Human-specific CpG “beacons” identify loci associated with human-specific traits and disease

Abstract

Regulatory change has long been hypothesized to drive the delineation of the human phenotype from other closely related primates. Here we provide evidence that CpG dinucleotides play a special role in this process. CpGs enable epigenome variability via DNA methylation, and this epigenetic mark functions as a regulatory mechanism. Therefore, species-specific CpGs may influence species-specific regulation. We report non-polymorphic species-specific CpG dinucleotides (termed “CpG beacons”) as a distinct genomic feature associated with CpG island (CGI) evolution, human traits and disease. Using an inter-primate comparison, we identified 21 extreme CpG beacon clusters (≥ 20/kb peaks, empirical p < 1.0 × 10⁻³) in humans, which include associations with four monogenic developmental and neurological disease related genes (Benjamini-Hochberg corrected p = 6.03 × 10⁻³). We also demonstrate that beacon-mediated CpG density gain in CGIs correlates with reduced methylation in these species in orthologous CGIs over time, via human, chimpanzee and macaque MeDIP-seq. Therefore mapping into both the genomic and epigenomic space the identified CpG beacon clusters define points of intersection where a substantial two-way interaction between genetic sequence and epigenetic state has occurred. Taken together, our data support a model for CpG beacons to contribute to CGI evolution from genesis to tissue-specific to constitutively active CGIs.

Introduction

The CpG dinucleotide is unique for its ability to carry both genetic and epigenetic information in the genome of a differentiated mammalian cell.^1,2 Variation in DNA methylation, facilitated by this two base pair motif, influences gene expression, and thereby enables tissue-specific function.^3-6 However, this dinucleotide is substantially depleted, to one fifth of the expected level, due to the hypermutability (~11-fold) of cytosines when methylated.^7-9 Yet, a minority of CpGs is retained against this strong tide of loss by a variable combination of: evasion of methylation in the germline, functional importance or chance. These are predominantly in CpG dense regions.¹⁰ Additionally, new CpGs are created by mutation through base substitution and as a by-product of the increase of GC in regions of biased gene conversion (BGC).¹¹

A high density region of unmethylated CpGs can recruit CpG binding proteins, such as Cfp1 and KDM2A, which modify histone tails.^12,13 Thomson et al. have shown that the experimental inclusion of a cluster of unmethylated CpGs, is sufficient to establish domains of H3K4me3.¹² This histone modification leads to genomic three-dimensional structure change and the acquisition of permissive chromatin regions within the expanse of repressed genome.¹⁴ CpG clusters, termed CpG Islands (CGIs), co-locate with 60–70% of human gene promoters, often those of housekeeping genes that are hypomethylated in the germ line,^15,16 but also 40% that are tissue-specific.^10,17 Methylation of CGI promoters acts as a durable silencing mechanism. However, the majority of CGIs are unmethylated in differentiated cells independently of their transcriptional activity.^10,18 The methylation state of CGI is strongly correlated with its CpG content, with high density CGIs being predominately constitutively unmethylated and “weak” low density islands the preferred target for tissue-specific methylation.^10,19 CpG gain that shifts an island from weak to strong status therefore affects its dynamic ability for methylation change.

CpGs located in the lower density regions surrounding islands, termed CpG shores (~2kb up- or down-stream), identify significant tissue-, cancer- and reprogramming-specific methylation variability.^6,20-22 Therefore, shore accretion and island erosion by subtle modulation in CpG density within these regions may have a disproportionate influence on the methylation levels and locations of these flanking regions. Additionally, an increase in methylation variance has been proposed to have an evolutionary important role, as well as being a potential influence on disease susceptibility.²³

The genetic loss and gain of CpG dinucleotides over evolutionary time will impact upon the epigenome. Genome-wide variation in GC content at the megabase scale led to the formation of isochores before mammalian radiation^24,25 with an increase in CpGs occurring ~90 million years ago (MYA).²⁶ A subsequent clock-like loss of CpGs, due to the time- not generation-dependent substitution rate of cytosine deamination,²⁷ has led to roughly similarly numbered, but differing sets of CpGs in primates. The mutability of individual CpGs can be determined by accounting for the influence of surrounding CpG density, as well as by sequence context and nucleosome position.^28,29 Regions of CpGs that remain hyperconserved have been found to co-locate with polycomb repressive complex 2 binding domains and developmental genes.²⁸

On the other hand, GC increase is influenced by primate recombination rates.^30,31 So much so, that regions of extreme substitutional divergence in the human genome^32-34 co-locate with recombination-associated BGC.^35-39 This process therefore negates or obscures any potential evidence of weak selection.^39,40 BGC can lead to the formation of CGIs¹¹ and, furthermore, Cohen et al. have recently shown that CGIs can evolve without the requirement of selective pressure,⁴¹ although a possible subtle influence on CpGs via gene body methylation may exist.⁴²

