Introduction

It is generally accepted that DNA methylation almost exclusively occurs in CpG dinucleotides in mammals [1–4]. Non-CpG methylation has been documented, but with limited extent (see Discussion) and in specific cell types—mainly stem cells. Using the bisulfite modification followed by PCR amplification, cloning and sequencing, we previously reported unexpectedly high non-CpG methylation in myogenin mouse promoter [5]. Moreover, in human PSEN1 promoter, we observed discordant methylation patterns when using PCR primers designed using the MethPrimer software [6] (assuming that non-CpG cytosines are modified by bisulfite; defined as “MethPrimers” from now on) or primers designed to be insensitive to cytosine-methylation status. When the bisulfite technique has been originally described [7, 8], it was recommended not to include cytosines in the recognition sequence of the primers to avoid possible mismatches depending on methylation status. Interestingly, the same authors were able to evidence some non-CpG methylation in the first applications of the technique [9]. However, in the following years, MethPrimers became the most (if not exclusively) used primers in bisulfite-based applications, due to the advantage of the software-assisted primer design and to the general assumption that non-CpG cytosines were mainly unmethylated. Although the lack of a specific name makes it difficult to retrieve in PubMed the number of papers in which they are used, looking at the citations of the original article describing the MethPrimer software [6], it is possible to infer at least one thousand citations. Furthermore, an on line search through Google Scholar evidences about 29300 articles in which the bisulfite approach is used; among these, about 80% reports the use of MethPrimer software or similar primer design strategy.

Despite this trend, we have always been using primers designed in regions without cytosines or, when this was not possible, primers with degenerated bases (G/A) to cope with the uncertain C/U conversion of the few (max. 3) cytosines residues included in the sequence of the primer [5,10]. These primers will be here defined as “methylation-insensitive primers” (MIPs). The high non-CpG methylation observed for myogenin and the discordant methylation profile observed when PSEN1 was analyzed using either MethPrimers or MIPs lead us to hypothesize that MethPrimers could negatively select non-CpG methylated DNA molecules also resulting in a biased outcome of the CpG methylation assessment.

In order to verify this hypothesis, we analyzed two genes (myogenin and PSEN1), each one in two experimental models: CD1 mouse embryos and C2C12 myoblast cells for myogenin and human brains and neuroblastoma SK-N-BE cells for PSEN1. In each model, we compared conditions with differential methylation profiles, as ascertained in previous studies [5, 10].

Results and Discussion

Samples were bisulfite-modified and then amplified by two primer sets, MethPrimers or MIPs, recognizing the same promoter region (S1 Fig.); each sample was divided in two aliquots after bisulfite conversion for amplification with the two primer sets and then processed in parallel until the final sequencing step. Several positive and negative controls were performed to avoid any possible technical bias (see Methods section).

Fig. 1 shows the methylation pattern of the 9 CpG sites investigated in myogenin promoter in C2C12 cells (Fig. 1A and 1B) and in mice tissues (Fig. 1C and 1D). When myogenin methylation is high as in cells grown in 10% FCS and in mouse embryonic brain, MethPrimers significantly underestimate DNA methylation levels. As a matter of fact, whereas Mann-Witney test (used to evidence differences between two samples analyzed with the same primers) results in a significant difference for all the cytosine moieties when comparing the hypermethylated (Fig. 1A and 1C) vs. hypomethylated (Fig. 1B and 1D) experimental condition using MIPs [Cells: U = 9.00, p<0.05; Tissues: U = 16.00, p<0.02], the same comparison using MethPrimers indicates non-significant differences for 4 cytosines (1145, 1266, 1339, 1342). Moreover, Wilcoxon test (used to evidence differences when comparing MIPs vs. MethPrimers assay in the same sample) indicates that 5 cytosines result significantly hypomethylated when analyzed with MethPrimers [cytosines 1145, 1266, 1350, 1355, 1368: Z = -1.6, p<0.05] in C2C12 myoblasts. This difference was much more evident in brain and muscle tissues: all the cytosines (except the 1149) resulted significantly hypomethylated when the methylation assay is performed using MethPrimers vs. MIPs in the most methylated condition [Z = -2.5, p<0.02]. It is therefore evident that in experimental conditions in which DNA methylation is high MethPrimers show low power of detection compared to MIPs, eventually failing to evidence differences between differential methylation in two samples.

Download:

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

Fig 1. MIPs and MethPrimers result in different CpG methylation patterns.

