In vitro expansion of mouse primordial germ cell‐like cells recapitulates an epigenetic blank slate

Abstract

The expansion of primordial germ cells (PGCs), the precursors for the oocytes and spermatozoa, is a key challenge in reproductive biology/medicine. Using a chemical screening exploiting PGC‐like cells (PGCLCs) induced from mouse embryonic stem cells (ESCs), we here identify key signaling pathways critical for PGCLC proliferation. We show that the combinatorial application of Forskolin and Rolipram, which stimulate cAMP signaling via different mechanisms, expands PGCLCs up to ~50‐fold in culture. The expanded PGCLCs maintain robust capacity for spermatogenesis, rescuing the fertility of infertile mice. Strikingly, during expansion, PGCLCs comprehensively erase their DNA methylome, including parental imprints, in a manner that precisely recapitulates genome‐wide DNA demethylation in gonadal germ cells, while essentially maintaining their identity as sexually uncommitted PGCs, apparently through appropriate histone modifications. By establishing a paradigm for PGCLC expansion, our system reconstitutes the epigenetic “blank slate” of the germ line, an immediate precursory state for sexually dimorphic differentiation.

Introduction

Primordial germ cells (PGCs) are the founding population of the germ cell lineage that ensures reproduction, heredity, and evolution of a given species. In mice, PGCs originate from the epiblast at around embryonic day (E) 6.25 (Ginsburg et al, 1990; Lawson et al, 1999; Saitou et al, 2002; Ohinata et al, 2005), expand their numbers during their migration and in embryonic gonads (from ~30 to 40 at E7.25 to ~25,000 at E13.5: ~600‐ to 800‐fold) (Tam & Snow, 1981; Kagiwada et al, 2013), and initiate entry into either spermatogenic or oogenic pathways in response to signals from embryonic gonads from around E11.5 [reviewed in (McLaren, 2003; Spiller & Bowles, 2015)]. Notably, PGCs undergo epigenetic reprogramming, including genome‐wide DNA demethylation and histone‐modification changes (Seki et al, 2005, 2007; Popp et al, 2010; Seisenberger et al, 2012; Kagiwada et al, 2013), effectively erasing the epigenetic memory of a previous generation to create an epigenetic blank slate, upon which either an androgenetic or gynogenetic epigenome for the next generation is established thereafter [reviewed in (Saitou et al, 2012; Lee et al, 2014)].

On the other hand, PGCs have been known to be refractory to propagation in vitro (Buehr, 1997; De Felici et al, 2004), which has been a major impediment for their analysis and for their application to reproductive technologies or medicine. It has been reported that mouse pluripotent stem cells (PSCs) are induced into epiblast‐like cells (EpiLCs) and in turn into PGC‐like cells (PGCLCs), which contribute to proper spermatogenesis and oogenesis and to fertile offspring, upon transplantation or aggregation with gonadal somatic cells followed by an appropriate culture (Hayashi et al, 2011, 2012; Hikabe et al, 2016; Ishikura et al, 2016). Accordingly, mPGCLCs, and more recently, human PGCLCs induced from human PSCs (Irie et al, 2015; Sasaki et al, 2015), have provided relatively abundant experimental materials and have been instructive for analyzing the mechanisms of PGC specification and epigenetic reprogramming upon PGC specification [reviewed in (Saitou & Miyauchi, 2016)]. On the other hand, the induction of PGCLCs, which acquire a property similar to migrating PGCs, is a transient process, and without the aggregation of gonadal somatic cells, PGCLCs have also been refractory to further proliferation or maturation. Thus, establishing a strategy for expanding PGCs/PGCLCs under a defined condition is a key challenge that will open a wide range of novel applications in reproductive biology and medicine. Taking advantage of a relatively large number of PGCLCs induced from PSCs, we here set out to perform a screening to identify chemicals that stimulate the proliferation of mouse PGCLCs.

Results

Screening for chemicals that expand PGCLCs

We first derived several novel male lines of mouse embryonic stem cells (ESCs) bearing the B limp1‐m V enus and S tella‐ E CFP (hereafter we designate Blimp1‐mVenus as BV and Stella‐ECFP as SC) transgenes (Ohinata et al, 2008) and evaluated their efficiencies for PGCLC induction and proliferation. All the ESC lines were induced into EpiLCs by activin A (ActA) and basic fibroblast growth factor (bFGF), and then into BV/BVSC‐positive (+) PGCLCs by bone morphogenetic protein 4 (BMP4), leukemia inhibitory factor (LIF), stem cell factor (SCF), and epidermal growth factor (EGF). As shown in Appendix Fig S1A–C, BV (+) PGCLCs increased their numbers in the floating aggregates until day (d) 6 or 8 of induction and waned thereafter, which is consistent with our previous reports (Hayashi et al, 2011). Among the lines we evaluated, the BVSC BDF1‐2 exhibited the most robust induction and proliferation in the floating aggregates (Appendix Fig S1A–C), and we decided to use this line for the subsequent screening.

We decided to plate d4 PGCLCs, which appear to be in the growing phase in the floating aggregates (Appendix Fig S1A–C), on 96‐well plates with m220 feeders, which express a membrane‐bound form of SCF known to support the survival of PGCs (Dolci et al, 1991; Majumdar et al, 1994), and to screen chemicals that enhance BV (+) PGCLC proliferation using a cell analyzer (Fig 1A). We reasoned that BV positivity distinguishes PGCLC proliferation from PGCLC dedifferentiation into embryonic germ cells (EGCs) (Matsui et al, 1992), since ESCs/EGCs express no/low Blimp1/BV (Durcova‐Hills et al, 2008; Ohinata et al, 2008; Hayashi et al, 2011). Since m220 cells were highly vulnerable to mitomycin C (MMC) treatment that was required for their preparation as feeders, we cloned and used a subline of m220 that was resistant to the MMC treatment (Appendix Fig S2A–C). Under the chosen conditions and with the use of LIF, a classical factor known to stimulate PGC proliferation (Matsui et al, 1991), the proliferation of PGCLCs was monitored successfully after 7 days of culture as a corresponding increase in BV fluorescence by a cell analyzer (Appendix Fig S2D and E). We confirmed that under this simple positive control condition, there was little to no dedifferentiation of PGCLCs into EGCs.

Figure 1. Identification of chemicals stimulating PGCLC proliferation.

1. Experimental procedure for chemical library screening using PGCLCs.
2. Scatter‐plot representation of the results of chemical library screening (10 μM). The fold differences in Blimp1‐mVenus (BV) signals for each compound as detected by a cell analyzer (d7/d1) were plotted. The average value (red line) and 3 SDs (standard deviations: red dotted lines) for the negative controls are indicated. Results for the PDE4 inhibitors, RAR agonists, and Forskolin are shown in orange, blue, and green, respectively.
3. Stimulation of PGCLC proliferation by a representative PDE4 inhibitor (GSK256066, 10 μM). A heatmap image of a 96‐well plate at d7 of a screening (top) with a well containing GSK256066 (blue square) magnified for BV fluorescence images (bottom, left, and right). Scale bars: (left) 1 mm; (right) 100 μm.
4. A pie chart classifying the categories of the top 25 compounds (> +3 SDs) in the screening (10 μM).
5. Pie charts classifying the categories of the 426 compounds having a negative effect on PGCLC proliferation/survival (< −3 SD) in the screening (10 μM).