Cytosine deamination is consequently the predominant single nucleotide mutational force, occurring at one order of magnitude higher in the genome than other single base substitutions. Conversely, a highly localized BGC-mediated increase of CpGs occurs, associated with recombination.^36,40 To discover the locations of potential species-specific regulatory modulation, due to CpG dinucleotide change, we identified a subset of human CpGs that were only present, either uniquely maintained or gained, in that lineage. While there are approximately ~40 million genetic differences between human and chimpanzee, the vast majority are due to random genetic drift.⁴³ Divergence at CpG sites between these two species is estimated to be at 15.2%, compared with 0.92% for other nucleotide substitutions.⁴³ Therefore, we used the additive collective power of multiple closely related species in a six-species inter-primate comparison.^44,45

The CpG sequence can itself act as a genomic signaling molecule via combinatorial transcription factor binding specificity,^1,46 facilitate epigenomic variation by influencing CGI promoter amenable chromatin structure and gene body methylation.^47,48 Consequently we hypothesized that by identifying human-specific CpGs we may find potential regions of species-specific differential regulation.^5,42,49,50 The sequence comparative approach would be blind to any potential causative mechanisms. The novel CpG clusters identified may highlight genes where a species-specific shift in epigenetic control has been enabled by this genetic change. These regions would potentially be enriched for human traits, as well as the possibility of associations with disease susceptibilities that have arisen as a by-product of human evolution.

Results

Human-Specific CpGs

To uncover the subset of CpGs present only in the human lineage, an inter-primate comparison was performed by examining the Enredo-Pecan-Ortheus Whole-Genome Multiple Alignments Sequences for human, chimpanzee, gorilla, orang-utan, rhesus macaque and common marmoset (Ensembl Compara.6_primates_EPO).^51,52 This set of sequences contains 19,198 blocks and has been able to align 84.54% of the human genome. We parsed the blocks of this alignment requiring non-duplicated sequence in both human and chimpanzee and sequence of at least one other primate, which reduced our quasi-genomic set (referred to as h1c1o1: human, chimpanzee and other primate) to 79.99% of the human genome. This contained 25,100,205 or ~88.95% of the total haploid human CpGs. Each of these remaining CpGs was then interrogated with the requirement that at its precise position none of the other primates had a CpG dinucleotide present. Furthermore the chimpanzee sequence and the closest nearest other primate present in the alignment block (96.6% Gorilla) were required to have aligned sequence at this position i.e. was not N or -. This led to an initial human-specific subset of 1,820,319 CpGs. These CpGs were then conservatively filtered for polymorphism utilizing 1000 genomes data removing any CpG with any evidence of variation, as a SNP, or within a copy number or structural variant,⁵³ leading to a final estimate of 1,192,484 human-specific CpGs.

CpG beacons

We define “beacons” as species-specific non-polymorphic DNA motifs able to carry both genetic and epigenetic information. According to the above analysis, we estimated the number of CpG beacons to be ~1.19 million in the human genome. In the future a definitive set will be able to be established following mass whole genome sequencing in a large number of these primates. However this current calculation will already be enriched for “true” human CpG beacons that can facilitate unique species-specific epigenomic variation. A user interface to view the human CpG beacons in the UCSC genome browser in the context of existing annotation is available at www2.cancer.ucl.ac.uk/medicalgenomics/humanCpGBeacons/trackList.php.

The density distribution of the human beacons in 1 kb windows was estimated, which showed more than half were singletons, ~2% were ≥ 5 beacons/kb, and 0.03% were ≥ 20 beacons/kb. To assess the significance of this long tail with higher density, we performed 1,000 permutations by choosing a set of random beacons from the CpG locations in the h1c1o1 genomic set. This simulation never exceeded the number of peaks that are ≥ 20 CpG beacons/kb in the observed genome set (peaks ≥ 20 CpG beacons/kb: simulation peaks range = 0–7, simulation average = 1.527 peak per genome, observed peaks = 21, empirical p < 1 × 10⁻³).

