DNA methylation dynamics of the human preimplantation embryo

Abstract

In mammals, cytosine methylation is predominantly restricted to CpG dinucleotides and stably distributed across the genome, with local, cell-type-specific regulation directed by DNA binding factors. This comparatively static landscape is in marked contrast with the events of fertilization, during which the paternal genome is globally reprogrammed. Paternal genome demethylation includes the majority of CpGs, although methylation remains detectable at several notable features. These dynamics have been extensively characterized in the mouse, with only limited observations available in other mammals, and direct measurements are required to understand the extent to which early embryonic landscapes are conserved. We present genome-scale DNA methylation maps of human preimplantation development and embryonic stem cell derivation, confirming a transient state of global hypomethylation that includes most CpGs, while sites of residual maintenance are primarily restricted to gene bodies. Although most features share similar dynamics to those in mouse, maternally contributed methylation is divergently targeted to species-specific sets of CpG island promoters that extend beyond known imprint control regions. Retrotransposon regulation is also highly diverse, and transitions from maternally to embryonically expressed elements. Together, our data confirm that paternal genome demethylation is a general attribute of early mammalian development that is characterized by distinct modes of epigenetic regulation.

Extended Data Figure 1. Isolation of human preimplantation embryos for DNA methylation mapping

a. Three replicates of D6 embryos, ranging in inputs from 3 to 5 embryos, were thawed, screened for proper morphology, independently isolated from the zona pellucida and pooled prior to RRBS profiling. Embryos are displayed prior to purification. b. Two replicates of 18 and 19 human D3 cleavage stage embryos were thawed, screened for proper morphology, assessed for embryonic stage/cell number, and purified from the zona pellucida prior to RRBS profiling. Embryos are displayed prior to purification. c. Cell numbers from thawed cleavage stage embryos ranged from 4-11 cells per embryo with a median of 8 (+/- 1.6 standard deviation) cells. Within each replicate, only 3 embryos demonstrated onset of compaction at the time of collection. Red line signifies the median, boxes and whiskers the 25^th/75^th and 2.5^th/97.5^th percentiles, respectively.

Extended Data Figure 2. Assembly of a genome wide DNA methylation time series through human early development and over ESC derivation

a. Summary of RRBS libraries generated, with number of biological replicates (n), number CpGs captured at 1×, 5×, and 10×, mean/median methylation values for 100 bp tiles estimated from CpGs covered at ≥5×, and mean Euclidean distance and Pearson correlation across biological replicates for these tiles. b. Pearson correlation matrix for sperm, early embryonic, and fetal tissue samples. c. Clustering of gametic, somatic, and preimplantation methylation profiles segregate according to their global DNA methylation landscape, with Sperm/Somatic tissue forming an adult methylation cluster that contrasts the unique epigenetic landscape present in preimplantation embryos. d. Summary of RRBS libraries generated for ESC derivation, with number of biological replicates (n), number CpGs captured at 1×, 5×, and 10×, mean/median methylation values for 100 bp tiles estimated from CpGs covered at ≥5×, and mean Euclidean distance and Pearson correlation across biological replicates for these tiles. hESC ref refers to a reference collection of previously assayed ESC lines as part of the NIH Roadmap Epigenomics Project (Methods). Human ICM/TE were isolated through laser-assisted microdissection. e. Pearson correlation matrix for human samples used to model ESC derivation. A consistent signature is rapidly acquired by the outgrowth stage (p0) and stably maintained over additional passages. f. Methylation histograms for 100 bp tiles for human blastocysts and dissected ICM/TE tissue show minimal global difference, which is also observed when comparing previously assayed, immunosurgically purified mouse ICM to mechanically dissected ICM and TE. g. Boxplots of the change in methylation (Δ methylation) for 100 bp tiles from cleavage to the blastocyst stage show passive demethylation of DNA methylation, particularly for regions that exhibit the highest methylation levels at this stage. Red line signifies the median, boxes and whiskers the 25^th/75^th and 2.5^th/97.5^th percentiles, respectively.

