Role of UHRF1 in de novo DNA methylation in oocytes and maintenance methylation in preimplantation embryos

The methylation of cytosine at CG sites in the mammalian genome is dynamically reprogrammed during gametogenesis and preimplantation development. It was previously shown that oocyte-derived DNMT1 (a maintenance methyltransferase) is essential for maintaining and propagating CG methylation at imprinting control regions in preimplantation embryos. In mammalian somatic cells, hemimethylated-CG-binding protein UHRF1 plays a critical role in maintaining CG methylation by recruiting DNMT1 to hemimethylated CG sites. However, the role of UHRF1 in oogenesis and preimplantation development is unknown. In the present study, we show that UHRF1 is mainly, but not exclusively, localized in the cytoplasm of oocytes and preimplantation embryos. However, smaller amounts of UHRF1 existed in the nucleus, consistent with the expected role in DNA methylation. We then generated oocyte-specific Uhrf1 knockout (KO) mice and found that, although oogenesis was itself unaffected, a large proportion of the embryos derived from the KO oocytes died before reaching the blastocyst stage (a maternal effect). Whole genome bisulfite sequencing revealed that blastocysts derived from KO oocytes have a greatly reduced level of CG methylation, suggesting that maternal UHRF1 is essential for maintaining CG methylation, particularly at the imprinting control regions, in preimplantation embryos. Surprisingly, UHRF1 was also found to contribute to de novo CG and non-CG methylation during oocyte growth: in Uhrf1 KO oocytes, transcriptionally-inactive regions gained less methylation, while actively transcribed regions, including the imprinting control regions, were unaffected or only slightly affected. We also found that de novo methylation was defective during the late stage of oocyte growth. To the best of our knowledge, this is the first study to demonstrate the role of UHRF1 in de novo DNA methylation in vivo. Our study reveals multiple functions of UHRF1 during the global epigenetic reprogramming of oocytes and early embryos.

CG methylation is dynamically reprogrammed during mammalian gametogenesis and preimplantation development.In mouse germ cell development, CG methylation is first erased in primordial germ cells, partly due to the low expression of UHRF1 (Ubiquitin-like with PHD and RING finger domains 1) [9], a factor that is essential for maintenance methylation at CG sites [10,11] (see below).The hydroxylation of 5mC also plays a role in removing CG methylation through active or passive demethylation, particularly at the imprinting control regions (ICRs) [12,13], which dictate the parental-origin-specific expression of the imprinted genes in somatic cells [14].Subsequently, in the female germline, an oocyte-specific methylation pattern is established in GOs by a de novo DNA methyltransferase (DNMT) complex composed of DNMT3A and DNMT3L [4,5,[15][16][17].De novo methylation continues until the fully grown oocyte (FGO) stage, with the genome of the oocyte accumulating both CG methylation and non-CG methylation [3][4][5].It is known that the highly methylated regions in oocytes correspond to the actively transcribed regions [4,18,19], which are marked by histone H3 lysine 36 trimethylation (H3K36me3) [20].Since DNMT3A recognizes H3K36me3 through its PWWP domain [21], the enzyme appears to methylate regions that are premarked with this modification.De novo methylation in oocytes also requires histone replacement, since the disruption of the H3.3 chaperone HIRA results in a global reduction in CG methylation [22].
After fertilization, with the exception of certain regions such as the ICRs and a subset of retrotransposons, the gamete-specific methylation patterns are globally erased [4,23].Notably, the paternal genome undergoes 5mC hydroxylation, followed by active or passive demethylation [24][25][26][27].Several factors, including ZFP57, TRIM28, DPPA3, and maintenance-type DNA methyltransferase DNMT1 are required for maintaining CG methylation at the ICRs [28][29][30][31][32][33] against the wave of hydroxylation-dependent and hydroxylation-independent demethylation.In a study using oocyte-specific Dnmt1 knockout (KO) mice, we previously reported that, although DNMT1 is mainly localized in the cytoplasm, a small amount of oocyte-derived maternal DNMT1 in the cell nucleus engages in CG maintenance methylation at the ICRs in preimplantation embryos [31].
Mouse UHRF1 is a multi-domain protein that is essential for the CG maintenance methylation in proliferating somatic cells [10,11].The conventional KO of Uhrf1 results in an embryonic lethal phenotype, which is accompanied by global genomic hypomethylation [10].UHRF1 contains five major protein domains: an ubiquitin-like (UBL) domain, a tandem Tudor domain (TTD), a plant homeodomain (PHD) finger, a SET and RING associated (SRA) domain, and a really interesting new gene (RING) finger [34][35][36].UHRF1 recognizes hemimethylated CG sites through its SRA domain [37][38][39], and ubiquitinates H3K18 and H3K23 via the E3 ubiquitin ligase activity of its RING finger [40,41].Subsequently, DNMT1 recognizes H3K18 and/or other ubiquitinated lysine residues and methylates the unmodified C of hemimethylated CG sites to copy the existing methylation patterns.In addition, the TTD and PHD finger of this protein recognize H3K9me2/me3 [35,36,42,43] and unmodified H3R2 [44,45], respectively, suggesting their roles in the proper localization of this protein.
Unlike DNMT1, however, the role of UHRF1 during oogenesis and preimplantation development has not been explored.We therefore generated oocyte-specific Uhrf1 conditional KO mice and examined the impact on DNA methylation in oocytes and preimplantation embryos.Our study revealed an expected role of maternal UHRF1 in preimplantation embryos and an unexpected new function in GOs.

