The dynamics of genome-wide DNA methylation reprogramming in mouse primordial germ cells

Abstract

Genome-wide DNA methylation reprogramming occurs in mouse primordial germ cells (PGCs) and preimplantation embryos, but the precise dynamics and biological outcomes are largely unknown. We have carried out whole-genome bisulfite sequencing (BS-Seq) and RNA-Seq across key stages from E6.5 epiblast to E16.5 PGCs. Global loss of methylation takes place during PGC expansion and migration with evidence for passive demethylation, but sequences that carry long-term epigenetic memory (imprints, CpG islands on the X chromosome, germline-specific genes) only become demethylated upon entry of PGCs into the gonads. The transcriptional profile of PGCs is tightly controlled despite global hypomethylation, with transient expression of the pluripotency network, suggesting that reprogramming and pluripotency are inextricably linked. Our results provide a framework for the understanding of the epigenetic ground state of pluripotency in the germline.

Figure 1

Demethylation Dynamics in PGCs (A) Global CG methylation levels for each data set assessed by 5 kb tiling probes. Open circles represent each data point; lines represent the median value for the two samples per time point. Note that the two replicates of the J1 data point are technical replicates, and all others are biological replicates from pooled samples. Note that E9.5 PGCs are already fairly hypomethylated followed by a further gradual loss of methylation toward E13.5. De novo methylation is only observed at E16.5 in male PGCs. See also Figure S1 and Table S1. (B) Distribution of CG methylation levels across the genome (5 kb probes). Note that the loss of methylation in PGCs is observed across the entire percentile distribution. (C) Correlation between CG density and methylation levels. (D) Methylation levels for the whole genome (5 kb probes), CGIs, and maternal and paternal DMRs (DMR coordinates were taken from E12.5 embryos [Tomizawa et al., 2011]). The dashed line indicates the expected 50% methylation levels for a germline DMR. Note that maternal DMRs retain more methylation than the rest of the genome during global loss of methylation. Outliers are not shown. (E) Example plots for Nanog promoter (left), maternally methylated Igf2r DMR (middle), and paternally methylated H19 DMR (right). Each bar represents a single CG dinucleotide. Note that while the Nanog promoter shows early demethylation kinetics, CGIs within imprint DMRs undergo delayed demethylation. See also Figure S1.

Figure 2

Methylation Heatmap for Various Genomic Features Each line represents a single probe within the indicated feature. High methylation levels are shown in red, low methylation levels are shown in green. Of all features analyzed, IAPs seem to retain most methylation at all time points, while all other features undergo substantial reprogramming.

Figure 3

CGIs with Late Demethylating Kinetics (A) Demethylation kinetics of CGIs across the genome (left) or on the X chromosome (right). Note that X-linked CGIs undergo slower demethylation than CGIs for the rest of the genome. Outliers are not shown. See also Figure S2. (B) Demethylation kinetics for CGI-containing promoters selected with >25% methylation in E6.5 epiblast and E9.5, E10.5, and E11.5 PGCs. These promoters show consistently higher methylation levels across all time points analyzed. See also Table S2, Table S3, Table S4, and Table S5. (C) Example plots for an X-linked CGI (left) and a CGI-containing promoter with slow demethylation kinetics. Methylation marks seem to be retained especially around the CGI. Each bar represents a single CG dinucleotide. See also Figure S2. (D) Gene ontology categories with a Bonferroni-corrected p value < 0.05 for promoters selected for >25% methylation in E9.5, E10.5, and E11.5 PGCs. CGI promoters of genes selected for higher methylation levels between E9.5 and E11.5 seem to have a functional connection. The number of genes in each group is indicated.

Figure 4

Demethylation Resistance (A) Methylation levels for CG dinucleotides within the consensus sequence of IAP1 (top left), IAP2 (bottom left), LINE1A (top right), and LINE1Tf (bottom right). Note that these elements retain substantial levels of methylation across all time points. Outliers are not shown. See also Figure S3. (B) Methylation levels for resistant CGIs selected with >25% methylation in E13.5 male (left) and female PGCs (right) without an IAP in close proximity (top) or near an IAP (bottom). Note that resistant CGIs with a distance of <2 kb to an IAP show consistently higher methylation levels across all time points, while CGIs without the presence of an IAP show variable methylation levels across all time points. See also Figures S3–S5. (C) Example figures for resistant CGIs without an IAP in close proximity (left) or with an IAP nearby (right). Each bar represents a single CG dinucleotide.

