Optimisation of follicle-free oocyte culture model
To set up an oocyte in vitro model from immature to germinal-vesicle (GV) oocytes and to be able to manipulate oocytes sufficiently early during their development and maturation, we adopted the protocol from Honda et al. [24], which took advantage of maturing arrested primordial oocytes from dissociated ovaries of 1-week-old mice. In this method, oocytes are allowed to grow in follicle-free cultures, but initially supported by theca-like cells in the primary culture, and we considered that this method would have advantages in allowing manipulations that would be difficult to achieve in follicle cultures. Although oocytes grown by this method have been assessed as completing meiosis upon induction and to establish DNA methylation correctly for the limited number of igDMRs tested, their wider epigenetic fidelity has not been evaluated. We sought to optimise this oocyte culture system further to produce a larger quantity of GV stage oocytes for experiments. We evaluated mouse strain and age at commencement of culture (data not shown), settling on 7-day-old F1(C57BL6/Babr background) females. We included fatty acid-free bovine serum albumin (fafBSA), epidermal growth factor (EGF), and follicle stimulating hormone (FSH) in the culture medium, as these factors have been shown to improve oocyte development and maturation [25,26,27]. We also considered the effect of oxygen tension. Human oocytes in vivo are in contact with 3–5% oxygen [8], while in other mammals the oxygen concentration has been reported as being between 2 and 8% [28]. Most follicle cultures for mouse oocytes have been carried out under normoxia, with only a few studies using lower oxygen levels, such as 5% O2 [9]. In this study, we compared 20% O2 and 5% O2 (Fig. 1a).
To evaluate the performance of the cultures, and for subsequent molecular analyses, we size-selected oocytes and categorised them into four classes by their diameters: I, < 40 µm; II, 41–49 µm; III, 51–59; µm and IV, 60–70 µm (Fig. 1a). The progression of de novo DNA methylation has been described in in vivo oocytes of similar size classes [23], which serves as a reference. In the optimised normoxia cultures, oocytes achieved a maximum diameter after 14 days, whereas in 5% O2, size IV was attained in only 9 days and after these 9 days, the oocytes started to degenerate. Under both conditions, oocytes developed into GV-like oocytes similar to in vivo oocytes (Fig. 1b), but the maximum sizes attained differed. An in vivo fully grown GV measures 70–80 µm [29], whereas in 20% O2, only 7% of the oocytes reached a size of 65 µm and in 5% O2 10% of oocytes achieved a size of 70 µm (Fig. 1c, p < 0.0001, 7 independent oocyte cultures for both conditions). The contingency graph describing the growth model showed that the number of oocytes under 20% O2 decreased slowly from class I–IV, while they grew, whereas in 5% O2 the oocytes maintained their number at a plateau, when the oocytes reached the size of 50–60 µm (from class II–III) and then slowly decreased in number to reach class IV (Fig. 1d , n = 7 independent oocyte cultures for both conditions).
During the final step of oogenesis, the oocyte nucleus is subject to large-scale chromatin modifications that correlate with transcriptional silencing. Oocytes that present uncondensed chromatin (NSN, non-surrounded nucleolus) are transcriptionally active, while oocytes with dense chromatin around the nucleolus are silent (SN, surrounded nucleolus); moreover, the transition to SN is required for developmental competence [30]. To define the pattern of chromatin organisation, the in vitro-grown oocytes were stained with DAPI and compared with in vivo GVs collected from 23- to 25-day-old females (3 independent oocyte cultures for both conditions, total of oocytes n = 68 for 5% O2 and n = 62 for 20% O2). Under either culture condition, IV-class oocytes comprised both NSN and SN GVs (Fig. 1d); however, the SN oocytes grown in 20% O2 had a bigger nucleolus than in 5% O2. In terms of the SN and NSN proportions, oocytes grown in 5% O2 condition were closer to those of in vivo GV oocytes (75% SN, compared with 71% in vivo GV, n = 31 oocytes from 23- to 25-day-old females; [30]), but only 40% of oocytes grown in 20% O2 were in the SN chromatin organisation.
