Introduction

DNA methylation is an important epigenetic mark involved in the regulation of gene expression. In mammals, it is mainly present in the context of 5-methylcytosine (5mC) followed by a guanine (CpG dinucleotide) [1]. Regions containing a high degree of CpG dinucleotides are called CpG islands (CGIs). They are often located in the promoter region of genes and are mostly unmethylated [2]. In their methylated state or by the formation of transcriptionally silent heterochromatin they can repress gene transcription [1]. Special cases of DNA methylation are X-inactivation and genomic imprinting. Imprinted loci show a monoallelic, parent-of-origin specific expression most often regulated by a differentially methylated region (DMR) [3]. So far, about 80-90 genes are known to be imprinted in humans, some of them only in specific tissues or at certain times during development [4,5]. In the last years several studies aimed to identify novel imprinted genes in mice and human using different approaches. Luedi and colleagues [6,7] applied a computational method based on DNA sequence characteristics of known imprinted genes for prediction in mice and human. With the development of new, large-scale array and deep sequencing technologies, genome-wide expression and methylation analyses were possible (for review see 8,9). For instance, genotyping microarrays were used with genomic and cDNA to detect differences in allele expression at multiple loci indicative of monoallelic expression and imprinting [10]. Nakabayashi and colleagues [11] combined Illumina Infinium HumanMethylation27 BeadChips (27k arrays) with the use of mosaic genome-wide uniparental disomy samples to identify new imprinted genes for analyses. However, among all predicted novel yet unknown imprinted genes only very few could be shown to be imprinted so far.

Methylation analyses of biparental hydatidiform moles (OMIM #231090) caused by mutations of the maternal effect gene NLRP7 revealed aberrant methylation at multiple imprinted loci, indicating that maternal NLRP7 mutations affect establishment and/or maintenance of methylation imprints [12,13]. Homozygous and compound heterozygous mutations in this gene have been identified as a major cause of recurrent biparental hydatidiform mole pregnancies, but also stillbirths and spontaneous abortions were frequently reported, while live births are extremely rare ([14-19] and others). Heterozygous mutation carriers were also described, but reports on reproductive outcomes vary [18,20-22].

Recently, Caliebe et al. (in preparation) identified a family, in which two fetuses and one child of a mother with a heterozygous NLRP7 mutation (p.A719V) showed altered DNA methylation patterns at many imprinted loci. This family was instrumental in identifying RB1 as being imprinted [23]. In order to identify novel imprinted genes, we analysed additional candidate genes by deep bisulfite amplicon sequencing. With this technique up to several thousand clonally amplified single sequence reads per sample can be obtained, thus enabling a highly quantitative methylation analysis, which makes it possible to detect even small differences in methylation levels as for example seen in mosaic imprinting defects.

Results and Discussion

We have studied the only affected live-born individual from a family with multiple imprinting defects. To determine the extent of aberrant DNA methylation in the patient and to identify novel imprinted genes, we performed high-throughput methylation profiling of blood DNA from the patient, his parents and 18 normal controls on Illumina Infinium HumanMethylation27 BeadChips (27k arrays), which interrogate 27,578 CpG sites. Based on the raw data (GEO accession number GSE47879) we used a pseudoproband (permutation) approach to identify aberrantly methylated CpGs on autosomal chromosomes (loci on the X and Y chromosomes were excluded to avoid possible confounding effects of CpG methylation associated with X-inactivation). For this we successively assigned the patient and each normal control as a proband and determined the number of CpGs showing an aberrant methylation in comparison to the mean methylation levels of all other individuals. As a threshold for aberrant methylation we used delta β ≥ 0.3, where β is the ratio of methylated signal in relation to the overall signal of methylated and unmethylated with a scale from 0.0 as unmethylated to 1.0 as fully methylated [24]. CpGs passing the threshold (n=131; 87 CpGs associated with genes not known to be imprinted, 44 CpGs associated with known imprinted genes) are listed in Table S1. To characterize the methylation defect at genes of unknown imprinting status, all genes known to be imprinted at the time of analysis (February 2011; based on http://www.geneimprint.com and http://igc.otago.ac.nz) were excluded from further analysis. The 18 normal controls showed an average of about 15 CpGs associated with 15 genes with aberrant methylation (standard deviation (SD) 4.84). A similar number was found in the patient’s parents (mother: 15, father: 11). In contrast, 87 CpGs associated with 85 genes were observed in the patient (Figure 1), which is more than 14 times the SD (+14.9 SDs). The empirical p-value is 0.05 and the upper limit of the 95 % confidence interval is 0.176 (see Material and Methods).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Results of the pseudoproband approach.

