Vitamin C Promotes Widespread Yet Specific DNA Demethylation of the Epigenome in Human Embryonic Stem Cells§


  • Tung-Liang Chung,

    1. Australian Institute for Bioengineering and Nanotechnology, The University of Queensland, Brisbane, Queensland, Australia
    2. Australian Stem Cell Centre, Melbourne, Victoria, Australia
    3. Monash Institute of Medical Research, Monash University, Melbourne, Victoria, Australia
    Search for more papers by this author
  • Romulo M. Brena,

    1. USC Epigenome Center, Keck School of Medicine, University of Southern California, Los Angeles, California, USA
    Search for more papers by this author
  • Gabriel Kolle,

    1. Institute for Molecular Bioscience, The University of Queensland, Brisbane, Queensland, Australia
    Search for more papers by this author
  • Sean M. Grimmond,

    1. Institute for Molecular Bioscience, The University of Queensland, Brisbane, Queensland, Australia
    Search for more papers by this author
  • Benjamin P. Berman,

    1. USC Epigenome Center, Keck School of Medicine, University of Southern California, Los Angeles, California, USA
    Search for more papers by this author
  • Peter W. Laird,

    1. USC Epigenome Center, Keck School of Medicine, University of Southern California, Los Angeles, California, USA
    Search for more papers by this author
  • Martin F. Pera,

    1. Eli and Edythe Broad Center for Regenerative Medicine and Stem Cell Research, University of Southern California, Los Angeles, California, USA
    Search for more papers by this author
  • Ernst Jurgen Wolvetang

    Corresponding author
    1. Australian Institute for Bioengineering and Nanotechnology, The University of Queensland, Brisbane, Queensland, Australia
    • Australian Institute for Bioengineering and Nanotechnology (AIBN), Corner College and Cooper Rds (Bldg 75), The University of Queensland, Brisbane Qld 4,072 Australia
    Search for more papers by this author
    • Telephone: 61-7-33463894; Fax: 61-7-33463973

  • Author contributions: T.-L.C.: conception and design, collection and/or assembly of data, data analysis and interpretation, and manuscript writing; R.M.B.: conception and design, collection and/or assembly of data, and data analysis and interpretation; G.K.: conception and design, collection and/or assembly of data, microarray data analysis and interpretation; S.M.G.: conception and design, and financial support; B.P.B.: collection and/or assembly of data, data analysis and interpretation, and manuscript writing; P.W.L.: conception and design, and financial support; M.F.P.: conception and design, data analysis and interpretation, and manuscript writing; E.W.: overall planning and design, financial support, and provision of study material or patients, data analysis and interpretation, manuscript writing, and final approval of manuscript.

  • Disclosure of potential conflicts of interest is found at the end of this article.

  • §

    First published online in STEM CELLS EXPRESS August 4, 2010.


Vitamin C (ascorbate) is a widely used medium supplement in embryonic stem cell culture. Here, we show that ascorbate causes widespread, consistent, and remarkably specific DNA demethylation of 1,847 genes in human embryonic stem cells (hESCs), including important stem cell genes, with a clear bias toward demethylation at CpG island boundaries. We show that a subset of these DNA demethylated genes displays concomitant gene expression changes and that the position of the demethylated CpGs relative to the transcription start site is correlated to such changes. We further show that the ascorbate-demethylated gene set not only overlaps with gene sets that have bivalent marks, but also with the gene sets that are demethylated during differentiation of hESCs and during reprogramming of fibroblasts to induced pluritotent stem cells (iPSCs). Our data thus identify a novel link between ascorbate-mediated signaling and specific epigenetic changes in hESCs that might impact on pluripotency and reprogramming pathways. STEM CELLS 2010;28:1848–1855