Characterisation of DNA methylation in in vitro-grown oocytes
DNA methylation acquisition in the oocyte correlates with increasing diameter during oogenesis [23]. Therefore, the DNA methylation level of the oocytes grown under the two culture conditions was compared with the methylation of in vivo size-selected oocytes. We collected oocytes representing the starting populations and the largest oocytes at the end of the cultures which, for practical purposes, corresponded to ≤ 50 µm and 55–65 µm for the 20% O2 cultures, and ≤ 40 µm and 60–70 µm in the 5% O2 cultures. Pools of ~ 250 for the class IV to ~ 500 oocytes for class I and II were collected in replicate and processed for reduced representation bisulphite sequencing (RBBS), which preferentially samples CG-rich regions in the genome, including CGIs [31]. RRBS data from the cultured oocytes were compared to published data from in vivo growing oocytes (I–IV [23] and GV [3]). When the replicates were merged depending on their size within the experimental groups, coverage was sufficient to evaluate between 56.4 and 60.1% of CGIs (≥ 5 CpG sites per CGI and ≥ 5 reads per CpG; CGI annotation based on [32]). Principal component analysis (PCA) indicated that replicates clustered together (Additional file 1: Fig. S1). Furthermore, PCA demonstrated a separation along PC1 of the in vivo groups corresponding to developmental stage. Both 20% O2 and 5% O2 samples also separated on PC1 according to oocyte size, indicative of a de novo methylation process similar to in vivo, but the 5% O2 samples also separated on PC2 from the other groups due to the effect of 5% O2. As previously reported [3, 4], total methylation of CpGs increased from 3.2% in group II in vivo oocytes to 10% in group IV and 26.3% at the fully grown GV stage (Fig. 2a). The in vitro-cultured oocytes also exhibited increases in methylation with size: from 5 to 8% for the 20% O2 cultures, and 9–21% for the 5% O2 cultures (Fig. 2a). These results suggest that growth in 5% O2 is more suitable for in vivo-like DNA methylation establishment than 20% O2.
We next considered the fidelity of DNA methylation in the in vitro-grown oocytes. Similar to most cell types, the majority of CGIs are unmethylated in oocytes, but fully grown GV oocytes in vivo have a defined number of highly methylated CGI [3, 21, 22]: in the in vivo GV dataset we used as reference 4.1% of CGIs (n = 661) are hypermethylated (> 75%). In comparison, 1.2% (n = 169) of the CGIs informative in our RRBS datasets of the IV-sized oocytes from the 5% O2 cultures were classed as highly methylated, and just 0.06% (n = 9) of CGIs in oocytes grown in 20% O2 (Fig. 2b). On the other hand, 631 CGIs (4.6%) had methylation > 25% in the IV oocytes grown in 20% O2, and 1022 (7.45%) in 5% O2, which might indicate that the CGIs normally fully methylated in vivo are partially methylated in the in vitro-grown oocytes. To evaluate the CGIs subject to de novo methylation more closely, we selected those that had ≥ 40% methylation in at least one sample, which were 1530 of the 13,710 CGIs for which RRBS data were available in size III–IV and GV in vivo oocytes, and IV 20% O2 and IV 5% O2 oocytes (Additional file 1: Fig. S2). When the CGIs > 40% methylated in at least one sample including in vivo samples were clustered with the CGIs > 50% from GV stage oocytes, we observed that ~ 40% of the CGIs were common between GV and both in vitro conditions (Fig. 2c). With the exception of a single CGI found in the sample 5% O2, all CGIs ≥ 40% methylated in in vitro-grown oocytes were found in > 25% methylated in GVs (Additional file 1: Fig. S3A). To confirm that CGIs > 40% methylated in at least one sample were a correct cut-off from the total CGIs for further analysis, we correlated CGIs > 40% methylated in at least one sample with CGIs of GV, IV in vivo, III in vivo, IV 5% O2 and 20% O2. The correlation coefficient between GV and IV in vivo was lower than the correlation coefficient between IV in vivo and III in vivo, confirming the results observed by Gahurova et al. [23]. The correlation coefficient between IV in vivo and 20% O2 was better than IV in vivo and 5% O2 confirming the PCA plot (Additional file 1: Fig. S3B). From unsupervised hierarchical clustering analysis of the 1530 CGIs, six major clusters were apparent (Fig. 3a). Cluster 1 generally comprised CGIs methylated in all samples; cluster 2 had methylated CGIs in in vitro conditions and Class III in vivo, but lacking the methylation in class IV and GV in vivo samples; clusters 3 and 4 were dominated by CGIs showing progressive methylation in vivo samples but lacking methylation in the 5% grown oocytes; clusters 5 contained CGIs methylated predominantly in the in vitro-grown oocytes; and cluster 6 grouped all the intermediate methylated CGIs in all samples except GV in vivo samples (Fig. 3b). For all of the samples, CGIs > 40% methylated were predominantly intragenic in location, compared with unmethylated CGIs (Fig. 3B), consistent with the dependence of CGI methylation on transcription events in oocytes [3, 4].
