulfite converts unmethylated cytosine bases to uracils, whereas methylated cytosine (mC) bases remain unchanged. Thus, after PCR amplification, unmethylated Cs are converted to thymines (T) while methylated Cs remain unchanged. This method provides quantitative, contiguous and base-pair resolution of genome methylation map, including methylation on CpG sites and non-CpG sites.19,24 Recent rapid advances in sequencing technologies have enabled profiling more and more genome methylation patterns at the single-base resolution, including more than 20 eukaryotic methylomes (Table 1).

Despite these advances in BS treatment-based sequencing, there are several drawbacks to using this technique. For example, the sample preparation steps associated with BS treatment are costly and time-consuming. Additionally, the BS treatment reduces sequence complexity by converting all mCs to Ts, which complicates the alignment of short reads to reference genomes. Finally, BS treatment-based sequencing cannot discriminate mC and a newly identified epigenetic marker, 5-hydroxymethylcytosine (hmC). The presence of hmC was recently reported in the adult mouse brain and human embryonic stem (ES) cells;25,26 its functional significance has yet to be determined.

Several new technologies may bypass these problems by detecting DNA modification directly on single DNA molecules without BS treatment. The pioneering single-molecule sequencing approaches, including nanopore-based methods and single-molecule real-time (SMRT) DNA sequencing, with the capacity to discriminate native from modified bases, may enable methylome profiling for hundreds of thousands of contiguous bases, or even for entire chromosomes, as single long reads in which both primary sequence and methylation are simultaneously determined.27–29 Although accuracy of the data generated by these approaches at the current stage is not satisfactory, it is likely that it will replace the current BS-conversion based methods in the foreseeable future. Such methods may advance the frontier of DNA methylome study from only a small handful of studies towards the examination of the patterning and dynamics of cytosine methylation in diverse populations, tissues and disease states. These might include the examination of diet/nutrition, distinct cell types, mutants or diseases and large numbers of individuals from geographical populations (population methylomes).

Potential Function of Non-CG Methylation in Mammals

A surprising and provocative discovery in recent methylome studies was the prevalence of non-CG methylation in human ES cells.30,31 Nearly 25% of all methylated cytosines in the human ES cells did not occur exclusively at CG sites. The prevalence of non-CG methylation in another human ES cell line H9 revealed that the non-CG methylation sites are conserved, suggesting that such methylation may be a general feature of human ES cells. A previous study also found substantial non-CG methylation in mouse ES cells, although its prevalence and genomic location was not clearly defined.32 Interestingly, non-CG methylation was found to be absent in fetal fibroblasts and mature peripheral blood mononuclear cells.30,31 The sharp contrast in non-CG methylation between ES cells and differentiated cells raises a critical question: is non-CG methylation functionally relevant and, if so, what is its primary importance in ES cells?

Examination of non-CG methylation patterns in human ES cells and comparison to DNA methylation among ES cells, differentiated cells and induced pluripotent stem cells (iPS) provided some insights into this question:30,31 (1) Non-CG methylation occurs more often in the gene bodies than in the promoter regions, where methylation is typically thought to suppress gene expression;30,31 (2) Non-CG methylation in gene bodies, however, is positively correlated with gene expression;31 (3) Non-CG methylation is noticeably enriched on the antisense strand of gene bodies and a significantly more intronic transcription can be generated from non-CG methylation enriched genes;30,31 (4) Non-CG methylation is significantly enriched in genes encoding proteins involved in RNA processing, splicing and RNA metabolic processes;31 (5) Although non-CG methylation is absent in fibroblasts, when differentiated cells were experimentally reprogrammed to an iPS, non-CG methylation was restored at the tested sites.31,33 This raises the possibility that non-CG methylation may play an important role in maintaining pluripotency and like DNA demethylation at specific pluripotency-related gene promoters, restoration of methylation at non-CG sites in gene bodies may be required for efficient reprogramming of induced pluripotent stem cells.31 It is well accepted that evolutionarily conserved methylation patterns tend to be functionally important, so it is necessary to compare the methylomes of human and mouse ES cells to reveal the evolutionarily conserved sites of non-CG methylation, but such an effort is yet to be seen.