Extreme CpG beacon clusters

Taking this ≥ 20 CpG beacons/kb as an initial threshold (which reflects an increase of ≥ 4% in CpG density per kb in human compared with the other primates) we identified 21 extreme genomic outliers of human CpG beacon density (see Table S1). Beacon density distribution is displayed across the genome in Figure 1. This initial observation revealed that the third highest peak on Chromosome 20 co-located with the promoter CGI of the HAR1A gene, a non-coding gene significant in cortical development discovered by Pollard et al.(Fig. S1).³² HAR1A was identified to be co-expressed by Cajal-Retzius neurons, with Reelin, a secreted glycoprotein that is fundamental in specifying the six-layer structure of human cortex.³² This gene had been first identified by mammalian and vertebrate comparative genomics for regions of high conservation but outlying substitution of any type in the human genome, with an extreme region of 118 bp containing 18 human changes since the Homo-Pan split. In fact eight of these substitutions are CpG beacon creating, from the total of 35 in this cluster that spans ~1.8 kb. This locus would still be a genomic CpG beacon cluster outlier with a peak of 24/kb even with this 118bp region removed. Therefore this critical non-coding gene was able to be identified without any recourse to longstanding vertebrate or mammalian conservation but purely by focusing on inter-primate CpG density change. Larger regions of bias identified across this locus have implicated recombination hotspot drift over time.^38,54

Figure 1. Human CpG beacons by 1 kb density score and genomic location. Loci greater than or equal to a threshold of 20/kb are indicated. A telomeric bias in peaks is evident, as well as in the historic chromosome 2q13 fusion point.³⁶

We were therefore interested in the significance of the other 20 CpG beacon clusters identified in the human genome. We preformed gene set enrichment with Ingenuity Pathway Analysis (IPA, ©2011 Ingenuity Systems, Inc.) on these genes overlapping an extreme beacon cluster (70%), or within 100 kb 5′ or 50 kb 3′ of their transcript, and found highest significant enrichment for the categories of developmental and genetic disorder, inflammatory response, reproductive system development and function, and neurological disease genes [p value with Benjamini-Hochberg correction (P_B-H) = 6.03 × 10⁻³, Fig. S2). These 20 clusters included the causative genes when mutated for four different monogenic mental retardation disorders (ANKRD11,⁵⁵ CHL1,⁵⁶ EHMT1⁵⁷ and VLDLR^58,59) and genes implicated in complex traits through GWAS and CNV analyses to phenotypes such as behavioral and psychiatric disorders including autism and bipolar disorder (ANKRD11,⁶⁰ DLGAP2⁶¹ and DPP10^60,62); synaptic transmission (SYT1⁶³); as well as total cerebral brain volume in a radiological examination (CDH4⁶⁴) (see Table S1). To take into consideration any possible gene size bias, we also performed an analysis using the regional based binomial test included in the GREAT gene enrichment analysis tool⁶⁵ (using default cut-offs but reducing the maximum distal extension from 1 Mb to 100 kb/ see methods) for these 20 CpG beacon clusters, excluding the known HAR1A result. This was significant for only three categories of biological process with an FDR Q ≤ 0.15: which included the categories cognition (binomial raw p = 1.43 × 10⁻⁵, FDR Q = 1.26 × 10⁻¹), and behavior (binomial raw p = 2.25 × 10⁻⁵, FDR Q = 9.87 × 10⁻²). Furthermore, as a negative control, we also calculated the locations of the Chimpanzee CpG beacon clusters that exceed the 20/kb threshold and these identified no genes implicated with developmental delay or mental retardation (see Table S2) and as well was non-significant with GREAT analysis (via liftOver to human).

The human CpG beacon clusters represent regions of potential regulatory modulation or change to the nearby genes that is human-specific. This correlation, and not causation within these regions, is of interest particularly as these monogenic disease genes have shown that genetic mutation within them is not lethal but carries significant developmental and neurological pathology. These important genes could therefore be the plausible targets of significant regulatory change between human and other closely related primates due to the similarity of their proteomes.⁶⁶

Biased-gene conversion overlap

The human extreme beacon clusters showed very strong overlap with the top 200 regions of BGC identified by Drezser et al. (57.1% of ≥ 20 CpG beacons/kb clusters, χ² p < 2.20 × 10⁻¹⁶),³⁶ thus implicating localized GC rise, which is thought to be a neutral process, with a consequent increase in CpGs. Therefore this implies the CpG beacon clusters associate with a recombination driven CpG increase in human, as opposed to regions of high CpG mutability in other primates. Moreover, we also identified the majority of these clusters in telomeric regions (52.3% in terminal chromosome bands), which are known to have elevated rates of recombination in males,⁶⁷ with hotspots associated with BGC.⁴⁰ A 15% greater divergence in terminal ends of chromosomes was identified in the chimpanzee sequencing project.⁸

Cohen et al. recently reclassified CGIs using evolutionary modeling into those that were classical hypo deaminated islands, with ~80% of these 10 kb from a transcription start site (TSS) and with strong overlap with H3K4me3, and those that had arisen as a by-product of BGC that were typically constitutively hypermethylated.⁴¹ However, on examination of the available sperm methylome data via MeDIP-seq,⁶⁸ which includes data for 18 of these extreme beacon cluster regions that co-located with CGI, these were found to be predominately hypomethylated. The average methylation level was 26.38%, therefore aiding the retention of CpGs by reduced mutability, enabling potential regulation by methylation to occur.¹⁰