The subcellular localization of UHRF1 in oocytes and preimplantation embryos
To reveal the role of UHRF1 proteins in mouse oocytes and preimplantation embryos, we first examined the expression and subcellular localization of UHRF1 by immunostaining.Like DNMT1 [31], UHRF1 was mainly localized in the cytoplasm of GOs and preimplantation embryos (Fig 1A and 1B).However, a weaker signal was detected in the nucleus.To examine the timing of the zygotic activation of the Uhrf1 and Dnmt1 genes, we reprocessed published polyA(+) mRNA-seq data from F1 hybrid embryos [46], where expressed alleles could be distinguished based on the presence of single nucleotide polymorphisms (SNPs).While the levels of maternal Uhrf1 and Dnmt1 mRNAs had decreased by the 16-cell stage, the total levels of both mRNAs increased after this stage (S1

The role of maternal UHRF1 in the CG maintenance methylation in preimplantation embryos
In preimplantation embryos, both the paternal and maternal genomes are globally demethylated, while the ICRs and certain retrotransposons (such as intracisternal A particle [IAP] elements) retain CG methylation [4,23,31,49,50].To examine whether the maternal UHRF1 proteins play a role in CG maintenance methylation in these sequences, whole genome bisulfite sequencing (WGBS) was performed with control [Uhrf1 2lox/+ ], Uhrf1 mat-KO, and Dnmt1 mat-KO blastocysts generated by IVF.Only well-developed Uhrf1 mat-KO blastocysts were pooled (S3A Fig, indicated as expanded) and subjected to the analysis.Since KO females were of the C57BL/6J background (Mus musculus domesticus) and the wild-type sperm was from JF1 males (Mus musculus molossinus), the parental alleles were distinguishable where SNPs were available.We used the post-bisulfite adaptor tagging (PBAT) method, which was amplification-free and applicable to a limited amount of DNA, to construct the WGBS libraries [51].
Our WGBS with replicate samples basically confirmed good reproducibility, but the Uhrf1 mat-KO blastocysts showed a lower correlation, likely due to slight differences in developmental stage and extremely low levels of CG methylation, which are sensitive to small variations (S1 and S2 Tables and S4 Fig) .The global CG methylation levels of the control, Uhrf1 mat-KO, and Dnmt1 mat-KO blastocysts were 14.3%, 3.5%, and 5.7%, respectively (Fig 3A and S2 Table, combined replicate data).Thus, the absence of maternal UHRF1 reduced the level of global CG methylation to one-fourth of the control level.The fact that the global methylation decreased by more than 50% suggests that not only the maternal genome but also the paternal genome, which was derived from the wild-type sperm, was affected after fertilization.Indeed, this was confirmed by the allele-specific methylation analysis using SNPs (see later).The loss of CG methylation was milder in Dnmt1 mat-KO blastocysts.The non-CG methylation level was basically unaffected by the absence of either maternal UHRF1 or maternal DNMT1 (S2 Table ).