Extended Data Figure 3. Characterization of ESC derivation to in vivo pluripotent tissues in mouse

a. Global methylation histograms of 100 bp tiles for human ICM and p5 ESCs (rows) compared against mouse preimplantation and postimplantation embryos (ICM, E6.5 Epi), as well as with extraembryonic ectoderm (ExE) (columns), demonstrate the rapid acquisition of an epiblast-like, somatic methylation pattern upon ESC derivation in human. b. Regions that discriminate mouse ICM from E6.5 epiblast were used to assign human ICM and ESCs to an equivalent in vivo pluripotent state for orthologous features. The position along the axis from preimplantation (ICM) to postimplantation (Epi) pluripotency represents the proportion of regions in a set that resemble one state versus the other. For all feature sets, human ESCs rapidly establish an epiblast identity, maintaining this signature from the outgrowth stage over ensuing passages. hESC Ref refers to a reference collection of previously assayed ESC lines as part of the NIH Roadmap Epigenomics Project (Methods).

Extended Data Figure 4. Inverse correlation between expression and promoter methylation is retained during human preimplantation

a. Scatterplots of oocyte, preimplantation embryo and ESC derivation gene expression compared to promoter methylation display a canonical negative correlation, even during preimplantation where the range of promoter methylation values is contracted by global hypomethylation. b. Box plots of gene expression values for genes significantly upregulated by ≥ 2 fold from oocyte to 8 cell compared to non-dynamic genes and categorized by promoter methylation dynamics. Genes that are both demethylated and upregulated are associated with induction from a silenced state, while those that are demethylated but not upregulated display only basal level transcription that is significantly lower than observed in promoters that are not demethylated. Bold line signifies the median, boxes and whiskers the 25^th/75^th and 2.5^th/97.5^th percentiles, respectively. c. Gene expression dynamics following fertilization for hypermethylated sperm promoters demethylated ≥0.5 by the cleavage stage compared to the rest of promoters (Other). 123 of 541 (22%) demethylated promoters demonstrate significant upregulation (≥2-fold) compared to only 13.6% of other promoters. Moreover, the ratio of upregulated to downregulated genes in the demethylated set substantially favors zygotic activation, while other promoters include more downregulated maternal transcripts (Odds Ratio = 1.877, p = 1.344 × 10⁻⁸, hypergeometric test). d. RNA-seq track of the pluripotency promoting, zygotically induced gene POU5F1, whose promoter is demethylated from 0.59 in sperm to 0.02 in cleavage, concurrent with its transcriptional induction.

Extended Data Figure 5. Local retention of DNA methylation is similar for introns over human and mouse preimplantation

a. Introns are clustered according to their dynamics in human and the methylation of orthologous regions are tracked in mouse. Divergence is predominantly restricted to intermediately methylated features in human sperm that are generally hypermethylated in mouse. Δ methylation heatmap displays the difference in methylation values between equivalent preimplantation timepoints, with ESCs in human serving as a proxy for comparison to the E6.5 epiblast in mouse. Deviation is most apparent for intermediately methylated human sperm introns, where they are less methylated than in mouse. RMSK included: repeat masker annotated regions included. b. When repetitive elements are removed from the calculation of intron methylation, the apparent divergence between mouse and human values is lost. Methylation and Δ methylation heatmaps are as in a. Gray denotes missing values (m.v.) where estimates for intronic methylation were exclusively derived from repetitive elements. RMSK-free: repeat masker annotated regions excluded. c. Violin plots of the two main dynamics (maintained vs demethylated) for sperm hypermethylated introns over human and mouse preimplantation after repetitive elements are removed. As observed for orthologous exons, regions that retain high methylation throughout human preimplantation are conserved, hypermethylated in both mouse gametes, and display maintained regulation as early as the zygote stage.