Positive CpG beacon clusters

To differentiate between specific CpG increase, as opposed to generalized regions of GC rise, we controlled by the concurrent formation of the exact inverse dinucleotide GpC; which lacks methylation ability. We defined Positive CpG Beacon Clusters (PBCs) as regions where CpG beacons outweighed their local human-specific GpC content. BGC increases regional GC content and therefore passively CpGs, but if CpGs are methylated in the germ line their continual loss will eventually lead to the acquired GpCs outweighing CpGs over time. We calculated this via a sliding window analysis with a window size of 1 kb and slide of 100 bp across the genome (see Fig. 2). The vast majority of the extreme beacon clusters were genomic outliers of PBC score, i.e. EHMT1 and CDH4 and all except two possessed positive scores (see Table S1). These two extreme negative scores were identified in loci known for extensive and continual gene conversion, the olfactory receptors, with PBC score peaks of -23/kb and -16/kb for OR2T3 and OR2T12, respectively.

Figure 2. Human positive CpG beacon scores calculated across the genome in 1 kb windows with 100 bp slide. Extreme positive or negative loci are indicated.

Extreme CpG beacon clusters appear to be strongly driven by BGC; therefore, PBCs indicate regions where the gained CpGs beacons are not as hypermutable as would be expected, likely due to a loss of methylation in germ line. By retaining from the 20 clusters only those with at least a +5 PBC score, more significant p values in both biological category enrichments of cognition and behavior were obtained (binomial p = 7.19 × 10⁻⁶ and 9.41 × 10⁻⁶, Q FDR value = 6.3 × 10⁻² and 4.1 × 10⁻², respectively). To explore the potential of this CpG beacon-specific increase genome-wide, we identified all the PBC ≥ +5 loci comprising 2,601 regions, that account for ~0.1% of the human genome. IPA analysis of associated PBC genes showed significant results for a large number of common disease categories (P_B-H < 1 × 10⁻²⁰) (data not shown), although this result will be biased disproportionally with larger gene regions. Examining these PBC loci with GREAT (genomic regions enrichment of annotation tool) analysis, which corrects for this issue of potential genomic space available to input signal, we identified a number of significant results for potential human phenotypes and traits (see Table S3, FDR Q < 0.05), such as cortical gyral simplification (binomial FDR Q-value = 1.94 × 10⁻⁴), atrophy/degeneration affecting the central nervous system (Q = 1.69 × 10⁻²), abnormality of the cerebral cortex (Q = 2.23 × 10⁻²) and poor speech (Q = 4.66 × 10⁻²). A large number of biological processes were implicated as significant, including the nucleus accumbens development (Q = 6.31 × 10⁻⁷), nose development (Q = 4.52 × 10⁻⁶) and neurotransmitter transport (Q = 1.27 × 10⁻²) (see Table S4).

These PBCs regions were also enriched in intragenic islands (CGI within transcripts but not at the classical 5′ promoter region), being found twice as commonly within these regions as would be expected for their genomic size (see Fig. S3, χ2 p < 2.20 × 10⁻¹⁶). In fact, with increasing CpG beacon density this intragenic enrichment became stronger (see Fig. S4, Beacons ≥ 10, χ2 p < 2.20 × 10⁻¹⁶). These islands have been implicated in a number of other studies for their significant role in developmentally important isoforms.^5,69,70 Examination of repeat classes identified enrichment in the hominid-specific SVA subclass^71,72 (see Fig. S5, χ2 p < 2.20 × 10⁻¹⁶), which arose ~20 MYA and has been extremely prolific during evolution of the primate genome.^73,74

Correlation between CpG density and CGI hypomethylation

While specific genetic methylation-determining regions (MDRs⁷⁵) have been identified within CGIs, a correlation with CpG density and hypomethylation has also previously been recognized.¹⁰ Therefore, CpG beacon clusters will lead to species-specific CpG density increases which may be associated with increased CGI hypomethylation and formation of permissive chromatin.⁷⁶ A CpG density of ~20% CpGs (or 10% methylatable cytosines) was proposed by Eckhardt et al.¹⁹ as a threshold beyond which CGI are highly likely to be constitutively unmethylated across all differentiated tissues. Examining the available data from two bisulphite sequencing experiments from Li et al.⁷⁷ in peripheral blood cells and Lister et al.² from fibroblasts, we find methylation within CGIs is strongly correlated in these sets (r² = 0.84), despite being confounded by different experiments, tissues and cell line effects. Furthermore, the same significant trend of reduced methylation when CGIs were categorized into subgroups of all, ≥ 15%, ≥ 20%, ≥ 25% CpG density is seen using both the Ensembl CGI definition and an alternate CGI set by Wu et al. identified via hidden Markov models.⁷⁸ (Kruskal-Wallis p < 2.2 × 10⁻¹⁶) (Fig. 3; Fig. S6A‒C).