The analysis of CG methylation  in 12 (ten maternally methylated and two paternally methylated) of the 15 ICRs examined in the control blastocysts (S3D Fig); SNPs were not available for the rest of the ICRs.As we reported previously [31], the level of CG methylation was decreased to half that of the control blastocysts at the ICRs in Dnmt1 mat-KO blastocysts.In contrast, the CG maintenance methylation at the ICRs was more severely affected in Uhrf1 mat-KO blastocysts, to less than onefourth the level observed in the control blastocysts (Fig 3E and S3D Fig).The loss of methylation from both parental genomes in mat-KO blastocysts indicates a role for maternal Uhrf1 in the maintenance of CG methylation.The role of UHRF1 in de novo CG methylation during oocyte growth We next examined whether this protein has any effect on CG methylation in FGOs.We therefore performed WGBS with control [Uhrf1 2lox/2lox ] and Uhrf1 KO FGOs (S1 and S2 Tables and S4 Fig) .We found that the global CG methylation level was 30.8% in Uhrf1 KO FGOs, which was 7.9% lower in comparison to control FGOs (38.7%) (Fig 4A and S2 Table).This was surprising as no DNA replication occurs during oocyte growth (meiotic prophase I), during which only de novo methylation is observed.Since the CG methylation level in NGOs was reported to be 2.3% [5], these results suggested that UHRF1 was required for approximately one-fourth of the total increase in CG methylation (Fig 4A  increased hemimethylation [5].Although the proportion of hemimethylated sites among the highly methylated CG sites was 6.1% greater in Uhrf1 KO FGOs (12.9%) than it was in Dnmt1 KO FGOs (6.8%) (S5A Fig) (see Materials and Methods), this difference was not sufficient to explain the decrease in CG methylation that was observed in the Uhrf1 KO FGOs.
We previously reported that, in FGOs, nearly 65% of all 5mCs occurred at non-CG sites [5].This was confirmed in our control FGOs (S5B Fig and S2 Table ) and, surprisingly, non-CG methylation was decreased to 85% of the level of control FGOs in Uhrf1 KO FGOs (S5B-S5D Fig and S2 Table ).This corresponded to a 0.6% reduction in non-CG methylation on a per site basis (from 3.2% to 2.6%) (S2 Table ).Dnmt1 KO FGOs showed an increased non-CG methylation level, possibly due to the compensatory upregulation of Dnmt3a [5].With regard to the sequence context, the CHG and CHH sites (where H represents the base A, C, or T) were equally affected (S5B Fig) .These results strongly suggest that UHRF1 plays a role in de novo CG and non-CG methylation, not only through the recognition of hemimethylated CG sites, but also through an unknown mechanism.
To examine whether any regions of the genome are preferred for the de novo CG methylation involving UHRF1, we determined the CG methylation levels in 1-Mb windows and found that the levels in Uhrf1 KO FGOs decreased across the genome in comparison to control FGOs (Fig 4B