Human embryonic stem cells (hESCs) and human induced pluritotent stem cell (iPSCs) display both indefinite self-renewal as well as the ability to differentiate into most, if not all, cell types of the human body. Because of these remarkable properties these cells have attracted great interest as they can serve both as useful model systems for human embryogenesis and disease and as the cell of choice in regenerative medicine approaches. Both the regulation of lineage-specific differentiation as well as the reprogramming of somatic cells into iPSCs is controlled through the concerted temporally controlled action of specific sets of transcription factors in conjunction with epigenetic remodeling of the stem cell genome. Indeed, comparison of the methylome of fibroblasts, iPSCs, and hESCs has revealed that DNA methylation and demethylation of CpG and non-CpG nucleotides outside of classical CpG islands (CGIs), in so-called CpG shores, are important determinants for tissue-specific gene expression, cancer, and reprogramming [1]. Here, we investigate the role of the antioxidant vitamin C, a commonly used medium supplement for the culture of hESCs under serum-free conditions (knock-out serum replacement [KOSR] medium, mTeSR1 medium, StemPro medium) on global DNA methylation using the Infinium DNA methylation array platform. We show that vitamin C, at the concentration used in commercial media formulations, promotes widespread yet remarkably specific DNA demethylation of the hESC epigenome that are biased toward demethylation of shore CGIs and that significantly overlaps with bivalent domains and gene sets that are demethylated during reprogramming. Our data identify a novel link between vitamin C and specific epigenetic changes in hESCs that have the potential to directly impact on the differentiation of hESCs and reprogramming of somatic cells.


Cell Lines and Culture Conditions

HES2 and HES3 hESC lines were maintained in 20% fetal calf serum (FCS) on mouse embryonic fibroblast feeder layers (6 × 104/cm2) with mechanical dissection every 7 days according to the protocols described [2, 3]. HES2 and HES3 were all transferred to the culture system described by Amit et al. [4]. Collagenase was used to disaggregate colonies of embryonic stem cells for subculture. The culture medium was supplemented with 20% KOSR (Invitrogen Australia, Melbourne, Australia) and 4 ng/ml basic fibroblast growth factor (b-FGF), and feeder cells were used at a density of 2 × 104/cm2. Passage number of mechanically transferred hESCs in 20% FCS stock cultures (Px) and subsequent weekly passage number in serum free culture (Py) is indicated as Px + y.

Flow Cytometry

hESC line HES2 and HES3 were harvested with cell dissociation buffer (Sigma-Aldrich, Castle Hill/NSW, Australia) and stained using TG30 (anti CD9) monoclonal antibody supernatant (Australian Stem Cell Centre, Melbourne, Asutralia). Detection was carried out with Alexa Fluor 647 goat anti-mouse IgM2a (TG30) (1:1,000). Appropriate isotype controls were included in all analyses as well as the addition of propidium iodide into all samples (0.5 μg/ml) for exclusion of dead cells. A BD FC500 analyzer (Beckman Coulter Australia, Gladesville/NSW, Australia) was used to measure TG30 expression. In some experiments, flow cytometry was performed using a BD FACSVantage-Diva sorter.

Reconstitution of KOSR Minus Vitamin C

Custom preparation of the various KOSR fractions was carried out by Millipore (Millipore Australia, North Ryde/NSW, Australia) according to the International Patent Application WO98/30679 with the exception that vitamin C was omitted. For medium reconstitution, the pH of the amino acid solution was raised to about 7.0–7.4, and then the albumin solution and transferrin were added. The pH of the albumin-amino acid–transferrin mixture was adjusted to pH 7.7–7.9, and the insulin and trace elements were added. Cell culture grade water was added to give the desired volume, and the solution was filter-sterilized. hESCs were cultured with KOSR minus vitamin C as per Amit et al. [4] with the modification that 100 ng/ml of b-FGF was used.

Infinium DNA Methylation Array Analysis

For DNA methylation analysis, HES2 and HES3 were cultured in KOSR or KOSR minus vitamin C plus 100 ng/ml b-FGF using collagenase for transfer. HES2 (passage 11, 12, 13) and HES3 (passage 16, 18, 20) were FACS (Fluorescence activated cell sorting) sorted to collect TG30+ hESCs for isolation of genomic DNA. Bisulfite conversion was performed using the EZ-96 DNA methylation kit (Zymo Research, Orange/CA, USA), and the Illumina Infinium DNA Methylation assay was performed as described at Briefly, bisulfite-converted DNA was whole genome amplified and enzymatically fragmented, then purified and hybridized to the BeadChip arrays. The array was fluorescently stained after extension, scanned, and the intensities of the methylated and unmethylated bead types for each CpG locus are measured. The beta value, a DNA methylation score, is calculated as the ratio of the signal of the methylated bead type compared with the total signal (M + U) for the locus, and is compiled for each locus using Illumina BeadStudio software (Illumina, Inc., San Diego/CA, USA). After background subtraction bead studio produces beta values, detection p values, as well as methylated and unmethylated signal intensities. We identify all data points with a detection p value > .05 as not statistically significantly different from background measurements and treat them as missing data for all subsequent analyses.