Extended Data Figure 6. Genomic characterization of transient maternally contributed imprint-like regions

a. Heatmap of 100 bp tiles in mouse preimplantation identified using the same criteria as applied to human (Methods). This criteria, which assumes limited de novo methylation, identifies 2,044 tiles in mouse where methylation is ≥ 0.2 in both 8 cell and the ICM, there is ≥0.2 methylation difference between the ICM and sperm, and this difference is significant via t test, (q-value < 0.05). 89% of those tiles that are captured in the mouse oocyte are monoallelically inherited and show significant differences between the gametes by t-test, providing an empirical upper bound on the False Discovery Rate for this strategy when applied to human of ≤ 0.11, assuming the underlying principles of imprint regulation are the same as in mouse. b. The proportion of 100 bp tiles, classified according to their resolution in ESCs, for each genomic feature presented in Figure 3a. c. Cumulative density function (CDF) plot of the distance to the nearest annotated TSS for CGI DMRs that resolve to hypomethylation, intermediate/variable methylation, or hypermethylation. There is a discrepancy in genomic location between those that resolve to hypomethylation, of which a sizable fraction are in the TSS, and those that do not, which are generally enriched further downstream. d. Boxplots of CpG density for CGI DMRs that resolve to hypomethylation, intermediate/variable methylation, or hypermethylation paired with comparable non-DMR CGIs (Somatic). Those resolving to hypomethylation have higher CpG densities than those that resolve to intermediate/variable or hypermethylation, but have slightly lower CpG density than non-DMR, constitutively hypomethylated CGIs. Alternatively, while CGIs that resolve to hypermethylation show a lower CpG density than other DMRs, they show higher density than non-DMR hypermethylated islands, suggesting some level of protection against deamination as an attribute of their uniquely hypomethylated status in the male germline. e. Pie charts of cross species alignment and CGI status of human CGI DMRs into mouse. Those that resolve to hypomethylation are more often conserved in mouse and more frequently retain their CGI status, whereas those resolving to hypermethylation are less conserved. Moreover, intermediate/variable and hypermethylation-resolving regions that do align are less frequently retained as CGIs suggesting that hypomethylation specific to the male germline is insufficient to protect these regions from progressive deamination over time. 368, 166, and 260 CGIs comprise the hypo, intermediate/variable, and hyper methylation sets, respectively,

Extended Data Figure 7. Generation of single blastocyst libraries confirm the monoallelic behavior of putative maternal DMRs

a. Summary of two single blastocyst RRBS libraries. Number of CpGs captured at 1×, 5×, and 10×, mean/median methylation values for 100 bp tiles estimated from CpGs covered at ≥5×, and mean Euclidean distance and Pearson correlation when single blastocyst replicates are compared to the pooled blastocyst timepoint. b. Histograms of DNA methylation for 100 bp regions captured for each single blastocyst replicate. c. The ratio of reference allele to alternative allele for single nucleotide polymorphisms (SNPs) called as heterozygous in each blastocyst replicate. d. For the 4,492 and 5,118 SNPs that were considered as heterozygous within each single blastocyst, 10,068 and 11,415 single CpGs could be assigned to alleles. Scatterplots depict untracked methylation values for these CpGs against their normalized methylation values, which are the average of their monoallelic methylation states.

Extended Data Figure 8. The somatic promoter of DNMT1 is maternally methylated in human and mouse

a. Plots of single CpG methylation for DNMT1, including a CGI over the somatic promoter that behaves as a transient, preimplantation-specific DMR in both human and mouse. In mouse, hypermethylation of this island corresponds to its transcriptional readthrough and exclusion as part of an oocyte-specific isoform (Dnmt1-o) that is not annotated in human. Annotated CGIs and species conservation tracks are included for reference below. b. Heatmap of orthologous ICR dynamics over human and mouse preimplantation. Of those that map between species and are captured by RRBS, all but one (PEG10) behave identically.