We therefore embarked on screening of a total of ~2,000 chemical compounds that target a diverse set of intracellular signaling molecules/pathways for their ability to expand BV (+) d4 PGCLCs after a 7‐day culture (Fig EV1A–C). Consequently, at a concentration of 10 μM, we identified 63 chemicals that expanded the BV (+) cells significantly compared to the negative control culture, with the fold differences in BV fluorescence between d1 and d7 of culture being more than 3 SDs (standard deviations) of the mean values for the negative controls (Fig 1B and C, Table EV1). Notably, among the top 25 hit compounds, five (20%) were selective inhibitors for phospho‐di‐esterase 4 (PDE4) [ibudilast, S‐(+)‐Rolipram, Rolipram, GSK256066, cilomilast], three (12%) were agonists for retinoic acid (RA) signaling (acitretin, TTNPB, retinoic acid), and one was Forskolin (Fig 1D). PDE4 catalyzes the hydrolysis of cyclic AMPs (cAMPs) to AMP and, therefore, the inhibitors of PDE4 increase the intracellular cAMP levels (Pierre et al, 2009; Keravis & Lugnier, 2012). Forskolin is a potent activator of adenylate cyclase and therefore also elevates the intracellular cAMP levels (Pierre et al, 2009). RA signaling and Forskolin are known to stimulate PGC proliferation (De Felici et al, 1993; Koshimizu et al, 1995). Selective inhibitors for other PDEs or non‐selective inhibitors for PDEs did not show positive effects on PGCLC proliferation (Fig EV1D). Figure 1C shows the proliferation of BV (+) cells in the presence of a PDE4 inhibitor, GSK256066, at d7 of culture, revealing the formation of multiple colonies with a unique, flat morphology (see below). We performed a similar screening using some of the same chemical libraries with a concentration of 1 μM, which resulted in the identification of the same classes of compounds (selective inhibitors for PDE4, agonists for RA signaling, and Forskolin) as potent stimulators for PGCLC proliferation (Fig EV1E and F).

Figure EV1. Screening for chemicals for PGCLC proliferation.

• A, B
  Scheme of the chemical library screening using PGCLCs. (A) Two hundred d4 Blimp1‐mVenus (BV)‐positive PGCLCs were plated on m220‐5 feeders in 96‐well plates by FACS, and (B) the effects of chemicals (80 chemicals/96‐well plate) on PGCLC proliferation were evaluated. Negative (basal medium) and positive (basal medium with LIF) controls were allocated to both sides of a 96‐well plate.
• C
  A list of chemical libraries used in this study.
• D
  The numbers of PDE inhibitors (PDE4‐selective, other PDE‐selective, non‐selective PDE inhibitors) and RAR agonists included in the libraries and of hit compounds among them.
• E
  Scatter plots of the results of chemical library screening (1 μM). Fold differences in the BV signals detected by a cell analyzer (d7/d1) for each compound were plotted. The average value for the negative control (red line) and its 3 SDs (red dotted lines) are indicated. Results for the PDE4 inhibitors, RAR agonists, and Forskolin are shown in orange, blue, and green, respectively.
• F
  A list of the top 15 compounds stimulating PGCLC proliferation. The compounds with effects > +3 SD are labeled red. Results for the PDE4 inhibitors, RAR agonists, and Forskolin are labeled orange, blue, and green, respectively.
• G
  Pie charts classifying the categories of all 178 compounds having negative effects on PGCLC proliferation/survival (< −3 SD) in the screening (1 μM).

We also identified 426 and 178 chemicals from 10 and 1 μM screenings, respectively, that had a negative impact on the proliferation or survival of BV (+) cells (the fold reductions in BV fluorescence between d1 and d7 of culture were more than 3 SDs of the mean values of the negative controls: Figs 1B and E, and EV1E and G, Datasets EV1 and EV2). Such chemicals include inhibitors of key signal transduction pathways, including those known to have a positive influence on PGC proliferation/survival, such as the pathways for receptor tyrosine kinase (RTK) signaling, phosphatidylinositol‐3 kinase (PI3K) signaling, mammalian target of rapamycin (mTOR) signaling, Janus kinase (JAK) signaling, and AKT signaling [reviewed in (Saitou & Yamaji, 2012)], as well as inhibitors for cell cycle/cell division and for DNA replication/repair. Collectively, these findings strongly indicate that our screening successfully identified chemicals that influence key pathways relevant for PGC proliferation/survival.

Synergistic effect of Rolipram and Forskolin on PGCLC expansion

We decided to focus on the effects of one of the PDE4 inhibitors, Rolipram, in subsequent studies, since PDE4 inhibitors were the most enriched chemicals among the hit compounds (Fig 1D, Table EV1) and Rolipram has been reproducibly used in diverse experiments as an efficient PDE4 inhibitor (Keravis & Lugnier, 2012). We evaluated the effects of Rolipram, Forskolin, and their combined use (both increase intracellular cAMP concentration by different mechanisms) (Pierre et al, 2009), on the proliferation of d4 PGCLCs induced from the BVSC BDF1‐2 ESCs in GMEM/10% KSR/2.5% FCS in the presence of SCF on m220 feeders. We decided not to include LIF in this culture, since it might enhance the possibility of the dedifferentiation of PGCLCs into EGCs when applied with other stimulators of PGC proliferation (Matsui et al, 1992). We found that the effect of Rolipram alone (10 μM) was relatively mild and similar to that of Forskolin alone (10 μM) (Fig 2A). However, when Rolipram and Forskolin were applied in combination, they effectively stimulated the proliferation of d4 PGCLCs: With both Rolipram and Forskolin at 10 μM (FR10), d4 PGCLCs exhibited a robust and steady growth at least until d7 of culture (d4c7) and increased their numbers more than 20‐fold, corresponding to 4 to 5 doublings (Fig 2A–C). Importantly, the expanded cells formed flat colonies, continued to express BVSC robustly, and exhibited characteristics of motile cells with prominent filopodia and lamellipodia (Fig 2B and D), suggesting that they maintain a property of migrating PGCs after expansion by FR10.

Figure 2. Establishment of a culture system for PGCLC expansion.

