Abstract
Stable maintenance of gene regulatory programs is essential for normal function in multicellular organisms. Epigenetic mechanisms, and DNA methylation in particular, are hypothesized to facilitate such maintenance by creating cellular memory1,2,3,4 that can be written during embryonic development5,6 and then guide cell-type-specific gene expression7. Here we develop new methods for quantitative inference of DNA methylation turnover rates, and show that human embryonic stem cells preserve their epigenetic state by balancing antagonistic processes that add and remove methylation marks rather than by copying epigenetic information from mother to daughter cells. In contrast, somatic cells transmit considerable epigenetic information to progenies. Paradoxically, the persistence of the somatic epigenome makes it more vulnerable to noise, since random epimutations can accumulate to massively perturb the epigenomic ground state. The rate of epigenetic perturbation depends on the genomic context, and, in particular, DNA methylation loss is coupled to late DNA replication dynamics. Epigenetic perturbation is not observed in the pluripotent state, because the rapid turnover-based equilibrium continuously reinforces the canonical state. This dynamic epigenetic equilibrium also explains how the epigenome can be reprogrammed quickly8 and to near perfection9 after induced pluripotency.
This is a preview of subscription content, access via your institution
Access options
Subscribe to this journal
Receive 51 print issues and online access
$199.00 per year
only $3.90 per issue
Buy this article
- Purchase on SpringerLink
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
References
Bird, A. DNA methylation patterns and epigenetic memory. Genes Dev. 16, 6–21 (2002)
Cedar, H. & Bergman, Y. Programming of DNA methylation patterns. Annu. Rev. Biochem. 81, 97–117 (2012)
Hemberger, M., Dean, W. & Reik, W. Epigenetic dynamics of stem cells and cell lineage commitment: digging Waddington’s canal. Nature Rev. Mol. Cell Biol. 10, 526–537 (2009)
Mohn, F. & Schübeler, D. Genetics and epigenetics: stability and plasticity during cellular differentiation. Trends Genet. 25, 129–136 (2009)
Smallwood, S. A. et al. Dynamic CpG island methylation landscape in oocytes and preimplantation embryos. Nature Genet. 43, 811–814 (2011)
Smith, Z. D. et al. A unique regulatory phase of DNA methylation in the early mammalian embryo. Nature 484, 339–344 (2012)
Gifford, C. A. et al. Transcriptional and epigenetic dynamics during specification of human embryonic stem cells. Cell 153, 1149–1163 (2013)
Rais, Y. et al. Deterministic direct reprogramming of somatic cells to pluripotency. Nature 502, 65–70 (2013)
Kim, K. et al. Epigenetic memory in induced pluripotent stem cells. Nature 467, 285–290 (2010)
Stein, R., Gruenbaum, Y., Pollack, Y., Razin, A. & Cedar, H. Clonal inheritance of the pattern of DNA methylation in mouse cells. Proc. Natl Acad. Sci. USA 79, 61–65 (1982)
Goll, M. G. & Bestor, T. H. Eukaryotic cytosine methyltransferases. Annu. Rev. Biochem. 74, 481–514 (2005)
Arand, J. et al. In vivo control of CpG and non-CpG DNA methylation by DNA methyltransferases. PLoS Genet. 8, e1002750 (2012)
Kohli, R. M. & Zhang, Y. TET enzymes, TDG and the dynamics of DNA demethylation. Nature 502, 472–479 (2013)
Wossidlo, M. et al. 5-Hydroxymethylcytosine in the mammalian zygote is linked with epigenetic reprogramming. Nature Commun. 2, 241 (2011)
Landan, G. et al. Epigenetic polymorphism and the stochastic formation of differentially methylated regions in normal and cancerous tissues. Nature Genet. 44, 1207–1214 (2012)
Kivioja, T. et al. Counting absolute numbers of molecules using unique molecular identifiers. Nature Methods 9, 72–74 (2012)