Extended Data Figure 9. Repetitive element regulation during human and mouse preimplantation

a. Violin plots for LTRs over human and mouse development. In human, LTRs demonstrate a bimodal distribution in sperm. Hypermethylated LTRs display a range of demethylation in the early embryo that reflects the dynamics of subfamilies. Upon ESC derivation, and within fetal tissues, LTRs become stably hypermethylated. Alternatively, during mouse preimplantation, LTRs are consistently hypermethylated in sperm and generally retain methylation over preimplantation. E6.5 Epi and E6.5 ExE refer to dissected epiblast and extraembryonic tissue from E6.5 embryos. b. Violin plots for LINEs over human and mouse development. In human sperm, LINEs are unstably hypermethylated, with discrete populations methylated with a mean of ∼0.75, ≥0.9, and a small subpopulation showing gametic escape from high methylation. Alternatively, LINEs are indiscriminately hypermethylated in mouse sperm. In both species, several populations of elements demonstrate different extents of demethylation during preimplantation, including many that retain higher levels in cleavage and only minor, passive depletion into blastocyst. Upon human ESC derivation or during mouse implantation, elements are generally remethylated, though only partially for those elements that are demethylated after fertilization. Hypermethylation is complete in fetal tissue. In human, these discrete dynamics can be attributed to to the unstable methylation for L1HS-L1PA3a subfamilies while, in mouse, the currently active L1Md_Tf and L1Md_Gf subfamilies are similarly demethylated and elements of the independently emerging L1Md_A lineage remain methylated. c. Violin plots for SINEs highlight intermediate methylation in sperm in both species, though more so for humans. After fertilization, SINE methylation rapidly diminishes to near complete hypomethylation over preimplantation, similar to what is observed for intergenic sequence, before complete hypermethylation during ESC derivation in human or in postimplantation mouse E6.5 embryos. Taken globally, SINEs appear to be uniformly regulated regardless of subfamily, though differences in regulatory status for specific SINE elements may be reflected by their surrounding genomic context. Unfortunately, such inferences require higher genomic resolution than is currently available to distinguish the dynamics of specific integrations. d-g. Violin plots of the four major LTR families present in mouse over the complete preimplantation timeline. ERV1 elements (d) are hypermethylated in sperm and display a range of demethylation following fertilization and prompt remethylation upon implantation. In mouse, ERVK elements (e) are emergent and largely consist of the dominating, constitutively hypermethylated IAP subfamilies. ERVL and MalR (ERVL-MalR) elements (f and g), the evolutionarily oldest mammalian LTRs, are hypermethylated in sperm and rapidly demethylated after fertilization frequently in association with their rapid zygotic induction. h. Distribution (as boxplots) of per element expression and CpG density at different methylation levels for LTR12c demonstrates negative correlation between methylation and expression. On average, LTR12C is hypomethylated in sperm and the early embryo, but demonstrates a consistent range of values at the level of single elements, with least methylated elments contributing the most to LTR12c expression. The CpG density of these elements corresponds to their degree of hypomethylation, suggesting that escape from de novo methylation during spermatogenesis and preimplantation is maintained for specific elements over generations. Once targeted, element expression is apparently restricted and its CpG density decays correspondingly. During ESC derivation, the kinetics of LTR12c methylation is more rapid for those of lower CpG density, as evident from p0 to p5 in the ESC lines. DNA methylation in the early embryo is therefore not exclusive to the regulation of different ERV1 subfamilies, but also affects the contribution of single elements to the broader transcriptional pattern. Bold line signifies the median, boxes and whiskers the 25^th/75^th and 2.5^th/97.5^th percentiles, respectively. Expression is calculated as the number of fragments per million that align to a given element divided by its length in kb (FPKM).