Figure 3. CpG density influence on methylation in Wu et al. CpG Islands from Li et al. bisulphite methylome data from peripheral blood mononuclear cell DNA (Kruskal-Wallis rank sum test p value < 2.2 × 10–16).^77,78

Next, we generated peripheral blood cell methylome data by MeDIP-seq of pooled samples from chimpanzee and rhesus macaque as well as pooled human samples. Examination of these data also supported the inverse correlation of CpG density with methylation in CGI across all three species in the Ensembl CGI set (see Fig. S7A, p < 2.2 × 10⁻¹⁶) and as well the Wu et al. CGIs that are proposed to have improved trans-species CGI prediction (Fig. S7B, p < 2.2 × 10⁻¹⁶).⁷⁸

We then investigated whether the influence on methylation change was still apparent in CpG density that changed over time in orthologous CGI between these species. We identified the orthologous CGI set between human and chimpanzee (Ensembl n = 13,999, Wu et al. n = 34,053), and human and macaque (Ensembl n = 4,654, Wu et al. n = 19,200) and chimpanzee and macaque (Ensembl n = 4,004, Wu et al. n = 18,747). For example, the orthologous DPP10 CpG beacons extreme cluster CGI, showed average methylation (RPM) of 0.51 and 4.80 in human and chimpanzee respectively, but not enough CpG density in macaque for a CGI to be defined even by the Wu et al. methodology. The subset of these orthologous islands that were ≥ 20% CpG density in one species and ≤ 19% in the other was then obtained. A significant difference was identified in the Ensembl set for human vs. chimpanzee CGI (Wilcoxon p = 1.954 × 10⁻¹²; data not shown) and in the larger Wu et al. set these groupings showed a small but significant reduction in methylation [expressed as average reads per million (RPM)] consistently in the higher density CGI group across all species comparisons (see Figure 4, all p Wilcoxon < 2.2 × 10⁻¹⁶).

Figure 4. Comparison of methylation in subset of orthologous CGI set from Wu et al.⁷⁸ The RPM in these islands was compared by subtraction. First Human[RPM] – Chimpanzee[RPM], in islands Human (Hs) > 20% CpG and Chimpanzee (Pt) < 19% CpG then Chimpanzee > 20% CpG and Human < 19% CpG. Then Human[RPM] – Macaque[RPM], Human (Hs) > 20% CpG and Macaque (Mm) < 19% CpG and Macaque > 20% CpG and Human < 19% CpG. Finally Chimpanzee [RPM] – Macaque [RPM] in islands Chimpanzee > 20% CpG and Macaque < 19% CpG then Macaque > 20% CpG and Chimpanzee < 19% CpG. All show a consistent pattern where the less CpG dense island (< 19%) is more methylated than the more CpG dense island (> 20%).

Therefore, we have shown that varying CpG density changes the methylation potential within CGIs. This change in likelihood of methylation between low and high density CGIs was found to occur across and between the three species. The most extreme human-specific genomic eruptions of CpGs occur in the identified “CpG beacon” clusters, which in turn have highlighted genes associated with human traits and disease.

Discussion

The proteome is similar across placental mammals; therefore, the creation of new protein coding genes is rare,⁷⁹ although not unknown.⁸⁰ Regulatory modification has long been proposed as critical in human acquired traits⁶⁶ and genomic data acquired in the last decade supports the hypothesis that species evolution is predominately via novel regulatory adaptation and subsequent altered gene expression.^79,81 Recently, the identification of human-specific loss of regulatory DNA revealed insights into human evolutionary divergence.⁸²

Epigenetic mechanisms, including DNA methylation, are critical in genome regulation and viable mammalian development requires the ability to methylate cytosines.⁸³ This chemical modification can lead to disparate effects depending on genomic location; repressive in CGI,¹⁸ activating in gene bodies⁴⁷ and splicing influence in CTCF binding sites.⁸⁴ Changes in developmental timing are significant in species-specific differences⁸⁵ and the epigenetic modulation of intragenic islands may direct developmentally critical isoforms.^5,86 Thus, this epigenomic extra layer of control enables additional axes to the adaptive landscape and aids in the evolution of complex phenotypes.^23,87 Cytosine methylation has also been suggested to be significant in karyotype evolution.⁸⁸ Even simply focusing on human higher cognitive functioning, notwithstanding all the other phenotypic differences, levels of brain tissue-specific imprinting,^89,90 distinctive neuronal DNA methylation profiles^91,92 and potential role in synaptic plasticity,⁹³ as well as pathogenic Methyl Binding Domain gene mutations in post-natal brain development disorders,^94,95 all postulate that the gain and loss of CpG may be fundamental in the human-specific phenotype.