Gu, H. et al. Preparation of reduced representation bisulfite sequencing libraries for genome-scale DNA methylation profiling. Nature Protocols 6, 468–481 (2011)
Halligan, D. L. & Keightley, P. D. Spontaneous mutation accumulation studies in evolutionary genetics. Annu. Rev. Ecol. Evol. Syst. 40, 151–172 (2009)
Gal-Yam, E. N. et al. Frequent switching of Polycomb repressive marks and DNA hypermethylation in the PC3 prostate cancer cell line. Proc. Natl Acad. Sci. USA 105, 12979–12984 (2008)
Schlesinger, Y. et al. Polycomb-mediated methylation on Lys27 of histone H3 pre-marks genes for de novo methylation in cancer. Nature Genet. 39, 232–236 (2007)
Guelen, L. et al. Domain organization of human chromosomes revealed by mapping of nuclear lamina interactions. Nature 453, 948–951 (2008)
Hansen, R. S. et al. Sequencing newly replicated DNA reveals widespread plasticity in human replication timing. Proc. Natl Acad. Sci. USA 107, 139–144 (2010)
Stadler, M. B. et al. DNA-binding factors shape the mouse methylome at distal regulatory regions. Nature 480, 490–495 (2011)
Lienert, F. et al. Identification of genetic elements that autonomously determine DNA methylation states. Nature Genet. 43, 1091–1097 (2011)
Kelly, T. K. et al. Genome-wide mapping of nucleosome positioning and DNA methylation within individual DNA molecules. Genome Res. 22, 2497–2506 (2012)
Deal, R. B., Henikoff, J. G. & Henikoff, S. Genome-wide kinetics of nucleosome turnover determined by metabolic labeling of histones. Science 328, 1161–1164 (2010)
Hansen, K. D. et al. Increased methylation variation in epigenetic domains across cancer types. Nature Genet. 43, 768–775 (2011)
Berman, B. P. et al. Regions of focal DNA hypermethylation and long-range hypomethylation in colorectal cancer coincide with nuclear lamina-associated domains. Nature Genet. 44, 40–46 (2012)
Coolen, M. W. et al. Consolidation of the cancer genome into domains of repressive chromatin by long-range epigenetic silencing (LRES) reduces transcriptional plasticity. Nature Cell Biol. 12, 235–246 (2010)
Lister, R. et al. Human DNA methylomes at base resolution show widespread epigenomic differences. Nature 462, 315–322 (2009)
Milyavsky, M. et al. Prolonged culture of telomerase-immortalized human fibroblasts leads to a premalignant phenotype. Cancer Res. 63, 7147–7157 (2003)
Krueger, F. & Andrews, S. R. Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications. Bioinformatics 27, 1571–1572 (2011)
Karolchik, D. et al. The UCSC Genome Browser database: 2014 update. Nucleic Acids Res. 42, D764–D770 (2014)
Bernstein, B. E. et al. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57–74 (2012)
Ernst, J. & Kellis, M. ChromHMM: automating chromatin-state discovery and characterization. Nature Methods 9, 215–216 (2012)
Cohen, N. M., Kenigsberg, E. & Tanay, A. Primate CpG islands are maintained by heterogeneous evolutionary regimes involving minimal selection. Cell 145, 773–786 (2011)
Acknowledgements
We thank E. Kenigsberg, E. Yaffe and the Tanay group for discussions. Research in the Tanay group was supported by the European Research Council (EVOEPIC), the EU BLUEPRINT project, the Israel Science Foundation (1050/12 and I-Core) the Israel Ministry of Science and the Helen and Martin Kimmel Award.
Author information
Authors and Affiliations
Contributions
Z.S., Z.M. and A.T. designed the study. Z.S. and Z.M. performed the experiments. Z.S., N.M.C. and A.T. developed the algorithms. Z.S., Z.M. and A.T. analysed the data. G.L. and E.C. helped in developing the experimental protocol and analytical framework. Y.C.F. and E.A. generated ES-cell clones. S.R.Z. and N.F. generated CD8+ T-cell clones. Z.S., Z.M. and A.T. wrote the paper.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing financial interests.