Extended Data Figure 10. L1PA subfamily dynamics during human early development

a. Expression composite averaged by genomic representation for L1HS through L1PA7 from oocyte through preimplantation and ESC derivation. Dynamic expression within the L1PA phylogeny is restricted to the same subfamilies that are demethylated by cleavage. The position of each respective 5′ UTR, the functional promoter for LINEs, is highlighted in the legend. Beneath these composites is the genomic representation to the full length consensus for each annotated L1PA subfamily, which demonstrates relative equivalence of 5′ UTR representation across different subfamilies, but an increasing proportion of truncated 3′ fragments with subfamily age (Methods). b. The frequency of CpGs within aligned L1PA subfamilies, including 5′ UTR, Orf1/2, and 3′ UTR. CpGs are primarily enriched within the 5′UTR/promoter and become progressively CpG depleted with element age. c. Complete composite plot of cleavage stage methylation values across aligned 5′ UTRs from L1HS through L1PA7 as in Figure 6d. The multiple sequence alignment for each subfamily to the assembled consensus is visualized below each composite, with blue corresponding to conservation, black to divergence, and white to gaps or deletions. The x-axis represents position along the 5′ UTR and a portion of ORF1 for the L1HS consensus. CpG Frequency describes the level of conservation for individual CpGs found within single elements to the consensus. The ∼130 bp sequence present from L1PA7 to L1PA3b and absent from L1PA3a to L1HS is highlighted in pink, while two older sequences specific to L1PA7 are highlighted in gray. d. Percent identity to the consensus for the extracted ∼130 bp insert sequence in elements from L1PA7 through L1PA3b. Mean nucleotide identity to the consensus is 85%, with a median of 89%.

Figure 1. Human preimplantation embryos are globally hypomethylated

a. Top. DNA methylation across 100 bp tiles for human sperm, preimplantation embryos, including the ICM and TE, ESC derivation from outgrowth to 5^th passage, and somatic fetal tissues representing all germ layers. Gray highlights the average. Bottom. Boxplots of methylationat different local CpG densities. Bulls-eye signifies the median, boxes and lines the 25^th/75^th and 2.5^th/97.5^th percentiles, respectively. b. Bar plots of 100 bp tiles segregated by non-repetitive (unique) or repetitive designation and binned by methylation status. Sp, sperm; Cl, cleavage; Bl, blastocyst;Som, somatic. “ESC” and “Som” refer to the average of these timepoints. c. Non-repetitive 100 bp tiles are clustered via k-means into 10 dynamics. Sperm hypermethylated sequences follow three general trajectories: persistent maintenance, incomplete or complete demethylation. Other dynamics include sperm specific hypermethylation and sperm/early embryonic hypomethylation that is de novo methylated in ESCs. Finally, 3,586 tiles are hypomethylated in sperm and ESCs but methylated in embryos, representing transient imprint-like signatures. d. Dynamics for sperm hypermethylated, non-repetitive tiles as clustered in (c). Left heatmap, per cluster average of tiles. Right heatmap, –log₁₀ p value of hypergeometric enrichment for each cluster for intergenic, exonic, intronic, CGI, or TSS annotations using sperm hypermethylated regions as the background. e. Violin plot forsperm hypermethylated intergenic (inter), exonic, and intronic features. f. The OBSCN gene exhibits high inter- and intra-genic methylation and an unmethylated promoter in sperm and ESCs. In cleavage embryos, a 130 kb region, highlighted in blue, remains specifically methylated while the periphery is demethylated.

Figure 2. Human preimplantation dynamics are globally similar to mouse

a. Histograms of methylation changes (Δ methylation) for 100 bp tiles across human and mouse preimplantation from fertilization (Sp to Cl) through preimplantation (Cl to Bl) to global remethylation at implantation, as measured from blastocyst to ESC in human and ICM to E6.5 epiblast in mouse (Bl to ESC/Epi). b. Exons clustered by dynamics in human with equivalent methylation values for orthologous sequences in mouse. Δ methylation heatmap displays the difference in methylation for matched timepoints. c. Violin plots of orthologous human sperm hypermethylated exons classified as maintained vs demethylated and measured over human and mouse preimplantation.