UHRF1 facilitates de novo methylation in inactive regions during late oocyte growth
As a first step to understand how UHRF1 is involved in de novo CG and non-CG methylation in oocytes, we attempted to explore the features of the regions affected in Uhrf1 KO FGOs in greater detail.We first identified the genomic regions (10-kb windows) with a CG methylation level of !40% in control FGOs and divided them into four groups: moderately (40-80%) methylated regions that were unaffected (decrease <20%) by Uhrf1 KO (Group 1); moderately methylated regions that were affected (decrease !20%) by KO (Group 2); highly (!80%) methylated regions that were unaffected by KO (Group 3); highly methylated regions that were affected by KO (Group 4) (Fig 5A and S3 Table).Of the moderately methylated regions (Groups 1 and 2), 35% of the cases were affected (Group 2) but, in contrast, only 9% of the highly methylated regions (Groups 3 and 4) were affected (Group 4) (Fig 5A ), confirming that the moderately methylated regions were preferentially affected.
Transcription-coupled de novo CG methylation is predominant in oocytes [4,18,19].To examine the relationship between transcription and UHRF1-dependent de novo CG methylation, we performed RNA-seq with replicate samples from control [Uhrf1 2lox/2lox ] and Uhrf1 KO FGOs (S4 Table ).Overall, the transcription profiles of these FGOs were very similar (R = 0.969, combined replicate data, S6D Fig) , and we confirmed that the highly methylated regions produced more transcripts (Fig 5B).Also, using previously published H3K36me3 chromatin immunoprecipitation-sequencing data from GOs (!30-65 μm) [20], we found that the highly methylated regions were preferentially marked with this transcription-coupled histone mark [52] (Fig 5B).We then examined the CG methylation changes in Uhrf1 KO FGOs in the transcribed regions and found that untranscribed or only lowly transcribed regions were the most severely affected (Fig 5C ).
Lastly, we used the published WGBS data from NGOs collected at the newborn stage [5] and GOs collected on postnatal days 12 (P12) and P15 [53] to examine the timing of de novo CG methylation in affected and unaffected regions.In comparison to the regions that were unaffected by Uhrf1 KO (Groups 1 and 3), the affected regions gained less methylation in the early phase of oocyte growth (between newborn NGO and P12 GO) and underwent greater The relative contribution to de novo CG methylation and maintenance CG methylation Since UHRF1 was found to be involved in de novo CG (and non-CG) methylation in GOs, we attempted to determine what proportion of the loss of CG methylation in Uhrf1 mat-KO blastocysts (Fig 3) was explained by the de novo activity and what proportion was explained by the maintenance activity.The de novo activity only affects the maternal genome, while the maintenance activity affects both parental genomes: we therefore determined the CG methylation level of each parental genome in control and Uhrf1 mat-KO blastocysts using SNPs.In the control blastocysts, the CG methylation level of the maternal genome was 14.3%, which was 2.3% lower than that in the paternal genome (16.6%) (S7A Fig) .(The observed levels were lower than those of the inner cell mass cells (20% and 21% for the maternal and paternal genome, respectively) [50], perhaps because trophectoderm cells display lower methylation.)In Uhrf1 mat-KO blastocysts, the CG methylation levels of the maternal and paternal genomes were decreased to 3.1% and 4.5%, respectively (S7A Fig) .(Since SNP-associated CG sites were more frequently found in the genic regions, where the CG methylation level was high, the average CG methylation level detected in the parental genomes (3.8%) was slightly higher than that in the whole genome (3.5%).) The CG methylation level in control and Uhrf1 KO FGOs was 38.7% and 30.8%, respectively (S2 Table ).These levels correspond to those of the maternal genome at the time of fertilization.Since the methylation level of the maternal genome was 14.3% in control blastocysts (see above), this genome undergoes a ×0.37 (14.3%/38.7%)methylation change, mainly due to replication-coupled dilution during preimplantation development (S7B Fig) .(Hydroxylationdependent demethylation selectively occurs on the paternal genome [24][25][26][27].)If there was no maternal effect caused by oocyte-specific Uhrf1 KO and if we apply the demethylation rate determined above to the maternal genome derived from Uhrf1 KO FGOs, then the expected methylation level would be 11.4% (= 30.8% ×0.37) (S7B Fig) .Thus, the de novo activity would account for 2.9% (= 14.3%-11.4%) of the total methylation loss of the maternal genome (11.2% [= 14.3%-3.1%]) in Uhrf1 mat-KO blastocysts).The rest of the loss (8.3% [= 11.2%-2.9%])would be attributable to the maintenance activity in preimplantation embryos (S7B Fig) .Although the precise values may be subject to variations between the samples (S4 Fig), these results show that the loss of CG methylation in the Uhrf1 mat-KO blastocysts can largely be explained by the loss of the maintenance activity in early embryos.