• A
  Effects of Forskolin and Rolipram on PGCLC proliferation. d4 PGCLCs were cultured in the basal medium (GMEM with 10% KSR, 2.5% FCS, and 100 ng/ml SCF) on m220‐5 feeders (NC; negative control) and the effects of 10 μM of Forskolin (F10), Rolipram (R10), and both Forskolin and Rolipram (FR10) on PGCLC proliferation were examined. The numbers of PGCLCs were counted on days 3 (c3), 5 (c5), 7 (c7), and 9 (c9) of culture. Average fold increases of the number of PGCLCs at each time point relative to the number of PGCLCs plated are shown with their SDs (n = 3).
• B
  A representative culture of d4 PGCLCs with FR10. Photographs [images for bright field (BF), Blimp1‐mVenus (BV), and Stella‐ECFP (SC)] and the BVSC FACS plots (all live and single cells in culture) were taken for ESCs and on days 1 (d4c1), 3 (d4c3), 5 (d4c5), and 7 (d4c7) of culture. Scale bar, 100 μm.
• C
  (Left) Expansion by FR10 of PGCLCs induced from male (BVSC R8, BDF1‐2, BCF1‐2) and female (H14, H18) ESC lines. Fold increases in the numbers of PGCLCs at d4c7 relative to the number of PGCLCs initially plated are plotted for each experiment for each ESC line. The average values are indicated as red bars. (Right) Expansion by FR10 of E9.5 PGCs, as measured in the left panel.
• D
  Cultured PGCLCs (BV: green) stained with phalloidin (red). Scale bars, 20 μm.
• E
  The effects of FR10 on the elevation of intracellular cAMP concentrations in d4 PGCLCs. The average values with SDs from three independent experiments are shown.
• F–H
  Cell‐cycle status of cultured PGCLCs (F) and male embryonic germ cells (G). Representative plots for the cell‐cycle status by FACS analysis of the indicated cell types are shown. The vertical axis represents BrdU incorporation and the horizontal axis represents DNA content (7AAD). Cells in S, G2/M, and G1 phase are shown in purple, blue, and red, respectively, along with the percentage of each population. The average values with SDs from three independent experiments are shown in (H).

To determine whether Forskolin and Rolipram indeed elevate cAMP concentrations in PGCLCs, we measured the increase of cAMP concentrations in PGCLCs in response to Forskolin, Rolipram, or both (Materials and Methods). As shown in Fig 2E, Forskolin and Rolipram independently increased the cAMP concentrations in PGCLCs by ~4 nM/1 × 10⁴ d4 PGCLCs. More remarkably, we found that simultaneous addition of Forskolin and Rolipram (FR10) elevated the cAMP concentrations in PGCLCs by more than ~40 nM/1 × 10⁴ d4 PGCLCs.

The FR10 was effective in expanding PGCLCs induced from other male as well as female ESC lines, with an average expansion rate of ~20‐fold at d7 of culture: In some cases, it expanded PGCLCs nearly 50‐fold, corresponding to 5–6 doublings (Fig 2C). The FR10 was also effective in expanding PGCs at E9.5, but to a somewhat limited extent (up to ~eightfold expansion, Fig 2C), which might have been due to the difference in survivability under the present condition between E9.5 PGCs directly isolated from embryos and d4 PGCLCs induced from PSCs in vitro. The cell‐cycle analysis revealed that a majority of cultured PGCLCs (> ~60%) were in the S phase, with a slight decrease and an apparent corresponding increase in the S and G2/M phase at d7 of culture, respectively (Fig 2F and H). This is in stark contrast to the cell‐cycle property of male germ cells in embryonic gonads, which ceased to proliferate and became arrested in the G0/G1 phase after E13.5 (Fig 2G and H) (Western et al, 2008). These findings demonstrate that Rolipram and Forskolin act synergistically to progress the cell cycles of PGCLCs, most likely via the robust activation of cAMP signaling.

Robust capacity for spermatogenesis of PGCLCs expanded in culture

We next went on to evaluate whether PGCLCs expanded by FR10 in culture maintain their function as PGCs/PGCLCs. For this purpose, we transplanted d4c7 and d4 PGCLCs induced from the BVSC BDF1‐2, BCF1‐2, or R8 (on a largely C57BL/6) ESCs (Appendix Fig S3A) into the testes of neonatal W/W ^v mice lacking endogenous germ cells. We found that the testes transplanted with d4c7 PGCLCs, as well as those transplanted with d4 PGCLCs, induced from the BVSC BDF1‐2 or BCF1‐2 ESCs, exhibited a significant enlargement in size after 7 months of transplantation (Figs 3A and EV2A), contained numerous seminiferous tubules with evidence of spermatogenesis, and indeed bore abundant spermatozoa (Figs 3B–F and EV2B). Remarkably, the restoration of the spermatogenesis by both d4c7 and d4 PGCLCs from the BVSC BDF1‐2 or BCF1‐2 ESCs became pronounced enough that the spermatozoa were transported into the epididymis, and such spermatozoa acquired a robust motility to be used for in vitro fertilization (IVF) to generate apparently normal offspring (Figs 3G–K and EV2B). Accordingly, the recipient males for d4c7 as well as d4 PGCLCs from the BVSC BDF1‐2 or BCF1‐2 ESCs were able to produce offspring with substantial litter sizes by natural mating and the resultant offspring exhibited apparently normal growth (Figs 3L–N and EV2B–D).

Figure 3. Robust spermatogenesis by cultured PGCLCs.

• A
  W/W ^v testes (left, untransplanted) 7 months after transplantation of d4c7 PGCLCs induced from BDF1‐2 (center) or BCF1‐2 (right) ESC lines.
• B–D
  The seminiferous tubules showing spermatogenesis (B, C) transplanted with d4c7 PGCLCs (BDF1‐2), and the resultant spermatozoa (D).
• E–G
  Hematoxylin and eosin (HE) staining of the sections of a transplanted testis (E, F) and cauda epididymis (G).
• H–K
  In vitro fertilization (IVF) using the sperm (H) retrieved from the cauda epididymis of the recipient mice. The resultant two‐cell embryos (I) and the offspring (J, K) with normal placenta (J) are shown.
• L–N
  Fertility of recipient W/W ^v mice transplanted with d4c7 PGCLCs (BDF1‐2). The fertility of recipient W/W ^v mice was confirmed by natural mating (L). (M) Litter sizes of recipient W/W ^v mice transplanted with d4 or d4c7 PGCLCs. The average values are indicated as red bars. (N) Genotype of the offspring from d4c7 PGCLCs for BV and SC transgenes.

Data information: Scale bars: (A) 1 mm; (B) 2 mm; (C, E) 0.5 mm; (D) 20 μm; (F, G, I) 100 μm; (H) 25 μm.

Figure EV2. Spermatogenesis by cultured PGCLCs.

1. Weights of W/W ^v testes 7 months after transplantation of d4 or d4c7 PGCLCs induced from the indicated ESCs, with averages indicated by red bars. Weights of the un‐transplanted W/W ^v testes and of testes from the wild‐type BDF1 mice (10 weeks) are also shown.
2. (i, ii) The seminiferous tubules showing spermatogenesis transplanted with d4c7 PGCLCs induced from BCF1‐2 ESCs. (iii–v) HE staining of the sections of the transplanted testis (iii, iv) and cauda epididymis (v). (vi) The fertility of the recipient mice was confirmed by natural mating. (vii) Genotype of the offspring from d4c7 PGCLCs for the Blimp1‐mVenus and Stella‐ECFP (BVSC) transgenes. Scale bars: (i) 2 mm; (ii) 200 μm; (iii) 500 μm; (iv, v) 100 μm.
3. Body weights of offspring from d4c7 PGCLCs induced from BDF1‐2 or BCF1‐2 ESCs and of the wild‐type mice. The mean values are indicated as bars.
4. Fertility of W/W ^v mice transplanted with d4c7 or d4 PGCLCs.
5. Transplantation of d4c7 PGCLCs harboring constitutive active LacZ transgene (BVSC R8) into neonatal or adult W/W ^v testes. LacZ staining (blue) was performed using the seminiferous tubules 6–10 weeks after transplantation. Spermatogenesis was observed in the seminiferous tubules when d4c7 PGCLCs were transplanted into neonate W/W ^v testes (i–iii), whereas no spermatogenesis was observed when they were transplanted into adult W/W ^v testes (iv, v). Spermatozoa found in the seminiferous tubules after LacZ staining are shown in (iii). Scale bars: (i, iv) 2 mm; (ii, v) 1 mm; (iii) 25 μm.
6. Summary of the transplantation.