CpG methylation pattern is expressed as percent methylation for each CpG in the investigated region of the mouse myogenin (a-d) and human PSEN1 (e-h) promoters. Light grey columns show the values obtained using MIPs, dark grey columns show the values obtained using MethPrimers. Time points for cell cultures are: 48 h for C2C12 10% FCS, SK-N-BE+SAM, SK-N-BE B def.; 24 h for C2C12 1%FCS. Symbols: * p<0.05 MIPs vs. MethPrimers; &: p<0.05 High Methylated vs. Low Methylated with MIPs; §: p<0.05 High Methylated vs. Low Methylated with MethPrimers. Y axes in histograms have a different scale (up to 100% for high methylated samples, up to 50% for low methylated samples) in order to better evidence intra-sequence differences in cytosine methylation analyzed with Methprimers vs. MIPs.

https://doi.org/10.1371/journal.pone.0118318.g001

Very similar results were obtained when studying DNA methylation pattern of PSEN1 in SK-N-BE cells grown in hypermethylating (S-adenosylmethionine supplemented) or hypomethylating (B vitamin deficiency) conditions [11] (Fig. 1E and F) and in frontal cortex samples from control subjects and patients with Alzheimer’s disease (AD) and control subjects (Fig. 1G and H). Mann-Withney test resulted in 10 out of 24 cytosines moieties significantly hypomethylated in PSEN1 promoter of low methylated (Fig. 1F) vs. high methylated (Fig. 1E) SK-N-BE cells when MIPs were used [cytosines 974, 1019, 1035, 1141, 1154, 1172, 1191, 1217, 1224, 1226: U = 9.00, p<0.05]. On the contrary, only 1 cytosine resulted significantly hypomethylated in the same samples when MethPrimers were used [cytosines 1075: U = 9.00; p<0.05]. This result was, also for PSEN1, more evident in tissues: all the cytosines analyzed were significantly hypomethylated in AD brains vs. controls [U = 16.00, p<0.05] whereas only 5 cytosines resulted hypomethylated when studied using MethPrimers [cytosines 1028, 1145, 1154, 1191, 1226: U = 16.00, p<0.05]. The inter-assay variation was confirmed by Wilcoxon test also for PSEN1; as a matter of fact, all the cytosines resulted significantly hypomethylated when the assay was performed using MethPrimers vs. MIPs [SK-N-BE cells: Z = -2.52, p<0.05; brain: Z = -2.52, p<0.02]. Therefore, assessing methylation with MethPrimers can result in underestimation of the high methylation levels eventually biasing the detection of differences.

The Sanger sequencing after bisulfite modification allows detecting the methylation level of any individual cytosine in the whole amplified region, including non-CpG cytosines [7, 8]. As previously observed for myogenin promoter [5] (S2 Fig.), when sequencing PCR products amplified using MIPs we were able at evidencing discrete non-CpG methylation also in PSEN1 promoter (S3 Fig.) at least in the conditions of high methylation [myogenin: Z = -2.91, p<0.01; PSEN1: Z = -3.13, p<0.01]. It is worth of note that, whereas MIPs allow discriminating between conditions with high/discrete and low/absent non-CpG methylation, MethPrimers invariably fail to evidence it (S2 and S3 Figs.). This observation raises two considerations: i) from the technical point of view it is evident that the described underestimation of DNA methylation when MethPrimers are used is due to the inability of these primers to bind non-CpG-methylated sequences; ii) from a perspective point of view, it can be inferred that the use of these primers has been causing general underestimation of the non-CpG methylation.

In order to confirm these observations by an independent approach, we also compared, by real-time PCR on bisulphite-modified genomic DNA, the amplification efficiency of each primer set respect to differential methylation statuses (Fig. 2). Fig. 2A and 2B show the overall non-CpG methylation levels measured, respectively, for myogenin and PSEN1 promoter, in different experimental conditions, ordered according to increasing methylation. Fig. 2C and 2D demonstrate that whereas myogenin and PSEN1 MIPs correctly amplify with similar efficiency all bisulfite-modified DNA samples independently on their original non-CpG methylation status (light gray columns), MethPrimers show significant inverse correlation [myogenin: r = 0.94, p<0.001; PSEN1: r = 0.89, p<0.001] between amplification efficiency and DNA methylation level. This result indicates that the more DNA is non-CpG methylated, the less MethPrimers are able to bind to the bisulfite-modified product indicating that these primers negatively select the high methylation (poorly modified) DNA fraction in the sample. As expected, MethPrimers failed to amplify untreated PCR products whereas MIPs show similar amplification efficiency for both treated and untreated PCR fragments.

Download:

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

Fig 2. MethPrimers amplification efficiency is affected by DNA methylation.

a) Myogenin and b) PSEN1 overall non-CpG methylation. c) and d) Amplification efficiency of primers on samples with differential overall non-CpG methylation was assessed by Real-Time PCR assay using MIPs (light grey columns) or MethPrimers (dark grey columns) for myogenin (c) and PSEN1 (d). Bisulphite-treated PCR products and untreated PCR products were used as control of amplification efficiency. Values are expressed as fold n° versus control (C2C12 in 1% 24 hours for myogenin, SK-N-BE in complete medium for PSEN1 used as respective calibrators); myogenin amplification values were normalized using PSEN1 PCR product amplification (added to mouse samples and used as exogenous reference) whereas PSEN1 amplification values were normalized to myogenin PCR product amplification (in this case added to human samples and used as exogenous reference). *: p<0.001 vs. ctrl.

https://doi.org/10.1371/journal.pone.0118318.g002

Finally, we used a third independent experimental approach to demonstrate that non-CpG methylation was present in the samples we analyzed. To this end, DNA samples with differential methylation levels were digested by non-CpG methylation-sensitive and methylation-insensitive endonucleases. Endonuclease assays for myogenin were already published [5], whereas the assay on PSEN1 is shown in S4 Fig. The results also indicate that the cutting efficiency is impaired in DNA samples with high methylation.