Figure 3. Transient maternal DMRs target a divergent set of CpG island promoters

a. Heatmap of 5,265 100 bp tiles consistent with maternally contributed monoallelic methylation (Methods). Tiles are partitioned according to their hypo (≤0.2), intermediate/variable (0.2<x<0.8) or hypermethylated (≥0.8) resolution in ESCs. Feature annotations are included as separate heatmaps. b. Boxplots of CGI DMR methylation for two independent single blastocysts, with heterozygous SNP-linked CpGs highlighted. Within each replicate, 31 and 33 CGI DMRs contain CpGs that could be assigned to parental loci. In each case, DNA methylation is restricted to only one of the two alleles. Untracked refers to the inferred methylation status prior to haplotype segregation. Red line signifies the median, boxes and whiskers the 25^th/75^th and 2.5^th/97.5^th percentiles. c. Single CpG track of a conserved preimplantation-specific DMR in human and mouse. Human blastocyst data includes information from the pooled sample as well as for a single blastocyst replicate (purple) with allele-tracked methylation for 10 CpGs highlighted in pink and blue. Annotated CGIs are included below. d. Resolution of CGIs that behave as maternal DMRs in human and mouse. e. Heatmap of orthologous ICRs over human and mouse preimplantation development. f. Orthologous hypomethylation-resolving CGID MRs in human and mouse share only 13 equivalently regulated regions. When methylation values of mouse or human specific DMRs are tracked in the alternate species, they are constitutively hypomethylated, indicating that oogenesis targets equivalent genomic features but at species-specific sequences. PreImp refers to the average value for cleavage and blastocyst in human or 8 cell and ICM in mouse.

Figure 4. LTR subfamily dynamics are divided into early and late preimplantation phases

a. Violin plots forthe four LTR families present in human over early development and ESC derivation. b. Pie charts of LTR family expression calculated as the number of fragments per million (FPM) that align to elements with in the family. c. Mean methylation of notable ERV1, ERVK, and MalR subfamilies. Three ERV1 subfamilies are included to represent discrete dynamics: gamete/early embryonic hypomethylation (LTR12c), constitutive methylation (HERV9-INT) and rapid demethylation (LTR7). The ERVK subfamily LTR5Hs is also demethylated. d. Expression dynamics for the same subfamilies in (c). LTR12c is expressed early and downregulated in the blastocyst. Alternatively, LTR7 is expressed throughout, but upregulated in the blastocyst and maintained in hESCs, where it accounts for the majority of ERV1 transcripts. Like LTR7, LTR5Hs is only intermediately methylated during ESC derivation and is embryonically induced. Alternatively, the ERV1 HERV9-INT remains repressed. MLT1H2 is the prevailing MalR transcribed in the oocyte and is lost after fertilization. Expression is the fragments per million that align to subfamily elements, divided by the kb annotated as the subfamily in the genome (FPKM).

Figure 5. Emergent L1PA subfamilies escape DNA methylation-based repression during preimplantation

a. Pie charts of the LINE expression divided into the L1PA subfamily, other LINE1 and LINE2 subfamilies. Total expression is calculated as the number of fragments per million (FPM) that align to family elements. b. Mean methylation values for the most recent L1PA subfamilies. In cleavage embryos, L1HS through L1PA3 are demethylated and maintain these levels through the blastocyst. c. Expression dynamics for the same subfamilies in (c) over preimplantation and in ESCs. The three youngest L1PA subfamilies are induced by the 8cell stage. Expression is the number of fragments per million that align to subfamily elements, divided the kb annotated as the subfamily in the genome (FPKM). d. Composite plot of cleavage stage methylation values across aligned 5′ UTRs in L1PA subfamilies. The composite for L1PA3 is split by the presence (red) or absence (blue) of a ∼130 bp sequence that distinguishes L1PA3b from L1PA3a and demarcates methylation values between older and newer subfamilies (highlighted in pink). Multiple sequence alignment for each subfamily to the assembled consensus is below each composite, with blue corresponding to conservation, black to divergence, and white to gaps or deletions. The x-axis represents position along the L1HS 5′ UTR and a portion of ORF1. CpG Frequency describes per CpG conservation within single elements to the consensus. Two older sequences specific to L1PA7 are highlighted in gray. e. Boxplot of L1PAmethylation in cleavage embryos, sorted by the presence of the ∼130 bp sequence for all elements and L1PA3 specifically. Preimplantation methylation is higher for elements that contain this insert. Bold line signifies the median, boxes and whiskers the 25^th/75^th and 2.5^th/97.5^th percentiles. f. Expression composite of full-length insert deleted L1PA3a and insert containing L1PA3b subfamilies in oocyte, 8 cell and blastocyst stage embryos. Transcriptional induction is not apparent until after fertilization and is specific to L1PA3a. Read count is the read coverage normalized by total reads (Methods).