Amongst the genomic elements for which de novo DNA methylation in oocytes does have a critical role are the igDMRs of imprinted genes. Oocytes grown either in 5% O2 or 20% O2 demonstrated a gain of methylation at igDMRs comparable to the corresponding sized in vivo oocytes (Fig. 3c), with the oocytes grown in 5% O2 showing a higher mean methylation level (58.1% compared with 42.9%), suggesting greater progression towards the complete methylation characteristic of fully grown GV oocytes. Together, these results confirm that DNA methylation starts to be established in the in vitro cultures, with critical elements gaining methylation appropriately, but there are also subsets of CGIs displaying anomalous methylation.
Transcriptome analysis of in vitro-grown oocytes
To investigate gene expression patterns in the oocytes grown in vitro, we generated deep, strand-specific libraries in duplicate from the same size populations of in vitro-grown oocytes as used in the methylation analysis. PCA of these RNA-seq libraries, in comparison with in vivo reference datasets [23], indicated that the in vitro-grown oocytes from either O2 concentration separated from in vivo growing oocytes and were closer to GVs on PC1, but were separated from all in vivo groups on PC2 (Additional file 1: Fig. S4).
To explore the gene expression differences between in vitro-grown and in vivo oocytes, we used DESeq to compare the in vitro RNA-seq data with in vivo GVs. The global gene expression correlations between the IV 5% O2 oocytes and GVs, and between the IV 20% O2 oocytes and GVs, were r = 0.883 and r = 0.917, respectively (Fig. 4a, b); furthermore, there were 890 differentially expressed genes between IV 5% O2 and in vivo GVs, and 2990 between IV 20% O2 and GVs (Fig. 4a, b, p < 0.05). Gene Ontology (GO) analysis using Panther revealed that the differentially expressed genes were most enriched in biological processes related to the unfolded protein response, response to starvation, and endoplasmic reticulum stress for IV 5% O2 oocytes, and terms related to female gamete and chromatin organisation for IV 20% O2 oocytes (Fig. 4c). We also interrogated the expression of oocyte-specific genes important for follicle development, reproduction, and early development [33]. In the 20% O2-grown oocytes, there was a greater abundance of transcripts for key oocyte transcription factors Spermatogenesis And Oogenesis Specific Basic Helix-Loop-Helix 1 and 2 (Sohlh1 and Sohlh2), whereas 5% O2-grown oocytes had reduced abundance of Folliculogenesis-Specific BHLH Transcription Factor (Figla) and LIM Homeobox 8 (Lhx8) transcripts where both transcription factors co-express in oocyte for its maturation, as well as for the oocyte markers Zona pellucida genes, Zp1, Zp2, and Zp3 (Fig. 4d). We also evaluated expression of Dnmt3A, Dnmt3Ll and Dnmt1 which are key enzymes for regulating the establishment of methylation on DNA in the RNA-seq datasets. Transcripts for all three genes had reduced abundance in the 5% O2 and 20% O2-grown oocytes, with a greater reduction in the 5% O2-grown oocytes (Fig. 5a, p < 0.05 for Dnmt1 gene, isoform Dnmt1o oocyte specific). For Dnmt3a, the principal catalytically active de novo methyltransferase in mouse oocytes, we also examined protein abundance and localisation by immunofluorescence, comparing IV 5%O2 oocytes (SN and NSN) with in vivo post-natal D15 growing oocytes and GV oocytes. Dnmt3a protein was observed both in the cytoplasm and nucleus in the NSN oocytes at D15, whereas in SN GV oocytes it was concentrated on chromatin, accompanied by an increase in total protein abundance (Fig. 5b, c). In the NSN oocytes from the 5% O2 cultures, Dnmt3a was predominantly cytoplasmic and retained some cytoplasmic localisation even in the SN-like oocytes (Fig. 5b, C).