Extended data figures and tables
Extended Data Figure 1 UMI-bis.
a, Protocol schematics. b, Size distribution of UMI-bis library before sequencing, including the sequencing adaptors. c, Top, size distribution of mapped sequenced fragments, after filtering the adaptors. Bottom, distribution of the number of reads covering each barcoded methylation pattern. UMIs that are covered by more than one read can be attributed to amplification bias, or to spontaneous relabelling of two molecules with the same UMI code. In any case, we retain only one pattern for each UMI. d, Comparison of average CpG methylation between standard Bis-seq30 and UMI-bis. e, Genomic annotation of the 290,279 MspI fragments that are covered by 20 or more molecules in the WI38 population data set. f, CpG content distribution of MspI fragments that are covered by 20 or more molecules in the WI38 population data set (low CpG content: <1.5%, mid: 1.5–5%; high: >5%). The number of individual CpGs within the fragments is indicated in parentheses (showing that high CpG content fragments cover many more CpGs). UMI-bis libraries cover a wide spectrum of genomic contexts, while enriching for high CpG content loci. g, Single CpG methylation average is compared between H9 human ES cells and WI38. h, Fragment average methylation comparison between H9 human ES-cell and K562/MCF7/HepG2 cancer cell lines. Cancer cell lines show considerable groups of loci changing their methylation identity altogether. i, Fragment average methylation comparison between mouse ES cells and mouse CD8+ T-cell data. General conservation of the bipolar methylation state in primary T cells is accompanied by some significant changes in average methylation observed mostly for methylated loci.
Extended Data Figure 2 Methylation pattern analysis defines epigenomic heterogeneity in stem cells and fibroblasts.
a, Quantification of average methylation for individual CpGs in regions with low (<1.5%), medium (1.5–5%) and high CpG (>5%) content reconfirms the known bipolar epigenomic organization in ES cells. Low CpG content regions are always methylated to near completeness, and high CpG content regions are strongly biased to low methylation levels. Regions with 20–80% average methylation are rare in ES cells. b, A similar analysis for fibroblasts shows that although the overall bipolar organization is retained, a proportion of the genome develops intermediate methylation levels. c, Colour-coded density map shows comparison of regional (MspI fragment) methylation average between ES cells and fibroblasts. Regions developing intermediate methylation in fibroblasts are typically fully methylated in ES cells. d, Methylation pattern analysis is based on estimating frequencies of patterns at loci. For each genomic locus with adequate coverage, we identify the most common pattern, other frequent patterns (if such exist), and all rare (<20%) patterns. The most frequent pattern (11/20) in this example is the unmethylated pattern, with a second pattern observed in five molecules and an additional four rare patterns. e, Fraction of loci dominated by unmethylated, partially methylated or methylated patterns, in ES cells (left) and fibroblasts (right). Note that fractions are computed relative to the pool of MspI fragments that are covered in both libraries. f, Shown are the distributions of total frequencies of rare patterns at loci dominated by methylated, partially methylated and unmethylated patterns. g, Comparison of the entropy of methylation pattern distributions between polyclonal human ES-cell (H1/H9) populations and non-clonal WI38 populations. Regions that are of generally high entropy are more heterogeneous in fibroblasts, while regions with generally low entropy are more heterogeneous in ES cells. h, Similar to g, but using epipolymorphism (the probability that two random epialleles are not the same). i, For each of four ES-cell clonal populations (two from H1 and two from H9), we compare the frequency of patterns in the clonal population (y-axis) to the frequency of the same pattern in the matching polyclonal ES-cell population. No notable differences are apparent. j, Similar to i, but using data on six WI38 fibroblasts clonal populations. For each clonal population we observe an independent group of patterns that appear at a frequency of approximately 50% but are rare in the founding polyclonal population.
Extended Data Figure 3 Clustering pattern distributions.
a, Schematic illustration of the comparative pattern clustering procedure. UMI-bis data can be represented as defining the frequencies of 2N possible patterns at each MspI fragment (where N is the number of CpGs covered in the fragment). These sampled frequencies, however, depend on the sequencing depth and number of CpGs in each fragment. b, We derive uniform coverage by sub-sampling at each fragment and library. c, We then determine, for each fragment, a set of common patterns, including all methylation patterns that are observed at 20% or more in at least one library, together with the special unmethylated and methylated patterns. This set of common patterns is now ordered arbitrarily such that the frequency vector for each fragment defines the unmethylated and methylated frequencies, following the frequencies at the K common patterns, and the total frequency of non-common (rare, or ‘noise’) patterns. This feature-selection strategy allows for simplified comparison of the distributions at each fragment between libraries. d, Clustering of the frequency vectors allows discovery of groups of loci with specific structures of methylation distributions, as demonstrated here with clusters showing a dominant unmethylated pattern with variable levels of noise, or a cluster showing fragments in which one library is a clear outlier of the other libraries. e, The genomic contexts of the fragments in each of these clusters can be studied by statistical enrichment.