To perform unsupervised clustering of Infinium methylation profiles, we first identified a set of highly variable probes by examining probe standard deviations across all 19 samples and picking those 1,472 with a standard deviation of 0.15 or greater. Linear hierarchical clustering was performed in MATLAB using the Euclidean distance metric and average linkage.

To screen for genes with significant differential methylation, we performed a two-tail rank sum (Mann Whitney U Test) comparing beta values of the 6 KOSR-A samples against the 11 KOSR samples for all 27,579 probes. Raw p values were corrected for multiple hypotheses using the linear step-up method of Benjamini and Hochberg [5]. For comparisons between HES2 and HES3, we used a stricter criteria that a differentially methylated probe must have all beta values of one group be at least 0.2 greater than all beta values of the other group (generally with one group containing three samples and the other group containing three samples). The value 0.2 was used as a separation cutoff because Illumina has found that extremely few probes have differences of 0.2 between replicate experiments (R.M. Brena, personal communication).

Profiling by Microarray

Total RNA was isolated from TG30+ HES2 cells (passage 17, 19, 20) cultured with or without vitamin C using the RNeasy mini kit (Qiagen Australia, Doncaster/VIC, Australia). llumina Human 6 BeadChips (V3) were hybridized, washed, stained, and scanned according to the manufacturer's protocol from 200 ng of starting RNA. Raw intensity measurements were calculated using Illumina BeadStudio software (Illumina, Inc., San Diego/CA, USA). The data was background adjusted by Robust Mutiarray Averaging (RMA) method and quantile normalized (normalize.quantiles) using the affy and BeadExplorer packages in R. Differential expression was measured using the empirical Bayes method [6], using the Limma package in R. Raw and processed data has been submitted to the GEO repository under accession number (GSE16590). Data was visualized in GeneSpring GX 7.3.1. p value of overlap to methylation data was generated using a hypergeometric test.

Western Blot Assays to Detect Global Histone H3K9 Modifications

Undifferentiated HES3 hESCs cultured with KOSR or KOSR minus ascorbate were harvested for western blot. hESCs were mixed with 6X SDS sample buffer for SDS-polyacrylamide gel electrophoresis (15%). Proteins were transferred to nitrocellulose membrane electrophoretically using iBlot Dry Blotting System (Invitrogen Australia, Melbourne, Australia). The nitrocellulose membrane was incubated in 10 ml blocking buffer (0.1% Tween-20 in tris-buffered saline buffer (T-BST) with 5% skim milk for detection of total H3 and acetylated H3K9; 5% BSA for monomethylated H3K9, dimethylated H3K9, and trimethylated H3K9) for 2 hours at room temperature. Wash the nitrocellulose membrane in 20 ml T-BST for 10 minutes each with gentle agitation. Samples were incubated with primary antibodies at 4°C for overnight in TBS-T. Antibodies were anti-histone H3 (catalog no. ab1791; Abcam, Cambridge/MA, USA), anti-H3K9 mono-methyl (catalog no. ab9045; Abcam), anti-H3K9 di-methyl (catalog no. ab1220; Abcam), and anti-H3K9 tri-methyl (catalog no. ab8898; Abcam). All antibodies were used in the concentration of 1 μg/ml. Blots were washed three times for 10 minutes each in TBS-T. Secondary antibodies (goat anti-rabbit-horseradish peroxidase conjugate; 1:2,500; Upstat/Millipore Australia, North Ryde/NSW, Australia) were applied in TBS-T for 1 hour at room temperature. Wash for 10 minutes twice as described previously. Blots were used for Pierce ECL Western Blotting detection according to the manufacturer's recommendation (Thermo Fisher Scientific Inc., Waltham/MA USA).

Real-Time Polymerase Chain Reaction of hypoxia-inducible factor (HIF) histone demethylases JMJD1, and JMJD2 in HES3 hESCs

HES3 hESCs were cultured in either KOSR or KOSR minus ascorbate for 7 days. Cells were harvested with cell dissociation buffer (Sigma-Aldrich, Castle Hill/NSW, Australia) for RNA isolation. Total RNA was prepared using the RNA easy kit (Qiagen Australia, Doncaster/VIC, Australia). One microgram of RNA was converted to cDNA with SuperScript III Reverse Transcriptase Kit (Invitrogen Australia, Melbourne, Australia) in a 20 μl reaction mix according to the manufacturer's instructions. Real-time polymerase chain reaction was performed using Fast SYBR Green Master Mix (Applied Biosystem Australia, Melbourne, Australia) and detected with ABI PRISM 7500 sequence detection system according to the protocol provided by the manufacturer. The primers in this study are as follows:



Vitamin C Is Required for Undifferentiated hESC Growth in KOSR Medium

Vitamin C is a well-known water-soluble antioxidant and a common medium supplement that enhances cell proliferation [7–10]. To examine the effect of vitamin C on DNA methylation in hESCs, we cultured hESCs in KOSR medium without vitamin C for prolonged periods of time. We observed that hESCs cultured in KOSR medium without vitamin C proliferated more slowly than in standard KOSR and failed to proliferate in culture after 2–3 passages (Fig. 1A). However, by increasing the b-FGF concentration to100 ng/ml, this reduced proliferation in the absence of vitamin C could be overcome. Indeed, we were able to maintain HES2 and HES3 hESCs in KOSR medium without vitamin C in the undifferentiated state for 15 and 23 weeks, respectively (Fig. 1B).

Figure 1.

Human embryonic stem cells (hESCs) growth in KOSR requires vitamin C that can be substituted by high concentration of b-FGF. (A): HES3 hESC bulk cultures were maintained using various medium compositions containing, complete KOSR + 4 ng/ml b-FGF for 7 days, KOSR minus ascorbate + 4 ng/ml b-FGF for 7 days, KOSR minus ascorbate + 100 ng/ml b-FGF for 7 days, and KOSR minus ascorbate + 100 ng/ml b-FGF for three passages. (B): High concentration of b-FGF (100 ng/ml) promotes undifferentiated hESCs growth in the absence of vitamin C in KOSR. HES2 and HES3 hESCs in KOSR minus vitamin C with 100 ng/ml b-FGF display 90%–95% TG30+ at passage 15 and 23, respectively. Abbreviations: b-FGF, basic fibroblast growth factor; KOSR, knock-out serum replacement medium.

Vitamin C Promotes DNA Demethylation of the hESC Epigenome

Using Infinium DNA methylation array analysis, which interrogates 27,578 probes located in the proximal promoter regions of 14,495 genes, we next compared genome-wide DNA methylation of flow sorted undifferentiated HES2 hESCs (passages 10, 11, 12) and HES3 hESCs (passages 16, 18, 20) cultured in KOSR + 100 ng/ml b-FGF with or without 50 μg/ml vitamin C. In examining genome-wide DNA methylation levels (Supporting Information Table S1), we observed a large number of CpG probes that were highly variable across samples as illustrated by the long right tail in the distribution of standard deviations (Fig. 2A). Isolating the 1,472 most highly variable CpG probes and performing unsupervised hierarchical clustering revealed a striking epigenomic shift from the methylated state in all samples cultured without vitamin C to an unmethlyated state in all hESC samples cultured with vitamin C (Fig. 2B). Although some DNA methylation differences were attributable to intrinsic differences between cell lines, global ascorbate-associated DNA hypomethylation accounted for the majority of changes between samples (Fig. 2C). HES3 cultured in KOSR plus vitamin C for 78 passages displayed significantly more DNA methylation loss than the rest of the lines cultured in KOSR with vitamin C for 10–20 passages, suggesting that prolonged exposure of hESCs to vitamin C leads to further DNA demethylation.

Figure 2.

Vitamin C promotes genome-wide DNA demethylation of a specific subset of human embryonic stem cell promoters. Infinium array analysis of DNA extracted from 5 HES2 in knock-out serum replacement medium (KOSR), 6 HES3 in KOSR, 3 HES2 KOSR-A, 3 HES3 KOSR-A sorted for TG30/CD9 cultured in KOSR with or without 50 μg/ml ascorbate, and two control samples (HES2 cultured for 111 passages under standard FCS conditions and HES3 passaged for 78 passages in KOSR). (A): A total of 1,472 probes of the 2,205 perfect discriminators had a standard deviation of greater than 0.15 among 19 samples investigated. (B): A clustergram showing the 1,472 most variable probes shows that the vast majority were significantly more methylated in ASC− samples than in ASC+ (blue, cyan) samples. A HES3 sample cultured for 78 passages in ascorbate (P78 KOSR) shows extremely little methylation. (C): Clustering of DNA methylation profiles. Euclidean distances were calculated for all sample pairs among 1,472 highly variable probes. Pseudocolor matrix shows the most similar pairs in red and the most differing pairs in blue. The dendrogram at the left shows ASC+ and ASC− samples form two distinct clusters, each of which is divided into HES2 and HES3 subclusters. (D): Of 1,036 probes with DNA methylation loss in HES2 fractions, 891 (86%) also showed loss in HES3, whereas these common probes only represented 52% of the 1,702 probes lost in HES3. Abbreviations: ASC−, ascorbate negative; ASC+, ascorbate positive; HES2 and HES3 are the names of hESC lines.