Discussion
In the present study, we found that the UHRF1 produced in oocytes has an important role in shaping the DNA methylation landscape in oocytes and preimplantation embryos.We first showed that, like DNMT1 [31], UHRF1 is mainly, but not exclusively, localized in the cytoplasm during oocyte growth and preimplantation development.Why should epigenetic regulators, which in most cases exert their function in the nucleus, exist in the cytoplasm during this period?For DNMT1, a few studies have addressed the mechanism of cytoplasmic localization in early embryos [54]; however its function in the cytoplasm is still unknown.Furthermore, CG maintenance methylation and other events that are known to involve UHRF1 (for example, DNA repair [55][56][57]) occur in the nucleus and thus the biological significance of cytoplasmic UHRF1 remains to be clarified.As expected, however, a small amount of UHRF1 did exist in the nucleus of oocytes and early embryos, as did DNMT1 [31].Interestingly, the nuclear localization of DNMT1 was dependent on UHRF1, as oocyte-specific Uhrf1 KO resulted in a greatly reduced DNMT1 signal in the nucleus.A previous study in ESCs showed that DNMT1 recruitment to specific target loci, but not nuclear localization, is dependent on UHRF1 [10,11].Thus, the nuclear entry and/or retention of DNMT1 is regulated differently in different cell types.
Uhrf1 mat-KO embryos showed partial preimplantation lethality, a phenotype that is much more severe than that of Dnmt1 mat-KO embryos.Our WGBS showed that the global CG methylation level is more greatly reduced in Uhrf1 mat-KO blastocysts (×0.25 in comparison to control blastocysts) than it is in Dnmt1 mat-KO blastocysts (×0.36 in comparison to control).The loss of CG methylation at the ICRs and IAP elements in the Dnmt1 mat-KO embryos was approximately ×0.50 (see also [31]); however, the loss in Uhrf1 mat-KO embryos was greater (×0.25 or less).These results strongly suggest that maternal UHRF1 is essential for imprint maintenance and IAP repression in preimplantation embryos.It is known that the repressed alleles of the ICRs and IAP elements are marked with H3K9me3 in ESCs and preimplantation embryos [58,59].These results suggest that UHRF1 may be recruited to specific targets through H3K9me3 recognition via its TTD [35,36,42,43].However, mice expressing a mutated UHRF1 protein with a defective TTD exhibit normal viability/fertility with an only ~10% reduction in CG methylation [60].Thus, the mechanism through which UHRF1 is recruited to the targets (particularly the ICRs and IAP elements) in the preimplantation embryos remains an open question.
An unexpected discovery of the present study was the involvement of UHRF1 in de novo CG and non-CG methylation in GOs.We previously reported that DNMT1 is important for the completion of de novo CG methylation in oocytes: DNMT1 appears to act on the hemimethylated CG sites left behind during the de novo methylation process and make them fully methylated [5].UHRF1 may contribute to this process by efficiently recognizing hemimethylated CG sites since the proportion of hemimethylated CG sites among the highly methylated sites was increased in Uhrf1 KO FGOs.This increase in hemimethylated CG sites was greater than that observed in Dnmt1 KO FGOs.Unexpectedly, but consistent with the above-mentioned findings, we found that not only CG methylation, but also non-CG methylation, was affected in Uhrf1 KO FGOs.Thus we speculate that UHRF1 may be directly involved in de novo CG and non-CG methylation, which are both mediated by the DNMT3A-DNMT3L complex in GOs [5].
Since UHRF1 can interact with DNMT3A and DNMT3B in ESCs and silence exogenous promoters [61], it may interact with these enzymes in oocytes as well and facilitate de novo methylation.Interestingly, however, UHRF1 only played a limited role in the establishment of CG methylation at the maternally methylated ICRs: rather, the protein was important for de novo CG and non-CG methylation in the moderately methylated genomic regions.These regions were mostly untranscribed, lacked H3K36me3 marking, and tended to be de novo methylated in a late stage of oocyte growth.Interestingly, among the maternally methylated ICRs, those that were less resistant to UHRF1 depletion normally gained CG methylation in the late stage of oocyte growth.At present, the mechanism through which UHRF1 is recruited to these regions and through which it engages in de novo CG and non-CG methylation remains to be elucidated.
In conclusion, our study showed the importance of maternal UHRF1 in CG maintenance methylation during preimplantation development and in de novo CG and non-CG methylation in transcriptionally inactive regions during oocyte growth.To the best of our knowledge, this is the first study to show the involvement of UHRF1 in de novo methylation in vivo.We believe that our findings give further insight into the epigenetic reprogramming in early development and provide a basis for further improvements in reproductive technologies and regenerative medicine.

Ethics statement
Mouse husbandry and all of the mouse experiments were carried out under the ethical guidelines of Kyushu University.The mice were euthanized by carbon dioxide asphyxiation.The protocols were approved by the Institutional Animal Care and Use Committee of Kyushu University.

Oocyte collection, IVF, and embryo culture
NGOs and GOs were obtained from the ovaries on P10.FGOs were obtained from the ovaries at 8-12 weeks.Preimplantation embryos were obtained by IVF.Females aged !8 weeks were injected with 7.5 U of pregnant mare serum gonadotropin; 48 h later, they were injected with 7.5 U of human chorionic gonadotropin to induce super ovulation.Cumulus-oocyte complexes were collected from the oviduct and fertilized with C57BL/6J sperm for immunostaining and Western blotting or with JF1 (Mus musculus molossinus) sperm for WGBS.Cumulus cells were carefully removed by washing with phosphate-buffered saline (PBS), and fertilized embryos were cultured in KSOM medium (EmbryoMax KSOM Medium (1X) w/ 1/2 Amino Acids, Merck Millipore) at 37˚C with 5% CO 2 .