In contrast, the testes transplanted with d4c7 or d4 PGCLCs from the BVSC R8 ESCs exhibited only a mild increase in size (Fig EV2A) and contained fewer seminiferous tubules with spermatogenesis, and the resultant spermatozoa did not reach the epididymis. We were nonetheless able to obtain apparently normal offspring with intracytoplasmic injection (ICSI) of the resultant spermatozoa, as we reported previously (Hayashi et al, 2011). Importantly, as in the case of d4/d6 PGCLCs or PGCs (Ohta et al, 2004), none of the d4c7 PGCLCs induced from any of the ESC lines were able to colonize adult testes (Fig EV2E and F). We did not find teratoma formation in any of the transplants. Collectively, these findings demonstrate that the PGCLCs expanded by FR10 in culture faithfully maintain their original functional property and that PGCLCs with a hybrid genetic background are powerful enough to reconstitute the spermatogenesis of infertile mice so that the recipient mice produce offspring with natural mating.

Transcriptomic properties of PGCLCs during expansion culture

We next went on to determine the detailed transcriptional properties of PGCLCs during their expansion in culture. First, immunofluorescence (IF) analyses revealed that, compared to male germ cells at E13.5, d4c7 PGCLCs, while expressing higher levels of OCT4, exhibited much lower levels of DDX4 and DAZL, key translational regulators that PGCs progressively up‐regulate after their colonization of embryonic gonads (Fujiwara et al, 1994; Cooke et al, 1996) (note that some d4c7 PGCLCs appeared to fully express DAZL) (Fig 4A). Second, we determined the transcriptomes of cultured PGCLCs [d4c3, d4c5, and d4c7 PGCLCs induced from the BVSC R8 ESCs (Appendix Fig S3B)] by an RNA sequencing (RNA‐seq) methodology (Nakamura et al, 2015, 2016) and compared them with those of ESCs, EpiLCs, d4/6 PGCLCs, and germ cells in vivo [PGCs at E9.5, E10.5, and E11.5; male/female germ cells at E12.5 and E13.5 (Kagiwada et al, 2013)] (Table EV2). The expression levels of key genes associated with germ cell development in these cell types are shown in Fig EV3. Remarkably, principal component analysis (PCA) clustered cultured PGCLCs closely with d4/6 PGCLCs and then with PGCs at E9.5, E10.5, and E11.5, but distantly with male/female germ cells at E12.5 and E13.5 (Fig 4B), indicating that PGCLCs grossly maintain their transcriptome during their expansion culture. On the other hand, unsupervised hierarchical clustering (UHC) and PCA among PGCLCs revealed a progressive transition of the properties of PGCLCs from d4 to d6, and then to d4c3, d4c5, and d4c7 PGCLCs (Appendix Fig S4A–C). Thus, during expansion culture, PGCLCs appear to undergo a directional transcriptional change, the initial phase of which also manifests in the floating aggregates.

Figure 4. Transcriptome of cultured PGCLCs.

• A
  Immunofluorescence (IF) analysis of the levels of DDX4 (top), DAZL (middle), and OCT4 (bottom) in d4c7 PGCLCs [Blimp1‐mVenus (BV)‐positive] compared to E13.5 male germ cells. The d4c7 PGCLCs are delineated by green dotted lines in the middle column. The ratios of the levels in d4c7 PGCLCs to the average levels in E13.5 male germ cells measured by densitometry [DDX4 (n = 48), DAZL (n = 77), OCT4 (n = 61)] are shown on the right (the averages were shown by red bars). Scale bar, 5 μm.
• B
  PCA of the transcriptome of the indicated cells.
• C, D
  Venn diagram showing the overlap of genes up (C)/down (D)‐regulated in d4c7 PGCLC and E13.5 male/female germ cells, compared to d6 PGCLCs. The numbers of genes in each category are indicated.
• E
  Box plots [the median (horizontal line), 25^th and 75^th percentiles (box), and 5^th and 95^th percentiles (error bars)] of the expression‐level differences compared to d6 PGCLCs (log₂ fold differences) of DEGs between d6 and d4c7 PGCLCs during PGCLC cultures or germ cell development (E9.5–E13.5). The color coding is as indicated.
• F
  Scatter‐plot representation of the log₂ expression‐level changes in E13.5 germ cells (the larger values in males or females) compared to d6 (x‐axis) and d4c7 (y‐axis) PGCLCs. Genes up‐regulated in d4c7 compared to d6 PGCLCs are indicated with red open circles (“d4c7/d6 > 2”), and if x > 1 (i.e., up‐regulated in E13.5 germ cells compared to d6 PGCLCs), they are classified according to the fold difference between E13.5 germ cells and d4c7 PGCLCs; “fully activated in E13.5” (up‐regulated in E13.5 germ cells, yellow); “fully activated in d4c7” (within twofold difference, cyan); and “over activated in d4c7” (down‐regulated in E13.5 germ cells, gray). Representative genes and selected GO terms are indicated. Previously reported “germline genes” (Weber et al, 2007; Borgel et al, 2010; Kurimoto et al, 2015) or other relevant genes are colored red or blue, respectively.

Figure EV3. Transcriptional dynamics during PGCLC expansion.

Expression levels [log₂ (RPM+1)] of key genes for pluripotency (Pou5f1, Sox2, Nanog and Klf4), early PGCs (Prdm1, Prdm14, Tfap2c, Dppa3, Nanos3, Klf2, and Klf5), late PGCs and meiosis (Ddx4, Dazl, Stra8, and Sycp3), mesoderm (T, Hoxa1, and Hoxb1), epigenetic regulation (Dnmt1, Dnmt3a, Dnmt3b, Uhrf1, Ehmt1, and Ehmt2), and cAMP signaling (Pde4a, Adcy6, and Adcy7) during PGCLC induction/expansion or germ cell development (E9.5–E13.5).