Global DNA Methylation Patterns in Eukaryotes

New approaches combining BS conversion and next-generation sequencing allow us to measure the DNA methylation level more accurately and comprehensively. In addition to detecting whether a particular cytosine site is methylated, multiple reads covering each methylcytosine can be used as a readout of the fraction of the sequences within the sample that are methylated at that site.34 For each strand of a DNA segment or the whole genome, there are many different measurements for the level of DNA methylation. Using different measurements to compare the diverse methylation patterns among species might help us to understand the regulation and function of DNA methylation from different angles. Here, we propose two novel measurements to estimate the methylation level of a DNA segment or the whole genome. One is methylation broadness, which represents the fraction of cytosine sites detected as methylated in a given DNA segment. It can be calculated as the proportion of methylated sites over the total sites in a sequence (i.e., #mCG sites/total CG sites). The other is methylation deepness, which represents the extent of methylation of the methylated cytosines from reads (i.e., #reads covering mCs/total reads covering the same sites). We obtained all the available methylome data and calculated their broadness and deepness. As shown in Table 1 and Figure 1, the methylation deepness of CG sites in plants is typically as strong as in vertebrates. However, the methylation broadness of CG sites is much weaker in plants than in vertebrates (Table 1 and Fig. 1), indicating that the methylation at CG sites is conserved in different cell types in both plants and vertebrates and that CG methylation tends to occur in particular regions of plant genomes. We also found that the deepness of CHH methylation is higher than that of CHG methylation in most animals and fungi. Interestingly, the opposite pattern was observed in plant genomes, suggesting that the plant-specific machinery for establishing and maintaining CHG site methylation might be more efficient than in other organisms (Fig. 1).

The expansion of eukaryotic methylome data prompted a more comprehensive comparison of whole-genome methylation profiles across the plant and animal kingdoms, revealing both conserved and divergent features of DNA methylation in eukaryotes. Although DNA methylation appears to be a widespread epigenetic regulatory mechanism, genomes are methylated in different ways in diverse organisms. In animals, DNA methylation occurs mostly symmetrically (on both strands) at the cytosines of a CG dinucleotide. An intriguing exception is that a substantial portion of methylated cytosines in human ES cells were found in a CHG and CHH context.30,31 However, in higher plants, DNA methylation can occur at cytosines in both symmetric sequence contexts of CG and CHG (H = A, T or C), and also in an asymmetric CHH context, with the latter directed and maintained by small RNAs (see Table 1).35,36 In the model plant Arabidopsis thaliana, methylation broadness at CG, CHG and CHH nucleotides are about 24.48, 14.47 and 6.47%, respectively.37 Some fungi show substantial CpG site methylation and strong non-CpG methylation, corresponding to their Dnmt1-like and fungispecific Dim-2-type enzymes, respectively (Table 2). Despite the differing methylation sequence contexts, cytosine methylation is established and maintained by a family of conserved DNA methyltransferases. 36,38 Not surprisingly, the absence of DNA methylation in some eukaryotes such as yeast, roundworm and fruit fly has been associated with the evolutionary loss of DNA methyltransferase homologs.39

Recent methylome studies showed that cytosines are methylated not only in repetitive sequences and transposable elements (TEs) but also in promoters and gene bodies and that DNA methylation is highly correlated with transcription (Table 2). The methylation of gene's promoter regions has long been considered as a suppressor of gene expression. Consistent with DNA methylation's function of repressing transcription, gene expression correlates inversely with methylation level in the proximal region of the transcript start site (TSS) in all species with methylated promoters except invertebrates (Table 2). Considering that DNA methylation in invertebrate genomes occurs preferentially in gene bodies rather than promoter regions,40–42 the absence of correlation between promoter methylation and gene expression level might reflect background methylation signal in the promoter regions; thus, the role of suppression of gene expression via promoter methylation in invertebrates may be very weak. Furthermore, in the rice genome, the negative correlation was also observed at the 3′ end of genes. This suggests that the lack of methylation around the transcription termination site is important for gene expression.24,42