The diagram shows the number of CpGs in all investigated individuals passing the threshold of delta β ≥ 0.3 when assigned as a pseudoproband. All imprinted genes were excluded.

a - t – healthy controls; M – mother of the patient; F – father of the patient; average – average number of CpGs showing a deviation in normal controls and the parents.

https://doi.org/10.1371/journal.pone.0076953.g001

For half of the aberrantly methylated loci (46 CpGs) the patient showed a hypomethylation as compared to the normal controls, whereas the other half were hypermethylated, supporting the hypothesis of a multilocus methylation defect rather than a multilocus hypomethylation. Out of the 85 candidate genes with unknown imprinting status, 19 were selected for in-depth methylation analyses by deep bisulfite sequencing (Table 1). We selected 16 hypomethylated and three hypermethylated genes. Eleven genes had an average methylation level (β-value) in the normal controls between or close to ~0.4 and 0.6 suggestive of a possible allele-specific methylation (Table 2). Other genes showed a high degree of hypo- or hypermethylation in the patient compared to the normal controls. Only loci were chosen where the CpG analysed on the 27k array could be investigated together with a neighbouring single nucleotide polymorphism (SNP), so that each amplicon included the CpG analysed on the array and a nearby SNP. Furthermore, at least two normal controls not investigated on the array had to be informative for these variants (Table S2) to enable the differentiation of the two parental alleles with the exception of HKR1, which was chosen as an example for high inter-individual variability. For all genes a literature research concerning previous methylation analyses was conducted at the time of analysis. A gene set enrichment analysis of the 85 candidate genes using GeneTrail (http://genetrail.bioinf.uni-sb.de/enrichment_analysis.php?js=1&cc=1) showed no significant enrichment for pathways or gene ontology categories after Bonferroni correction (data not shown).

Among the 19 candidate genes, only FAM50B had been proposed before to be possibly imprinted by Luedi et al. [6]. FAM50B is a retrotransposed gene located on human chromosome 6p25 and was recently reported to be imprinted [11,25]. They showed that FAM50B is subject to genomic imprinting in various tissues being methylated on the maternal allele and exhibiting expression from the paternal allele only. We studied the methylation status of FAM50B inside a CpG island in the promoter region in peripheral blood of two normal controls (Table 1). The detected degree of methylation was about 50 % confirming the 27k array data (average β in normal controls 0.51; Table 2). Separation of the two parental alleles showed an unmethylated and a methylated allele thus confirming allele-specific methylation (Table 3; Figure 2a; for SNP rs number see Table S2). Furthermore, expression analyses in blood displayed monoallelic expression indicating that FAM50B is subject to genomic imprinting in blood (Figure 2b), too.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Results of methylation and expression analyses of FAM50B in blood.

a) Methylation analysis.

The figure shows the results of the methylation analyses by deep bisulfite sequencing of two normal controls (NC 4 and NC 15) and the patient (top). Heterozygosity for a SNP in both normal controls allowed to separate the alleles (bottom).

Mean methylation levels, SNP allele and number of reads are given below each pattern. Lines represent reads; columns represent CpG dinucleotides; blue squares – unmethylated CpGs; red squares – methylated CpGs; white squares – missing sequence information; * – CpG investigated on the 27k array (CpG 17).

b) Expression analysis.