These experiments show that MethPrimers, commonly used in the PCR step of several bisulfite-based assays for the study of DNA methylation, can give false negative results (no methylation differences) due to their selective binding to the DNA fraction with lower methylation. This problem particularly affects highly methylated samples whereas samples with low methylation are unaffected; the use of MIPs allows bypassing this deficiency. We do not affirm that all the data so far obtained using MethPrimers are biased, but suggest being more cautious when “no differences” in methylation patterns are found. It is likely that a number of (mainly negative) findings obtained with MethPrimers could benefit from a verification using MIPs. These data were obtained for two independent genes, in two different organisms, each one in two experimental models; this approach appears sufficient to sustain the disclosure of the technical bias. A genome-wide confrontation between MIPs and MethPrimers appears beyond the scope of this technical report and could be probably not resolutive, since we demonstrated that the bias is evident only in conditions of high DNA methylation. However, the comparison of the epigenome as evaluated by the two kinds of primers would be a natural prosecution of the experiments here reported.

Finally, although the aim of this paper is not to evidence discrete non-CpG methylation in differentiated cells, we cannot ignore that our data strongly point out the idea that non-CpG methylation extent and role are probably underestimated. Indeed, recent evidences using unbiased techniques like Whole Genome Bisulphite Sequencing (WGBS) stress a functional role for non-CpG methylation [12–15]. Noteworthy, non-CpG methylation was demonstrated not just at genomic but at specific gene level [16–18] and particularly in brain [19, 20]. We therefore suggest that unbiased approaches, not based on the use of MethPrimers, are used to further investigate the role of non-CpG methylation in gene-specific and genome-wide analyses. As CpG and non-CpG DNA methylation patterns are largely variable depending on the species, developmental stage, tissue and cell type, a variable extent of the bias is expected. The more the methylome is methylated, the greater is the extent of the bias, with a prevalent effect of non-CpG methylation.

DNA methylation patterns and quantitative levels of methylation are largely variable in different tissues and cell types, as well as during development and differentiation. Several experimental evidences have been published on the fact that, despite similar levels of CpG methylation, mouse embryonic stem cells (ESCs) have considerably higher levels of non-CpG methylation (from 15 to 20% of total cytosine methylation), depending to the ability of DNMT3a, DNMT3b and DNMT3L to catalyze and regulate it [1, 14]. Even though these “first generation” studies were based on the analysis of a small fraction of the genome, they have been fully confirmed by subsequent “next generation” studies.

Genome-wide single-base-resolution maps of cytosine methylation in humans demonstrated that in ESCs, as well as in induced pluripotent stem cells (IPSCs), the 12–25% of the cytosine methylation is present in non-CpG context, is functionally linked to expression and is more prevalent in gene bodies than in protein binding sites and enhancers [15, 21, 22].

Several papers have demonstrated that from the 25% to the 35% of total DNA methylation in adult mouse and human brain resides in non-CpG sites, as measured by base-resolution analyses of respective methylomes [19, 23–26]. These researches also demonstrated that this kind of methylation is usually established de novo during neuronal maturation, conserved and correlated with gene expression. It has also been demonstrated that a highly conserved non-CpG methylation accumulates in human neurons during fetal to young adult development, to even become, at some developmental stages, the dominant form of methylation (53%) of the human neuronal genome [25].

Subsequent studies compared DNA methylation across diverse human cell line and tissue. It is worth of note that, after ESCs and brain, the tissue with the highest level of non-CpG methylation appeared to be the skeletal muscle [27]. However, quantitative data about the ratio between CpG and non-CpG residues in skeletal muscle methylome are poor and affected by the bias introduced from the study of a reduced number of non-CpG and/or by the use of restriction digestion-based approaches, such as RRBS [15] or LUMA [12].

From methodological point of view, bisulfite treatment is, at moment, an unavoidable step for studying DNA methylation at single-C level. If the subsequent experimental steps are based on strand-specific PCR performed by methylation-specific primers, the approach is potentially biased by non-CpG methylation. It depends on the primer design. Among recent approaches and analytical platforms used for genome wide methylation analysis (for a review see [28]), those based on methods of enrichment for methylated DNA able to recognize also non-CpG methylation and using bisulfite-insensitive adaptors, are expected to be unbiased. The comparison between data obtained by previous single-gene methods potentially biased and those obtained by the new unbiased approaches at methylome level, is likely to deserve amazing differences about the extent, dynamics and role of non-CpG methylation.

Materials and Methods

Supporting Information