We therefore identified a subset of species-specific CpGs by inter-primate comparison, impartial to mechanistic cause, which we have termed CpG beacons. Focusing initially on extreme human CpG beacon clusters, we showed they are enriched for neurological disease genes and, additionally, co-locate with the evolutionary accelerated HAR1A nc-RNA gene. A strong correlation between accelerated genomic loci and bias toward increased GC content was observed previously,³⁵ due to the effects of recombination.³⁰ Fine scale recombination hotspots show high diversity between human and chimpanzee, as they are short-lived relative to divergence times,^96,97 potentially strongly influenced by the variation in zinc finger binding of PRDM9.^98,99 GC-coupled CpG increase due to recombination has been suggested to have played a considerable role in CGI formation¹¹ and thus may be a strong driver in the formation of CpG beacon clusters and thus species-specific regulation.

However, on top of the localized strong effect of BGC on increased GC, multiple subtle substitutions have been shown to have a morphological evolution effect altering the timing and level of expression.¹⁰⁰ We looked for potential CpG-specific signatures by identifying where human-specific CpG exceeded human-specific GpC, defining Positive CpG Beacon Clusters, which identified potential human traits that may have arisen during human evolution.¹⁰¹

Recent comparative methylome analyses have revealed species-specific differences.^102,103 Molaro et al. examined chimpanzee and human sperm and supported the link between genome and epigenome, by identifying strong CpG decay correlated with methylation over brief evolutionary periods. They also found extensive species-specific methylation differences in SVA repeats, with significantly increased numbers of orthologous hypomethylated SVAs within humans. Interestingly, this is the same subtype in which we identified high enrichment of positive beacon clusters, which could be driving this hypomethylation. Additionally SVAs have recently been identified to be involved in active somatic retrotranspositon in human brain.¹⁰⁴

Lienart et al. have recently identified the existence of methylation-determining regions (MDRs) due to sequence characteristics within CGI, but also acknowledged the necessity of a critical CpG density within these regions.⁷⁵ Increasing CpG density forms low density CpG islands which are more likely to have variable methylation and hence to be tissue-specific or developmental time-dependent. Further density rise will create high density islands that eventually will become increasingly likely to become constitutively hypomethylated above a threshold identified earlier by Eckhardt et al.¹⁹ This correlation can be seen by re-examination of the Li et al.⁷⁷ and Lister et al.² bisulphite sequencing data and is supported across three species in this study via MeDIP-seq. Either MDR-induced CpG retention, BGC created CpGs, or potentially both can lead to clusters of CpGs resulting in hypomethylated islands that then recruit factors such as Cfp1. While this process does not lead to explicit expression per se, it establishes open permissive chromatin via H3K4me3 marked domains.^12,76 Therefore, CpG beacons have the capacity to move, or indicate the movement of, these loci along a continuum from no island to tissue-specific island to constitutively active island, as illustrated in the model shown in Figure 5.

Figure 5. Model for beacon-mediated CGI evolution and example for possible association with disease. The left panel illustrates a model for CpG beacon-mediated increase of CpG density. A moderate CpG beacon increase lead to the formation of low density CGIs which are predominantly methylated and prone to tissue-specific methylation. Further CpG beacon increase can eventually lead to high density CGIs with increased likelihood of becoming constitutively hypomethylated. Variation in such CGIs could then result in species-specific phenotypic changes as discussed in this study for CGIs containing extreme clusters of > 20 beacons per kb. CpG beacon gain may indicate reduced mutability due to acquisition of new methylation determining regions⁷⁵, direct gain due to accumulated substitution or biased gene conversion, or a combination of all of these processes. The right panel shows an example of one of the identified extreme CpG beacon cluster containing CGIs that is associated with the human ANKRD11 gene, implicated in Autism. In this case, the CpG density and methylation state of the orthologous CGIs in human, chimpanzee and macaque are concordant with the proposed model. Methylation values were determined by MeDIP-seq and are given as reads per million (RPM).

In conclusion, the CpG dinucleotide is vital for regulation and not only transmits genomic data but also enables epigenomic variation. Thus, genomic change in this dynamic dinucleotide required for DNA methylation, influencing CGI methylation, gene body methylation, imprinting and splicing, is fundamental to understanding our evolutionary acquired traits and vulnerabilities to disease.