To identify all ascorbate-sensitive gene promoters, we performed a two-tail rank sum test (Mann-Whitney) between the 6 ascorbate negative (ASC−) samples and 11 ascorbate positive (ASC+) samples, using all probes on the array and adjusting for multiple hypotheses using Benjamini-Hochberg correction [5]. We identified 2,958 differentially methylated probes with a false discovery rate (FDR) adjusted p value of .01 or less, and 2,205 of these had the lowest possible p value of .002 indicating perfect segregation of DNA methylation values based on ASC status (Supporting Information Fig. 1). For 2,194 of these (representing 1,847 genes), the highest DNA methylation level among all 11 ASC+ samples was lower than the lowest DNA methylation level in any of the 6 ASC− samples (the remaining 11 probes exhibited complete DNA methylation gain in ASC+ samples, a number consistent with the fraction expected by chance at this FDR). This large number of genes shifting DNA methylation all in the same direction indicates that loss of promoter methylation is a strikingly widespread consequence of exposure to vitamin C. Database for Annotation, Visualization and Integrated Discovery (DAVID) analysis [11, 12] revealed a significant overrepresentation of integral membrane proteins as well as genes of secreted proteins (Supporting Information Table S2). The significance of these observations remains to be determined. Ingenuity analysis [13] further indicates that of the 1,847 ascorbate-responsive hypomethylated genes in both HES2 and HES3, 512 genes are associated with cancer, 407 with cellular growth and proliferation and 327 with tissue development (Supporting Information Table S3), including important hESC genes such as Jagged-2, Lefty2, Piwil2, Cdx4, and Fox12.

To select only those probes with the most robust DNA methylation changes, we used a stringent criteria requiring the lowest beta value among ASC− samples to be at least 0.2 greater than the highest beta value among ASC+ samples. A Comparison between the HES2 and HES3 cell lines showed that 891 of the 1,036 probes (86%, representing 733 genes) losing DNA methylation in HES2 were also lost in HES3, illustrating the overwhelmingly consistent nature of this process (Fig. 2D and Supporting Information Table S4). HES3, however, showed significantly more loss than HES2 (Fig. 2D), possibly due to the fact that this line experienced more overall passages in culture.

Ascorbate-Induced DNA Hypomethylation Is Associated with Altered Gene Expression

We next investigated how the ascobate-dependent-specific global demethylation affects gene expression in hESCs. RNA was isolated from undifferentiated TG30/CD9 sorted HES2 hESCs cultured for 17, 19, and 20 passages in KOSR with or without 50 μg/ml vitamin C, and arrayed on the Illumina Human V3 beadArray platform. After applying an empirical Bayes method to determine differentially expressed genes, we found 295 genes (317 individual probes) upregulated and 233 genes (250 probes) downregulated on culturing in vitamin C containing media (Fig. 3A and Supporting Information Table S5).

Figure 3.

Culturing human embryonic stem cells in vitamin C leads to strong changes in gene expression. (A): Gene and condition tree showing the relative expression levels for all genes that are significantly differentially expressed (B-statistic > 0) between cells grown in ASC+ and ASC− media (three replicates [P17, P19, and P20] of each). Genes colored in relation to the median expression for each gene (value = 1) as shown in the scale bar. (B): Venn diagram displaying the overlap between gene expression and methylation data. Of a total of 12,829 genes that are in common between the two platforms, 1,614 of these are downmethylated and 9,112 are expressed at a detectable level on the gene expression microarray. Importantly, 44 genes are both upregulated and downmethylated in ASC+ cells (p value of enrichment = 4.0 × 10−13). Abbreviations: ASC−, ascorbate negative; ASC+, ascorbate positive.