Extended Data Figure 4 Controls and annotation of methylation pattern clusters.
a, Propidium iodide intensity distribution of H9 human ES cells, illustrating the S and G phase gating we used in order to compare methylation distributions along the cell cycle. b, UMI-bis was performed on H9 human ES cells and WI38 cells that were sorted into G1 and S populations. Shown is average methylation for well covered fragments in the two data sets. This analysis confirms that the heterogeneity we observe in the methylation patterns is not linked directly to cell-cycle effects. c, Annotation of the pattern clusters in the high and intermediate methylation classes, as shown in Fig. 2. The different classes are not strongly linked to transcription start sites (TSSs), exons or H3K27me3 occupancy. d, Cancer cell line methylation distributions for loci classified according to the WI38 pattern clusters. e, Distributions of lamina association and time of replication (Imr90 Repli-Seq data) for loci associated with the different pattern clusters discussed in Figs 1 and 2.
Extended Data Figure 5 Clonal analysis in K562 cells.
a, K562 pattern clusters of low methylation regions. Data for three clonal and one polyclonal population are shown. Visualization is as described for Figs 1 and 2. b, Annotation of low methylation K562 pattern clusters. c, High methylation K562 pattern clusters. d, Intermediate methylation pattern clusters. e, Distribution of replication time (K562 Repli-Seq) for the pattern clusters defined earlier. Overall the K562 methylome is defined by the existence of persistent methylation clusters at both the low and high methylation regimes, and by an additional large subset of genomic loci that show highly heterogeneous methylation at late replicating domains.
Extended Data Figure 6 Clonal analysis in MCF7 and HepG2 cancer cell lines.
a, HepG2 pattern clusters of low methylation regions. Data for one clonal and one non-clonal population are shown. Visualization is as in Figs 1 and 2. b, Annotation of low methylation HepG2 pattern clusters. c, High methylation HepG2 pattern clusters. d, Intermediate methylation pattern clusters. Overall the HepG2 methylome is defined by a large set of intermediate methylation regions. e, Distribution of replication time (K562 Repli-Seq) for HepG2 pattern clusters. Clusters with intermediate methylation show a clear bias towards late replication. f, Pattern clusters of low methylation regions, comparing non-clonal and clonal MCF7 populations. MCF7 methylomes are saturated for high methylation, so persistent low methylation clusters are not observed. g, Annotation of low methylation clusters. High noise regions are enriched for H3K27me3 ES-cell occupancy. h, High methylation pattern clusters. i, Intermediate methylation MCF7 pattern clusters. Note that since only one clonal population is analysed the number of common partial patterns cannot be large. j, Distribution of replication times (K562 Repli-Seq data) on all MCF7 classes.
Extended Data Figure 7 Methylation dynamics in mouse T cells and in human tissues.
a, Pattern clusters for low methylation loci, using one non-clonal and two clonal CD8+ T-cell populations. Persistent and high noise clusters (III) are identified. b, Annotation of pattern clusters in a, showing enrichment of H3K27me3 loci at the persistent classes. c, Pattern clusters for high methylation loci. d, Pattern clusters for intermediate methylation loci. e, Distribution of replication time (using ENCODE Repli-chip data on MEL cells, quantile normalized), on loci classified according to a–d. As observed for WI38 cells, persistent and intermediate methylation loci tend to be observed at late replicating domains. f, Comparison of human ES-cell average methylation and methylation in three different human tissues, stratified according to ES-cell time of replication (late replicated, left; early replicated, right). These data show that methylation in adult tissues is massively degraded compared to ES cells, in a replication-time-dependent fashion, as also predicted by our clonal analysis. g, Shown are average epipolymorphism values for loci that show high methylation in human ES cells (>0.9) and lose methylation in adult tissues (x-axis). This analysis shows that methylation loss in adult tissues is correlated with extremely high heterogeneity, suggesting that the stochastic dynamics we observed in in vitro clonal populations widely affect methylation in vivo.