We identified differentially expressed genes (DEGs) between d4c7 and d6 PGCLCs, and between male/female germ cells at E13.5 and d6 PGCLCs. d4c7 PGCLCs up/down‐regulated 478 and 409 genes, respectively (Appendix Fig S4D), and the up‐regulated genes were enriched with those bearing gene ontology (GO) functional terms such as “intracellular signaling cascade” and “pattern specification process”, whereas the down‐regulated genes were enriched with those for various metabolic/biosynthetic processes (Appendix Fig S4E). Consistent with the PCA, the DEGs between male/female germ cells at E13.5 and d6 PGCLCs were much larger in number (Fig 4C and D): male/female germ cells at E13.5 up/down‐regulated 2,381 and 1,705 genes, respectively, and these DEGs exhibited enrichments of GO terms reflecting key developmental progressions during germ cell development (Appendix Fig S5A and B) (Dataset EV3). For example, the genes specifically up‐regulated in males were enriched with those for “transcription” (Foxo1, Utf1, Pou6f1) and “chromatin organization” (Ezh1, Prmt5, Kdm2a), the genes specifically up‐regulated in females were enriched with those for “regulation of transcription” (Gata2, Msx1, Cdx2) and “gamete generation” (Figla, Nr6a1, Rec8), and notably, the genes commonly up‐regulated in both males and females were enriched with those for “meiosis” (Spo11, Mael, Sycp1), “chromosome organization” (Ehmt1, Suv39h1, Smarcc1), and “methylation” (Piwil4, Satb1, Mll3), and accordingly, for the so‐called “germline genes” that have previously been identified as genes involved in germline functions such as meiosis and transposon repression and repressed in somatic cells primarily by DNA methylation (Weber et al, 2007; Borgel et al, 2010).

Consistent with UHC and PCA among PGCLCs, the DEGs between d4c7 and d6 PGCLCs acquired such states progressively during the culture (Fig 4E), and notably, they also exhibited progressive up/down‐regulation during germ cell development in vivo (Fig 4E): They comprised a minor part of—and a majority of them were included in—the DEGs between male/female germ cells at E13.5 and d6 PGCLCs (Fig 4C and D). Importantly, they did not exhibit a bias for sex‐specific regulation (Fig 4C and D). To gain more quantitative insight into the relationship between the DEGs between d4c7 PGCLCs and d6 PGCLCs and those between male/female germ cells at E13.5 and d6 PGCLCs, we plotted expression‐level differences between male/female germ cells at E13.5 and d4c7 PGCLCs against those between male/female germ cells at E13.5 and d6 PGCLCs (Fig 4F, Appendix Fig S5C). This analysis revealed that among the 306 genes commonly up‐regulated in male/female germ cells at E13.5 and d4c7 PGCLCs compared to d6 PGCLCs, 104 genes were only partially activated (E13.5−d4c7 > 2 folds) and 197 genes were fully activated (−2 folds < E13.5−d4c7 < 2 folds) in d4c7 PGCLCs (Fig 4F). The former included genes such as Ddx4, Dazl, Brdt, Asz1, Dmrt1, Stra8, Sycp3, Syce1, and Smc1b, and were enriched with the “germline genes”, whereas the latter included genes such as Piwil2, Rpl10 l, Rpl36, and the Rhox genes (Fig 4F). On the other hand, among the 252 genes commonly down‐regulated in male/female germ cells at E13.5 and d4c7 PGCLCs compared to d6 PGCLCs, 68 genes were only partially down‐regulated (E13.5−d4c7 < −2 folds) and 180 genes were fully down‐regulated (2 folds < E13.5−d4c7 < −2 folds) in d4c7 PGCLCs (Appendix Fig S5C). These findings demonstrate that, during expansion, PGCLCs, while essentially maintaining the characteristics of migrating PGCs, gradually acquire a part of the program for germ cell maturation that manifests in embryonic gonads prior to overt sex differentiation.

Epigenetic properties of PGCLCs during expansion culture

To explore the mechanism underpinning the characteristics of cultured PGCLCs, we next went on to determine their epigenetic profiles. First, we performed IF analyses, which revealed that, compared to EpiLCs, d4c7 PGCLCs exhibited much lower levels of 5‐methylcytone (5mC) (Fig 5A). Consistent with the results of the transcriptome analyses (Fig EV3), compared to EpiLCs, d4c7 PGCLCs expressed a similar level of DNMT1, but much lower levels of DNMT3A/3B and UHRF1 (Fig 5B). Furthermore, compared to EpiLCs, they exhibited higher and lower levels of histone H3 lysine 27 tri‐methylation [H3K27me3: representing repression by polycomb complex 2 (PRC2)] and H3K9 di‐methylation [H3K9me2: representing repression by G9A/GLP], respectively (Fig 5A). Thus, the epigenetic properties of d4c7 PGCLCs appeared to be grossly similar to those of d6 PGCLCs (Hayashi et al, 2011; Kurimoto et al, 2015), except that d4c7 PGCLCs appeared to bear much lower 5mC levels than d6 PGCLCs.

Figure 5. Key epigenetic properties of cultured PGCLCs.

1. IF analysis of the levels of 5mC (top), H3K27me3 (middle), and H3K9me2 (bottom) in d4c7 PGCLCs [Blimp1‐mVenus (BV)‐positive] compared to EpiLCs. The d4c7 PGCLCs are delineated by green dotted lines in the middle column. The relative levels in d4c7 PGCLCs compared to the average in EpiLCs measured by densitometry [5mC (n = 49), H3K27me3 (n = 44), H3K9me2 (n = 46)] are shown on the right (the averages were shown by red bars). Scale bar, 5 μm.
2. IF analysis of the levels of DNMT1, DNMT3A, DNMT3B, and UHRF1 in d4c7 PGCLCs (BV‐positive) compared to EpiLCs. The d4c7 PGCLCs are delineated by green dotted lines in the middle column. The relative levels in d4c7 PGCLCs compared to the average in EpiLCs measured by densitometry [DNMT1 (n = 57), DNMT3A (n = 56), DNMT3B (n = 51), UHRF1 (n = 55)] are shown on the right (the averages are indicated by red bars). Scale bar, 5 μm.
3. ChIP‐seq (H3K4me3, H3K27ac, and H3K27me3) and 5mC‐level tracks in 100‐kb regions around the Prdm14 (left) and Hoxb cluster (right) in the indicated cell types. The d4c7 PGCLCs are shaded in pink. Transcription start sites (TSSs) are indicated with dotted lines.

We therefore next quantified the genome‐wide levels and distributions of DNA methylation in d4c3 and d4c7 PGCLCs (induced from the BVSC R8 ESCs) (Appendix Fig S3) by whole‐genome bisulfite sequencing (WGBS) (Table EV3) and of H3K4me3 (representing promoter activity), H3K27 acetylation (ac) (representing active enhancers), and H3K27me3 in d4c7 PGCLCs (induced from the BVSC R8 or BDF1‐2 ESCs) by chromatin immunoprecipitation followed by massively parallel sequencing (ChIP‐seq) (Table EV4), and analyzed these data in comparison with those of key cell types during PGCLC induction (ESCs, EpiLCs, and d2, d4, and d6 PGCLCs), which we reported recently (Kurimoto et al, 2015; Shirane et al, 2016). Since bisulfite sequencing does not discriminate 5mC and 5‐hydroxymethylcytone (5hmC) (Hayatsu & Shiragami, 1979), and since the 5hmC level during PGCLC induction is almost negligible (Shirane et al, 2016), we hereafter designate 5mC and 5hmC collectively as 5mC. Figure 5C shows WGBS and ChIP‐seq track transitions around the Prdm14 locus and the Hoxb cluster. Consistent with the IF analyses, remarkably, the 5mCs were almost fully erased in both loci during the PGCLC culture, whereas both active (H3K4me3 and H3K27ac) and repressive (H3K27me3) histone modifications exhibited relatively similar distributions between d6 and d4c7 PGCLCs (Fig 5C), suggesting that the PGCLC expansion is a process to progressively erase 5mCs, while maintaining the histone modifications.