Expression analysis in four normal controls heterozygous for a SNP (rs6597007 C/G). For each normal control (NC) sequences obtained from peripheral blood DNA and RNA are shown. In the RNA only one allele is present, which indicates monoallelic expression.

https://doi.org/10.1371/journal.pone.0076953.g002

By deep bisulfite sequencing, the patient had a methylation level of about 1.4 %, thus confirming the 27k array data (β = 0.20; Figure 2a) and showing an even more pronounced hypomethylation compared to the array. Separation of alleles was not possible due to the lack of an informative variant. The identification of this second imprinted gene - FAM50B - after the previously described RB1 gene [23] emphasises the potential of this analysis to discover new imprinted genes.

Methylation analyses of TRPC3 (Transient receptor potential channel subfamily C, member 3) were performed inside CpG island 55 located in the promoter region of transcript 2/isoform b (UCSC, hg18 uc003ief.1; NM_003305). The 27k array data showed a hypomethylation in the patient (β = 0.23), which we confirmed by deep bisulfite sequencing showing an even more pronounced hypomethylation (8 %). The methylation patterns of eight normal controls showed a high inter-individual variability in the degree of overall methylation ranging from 47 to 72 %, which again confirms the range observed in the 18 normal controls and the patient‘s parents on the 27k array (β = 0.41 - 0.78; Figure 3a, Figure S1 and Table 2). After allele separation in five normal controls heterozygous for a present SNP (rs13121031), the methylation observed showed a variable degree across both alleles (Figure 3a). Interestingly, in one individual (NC 12), a nearly allele-specific methylation pattern was present, showing methylation levels of 16 % and 92 % for the separated alleles, respectively, highlighting the high variability of the methylation (see Figure 3a). Investigation of three individuals homozygous for one allele showed degrees of methylation of 55 % and 59 % (C allele) and 69 % (G allele), respectively, which is in the range shown by the heterozygous individuals.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Results of methylation and expression analyses of TRPC3 in blood.

a) Methylation patterns of TRPC3.

Results of the methylation analyses of 24 CpG dinucleotides obtained by deep bisulfite sequencing. Peripheral blood samples of eight normal controls (NC) and the patient were investigated. The present SNP at CpG 6 indicates the distribution of methylation across the two alleles. As it disrupts the CpG site 6, the white squares refer to the C allele, while filled squares refer to the G allele.

Mean methylation levels, SNP allele and number of reads are given below each pattern. Lines represent reads; columns represent CpG dinucleotides; blue squares – unmethylated CpGs; red squares – methylated CpGs; white squares – missing sequence information, at CpG 6 due to a SNP (rs13121031) that disrupts the CpG dinucleotide; * – CpG investigated on the 27k array (CpG 21).

b) Expression patterns of TRPC3 in blood.

Expression analyses in three normal controls heterozygous for a SNP (rs11732666 G/A). For each normal control (NC) sequences obtained from peripheral blood DNA and RNA are shown. Expression varies from biallelic to skewed.

https://doi.org/10.1371/journal.pone.0076953.g003

TRPC3 was also identified as a possible candidate gene for imprinting in a screen by Nakabayashi et al. [11]. Further analyses conducted by Martin-Trujillo et al. [26] via bisulfite PCR and cloning in four different tissues including leukocytes showed that the methylation seemed to be influenced by the genotype of the present SNP rs13121031 and that the major allele showed high methylation levels while the minor allele did not show methylation. We could not detect such an influence of the genotype of rs13121031 on the methylation by deep bisulfite sequencing in peripheral blood samples of eight homo- and heterozygous normal controls. In contrast to the previous report we investigated the methylation on the antisense strand from which TRPC3 is transcribed. Therefore we cannot rule out a strand-specific difference in methylation although such reports were rare [27,28].

Methylation of the CpG Island 160 located in the promoter region of transcript a/isoform 1 (UCSC hg18, uc003ieg.1; NM_001130698) was investigated by Sanger sequencing of bisulfite treated DNA and showed only unmethylated CpG sites (data not shown) corresponding to results reported before [26].