Immunostaining
Embryos and oocytes were fixed with 4% paraformaldehyde in PBS at room temperature for 30 min and washed with PBS three times.After incubation with a pretreatment buffer (1% bovine serum albumin and 2% Triton-X100 in PBS) at room temperature for 15 min, the embryos and oocytes were incubated at 4˚C overnight with anti-UHRF1 rat monoclonal (Th-10a, MBL) antibody (1:500 dilution) or anti-DNMT1 rabbit polyclonal antibody (1:1000 dilution).After washing three times, the embryos and oocytes were incubated with fluorescencelabeled secondary antibody (1:1000 dilution) at room temperature for 30 min.The secondary antibody was CF488 donkey anti-rabbit IgG (H+L) antibody (20015, Biotium Inc), CF488A donkey anti-rat IgG (H+L) antibody (20027, Biotium Inc), or CF594 donkey anti-rabbit IgG (H+L) antibody (20152, Biotium Inc).After washing three times, oocytes and embryos were mounted in VECTASHIELD medium with DAPI (Vector Laboratory) and observed using an LSM510 or LSM700 confocal laser scanning microscope (Carl Zeiss) with a 63× objective lens and a 10× ocular lens.The signal intensities were measured using the ZEN 2012 (blue edition) software program (Carl Zeiss).The nuclear/cytoplasmic signal ratios were calculated using the mean nuclear and cytoplasmic signal intensities.

The qRT-PCR
Total RNA was extracted from 100 FGOs using AllPrep DNA/RNA Mini kit (QIAGEN) and reverse transcribed using a PrimeScript RT reagent kit with gDNA Eraser (TaKaRa).A qRT-PCR was carried out using a KAPA SYBR Fast qPCR kit (Kapa Biosystems) in Thermal Cycler Dice Real Time System Single (TaKaRa) in accordance with the manufacturer's protocol.The primers that were used are listed in S5 Table .The relative gene expression was quantified using the comparative cycle threshold method, with the Gapdh expression used for normalization.

WGBS and the data analysis
WGBS libraries were prepared using the PBAT method as described in our previous reports [5,51,64].Five hundred to 1,000 oocytes or 30-40 blastocysts were spiked with 0.1 ng of lambda phage DNA (Promega) and subjected to bisulfite conversion.The concentrations of the PBAT libraries were measured by a qPCR using a KAPA Illumina Library Quantification kit (Kapa Biosystems).Cluster generation and sequencing were performed in a single-read mode using a Tru-Seq SR Cluster kit v3-cBot-HS (Illumina) and TruSeq SBS kit v3-HS (Illumina) according to manufacturer's protocols.The libraries were sequenced on a HiSeq 2500 or HiSeq 1500 equipped with HCS v2.0.5 and RTA v1.17.20 (suitable for WGBS) [65] to generate 101-nucleotide singleend reads.We trimmed the raw sequence reads to 96 bases by removing the adapter sequences from the 5' end and one base from the 3' end.The resulting reads were aligned to the reference mouse genome (mm10) using Bismark v0.14.2 [66].The seed length was 28, the maximum number of mismatches permitted in the seed was 1, and the "-pbat" option was selected.Only uniquely aligned reads were analyzed.We estimated the bisulfite conversion rate using reads that were uniquely aligned to the lambda phage genome (accession no.J02459).The sequences and information of the chromosomes, RefSeq genes, CGIs, and repetitive elements were downloaded from the UCSC genome browser [67].Throughout this paper, the global CG methylation levels refer to the weighted levels, which take the sequencing depth into account [68].

Estimating the frequency of hemimethylated CG sites
The proportion of hemimethylated sites among the highly methylated CG sites was estimated as previously described [5].In brief, we selected CG sites with a methylation level of !70% on either strand that showed a read depth of !10.The CG sites with a significant difference in methylation (Fisher's exact test, p <0.05) were defined hemimethylated.

RNA-seq and the data analysis
Total RNA was extracted from 30 oocytes.Whole transcript amplification and library construction were performed as described previously [69].The libraries were sequenced on a HiSeq 2000 to generate 50-nucleotide single-end reads.We trimmed the raw sequence reads by removing poly-A tails and low quality bases from the 3' end.The resulting sequences were aligned to the reference mouse genome (mm10) using the TopHat software program (v2.0.14) [70].We used a gtf file of published oocyte transcripts [19] for mapping.The mapped reads were counted using featureCounts [71] to calculate the reads per kb per million mapped reads (RPKM) values.

Accession numbers
The sequence data sets supporting the results of this article are available in the DDBJ Sequence Read Archive under accession number DRA005849.