Comprehensive DNA methylation erasure in cultured PGCLCs

Next, we performed more detailed analyses on the 5mC‐level dynamics during the PGCLC culture. We focused on the 5mCs in CpG contexts, since CpH (where H = A, C, or T) methylation is limited during PGCLC induction (Shirane et al, 2016) and has not been found to exhibit any clear biological role in mammalian cells (Schubeler, 2015). As a key parameter for the genome‐wide DNA methylation state, we determined the average 5mC levels of the total unique sequence regions (unique regions: 2‐kb sliding windows with 1‐kb overlap). We separately determined the 5mC levels of promoters (high, intermediate, and low CpG‐density promoters: HCPs, ICPs, and LCPs, respectively) (Weber et al, 2007), the consensus sequences of repetitive elements [long interspersed nuclear element 1 (LINE1); intracisternal A particles (IAPs); endogenous retrovirus sequences (ERVs) other than IAPs; and major and minor satellites], and imprint control regions (ICRs) of imprinted genes (Dataset EV4). We also determined the 5mC levels of non‐promoter CpG islands (CGIs), exons, introns, intergenic regions, cell type‐specific enhancers (Kurimoto et al, 2015), and the CGIs of “germline genes” (Weber et al, 2007; Borgel et al, 2010; Kurimoto et al, 2015) (Dataset EV4).

As we reported recently, PGCLCs exhibit a progressive dilution of the 5mCs established in EpiLCs, which bear a DNA methylome highly similar to that of the epiblast, and consequently, d6 PGCLCs acquire an average 5mC level of ~37%, a state considered to be similar to that in migrating PGCs at ~E9.0–9.5 (Seisenberger et al, 2012; Kobayashi et al, 2013; Shirane et al, 2016) (Fig 6A). Strikingly, in agreement with the analyses of Prdm14 and Hoxb loci, in the cultured PGCLCs, there was a progressive dilution of the 5mC level of the d4/d6 cells in essentially all genomic regions, with differing kinetics for unique regions, repeats, and distinct regulatory elements, and as a result, there was an average 5mC level of only ~6% in the d4c7 cells (Figs 6A and EV4, Appendix Fig S6), a level equivalent to that in E13.5 germ cells, which bear the lowest 5mC level throughout the germline cycle (Seisenberger et al, 2012; Kobayashi et al, 2013). Importantly, the 5mC distribution patterns in essentially all genomic elements, including the repeats, the promoters of demethylation‐resistant “germline genes”, and the ICRs of imprinted genes, were remarkably similar between d4c3 PGCLCs and E10.5 PGCs, and between d4c7 PGCLCs and E13.5 germ cells (Fig 6B), whereas those between d4c3 PGCLCs and ESCs cultured with two kinase inhibitors (2i) (Habibi et al, 2013; Shirane et al, 2016), which exhibited similar 5mC levels, were divergent (Appendix Fig S7), suggesting that DNA demethylation in cultured PGCLCs involves similar, if not identical, mechanism as that in PGCs in vivo, but not that evoked by 2i in vitro. We analyzed the “escapees” that evade the DNA demethylation (5mC > 20%) (Seisenberger et al, 2012) and found that the escapees were also largely overlapped between d4c7 PGCLCs and E13.5 germ cells (Appendix Fig S8), and a majority of them were in the vicinity of IAPs (Appendix Fig S8). Thus, the induction of PGCLCs and their culture with FR10 on m220 feeders reconstituted DNA methylation reprogramming in PGCs in a comprehensive fashion. Nonetheless, as we demonstrated above, the cultured PGCLCs basically maintained the transcriptional state of migrating PGCs (Fig 4B).

Figure 6. DNA methylation erasure in cultured PGCLCs.

1. Scatter‐plot comparisons of 5mC levels in d6, d4c3, d4c7 PGCLCs, and E10.5 and E13.5 male germ cells against those in EpiLCs. The 5mC levels of 2‐kb unique genomic regions (contour plot, top), ICRs and “germline genes” (n = 102) (middle), and repeat consensus sequences (bottom) are shown, and the latter two are shown along with the 5mC levels of promoters. Contour lines are drawn with an interval of 100 regions, the yellow dotted line connects the origin and the peak, and the slopes are shown. The color coding is as indicated.
2. Scatter‐plot comparisons of 5mC levels between E10.5 male PGCs and d4c3 PGCLCs and between E13.5 male germ cells and d4c7 PGCLCs. The color coding is as in (A).
3. Definition of promoters demethylated between d6 and d4c7 PGCLCs; 5mC > 20% in d6 and < 20% in d4c7 (red open circles, n = 7,737).
4. Venn diagram showing overlap among promoter DNA demethylation and differentially expressed genes between d6 and d4c7 PGCLCs. The promoters demethylated between d6 and d4c7 PGCLCs are classified in LCPs and non‐LCPs. The numbers of genes in each category are indicated.

Figure EV4. The 5mC‐level dynamics of key genomic elements during PGCLC induction/expansion.

• A–C
  Box plots [the median (horizontal line), 25^th and 75^th percentiles (box), and 5^th and 95^th percentiles (error bars)] of the 5mC levels in (A) genes [promoters (HCPs, ICPs, and LCPs), exons, and introns], (B) cell type‐specific enhancers (Kurimoto et al, 2015), (C) randomly sampled 2‐kb unique genomic regions (n = 19,902), randomly sampled intergenic regions (n = 3,087), and unique regions overlapped with repeats defined using RepeatMasker (LINEs, SINEs, and IAPs). d4c3 PGCLCs and E10.5 male PGCs are colored yellow, and d4c7 PGCLCs and E13.5 male germ cells are colored orange.
• D
  Fold changes of the averaged 5mC levels compared to EpiLCs in PGCLCs and male germ cells (E10.5, E13.5) for repeat consensus sequences (IAPs, non‐IAP ERVs, LINE1, major and minor satellites) and paternal and maternal ICRs (left), non‐promoter CGIs and cell type‐specific enhancers (middle), and randomly sampled intergenic regions, exons, and introns (right). Averaged 5mC levels of randomly sampled 2‐kb unique genomic regions are also plotted to show the global changes. The color coding is as indicated.
• E
  The transitions of the 5mC levels of individual imprint DMRs during PGCLC induction/expansion and in male germ cells (E10.5, E13.5). The color coding is as indicated.

We therefore examined the impact of promoter demethylation on transcriptional activation in d4c7 PGCLCs. Reflecting the global DNA demethylation in cultured PGCLCs, as many as 7,737 promoters were demethylated between d6 and d4c7 PGCLCs (5mC > 20% in d6 and < 20% in d4c7) (Fig 6C). Among the 478 genes up‐regulated in d4c7 PGCLCs compared to d6 PGCLCs, 96 genes were promoter‐demethylated, 27 were among the 104 partially up‐regulated genes (E13.5−d4c7 > 2 folds) (Ddx4, Dazl, Brdt, Asz1, Dmrt1, Stra8, Sycp3, Syce1, Smc1b, etc.), and 34 were among the fully up‐regulated 197 genes (−2 folds < E13.5−d4c7 < 2 folds) (Piwil2, Rpl10l, Rpl36, the Rhox genes, etc.) (Fig 6D, Appendix Fig S9A). The proportions of the promoter‐demethylated genes among the partially/fully up‐regulated genes were higher than those of the promoter‐demethylated genes among the partially/fully down‐regulated genes (Fig 6D, Appendix Fig S9A and B). We conclude that promoter demethylation itself contributes partially to the activation of only a limited number of specific genes in cultured PGCLCs.