To determine the potential that these differentially expressed genes are mediated by methylation state, we overlapped expression data with methylation data from above. From a total of 12,829 genes that were present in both platforms, we found 1,162 genes that were both methylated and determined as present in at least three replicates on the array (Fig. 3B). Importantly, of the genes that were upregulated (148 genes), a significant number (29%) also show loss of methylation (44 genes, p value of enrichment = 4.0 × 10−13). In contrast, of the genes that were downregulated (178 genes) only 21 (12%) showed loss of methylation (p value of enrichment = .03).

Ascorbate-Mediated DNA Hypomethylation Occurs Preferentially at CGI Edges and Away from the Transcription Start Sites

We next sought to determine if the CpG genomic context played a role in the observed epigenetic and gene expression changes. Although all 25,578 Infinium probes fall within a tight 1,500 bp interval upstream or downstream of the gene transcription start sites (Supporting Information Fig. 2A), it is nevertheless possible to observe global trends within this limited context. CpGs hypomethylated in the ascorbate condition tended to occur more frequently in CGI than in non-CGI gene promoters and, remarkably, were strongly biased toward occurring away from the transcription start site (Fig. 4A). Hypomethylation appears to occur most frequently at the CpGs near the edges of a CGI (Fig. 4B), with only 10% of probes within the middle 50% of CGIs being hypomethylated (n = 7,375) compared with 29% of probes in the 500 bp flanking either edge of a CGI (n = 1,987; p < 10−16 by χ2 test).

Figure 4.

CpGs near gene promoters are less frequently hypomethylated but better predictors of ascorbate-induced gene expression changes. (A, B): Show the frequency of hypomethylated CpGs by their position relative to promoter features. (A): Ascorbate-induced hypomethylation occurs preferentially outside of the core promoter region in CpG island promoters (blue), but not in non-CpG island promoters (green). When plotted relative to Takai-Jones and Gardener-Gardin CGI features, it is apparent that hypomethylation peaks at CGI boundaries (B). (C, D): Show frequency of differential gene expression associated with CpGs at different genomic positions, stratified by ascorbate-induced methylation status (blue for hypomethylated CpGs, green for no change in methylation). Downregulated genes (C) associate best with hypomethylation at or just downstream of promoter cores, while induced genes (D) associate best with CpGs deep within the gene body. Each data point in (A–D) represents the average level of all CpG probes within 300 bp of the x coordinate. Abbreviations: CGI, CpG islands; TSS, transcription start site.

Ascorbate-Mediated Hypomethylation of CpGs Around the TSS Is Strongly Corrrelated with Increased Gene Expression

Clearly from these observations, CpG position within a gene promoter influences its chance of undergoing a shift in DNA methylation, but is this context relevant to the observed changes in gene expression? To investigate this, we stratified CpGs based on DNA methylation status and differential gene expression of the associated gene. DNA hypomethylation was associated with increased gene expression when the CpG was in a central range between 200 bp upstream of the gene promoter and 600 bp downstream (2.9% of 1,532 genes upregulated), but not when the CpG was outside of this central range (1.1% of 1,941 genes upregulated) as shown in Figure 4C, and this observation is statistically significant (p = 7 × 10−5 by χ2 test). DNA hypomethylation of CpGs in this central range was not correlated with decreased gene expression (Fig. 4D; p = .15 by χ2 test). DNA hypomethylation of CpG promoters farther downstream of the promoter (> 800 bp) may be correlated with decreased gene expression (Fig. 4D), but due to the very small number of probes in this region, this observation lacked a high degree of significance (p = .04) and requires additional confirmation. If confirmed, this set of spatial patterns would be consistent with reports of the inverse roles of DNA methylation in gene promoters and gene bodies [14–16], and they highlight the importance of taking base pair level genomic context into account even when considering a relatively controlled set of genomic elements (in this case, promoter regions).