Supporting information
Fig).The Uhrf1 and Dnmt1 mRNAs, which were derived from the paternal allele, were first detected at the mid 2-cell stage (S1 Fig), pinpointing the timing of the zygotic activation of these genes.We generated oocyte-specific Uhrf1 KO mice carrying a zona pellucida glycoprotein 3 (Zp3)-Cre transgene [Uhrf1 2lox/2lox , Zp3-Cre] (S2A and S2B Fig).The Zp3-Cre transgene is expressed exclusively in GOs [47].No significant defects were observed in the morphology, size, or number of Uhrf1 KO [Uhrf1 1lox/1lox ] FGOs (S2C Fig).As expected, virtually no Uhrf1 mRNA (by a quantitative reverse transcription-polymerase chain reaction; qRT-PCR) or UHRF1 protein (by Western blotting) was detected in the KO FGOs (S2D and S2E Fig).The predicted truncated protein (amino acid 1-133) was not expressed, as its coding exons (2-3) were not detected at the mRNA level (S2D Fig), probably due to nonsense-mediated mRNA decay [48].The loss of UHRF1 protein was confirmed by immunostaining as early as the nongrowing oocyte (NGO) stage (Fig 1C and S2F Fig).The expression of Dnmt1 was not affected by the Uhrf1 KO (S2E Fig).When Uhrf1 KO oocytes were fertilized with wild-type sperm in vitro (in vitro fertilization, IVF), only 20% of the embryos (Uhrf1 maternal-KO or mat-KO [Uhrf1 1lox/+ ]) developed to the expanded blastocyst stage (S3A Fig), suggesting that maternal UHRF1 plays a critical role in preimplantation development.This phenotype was much more severe than that of Dnmt1 mat-KO [Dnmt1 1lox/+ ] embryos [31,32] (S3A Fig).Virtually no UHRF1 protein was detected by immunostaining even in the surviving embryos (Fig 1D).Interestingly, while the absence of DNMT1 did not affect the subcellular localization of UHRF1, the absence of UHRF1 clearly decreased the level of DNMT1 in the nucleus of FGOs (Fig 2A).The absence of maternal UHRF1 also decreased the level of nuclear DNMT1 in 2-cell embryos (Fig 2B), suggesting that the nuclear localization of DNMT1 is dependent on UHRF1.

Fig 3 .
Fig 3.The role of maternal UHRF1 in CG maintenance methylation in preimplantation embryos.(A) Violin plots showing the distribution of the 10-kb genomic regions with different CG methylation levels in control [Uhrf1 2lox/+ ], Uhrf1 mat-KO, and Dnmt1 mat-KO blastocysts.The horizontal bars indicate the mean CG methylation levels.(B) The genomic distribution of CG methylation across chromosome 14 in control and mat-KO blastocysts.The CG methylation levels of 1-Mb windows are shown.GC contents, RefSeq genes, and CGIs are shown at the bottom.(C) The correlations between the CG methylation levels of the 10-kb genomic windows in the control and mat-KO blastocysts.(D) The CG methylation levels of respective portions of the genic regions, intergenic regions, and various repetitive sequences in control and mat-KO blastocysts.(E) The CG methylation levels of the ICRs in control and mat-KO blastocysts.https://doi.org/10.1371/journal.pgen.1007042.g003

Fig 4 .
Fig 4. The role of UHRF1 in de novo CG methylation during oocyte growth.(A) Violin plots showing the distribution of the 10-kb genomic regions with different CG methylation levels in NGOs [5], GOs (P12 and P15) [53], and control [Uhrf1 2lox/2lox ], Dnmt1 KO, and Uhrf1 KO FGOs.The horizontal bars indicate the mean CG methylation levels.(B) The genomic distribution of CG methylation across chromosome 14 in control and KO FGOs.The CG methylation levels of the 1-Mb windows are shown.GC contents, RefSeq genes, and CGIs are shown at the bottom.(C) The correlations between the CG methylation levels of the 10-kb genomic windows in control and KO FGOs.(D) The CG methylation levels of respective portions of the genic regions, intergenic regions, and various repetitive sequences in control and respective KO FGOs.(E) The CG methylation levels of the ICRs in control and KO FGOs.https://doi.org/10.1371/journal.pgen.1007042.g004 and S6A Fig).However, subsequent analyses in 10-kb windows revealed that the loss of CG methylation was most remarkable in genomic regions that showed moderate levels of CG methylation (Fig 4C).In addition, there was a positive correlation between the losses in CG and non-CG methylation in the 10-kb genomic windows (S6B Fig).Various repetitive elements, including IAPs, were moderately affected (Fig 4D).Many CGIs were known to be methylated in FGOs [49] and, again, moderately methylated CGIs were the most affected in Uhrf1 KO FGOs (S6C Fig).In contrast, the maternally methylated ICRs, which represent the most highly methylated regions in the FGO genome, were only slightly affected by the absence of UHRF1, except the Gnas1A, Peg10, and Mest ICRs, which showed significant reduction in CG methylation (p = 6.49x 10 −121 , 1.85 x 10 −41 , and 2.68 x 10 −81 , respectively; kai-squared test) (Fig 4E).