Compensatory up‐regulation of H3K27me3 in demethylated promoters in cultured PGCLCs

We next analyzed the dynamics of the H3K4me3, H3K27ac, and H3K27me3 distributions during the PGCLC induction and expansion. As in the cases of the key cell types during the PGCLC induction, high levels of H3K4me3 were associated predominantly with HCPs in d4c7 PGCLCs (Fig EV5A and B), the H3K4me3 levels around the transcription start sites (TSSs) were positively correlated with the expression levels of the associated genes (Fig EV5C), and reflecting a transcriptomic similarity between d6 and d4c7 PGCLCs, H3K4me3 profiles across the genome were similar between d6 and d4c7 PGCLCs (Fig EV5D) (Dataset EV5). Importantly, the distributions of H3K27ac, which represent enhancer usage and exhibit highly dynamic alterations across different cell types (Calo & Wysocka, 2013) and during the PGCLC induction (Kurimoto et al, 2015), were also similar between d6 and d4c7 PGCLCs (Fig 7A and B) (Dataset EV6), indicating that not only the gene expression itself but also the regulation of gene expression is grossly preserved during the PGCLC culture. Nonetheless, we identified H3K27ac peaks (< 15 kb from the TSSs and gene bodies) specific to either d6 or d4c7 PGCLCs, which would represent potential regulatory differences between d6 and d4c7 PGCLCs (Fig EV5E and F).

Figure EV5. Histone‐modification profiles in cultured PGCLCs.

• A, B
  The log₂ H3K4me3 level‐frequency plots for the HCP (blue), ICP (green), and LCP (orange) genes in d6 (A) and d4c7 PGCLCs (B).
• C
  Scatter plots of log₂ gene expression (x‐axis) and H3K4me3 (y‐axis) levels in d6 (top) and d4c7 (bottom) PGCLCs.
• D
  Heat‐map representation of the correlation coefficients of all H3K4me3 peaks in the indicated pairwise comparisons. The color coding is as indicated.
• E, F
  Selected GO terms enriched in genes associated with H3K27ac peaks biased to d4c7 (E) or d6 (F) PGCLCs. Representative genes for each term are listed on the right.
• G
  Definition of promoters demethylated only in d4c7 PGCLCs (5mC > 20% in d6 and < 20% in d4c7, red open circles), demethylated in d6 PGCLCs but not in EpiLCs (5mC > 20% in EpiLCs and < 20% in d6 PGCLCs, green open circles), and demethylated/unmethylated in both EpiLCs and d6 PGCLCs (5mC < 5% in both).
• H
  Scatter plots of log₂ H3K27me3 levels of EpiLCs (x‐axis) and d6 PGCLCs (y‐axis) around TSSs of all genes (left), demethylated only in d4c7 PGCLCs (middle left), demethylated/unmethylated in EpiLC and d6 PGCLCs (middle right), and demethylated in d6 PGCLCs (right). The color coding is as in (G).

Figure 7. Histone‐modification dynamics in cultured PGCLCs.

1. Scatter‐plot comparisons of log₂ H3K27ac levels between EpiLCs and d6 PGCLCs (left), and between d6 and d4c7 PGCLCs (right). The d4c7‐ and d6‐biased H3K27ac peaks are shown in orange and cyan, respectively.
2. Heat‐map representation of correlation coefficients of H3K27ac levels in the indicated pairwise comparisons. The color coding is as indicated.
3. Definitions of promoters demethylated/unmethylated only in d4c7 PGCLCs (5mC > 20% in d6 and < 20% in d4c7, red open circles) and in both d6 and d4c7 PGCLCs (5mC < 5% in d6 and d4c7, blue open circles).
4. Scatter‐plot comparisons of log₂ H3K27me3 levels between d6 and d4c7 PGCLCs around the TSSs of all genes (left, black), genes with promoters demethylated only in d4c7 PGCLCs (middle, red), and genes with promoters demethylated/unmethylated in both d6 and d4c7 PGCLCs (right, blue).
5. Numbers of bivalent genes in ESCs, EpiLCs, and d6 and d4c7 PGCLCs.
6. Transition of the indicated GO term enrichments during PGCLC induction and expansion.

The distribution patterns of H3K27me3 also appeared to be grossly similar between d6 and d4c7 PGCLCs (Fig 7C and D). Importantly, however, we noted that the promoters with substantial demethylation between d6 and d4c7 PGCLCs (5mC > 20% in d6 and < 20% in d4c7, 7737 promoters), but not those showing no 5mC‐level changes, exhibited a general trend for higher enrichment levels of H3K27me3 in d4c7 PGCLCs than in d6 PGCLCs (Fig 7C and D). We also noted that such promoters did not show overall changes of H3K27me3 enrichment levels between EpiLCs and d6 PGCLCs, even though they also exhibited substantial demethylation between EpiLCs and d6 PGCLCs (Fig EV5G and H). These findings indicate that extensive and nearly complete promoter demethylation in cultured PGCLCs is at least in part compensated by concomitant up‐regulation of H3K27me3 enrichment levels, which would contribute to the maintenance of a transcriptional state of migrating PGCs in cultured PGCLCs. Examples of genes exhibiting such trends are shown in Appendix Fig S10.

We evaluated the bivalent promoters with both activating H3K4me3 and repressing H3K27me3, which may represent a state poised for activation upon appropriate developmental cues (Voigt et al, 2013) (for the definition of bivalency, see Materials and Methods). Consistent with our previous report (Kurimoto et al, 2015), the numbers of the bivalent genes were largest in EpiLCs among the key cultured cell types, and d6 and d4c7 PGCLCs exhibited similar numbers of bivalent genes (Fig 7E). The overlap of the bivalent genes between d6 and d4c7 PGCLCs, however, was relatively moderate [~519/1,058 (~49%)] (Appendix Fig S11A), which was in part due to an inherent technical difficulty in precisely comparing the combinatorial levels of two histone modifications in two different samples. Nonetheless, the bivalent genes in d4c7 PGCLCs exhibited higher enrichments in GO terms such as “pattern specification process” and “embryonic morphogenesis” compared to those in d6 PGCLCs (Fig 7F, Appendix Fig S11A); for instance, d4c7 PGCLCs acquired elevated H3K4me3 levels around the Hoxc cluster, even though the cluster was repressed by high levels of H3K27me3 (Appendix Fig S11B). We conclude that d4c7 PGCLCs bear an epigenome highly poised for developmental regulators, with extremely low levels of genome‐wide 5mC and H3K9me2 levels (Figs 5A and 6) (Kurimoto et al, 2015), representing an epigenetic “blank slate”.