Finally, because of the role of transcription events in determining DNA methylation in growing oocytes [4, 21], we integrated results from the RRBS and RNA-seq from the cultured oocytes. After stratifying intragenic CGIs for level of DNA methylation attained, we detected a positive correlation between gene expression level and methylation, particularly for the oocytes grown in 20% O2 (Additional file 1: Fig. S5), suggesting that CGIs more advanced in de novo methylation are located within more highly expressed genes. Therefore, although the in vitro-grown oocytes do not completely recapitulate the DNA methylation profile of in vivo counterparts, the de novo methylation mechanism is likely to follow the same mechanistic principles.
Testing elements of the de novo methylation mechanism in the in vitro culture system
To investigate the utility of the oocyte culture system and to investigate mechanisms of DNA methylation establishment, we conducted an experiment using two drugs: tranylcypromine hydrochloride [34] and lithium chloride (LiCl) [35]. Tranylcypromine hydrochloride is a non-selective monoamine oxidase inhibitor (MAO) and can be used to inhibit an amine oxidase reaction essential for the activity of the histone lysine demethylases of the KDM1 family [36, 37], which have activity towards H3K4me2. Both Kdm1a and Kdm1b have roles in promoting normal de novo DNA methylation in growing oocytes [38, 39], in part by demethylation of H3K4me2 which is antagonistic to Dnmt3a binding and activity [40]. LiCl was chosen because it blocks activity of the oocyte-specific transcription factor Foxo3a by inhibiting its phosphorylation and translocation to the nucleus [35]. In view of the greater apparent fidelity of methylation in oocytes grown in 20% O2, we selected to culture the oocytes under normoxia. To evaluate the efficacy of MAO treatment, we performed immunofluorescence for H3K4me2 and H3K4me3 on oocytes after 14 days of culture, observing that H3K4me2 and H3K4me3 staining was reduced (Additional file 1: Fig. S6A, n = 10 oocytes for control and treated samples). For the LiCl-treated oocytes, the nuclear staining of Phospho-FOXO3A, that is seen in the untreated oocytes, is much diminished (Additional file 1: Fig. S6B, n = 10 oocytes for control and treated samples).Along with the staining of the cultured oocytes, GV oocytes (n = 8 for each condition MAO, LiCl, and control) were stained to observe the correct localisation of proteins of interest to assure that the 20% O2 had no effect on the cultured oocytes (Additional file 1: Fig. S6C).
To quantify the effect of MAO and LiCl treatments on DNA methylation, we collected oocytes after 14 days of culture, grown under 20% O2, in the absence or presence of the inhibitors in triplicate pools of between 70 and 110 oocytes, and processed the oocytes for whole-genome methylation analysis by low-cell post-bisulphite adaptor-tagging (PBAT) [38, 41]. After alignment and removal of reads from duplicates between 16,798,787 and 45,690,308 uniquely mapped reads were obtained per library. (One LiCl-treated sample was discarded because of low coverage, Additional file 2: Table S1.) For the clustering analysis, we included in this analysis, the Class IV in vivo oocyte from Gahurova et al. [23] to identify how the samples will group, as the class IV oocytes is our aimed size and present the maximum level of methylation. The PBAT libraries clustered according to the treatment (Additional file 1: Fig. S7). For most subsequent analysis, replicates were pooled.