Vitamin C has been included in cell culture media for a considerable time because of its recognized beneficial effect on cell proliferation. In addition to its role as a water-soluble antioxidant, it is also a cofactor for important enzymes such as prolyl-hydroxylases and histone demethylases [17]. By examining genome-wide DNA methylation profiles in hESC lines cultured in the absence or presence of vitamin C, we have uncovered an unexpected and novel DNA demethylating effect of vitamin C. Ascorbate-mediated DNA methylation occurs genome-wide, but specifically affects a subset of 1,847 genes and causes altered expression of a subset of these genes. Interestingly, ascorbate-mediated DNA demethylation occurs most frequently at CpGs near the edges of a CGI. This is an intriguing observation in light of Irizarry et al. who showed that most tissue-specific and cancer-specific DNA methylation occurs in such “CGI shores” [18]. Indeed, 50% of demethylated genes that show a concomitant alteration in gene expression (44 up and 21 down (Supporting Information Table S6) in response to vitamin C treatment were also identified by Irizarry et al., as tissue differential DNA methylation regions (T-DMR) that undergo shore CpG methylation changes. The Feinberg laboratory subsequently extended these observations to show that T-DMRs significantly overlap with both differentially methylated regions (r-DMRs) in reprogrammed cells and cancer-specific DMRs (c-DMRs). Their data further show that hypomethylated R-DMRs were associated with bivalent chromatin marks. A comparison of our set of ascorbate-hypomethylated genes (1,780) with genes exhibiting r-DMRs (3,017) [1] and genes marked by bivalent marks (3,292) [19], shows 353 genes overlapping with r-DMR, 778 with bivalent marked genes and 201 genes with both (Supporting Information Fig. 2B, Supporting Information Table S7). Our data therefore suggest that vitamin C specifically alters the methylation of shore CGIs that are associated with tissue-specific DNA methylation and reprogramming. Indeed, in strong support of this proposition, vitamin C was recently shown to enhance reprogramming of mouse and human fibroblasts into iPSCs and this was suggested to be independent from the antioxidant properties of vitamin C [20]. We suggest that the DNA demethylation effect of vitamin C contributes to the reprogramming accelerating effect of vitamin C and that in particular genes residing within the overlap between ascorbate-demethylated genes and reprogramming r-DMRs are candidate genes involved in this effect.

Ingenuity and DAVID analysis shows that exposure of hESCs to vitamin C leads to the DNA demethylation of genes that have important functions in pluripotency and tissue-specific differentiation, including genes such as Jagged-2, lefty2, piwil2, CDX4, and Fox12 (Supporting Information Table S5). Culture of hESCs in the presence of vitamin C therefore could affect the biological behavior of hESCs.

Previously, Allegrucci et al. [21] reported that switching BG01 hESC cultures from serum-containing (ascorbate-free) standard conditions to serum-free or feeder-free (ascorbate-containing) conditions resulted in epigenetic changes. These mainly comprised DNA methylation losses. Our data that shows vitamin C induces DNA demethylation provides a rationale for this observation. A separate study found that 14 of 20 genes that were demethylated with increased passage in multiple hESC lines [22] (1,536 CpG sites from 371 genes examined) overlap with the 733 genes, we identified as commonly demethylated by vitamin C in two hESC lines (Supporting Information Table S8). We therefore hypothesize that ascorbate-induced DNA demethylation may play a role in the adaptation of hESCs to serum-free culture conditions.

The reason for the specificity of the ascorbate-induced DNA methylation changes and the bias of ascorbate-mediated DNA demethylation for CGI boundaries remains unclear at present. These aspects, and the upstream molecular mechanism of ascorbate-induced demethylation, are the subject of ongoing investigation in our laboratory. We speculate that vitamin C may indirectly change DNA topology by affecting the activity of histone demethylases [23, 24] and/or histone-acetylation and thus modify the accessibility of DNA methylases and/or demethylases. Our preliminary data showing that ascorbate leads to increased histone acetylation, decreased H3K9-methylation as well as increased expression of JMJD 1A, 1B, 1C, and 2B supports this idea (Supporting Information Fig. 3 and 4). Alternatively vitamin C may also directly affect DNA methylation through an ascorbate-stimulated DNA demethylase such as FTO (fat mass and obesity associated), a 2-oxoglutarate-dependent nucleic acid demethylase [25] and TET1 [26].


Despite the long standing and widespread use of vitamin C in cell culture media, this is to our knowledge the first report on the CpG demethylating effect of vitamin C on human pluripotent stem cells. The genome-wide but specific DNA demethylation of genes by vitamin C alone underscores the importance of appreciating the potential effects of cell culture conditions on the epigenetic make-up of somatic, hESCs and iPSCs.


We thank the human ES cell Core facility of the Australian Stem Cell Centre for their technical support. This work was supported by the Australian Stem Cell Centre, the California Institute for Regenerative Medicine (RS1-00408, to P.W.L.), and postdoctoral fellowship from the Susan G. Komen Foundation (KG080103, to R.M.B.).


Peter W. Laird acted as a consultant for Epigenomics, AG. The other authors have no financial interests to disclose.