Extended Data Figure 8 Inference of turnover rates.
a–f, Schematic illustration of the algorithmic approach we use in order to infer turnover rates from the distribution of methylation patterns in clonal populations. Patterns are first split into two epiallelic groups, and then modelled as the outcome of a coalescent process in an exponentially growing population. This enables inference of gain and loss events together with the total time (in generations) in a methylated and unmethylated state for each CpG. These statistics can then be used to estimate the gain and loss rates, in particular when pooling together data from thousands of CpGs, with or without stratification for genomic and epigenomic factors of interest. We note that our implementation of the above approach uses several heuristics for enabling practical computation over millions of CpGs. For example, we use maximum-parsimony assumptions to approximate the posterior probabilities of ancestral CpG states for each locus. More details are available in Methods.
Extended Data Figure 9 Validating the turnover process.
a, Six clonal populations from two WI38 passages were generated. b, The passage 23 clones are described earlier, and are consistent on average with the matching passage 23 non-clonal population. Passage 27 clones are sampled from a population of cells that are on average 16 cell cycles older than passage 23 cells. As shown in the density plot, this affects the average population in these cells as predicted by our model (decreased methylation at high methylation loci, some increased methylation in low methylation regions). c, Stratification of the turnover rates inferred independently from passage 27 clones given replication time, lamina binding and nucleosome occupancy. The trends are indistinguishable from those observed for passage 23 clones (Fig. 3). d, K562 methylation gain and loss rates stratified according to nucleosome occupancy (light blue, depleted; blue, occupied), time of replication (x-axis of each panel), and LaminB1 interaction (different panels organized from low (left) to high (right)). e, Design of a time series clonal expansion experiment that was based on sampling a clonal population at three time points, each time extracting approximately 500,000 cells for profiling and 400,000 for continuous expansion. f, Average turnover rates (estimated from pooled statistics of all consistently covered CpGs) in independent UMI-bis libraries from two clones and three time points. Each time point was analysed using the same approach with the only difference being the assumed depth of the exponential coalescent model (22, 27 and 30 cell divisions). We recovered remarkably consistent rates from each of the time points. Interestingly, we also observed global differences in the rates in WI9 and WI11 clones (for both gain and loss). One possible explanation for this may be the existence of a rapidly growing subpopulation (which is sampled at higher depth than expected by the symmetrical exponential growth model we assumed).
Extended Data Figure 10 Bimodality and monoallelic distributions.
a–c, Three scenarios of emergence of bimodal methylation pattern distributions. A monoallelic pattern is predicted to be maintained in monoclonal populations, while loci that are differentially methylated in a subpopulation may become coherent following a clonal bottleneck. Alternatively, rapid methylation turnover can combine with slow fluctuation in a local chromosomal state to generate a methylation landscape that is either methylated or unmethylated, without clonal stability. d, Examples of monoallelic patterns that are conserved in clonal populations for both ES cells and fibroblasts. e, Chromosomal distributions of monoallelic patterns. f, Examples of loci developing bimodal methylation in the fibroblast population but not in ES cells. g, Chromosomal distribution of fibroblast-specific monoallelic loci, reflecting strong enrichment of the X chromosome, as well as clustering in several autosomal loci. h, Epiallelic noise at 0 is defined as the fraction of partially methylated patterns in the loci with methylation content less than 50%. The noise at 1 is defined reciprocally (Methods). Shown are the cumulative distributions for epiallelic noise in high variance loci (blue). Controls are computed using loci with less than 20% average methylation (for controlling noise 0) and over 80% average methylation (for controlling noise 1).
Supplementary information
Supplementary Information
This file contains Supplementary Tables 1-2. (PDF 105 kb)
Rights and permissions
About this article
Cite this article
Shipony, Z., Mukamel, Z., Cohen, N. et al. Dynamic and static maintenance of epigenetic memory in pluripotent and somatic cells. Nature 513, 115–119 (2014). https://doi.org/10.1038/nature13458
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/nature13458
This article is cited by
-
Antagonistic interactions safeguard mitotic propagation of genetic and epigenetic information in zebrafish
Communications Biology (2024)
-
Extensive DNA methylome rearrangement during early lamprey embryogenesis
Nature Communications (2024)
-
Genomic hypomethylation in cell-free DNA predicts responses to checkpoint blockade in lung and breast cancer
Scientific Reports (2023)
-
Potential biomarkers for immunotherapy in non-small-cell lung cancer
Cancer and Metastasis Reviews (2023)
-
Maintenance of methylation profile in imprinting control regions in human induced pluripotent stem cells
Clinical Epigenetics (2022)