Furthermore, expression analyses of TRPC3 were conducted in RNA from peripheral blood of six informative normal controls for both isoforms together utilizing a SNP (rs11732666 in exon 8 of transcript 2/isoform b (UCSC hg18, uc003ief.1; NM_003305) and exon 9 of transcript 1/isoform a, respectively (UCSC hg18, uc003ieg.1; NM_001130698; Table S2). The expression patterns observed varied from biallelic to skewed expression in favour of one allele (Figure 3b). These results correspond to the findings of a variable methylation across both alleles and to the results by Martin-Trujillo et al. [26], where the expression in one fetal brain sample was investigated using the SNP rs13121031, which showed a skewed expression.

As the array-based methylation data indicated that the patient has a more general methylation defect (Figure 1; Table S1) we additionally investigated three genes at which the patient showed a hypermethylation compared to the normal controls. One of these genes is AKR1C3 (aldo-keto reductase family 1, member C3), where the patient showed a hypermethylation (β = 0.75) compared to the normal controls with an average β of 0.10 (range 0.07 - 0.18) in the 27k data By deep bisulfite sequencing a methylation level of 71 % was observed in the patient confirming the strong hypermethylation. In blood of two investigated normal controls a low level of methylation of 2 % and 7 % respectively was observed (Figure 4; Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Methylation patterns of AKR1C3.

For this locus the patient shows hypermethylation, whereas the two normal controls (NC) display very low levels of methylation.

Mean methylation levels and number of reads are given below each pattern. Lines represent reads; columns represent CpG dinucleotides; blue squares – unmethylated CpGs; red squares – methylated CpGs; white squares – missing sequence information; * – CpG investigated on the 27k array (CpG 7).

https://doi.org/10.1371/journal.pone.0076953.g004

The extensive methylation analyses conducted here also revealed and highlighted some interesting methylation patterns. One of these genes was ACTN3 (Actinin alpha 3), which is located on chromosome 11q13.2. At this locus, a hypomethylation (β = 0.32) was detected in the patient compared to the normal controls (average β-value = 0.72) on the 27k array. The methylation values were verified by deep bisulfite sequencing, which showed 39 % methylation in the patient, whereas two normal controls showed methylation levels of 70 % and 74 % in blood, respectively. No significant differences in the methylation level or distribution across the alleles were present thus no allele-specificity. Interestingly, the degree of methylation between adjacent CpGs within the investigated area showed large differences. Whereas the CpGs 1, 3 and 5 of the amplicon exhibited methylation levels of about 80 - 90 %, the two CpGs in between (CpG 2 and 4) showed levels of only about 40 - 50 % (Figure 5 and Table 3). Such differences may be rare (as this was in one out of 19 genes) but should be borne in mind for analyses based on single CpGs only.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Methylation patterns of ACTN3.

The two normal controls investigated display a high variability in methylation levels of adjacent CpGs. The same pattern is visible in the patient in a hypomethylated form.

Mean methylation levels, SNP allele and number of reads are given below each pattern. Lines represent reads; columns represent CpG dinucleotides; blue squares – unmethylated CpGs; red squares – methylated CpGs; white squares – missing sequence information; * – CpG investigated on the 27k array (CpG 3).

https://doi.org/10.1371/journal.pone.0076953.g005

Another interesting example is HKR1 (Krueppel-related zinc finger protein 1) on chromosome 19q13.13. Here the array data showed a broad range of methylation in the normal controls between β 0.23 and 0.67 while the patient showed a nearly complete loss of methylation with β of only 0.02. Analyses performed on the Roche/454 Genome Sequencer confirmed the unmethylated state in the patient, obtaining nearly exclusively unmethylated reads, while residual methylation was present at a few single CpGs only (Figure 6). Due to the variability in methylation seen on the array, three normal controls were investigated showing methylation levels of 9.9, 25.6 and 38.7 % respectively. Allele-separation could only be performed in one of these normal controls (NC 3) due to the absence of an informative variant in the others (Table S2). The methylation was evenly distributed across both alleles and thus no allele-specific pattern was present (Figure 6 and Table 3). These results demonstrate that there can be large inter-individual differences [29-31].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Methylation patterns of HKR1.