Methods

CpG beacon identification

CpG loss is time- not replication-dependent; therefore, there are almost equal counts of CpG in human and chimpanzee.⁴³ Recent estimates from whole genome sequencing for mutation rate is ~1 × 10⁻⁸ per generation¹⁰⁵ with the CpG dinucleotide approximately ten times this. Due to the rapid turnover in regulatory elements,¹⁰⁶ most are too weakly conserved to mouse to distinguish⁴⁵ which will particularly be the case with the highly mutable CpGs. By utilizing the additive collective divergence of multiple primates^44,45 a CpG’s human-specific state was attributed. The comparatively inferior sequence quality of these individual non-human primate sequences was balanced by the combined multispecies comparison vs. human. Assignment of ancestral state by use of chimpanzee alone was found to have an error rate of 0.65% utilizing macaque as a second out-group.¹⁰⁷ Furthermore, most SNPs have been calculated to be < 1 million years of age, compared with the minimum divergence time of chimpanzee and human which is estimated at 5 million years.¹⁰⁸

Therefore, to identify the human-specific changes we utilized the Ensembl Compara.6_primates_EPO six primates alignment^51,52 build 58. This includes the species Homo sapiens, Pan troglodytes, Gorilla gorilla, Pongo abelii, Macaca mulatta and Callithrix jacchus. The builds included are human GRCh37, chimpanzee Mar. 2006 (CGSC 2.1/pan Trog2), gorilla (gorGor3); orang-utan Jul. 2007 (WUSTL version Pongo_albelii-2.0.2) ; macaque Jan 2006 (MGSC Merged 1.0/rheMac2) ; and marmoset Mar. 2009 (WUGSC 3.2 (GCA_000004665.1)) . The Ensembl Compara.6_primates_EPO alignment blocks was then reduced in a stepwise fashion due to the requirement of: unique human sequence, i.e. does not align to greater than one location in human genome (82.71% of human genome remaining); unique chimpanzee sequence (80.54%); and contains at least one other primate (79.99%) in order to utilize the strength of the inter-primate comparison.⁴⁵ Within this ~80% of the human genome that was able to be aligned, an algorithm was devised to identify the location of each human CpG site within these blocks and then compared with the corresponding bases in other species. To be identified as a potential human-specific CpG, the requirement in chimpanzee sequence was that it did not match CG and did not contain N or –. All other species sequences at this position also did not match CpG and the closest other primate (gorilla in 96.64% of sites) did not contain N or –. If this was the case then it was recorded as a human-specific CpG site, which led to a set of 1,820,319 CpGs. All human cluster locations are given in build Human GRCh37 coordinates. The chimpanzee beacons were calculated using the exact same methodological algorithm as the human, but instead changing the focused species to Pan troglodytes.

Polymorphic filter

Any CpG with evidence of polymorphism from 1,000 Genomes data for SNP, CNV and Indel was then removed from the set. The December 2010 Data update Full Project Genotype Release from calls on 629 individuals from the 20100804 sequence was used. Indel data was from the February 2011 update, which were calls from Dindel on the same 629 individuals from the 20100804 sequence and alignment, and also available CNV data from 179 unrelated individuals.¹⁰⁹ This resulted in a non-polymorphic human set of 1,192,484 CpGs.

CpG beacon density calculation and permutations

Initial density permutations were calculated for each CpG beacon by counting the number of CpG beacons within a region of 499bp downstream and 500 bp upstream. Random beacon sets of the same number 1,192,484 were generated from the total set of locations of CpGs in the h1c1o1 genome set of 20,207,732 and density calculated.

Positive CpG beacon clusters

Positive beacon clusters were calculated, via a sliding window of 1 kb and slide of 100 bases across the genome. The total non-polymorphic human-specific GpCs within the 1kb region was subtracted from the total CpG beacons within this region.

Gene set enrichment analyses

Ingenuity Pathway Analysis ©IPA was performed for gene set enrichment. The location of clusters was assigned to genes if it mapped to within 100k 5′ and 50k 3′ of the transcript. The following IPA analysis settings were used: Reference set: Ingenuity Knowledge Base (Genes Only), Relationship to include: Direct and Indirect, Includes Endogenous Chemicals. Optional Analyses: My Pathways My List. Filter Summary: Consider only molecules and/or relationships where (species = Human) AND (confidence = Experimentally Observed). Benjamini-Hochberg multiple test corrected p values are shown only. The region-based binomial analysis of GREAT analysis 2.0.1⁶⁵ was utilized to identify genome regional enrichments from the location of the extreme beacon clusters, as well as the moderate positive beacons clusters ≥ 5. This takes into consideration the bias of potential genomic space available compared with the traditional hypergeometric test. The parameters used were association by Basal plus extension with default values of proximal 5 kb upstream, 1 kb downstream, but a reduced distal limit to 100 kb from 1 Mb and significance assessed by the regional based Binomial test FDR Q value.

Publicly available bisulphite sequencing data

The Lister et al. fibroblast methylome data was downloaded from http://neomorph.salk.edu/human_methylome/data.html. The Li et al. PBMC methylome data was downloaded from the NCBI Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc = GSE17972). For the Lister et al. and Li et al. data sets individual CpG methylation was calculated by combining the reads from the two strands and subsequently requiring a 5 read minimum coverage for inclusion.