In plants and vertebrates, most methylated cytosines are found over repetitive elements and loss of this modification is associated with a transcriptional reactivation, as well as an increased mobilization of transposable elements. These observations likely reflect the ancestral role of cytosine methylation in the defense against invasive DNA. While DNA methylation in plants, fungi and vertebrates is concentrated in transposons, invertebrates showed an opposite pattern, with modifications occurring mainly in active genes (Fig. 1), suggesting that the use of DNA methylation to repress deleterious transposons in genomes may have evolved independently in plants and vertebrates, while this function was lost in the invertebrate lineage.24,42

As shown in Table 2, methylation of the gene body is highly conserved and was likely present in the last common ancestor of eukaryotes (Fig. 1).24,42 Gene body methylation is not associated with gene expression level, thereby leaving its biological role an open question. Several methylome studies have revealed an interesting parabolic relationship between gene-body methylation and transcription levels.23,24,41,42 Whereas modestly expressed genes are more likely to be methylated, genes expressed at either the lowest level or the highest level are usually less methylated. This suggests the coexistence of two opposing targeting mechanisms that increase or decrease methylation with gene expression. It has been suggested that extreme transcription rates may affect the balance between chromatin disruption and polymerase association, both of which could prevent the generation of aberrant transcripts that might drive methylation via a small RNA dependent pathway.23 Other interesting findings related to gene body methylation function are summarized as follows: (1) The gene body CHG and CHH methylation in human ES cells is correlated with gene expression level;31 (2) The methylation level of 3′ gene region is negatively correlated with gene expression in rice;42 (3) Exons tend to be more highly methylated than introns31 and (4) In gene bodies, methylated cytosines are more conserved than unmethylated cytosines and other nucleotides.43 These observations imply general roles of DNA methylation in transcriptional elongation, termination and perhaps in alternative splicing. Consistent with these findings, a recent study found exons were densely methylated at CpGs and packaged into nucleosomes.44 Exon enrichment of DNA methylation was specifically found in spliced exons and in exons with weak splice sites. Based on a high-resolution DNA methylation mapping of 24.7 million CpG sites in human brain, Maunakea et al.45 found 34% of all intragenic CpG islands were methylated, compared with only 3% of CpG islands in the 5′ promoters. Interestingly, tissue-specific methylation was more common in intragenic CpG islands than in 5′ promoters. Intragenic CpG islands also overlapped with markers of transcription initiation and unmethylated CpG islands overlapped with trimethylated H3K4 (a histone mark enriched at promoters). Altogether, these findings suggest that gene body methylation might play an important role in regulating tissue-specific or cell-specific alternative promoters.45 Although many assumptions have been proposed about the function of intragenic methylation, further genomewide comparison of DNA methylation patterns and experimental verification are warranted to illustrate the exact function of gene body methylation.

Perspectives

Although the latest studies have greatly improved our understanding of the evolutionary adaptations and conservation of DNA methylation, these studies nevertheless raised more questions than they could answer, as we are eager to further advance our knowledge in cellular systems. For example, the function of conserved gene body methylation is still unclear, although it has been proposed to suppress aberrant transcription from cryptic promoters inside the genes and regulate the alternative promoter.31,45 Furthermore, we still do not understand the mechanism by which DNA methylation regulates gene expression, especially with regard to the phenomenon of higher methylation of modestly transcribed genes than those expressed at the highest or lowest level. As the cost of sequencing drops rapidly and dramatically and single-molecule sequencing approaches evolve to be practical in the near future,27,29 we expect that much more methylome data will be generated from different samples and organisms. Subsequent comprehensive studies over a broad range of these methylomes will ultimately provide us a detailed view of the function and evolution of DNA methylation and ideally, a DNA methylation atlas.