Methylation patterns of three normal controls in blood displaying inter-individual variability. The methylation level in the patient is very low. In NC 3 the alleles could be separated (see lower part of the figure).

Mean methylation levels, SNP allele and number of reads are given below each pattern. Lines represent reads; columns represent CpG dinucleotides; blue squares – unmethylated CpGs; red squares – methylated CpGs; white squares – missing sequence information; * – CpG investigated on the 27k array (CpG 5).

https://doi.org/10.1371/journal.pone.0076953.g006

For the candidate genes investigated in more detail by deep bisulfite sequencing we could not detect allele-specific methylation or imprinting except for the previously described FAM50B [11,25]. Analyses after allele-separation did not reveal significant differences in level or distribution of methylation between the two parental alleles. This is also the case in the four genes CHP2, KCNAB3, MAMDC2 and TSPO (Figure S2), where allele-separation was possible in the patient. This emphasises that the methylation defect in the patient affects both parental alleles. Previous reports suggested that mutations in NLRP7 lead to a failure in the establishment of the maternal imprint in the female germline or in the postzygotic maintenance, as the investigated maternally methylated imprinted genes showed - if affected - a hypomethylation while the investigated paternally imprinted genes exhibited a hypermethylation [12,13,20,32,33]. Our and previous data ([23]; Caliebe et al. in preparation) suggest that not only the establishment and maintenance of the maternal methylation patterns can be affected, but also the methylation at imprinted and non-imprinted loci on the paternal allele . The results obtained for four genes (CHP2, KCNAB3, MAMDC2 and TSPO) where we could separate the alleles in the patient showed hypomethylation of both parental alleles. This supports the hypothesis that methylation maintenance is affected and that this failure in methylation maintenance also affects non-imprinted loci.

Most of the genes investigated (11 out of 19) showed an average level of methylation between 40 - 60 % in normal controls, as would be the case for an imprinted locus. We could verify allele-specific DNA methylation and expression of FAM50B in blood, but could not detect any other imprinted gene by this approach.

The investigated regions were located within CpG islands in the majority of genes (11 out of 19 based on UCSC, hg18). Although most CGIs in promoter regions are unmethylated to allow transcription, alterations have frequently been reported in cancer [1,2,34]. The genes investigated here in more detail display a broad variety of different methylation levels within CGIs and in the 5' regions of genes and may thus influence transcription and contribute to phenotypic variance as has recently been discussed [35].

Only TSPO and AKR1C3 showed very low methylation levels in the normal controls, which is in accordance with the frequently observed unmethylated state of CGIs in promoter regions. Although no CGI is present in AKR1C3 (according to the parameters used by UCSC hg18 for CGI detection [36]) the investigated region overlaps with an alternative transcription start site of known isoforms (UCSC).

As methylation can be influenced by or even depend on different cis- and trans-acting factors such as a present genotype (for review see 37) we analysed our data regarding the influence of the investigated SNPs. Evidence for a genotype dependent methylation based on the SNPs present in the investigated amplicons could not be observed (Table 3). Additionally, the normal controls investigated alongside the patient have been chosen to represent both sexes when applicable to be able to detect possible sex-specific differences in methylation as have been described before [38,39]. At the loci investigated such an effect was not present neither before nor after separation of the two alleles.

The patient displayed moderate to severe hypo- and also hypermethylation at the investigated loci with differences in the degree of methylation ranging from about 10 - 60 % as compared to the matching normal controls. Why different loci are affected to a varying extend by methylation changes needs to be elucidated. A stochastic effect or underlying sequence differences including binding sites may be possible contributors. Similar effects have been observed e.g. for mutations in ZFP57 or TRIM28 [40-42].

In summary, our data show that some loci have considerable variation in DNA methylation and that the patient has aberrant methylation at imprinted and apparently non-imprinted loci.

Materials and Methods

Supporting Information

Acknowledgments

We thank the family for participation in this study, Susanne Bens and Deniz Kanber for helpful discussion, Sabine Kaya and Stephanie Groß for expert technical assistance, Andre Scherag for statistical analyses and Peter Horn and Marianne Holtmann for providing samples from blood donors.