Figure 5

Mechanisms for Demethylation (A) Hairpin bisulfite heatmap of LINE1Tf. Total methylation levels are shown at the top. For each time point analyzed, each column represents one CG dyad along the LINE1Tf consensus sequence, and each row represents one sequencing read. The bars next to the heatmap represent the average distribution of fully methylated, hemimethylated, and unmethylated sites. Note that E9.5 and E10.5 PGCs have high levels of hemimethylated sites, which are then reduced to almost-complete hypomethylation at E13.5. (B) Shown is the distribution of methylated CG dinucleotides (meCG) at hemimethylated sites across the top (x axis) and bottom (y axis) strands of the LINE1Tf consensus sequence assessed by hairpin bisulfite sequencing. The LINE1Tf consensus sequence contains five CG dinucleotides, and the numbers 0–5 on the axis refer to the amount of meCGs on each strand and contain no position information. The values in the heat diagram represent the number of instances with the respective number of meCGs observed on the top and bottom strands. Shown is a simulation of the distribution of meCG within hemimethylated sites in the case of passive DNA demethylation (left) and active demethylation (middle). Note that with passive demethylation, all meCGs are located on the top strand, while the bottom strand is completely unmethylated and contains 0 meCGs and vice versa. For active DNA demethylation, a strand-independent distribution was simulated that leads to methylated and unmethylated CGs randomly distributed across both strands. The hairpin bisulfite data for E9.5 PGCs are shown in the right panel, and there is a strong strand bias for meCGs toward either top or bottom strand highly similar to the outcome for the simulation of passive DNA demethylation. Instances with meCGs distributed across both strands are rare in E9.5 PGCs. See also Figure S6. (C) Expression analysis of the DNA methylation machinery. Single-cell microarray data for ESCs (Vincent et al., 2011) and PGCs (Kurimoto et al., 2008) were reanalyzed (left, see the Experimental Procedures for more detail). RNA-Seq data for ESC and embryoid body (EB) (Cloonan et al., 2008) and PGCs of various time points are shown on the right. Whiskers represent the interquartile range of variation between replicates. Note that while Dnmt1 is continuously expressed, Np95 is transcriptionally downregulated in early PGCs. See also Figure S7. (D) Immunofluorescence staining for DNA (blue), Dnmt1 (green), and Np95 (red). Shown are immunostainings and RGB profiles created with Zeiss LSM software. Scale bars represent 10 μM in all images, and where RGB profiles are shown, the red line across a cell represents the midline along which the signal intensity is traced for each pixel and the profile is plotted below. Shown are stainings for Np95 and Dnmt1 in E13.5 male PGCs and E14 ESCs. In cycling PGCs, Dnmt1 localizes to the nucleus while Np95 is preferentially located in the cytoplasm. In ESCs, both Dnmt1 and Np95 localize to the nucleus. This suggests that in ESCs, the subcellular localization of Dnmt1 and Np95 is linked during S phase, while this dynamic pattern may be uncoupled in PGCs. See also Figure S7.

Figure 6

Dynamics of Transcriptomic Reprogramming (A) Distribution of expression values for the PGC RNA-Seq data sets. ESC and EB data sets were included for comparison (Cloonan et al., 2008). Even the most hypomethylated samples with E13.5 PGCs have an orderly expression program similar to that of all other PGC samples and the ESC and EB data sets. (B) Expression profiles for the pluripotency cluster (left) and the meiosis cluster (right). Error bars represent the standard deviation of measure across all probes within the cluster, and example genes are shown underneath for each cluster. Tet1 targets within the pluripotency cluster are highlighted by an asterisk. See also Figure S8 and Table S6. (C) Expression of LINE1s. Shown is the percentage of all RNA-Seq reads that map to the LINE1Tf and LINE1A consensus sequence and also as a comparison to the sequence of Cnpy3 and Pdia5, two single-copy genes that are expressed at constant levels across the time course. Note that the two LINE1 elements have higher expression levels at E16.5 in female PGCs than in any other data set and also than the two single-copy genes. Results for RNA-Seq data from ESCs and mouse lung fibroblasts (MLFs) (Guttman et al., 2010), J1 (Ficz et al., 2011), and Dnmt TKO ESCs (KO for Dnmt1, Dnmt3a, and Dnmt3b [Karimi et al., 2011]) are shown for comparison.