Comparative MeDIP-seq

DNA was extracted from peripheral blood of 5 chimpanzees (Pan troglodytes, 3 males, 2 females) and 5 rhesus macaques (Macaca mulatta, 3 males, 2 females). Samples were taken from captive individuals at Tierpark Nordhorn, Basel Zoo, Leipzig Zoo and at the German Primate Center during routine health checks and not specifically for this study. Microsatellite analysis conducted at the German Primate Center verified that respective individuals are not related. Sample collection adhered to the American Society of Primatologists (ASP) Principles for the Ethical Treatment of Non-Human Primates (www.asp.org/society/resolutions/EthicalTreatmentOfNonHumanPrimates.cfm).

To obtain averaged methylomes and reduce individual genotype effects, DNAs were pooled for each species at equal concentration for each individual. MeDIP was then executed according to Auto-MeDIP-seq protocol as described previously¹¹⁰ and sequenced on Illumina GAIIx. This was performed with paired end reads of 36 bp with average fragment sizes of: 197 bp in human, 222 bp in chimpanzee, and 217 bp in macaque. The corresponding methylome data are available from the authors on request. A comprehensive analysis of these methylomes will be described elsewhere (Wilson G.A. et al., manuscript in preparation).

MeDIP-seq data was processed using MeDUSA (Methylated DNA Utility for Sequence Analysis).¹¹¹ This computational pipeline performs a full analysis of MeDIP-seq data by utilizing a number of freely available software packages. Raw sequence data in fastq format were aligned to the reference genomes (Human GRCh37, panTro2 and rheMac2) using alignment software BWA (v0.5.8),¹¹² with default parameters, to generate a SAM format alignment file. Aligned reads were filtered using SAMtools (v0.1.9)¹¹³ to remove reads that failed to form a correctly aligned pair (forward and reverse templates). Further filtering based on mapping score was also performed (read pair must contain read with mapping quality ≥ 10). Potential PCR artifacts were removed by discarding all but one read within groups of non-unique reads (i.e., reads aligned to the exact same start and stop position on the same chromosome). FastQC (www.bioinformatics.bbsrc.ac.uk/projects/fastqc/) was used to determine sequence data was of acceptable quality and the Bioconductor (v2.7)¹¹⁴ package MEDIPS (v1.0.0)¹¹⁵ performed enrichment and coverage analyses. Reads per million (RPM) was calculated within regions as (reads/total reads) times 10⁶ for each species (total human reads = 40,797,356, chimpanzee = 32,910,189 and macaque = 24,933,164).

Genetic influence on the DNA methylome

The MEDIPS package was used to approximate absolute methylation scores from relative MeDIP results.¹¹⁵ This enables regional methylation to be compared over features, i.e., CpG Islands utilizing the appropriate genome sequence for each species. LiftOver¹¹⁶ was used to calculate orthologous CGI sets with overlap of at least 95% required. Greater than 20% and less than 19% orthologous sets were chosen from the orthologous island sets for each grouping with the following numbers: Hs ≥ 20 and Pt ≤ 19, Ensembl = 375, Wu = 557; Pt ≥ 20 and Hs ≤ 19, Ensembl = 150, Wu = 614; Hs ≥ 20 and Mm ≤ 19, Ensembl = 248, Wu = 605; Mm ≥ 20 and Hs ≤ 19 Ensembl = 62, Wu = 1530; Pt ≥ 20 and Mm ≤ 19, Ensembl = 256, Wu = 700; Mm ≥ 20 and Pt ≤ 19 Ensembl = 118, Wu = 1451 (Hs = Homo sapiens, Pt = Pan troglodytes and Mm = Macaca mulata).

Statistical analysis

Statistical calculations were performed in the R statistical environment.¹¹⁷ Empirical p values were calculated as the excess of simulation vs. observed values. Kruskal-Wallis rank sum test was used to compare methylation in CGI density sets and Wilcoxon test for comparison between average RPM values for orthologous island sets. Chi-square calculations for enrichments were performed for PBC by bases covered of PBC vs. total bases of category and CpG beacon vs. total CpGs.

Glossary

Abbreviations:

BGC

biased gene conversion

CGI

CpG islands

MeDIP

methylated DNA immunoprecipitation

PBC

positive beacon clusters

PBMC

peripheral blood mononuclear cells

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Author Contributions

Conceived and designed the experiments: C.G.B., G.A.W. and .SB. Performed the experiments: C.G.B., L.M.B. and G.A.W. Analyzed the data: C.G.B. and G.A.W. Contributed reagents/materials/analysis tools: C.R. and L.W. Wrote the paper: C.G.B. and S.B.

Supplemental Material

Supplemental materials may be found here: www.landesbioscience.com/journals/epigenetics/article/22127