Figures and Tables

Figure 1 Evolution of DNA methylation level of the available eukaryote methylomes. The phylogenetic tree was based on NCBI Taxonomy Browser (www.ncbi.nlm.nih.gov/taxonomy/taxonomyhome.html). Only the topology is shown and the branch lengths are not proportional to evolutionary divergence time. Green and red boxes indicate high methylation of gene body and transposon elements (TEs), respectively. On the right, a heatmap shows DNA methylation level (broadness and deepness for each kind of sequence context). The methylome data of human IMR90 fetal lung fibroblasts was used to represent the human.

Table 1 Statistics and features of DNA methylation in vertebrates, invertebrates, plants and fungi

Species	Coverage (X-fold)	Methylation broadness (%)				Methylation deepness (%)				References
		C	CG	CHG	CHH	C	CG	CHG	CHH
Vertebrates
Human H1 cells	28.4	5.67	88.79	4.64	0.55	63.91	84.66	21.94	22.12	Lister, et al.31
Human IMR90 cells	29.6	4.13	85.58	0.00	0.00	71.17	71.20	-	-	Lister, et al.31
Human neonatal fibroblasts	9.1	4.42	69.31	1.20	1.15	48.69	67.12	-	-	Laurent, et al.30
WA09 hESCs	8.0	7.26	74.41	7.28	2.91	39.91	75.03	9.76	20.94	Laurent, et al.30
hESC-derived fibroblasts	9.8	4.52	68.87	1.41	1.26	49.66	74.22	-	-	Laurent, et al.30
Tetraodon nigroviridis	7.3	11.03	44.46	0.78	0.85	70.18	76.64	-	-	Zemach, et al.42
Invertebrates
Ciona intestinalis	15.3	6.42	26.08	1.65	1.31	23.65	44.36	1.81	2.42	Zemach, et al.42
Apis mellifera	11.3	0.93	1.52	0.72	0.68	6.45	11.28	-	-	Zemach, et al.42
Tribolium castaneum	2.3	0.22	0.22	0.22	0.22	-	-	-	-	Zemach, et al.42
Bombyx mori	4.7∼32.8	0.50	1.36	0.28	0.24	21.45	29.59	-	-	Zemach, et al.42 Xiang, et al.41
Drosophila melanogaster	5.6	0.42	0.49	0.44	0.39	-	-	-	-	Zemach, et al.42
Nematostella vectensis	10.8	3.04	13.20	0.94	0.71	42.21	56.76	-	-	Zemach, et al.42
Plants
Oryza sativa	9.6	20.75	36.09	28.01	11.84	58.00	83.80	60.97	27.86	Zemach, et al.42
Arabidopsis thaliana	12.0∼20.8	10.60	24.48	14.47	6.47	31.24	58.63	34.70	12.93	Lister, et al.31 Cokus, et al.34 Hsieh, et al.37
Selaginella moellendorffii	9.2	9.32	19.61	16.37	2.70	26.75	38.56	32.79	7.58	Zemach, et al.42
Physcomitrella patens	6.0	26.40	25.66	24.53	26.76	54.18	70.58	76.40	49.36	Zemach, et al.42
Chlorella sp. NC64A	3.1	18.03	45.77	2.69	0.49	88.31	93.40	41.79	-	Zemach, et al.42
Volvox carteri	7.9	1.70	5.30	0.39	0.36	26.50	30.66	-	-	Zemach, et al.42
Fungi
Phycomyces blakesleeanus	11.8	2.57	6.75	1.40	2.02	9.37	17.84	3.55	6.62	Zemach, et al.42
Coprinopsis cinerea	14.3	6.05	11.87	4.18	3.86	42.75	70.49	15.35	18.35	Zemach, et al.42
Laccaria bicolor	11.3	6.88	20.12	2.48	2.80	34.39	56.30	6.80	10.85	Zemach, et al.42
Postia placenta	15.1	8.47	13.16	7.14	6.28	38.27	71.97	20.34	17.73	Zemach, et al.42
Uncinocarpus reesii	23.0	6.08	4.62	6.71	6.42	19.87	8.18	19.47	25.05	Zemach, et al.42

Table 2 Genome features and DNA methylation patterns of the available eukaryote methylomes