First, we assessed the effects of the inhibitors on overall DNA methylation in the oocytes and observed a decrease in CpG methylation in the three different conditions from 39.7% (MAO), 28.8% (LiCl) to 19.6% (MAO + LiCl) compared to 46.6% in in vivo IV oocytes, and 38% in 20% O2 condition (Fig. 6A). As one of the treatments affects the activity of Kdm1 enzymes, we included the PBAT methylation datasets of Kdm1a and Kdm1b M2 knock-out oocytes from Stewart et al. [38] in the analysis. Treatments of MAO or MAO + LiCl showed a reduced CpG methylation with a pattern close to Kdm1b (data not shown). However, here we compared oocytes from different sizes (Class IV in vitro oocytes 20% O2 with M2 ovulated oocytes) and could bias the results. Focussing on CGIs that normally become methylated in oocytes, we then access the total CGIs methylation which showed a reduction in highly methylated CGIs from 1.3% (n = 76) in 20% O2 CON to nearly 0% in MAO (n = 32), LiCl (n = 8), and MAO + LiCl (n = 2). The number of methylated CGI < 25% was increasing in all treated samples and was of 10% between 20% O2 CON and MAO + LiCl (Fig. 6B). Then, we used the same strategy as for analysing the RBBS data, and a cut-off of > 40% methylation was used on the PBAT data set. This resulted in 1138 CGIs which clustered within 3 groups and were mostly hypomethylated in the cultured oocytes under the inhibitory drugs (Fig. 6c). They failed to properly methylate, and the proportion of CGIs localised on the promoter and intergenic regions increased compared to oocyte 20% O2 CON. Most of the methylation is observed on the intragenic regions in GV oocytes [4], and the cultured oocytes presented a significant decreased in CGIs at the intragenic regions from 48% in 20% O2 CON to 40% in cultured oocytes with Mao + LiCl (Fig. 6d). We also specifically evaluated effects on igDMRs, as it has previously been described that genetic ablation of Kdm1b in oocytes [32] impairs de novo methylation of igDMRs. The heatmap of methylation difference of the igDMRs by comparing 20% O2 CON with each drug presented a various aberrant igDMRs methylation for all the cultured oocytes (Fig. 6e). For example, Igf2r was highly methylated (70.1%) in IV 20% O2 CON and the cultured oocytes under treatment reduced their methylation (MAO 56.5%, LiCl 38.2%, and MAO + LiCl 33.9%, respectively, Additional file 1: Fig. S7). These results suggest that the igDMRs did not gain a correct level of DNA methylation but instead obtained various levels of intermediate methylation.
Finally, we sought to assess the extent to which the treatments in vitro mimicked the ablation of Kdm1a and Kdm1b in vivo [38]. We took advantage of the published PBAT data from M2 oocytes of Kdm1a and Kdm1b knock-out mice [38]. We compared the defined hypermethylated domains (CGI over 75% methylation) of M2 oocytes knock-out mice, parameters defined in Stewart et al. [38], with our data at IV 20% O2 CON oocytes > 25% methylation and the 3 other in vitro conditions (MAO, LiCl, and MAO + LiCl) over 10% of methylation. We have chosen a methylation threshold of ≥ 25% methylation for 20% O2 CON (n = 1433) as most of the CGIs had an intermediate methylation. For the other in vitro conditions MAO, LiCl, and MAO + LiCl, the methylation threshold was chosen ≥ 10% methylation to gain the maximum number of CGIs in the different culture conditions (MAO n = 1983, LiCl n = 2243, and MAO + LiCl n = 1817). We have generated Venn diagrams to identify the overlapping CGIs between the effect of the in vitro conditions and the in vivo knock-out mice dataset. While comparing the CGIs in hypermethylated domains with in vivo CGIs of Kdm1a knock-out mouse hypermethylated domains, only 12 CGIs were common with all the in vitro conditions. When same comparison was carried out with the in vivo CGIs of Kdm1b knock-out mouse hypermethylated domains, only one CGI was common (Fig. 7a–c). In the case of the comparison with hypermethylated domains Kdm1a knock-out mouse, the 12 common CGIs are located mostly on intragenic regions (75%) and the remaining 25% are located on the promoter (Additional file 3: Table S2). However, in the case of non-common CGIs, we could observe an increase of CGIs located on intergenic regions for all conditions compared to Kdm1a and Kdm1b M2 knock-out oocytes (Additional file 3: Table S2). Interestingly, the CGIs in hypermethylated domains of 20% O2 CON compared with hypermethylated domains of Kdm1a knockout oocytes (n = 96) and the CGIs from 20% O2 CON compared with hypermethylated domains of Kdm1b knock-out oocytes (n = 98) were all common (Additional file 3: Table S2). These results suggest that the drugs against Kdm1 enzyme activity slowed down the gain of methylation on CGIs in grown oocytes.