Fig 5 .
Fig 5. UHRF1 facilitates de novo methylation in inactive regions during late oocyte growth.(A) The classification of the 10-kb genomic regions that were moderately (40-80%) or highly (!80%) methylated in the control FGOs, based on the extent of methylation change (!20% or not) in Uhrf1 KO FGOs (Groups 1-4).The right panel shows the proportion of moderately and highly methylated 10-kb regions belonging to each group.(B) The distribution of the transcript levels (RPKM) and H3K36me3 enrichment levels [20] (corrected read count) of the 10-kb regions belonging to each group.The bar in each box indicates the median value.(C) Differences in CG methylation across the transcribed regions between control and Uhrf1 KO FGOs.The transcribed regions were classified into four categories based on the expression levels (RPKM) in control FGOs and the results for the respective groups are shown separately.(D) The changes in CG methylation occurring in the three different stages of oocyte growth were determined for the 10-kb regions of the respective groups.https://doi.org/10.1371/journal.pgen.1007042.g005 Fig. The expression of Uhrf1 and Dnmt1 mRNAs in preimplantation embryos.The expression of Uhrf1 and Dnmt1 mRNAs from the maternal and paternal alleles was analyzed separately using published data from (CAST/EiJ × C57BL/6J) F1 hybrid embryos (GSE45719) [46].An enlarged view of the Uhrf1 and Dnmt1 expression from the zygote to the late 4-cell stage is shown on the right.The expression levels are shown by the RPKM values.Error bar, standard error.(EPS) S2 Fig.The establishment of oocyte-specific Uhrf1 KO mice.(A) A schematic representation of the genomic organization of the wild-type (WT), 2lox, and 1lox Uhrf1 alleles.The filled boxes indicate the protein-coding regions.(B) The predicted domain architecture of the WT and mutated UHRF1 proteins.The immunogen used to generate the anti-UHRF1 antibody M-132 (used for Western blotting) is shown.(C) The morphology of control [Uhrf1 2lox/2lox ] and Uhrf1 KO FGOs.Scale bar, 100 μm.(D) The expression of Uhrf1 and the Dnmt family mRNAs in control and Uhrf1 KO FGOs was measured by qRT-PCR.Two portions of Uhrf1 mRNA were amplified for quantification.The expression of Gapdh was used for normalization.(E) The detection of UHRF1 and DNMT1o proteins in control and Uhrf1 KO FGOs by Western blotting using M-132 and anti-DNMT1 antibodies.β-actin was used as a loading control.(F) The immunostaining signal intensities obtained with an anti-UHRF1 antibody (Th-10a) in control and Uhrf1 KO developing oocytes.(EPS) S3 Fig. CG methylation in Uhrf1 mat-KO blastocysts.(A) The morphology of control [Uhr-f1 2lox/+ ], Dnmt1 mat-KO, and Uhrf1 mat-KO embryos at 96 h after IVF.Among the Uhrf1 mat-KO embryos, those that developed to the expanded blastocyst stage are shown on the left.Only well-developed blastocysts were used for WGBS.Scale bar, 100 μm.(B) The genomic distribution of CG methylation across all chromosomes in the control and mat-KO blastocysts.The CG methylation levels of the 1-Mb windows are shown.(C) The correlations between the CG methylation levels of the CGIs in control and mat-KO blastocysts.(D) Allele-specific CG methylation at the ICRs in control and Uhrf1 mat-KO blastocysts.(EPS) ).We previously reported that the slight decrease in CG methylation in Dnmt1 KO FGOs (36.0%global CG methylation level compared to 38.7% in the control FGOs) (Fig 4A and S2 Table) could be explained by