Species	Genome sizea (Mb)	Gene numbera,b	GC content (%)	CpGO/E	TEsc (%)	Other repeat elementsd (%)	DNMTe	Contextf,g	Gene body methylationg	Transposonsg (repeats)	Promoter region methylation - gene expressiong
Vertebrates
Homo sapiens	3080.0	30210	40.9	0.24	44.4	5.0	1,3	CG, CHG, CHH	CG, CHG, CHH	CG	Negative correlation
Tetraodon nigroviridis	208.7	20562	46.0	0.60	0.1	2.1	1,3	CG	CG	CG	Negative correlation
Invertebrates
Ciona intestinalis	141.2	13922	35.7	0.87	11.7	1.5	1,3	CG	CG	Hypomethylated	No correlation
Apis mellifera	183.3	9450	34.9	1.54	0.1	8.2	1,3	CG	CG	Hypomethylated	No correlation
Tribolium castaneum	151.5	10159	33.9	1.15	2.9	2.3	1	No	No	No	No
Bombyx mori	431.7	18510	37.7	1.12	15.8	2.2	1	CG	CG	Hypomethylated	No correlation
Drosophila melanogaster	120.0	14869	42.4	0.93	11.0	2.3	No	No	No	No	No
Nemalostella vectensis	297.4	24780	40.6	0.78	11.7	2.0	1,3	CG	CG	Hypomethylated	No correlation
Plants
Oryza sativa	370.4	29070	43.6	0.87	37.2	1.2	1, 3, C	CG, CHG, CHH	CG	CG, CHG, CHH	Negative correlation
Arabidopsis thaliana	119.0	33282	36.0	0.72	13.9	2.4	1, 3, C	CG, CHG, CHH	CG	CG, CHG, CHH	Negative correlation
Selaginella moellendorffii	208.5	34747	45.2	0.73	0.1	1.6	1,3,C	CG, CHG, CHH	No	CG, CHG, CHH	No
Physcomitrella patens	453.9	36007	33.6	0.60	22.2	5.3	1, 3, C	CG, CHG, CHH	No	CG, CHG, CHH	No
Chlorella sp. NC64A	42.2	9791	67.1	0.74	0.0	4.9	1, C	CG, CHG	CG	CG, CHG	Negative correlation
Volvox carteri	125.5	14436	56.0	0.85	3.5	5.2	1, C	CG	CG	CG	Negative correlation
Fungi
Phycomyces blakesleeanus	50.9	14792	35.7	0.68	0.1	5.8	1, M2	CG, CHG, CHH	No	CG, CHG, CHH	No
Coprinopsis cinerea	36.1	13619	51.7	0.94	0.9	0.4	1	CG, CHG, CHH	No	CG, CHG, CHH	No
Laccaria bicolor	58.7	18215	47.0	0.81	0.7	1.0	1	CG, CHG, CHH	No	CG, CHG, CHH	No
Postia placenta	69.0	9083	53.8	1.02	2.0	0.1	1	CG, CHG, CHH	No	CG, CHG, CHH	No
Uncinocarpus reesii	22.1	7856	48.7	0.90	0.5	0.6	M2	CG, CHG, CHH	CG	CG, CHG, CHH	No

a The most recent versions of genome sequences and annotations were downloaded from NCBI, ENSEMBL, UCSC Genome Browser or the websites of the sequencing consortiums. Mitochondrial genome, chloroplast genome and “N” letters were excluded.

b Including protein-coding genes and non-coding RNA genes.

c Transposon elements (TEs) were annotated using RepeatMasker (www.repeatmasker.org) and repbase (release 20090604, www.girinst.org).

d Including small RNA, satellites, simple repeats and low complexity.

e DNMT, DNA methyltransferase; 1, Dnmt1; 3, Dnmt3; C, chromomethylase 3; M2, Dim-2.

f H, A, C or T.

g No, no methylation detected.

Acknowledgements

We would like to thank Jeffery Ewers for critical reading and improving the manuscript. This work was supported by a NIH grant (R03LM009598) from the National Library of Medicine, Vanderbilt's Specialized Program of Research Excellence in GI Cancer grant (P50CA95103) and the VICC Cancer Center Core grant (P30CA68485).