X chromosome dynamics in cultured female PGCLCs

Female ESCs bear two active X chromosomes (XaXa) and show lower genome‐wide 5mC levels compared to male ESCs (Zvetkova et al, 2005; Shirane et al, 2016). A majority of female EpiLCs bear XaXa, and upon floating aggregate formation, they undergo X inactivation, with female mPGCLCs exhibiting one Xa and one inactive X chromosome (XaXi) (Hayashi et al, 2012; Shirane et al, 2016). We next evaluated the X chromosome dynamics during female PGCLC expansion in culture. Since female ESCs are vulnerable to losing one X chromosome to become XO (Robertson et al, 1983; Zvetkova et al, 2005), we first examined the extent of the maintenance of two X chromosomes during PGCLC induction and expansion from two lines of female ESCs, BVSC H14 and H18 (129sv/C57BL/6 background) by DNA fluorescence in situ hybridization (FISH) analysis of an X‐linked gene, Huwe1. As shown in Fig 8A, the two Xs were relatively well maintained during the ESCs‐to‐EpiLC differentiation, but upon induction of PGCLCs, one X tended to be lost (XO: ~60 and ~40% for H14 and H18 d4 PGCLCs, respectively), and the XX/XO ratios were well preserved during the expansion of PGCLCs in culture (XO: ~60 and ~40% for H14 and H18 d4c7 PGCLCs, respectively).

Figure 8. X chromosome re‐activation in cultured female PGCLCs.

1. Loss of one X chromosome in female PGCLCs during PGCLC induction. (Left) Representative images of female EpiLCs stained for Huwe1 by DNA FISH. XX and XO EpiLCs are delineated by blue and orange dotted circles, respectively. (Right) Numbers of X chromosomes during PGCLC induction/expansion from two female ESC lines (H14 and H18). Scale bar, 25 μm.
2. Evaluation of X chromosome re‐activation in cultured female PGCLCs by double staining for Huwe1 and H3K27me3. (Left) Representative images for DNA FISH of Huwe1 and immunofluorescence of H3K27me3. X chromosome re‐activation was evaluated in cells retaining two X chromosomes. (Right) Analyses of the Huwe1 and H3K27me3 signals in female MEFs and d4/d4c3/d4c7 PGCLCs. Scale bar, 5 μm.
3. A model for epigenetic regulation during PGCLC induction/expansion. (Left) In vivo, from E9.5 to E13.5, both male and female germ cells propagate substantially (˜>100‐fold expansion) (Tam & Snow, 1981; Kagiwada et al, 2013) and comprehensively erase their DNA methylome. Meanwhile, from around E11.5, they receive signals from gonadal somatic cells, fully acquire “germline” genes, and initiate male or female differentiation [reviewed in (Spiller & Bowles, 2015)]. (Right) During the expansion culture from d4 to d4c7 PGCLCs (˜20‐fold expansion), PGCLCs comprehensively erase their DNA methylome as PGCs/germ cells in vivo. However, due to the lack of cues corresponding to the signals from gonadal somatic cells, PGCLCs essentially preserve their initial transcriptional properties, at least in part through a compensatory up‐regulation of H3K27me3 levels around key genes, and thus acquire “germline genes”, and male/female properties only moderately. See Dataset EV7 for the summary of expression, 5mC, H3K4me3, and H3K27me3 levels of genes analyzed in this study.

We then examined the Xa/Xi state of PGCLCs during their expansion by evaluating the H3K27me3 positivity on the X chromosomes. As shown in Fig 8B, ~95% of female mouse embryonic feeders (MEFs) had two Xs and ~90% of such cells exhibited a single H3K27me3 spot on one X, indicating their XaXi state. We found that ~50% of the XX d4 PGCLCs exhibited the XaXi state, whereas ~30% of the XX d4 PGCLCs did not show an H3K27me3 spot on the X chromosomes (Fig 8B). Remarkably, only a minority (< 5%) of XX d4c3 and d4c7 PGCLCs exhibited the XaXi state and ~70% of them exhibited no H3K27me3 spot on the X chromosomes (Fig 8B), indicating that one Xi was re‐activated or in the process of re‐activation during the expansion culture of PGCLCs, which is in good agreement with the notion that d4c7 PGCLCs acquire an epigenetic “blank slate”.

Discussion

We have shown here that Rolipram and Forskolin, which elevate intracellular cAMP concentrations via a synergistic mechanism, expand PGCLCs up to ~50‐fold in the presence of SCF on the m220 feeders (Fig 2), and the expanded PGCLCs robustly contribute to spermatogenesis (Fig 3). Thus, cAMP signaling and SCF‐KIT signaling are the key to PGCLC/PGC proliferation. On the other hand, we have not been successful in extending the PGCLC expansion culture for more than 7 days: Under the current condition, cultured PGCLCs ceased to proliferate and began to decline after ~d9, and they were not maintained upon passaging. In this regard, it would be interesting to examine the effect of other chemicals/signaling pathways identified by our screen on further proliferation of PGCLCs. Rolipram and Forskolin also propagated E9.5 PGCs, but to a limited extent (up to ~eightfold expansion, Fig 2C). Thus, a condition to extend the PGCLC/PGC expansion culture as well as the mechanisms by which cAMP SCF‐KIT signaling and SCF‐KIT signaling activate the cycling of PGCLCs/PGCs warrant future exploration.

A striking finding was that, upon proliferation, PGCLCs progressively erase their 5mCs apparently at a constant speed and in a manner highly parallel to that in PGCs for all genomic elements, including unique regions, repeats, and distinct regulatory elements such as the ICR of the imprint genes and the promoters of “germline genes” (Fig 6). Consequently, d4c7 PGCLCs establish a DNA methylome essentially identical to that in male/female germ cells at E13.5 (Fig 6). This was clearly a specific event in PGCLCs, since ESCs cultured with FR10, despite their robust propagation, did not show such demethylation (Appendix Fig S12). These observations provide strong evidence in support of two key concepts. First, the dilution of the 5mC patterns established in the epiblast, most likely through a replication‐coupled passive mechanism, is the principle for DNA methylation erasure in PGCs. This concept is also supported by the fact that knockout mice of TET1, a primary candidate for the mediation of active DNA demethylation, undergo normal genome‐wide DNA demethylation in PGCs (Yamaguchi et al, 2012), and by other studies supporting a replication‐coupled passive demethylation in PGCs (Seisenberger et al, 2012; Kagiwada et al, 2013; Arand et al, 2015). Second, for the genome‐wide DNA methylation erasure to occur, cues from embryonic gonads are essentially dispensable. Thus, it is likely that genome‐wide DNA demethylation and sexual differentiation of germ cells that manifests in embryonic gonads by E13.5 are genetically separable events (Fig 8C).

It has previously been reported that an erasure of global H3K27me3 patterns in PGCs, possibly through genome‐wide histone replacement, accompanies genome‐wide DNA demethylation in PGCs at around E11.5 (Hajkova et al, 2008; Mansour et al, 2012). After careful examination, however, we did not observe such erasure of H3K27me3 in PGCs in a previous study (Kagiwada et al, 2013), and we did not observe it in PGCLCs in this study. To the contrary, PGCLCs grossly maintained their histone‐modification patterns during their expansion, with up‐regulation of H3K27me3 enrichment levels around demethylated promoters, and this would be expected to contribute to the relative stability of their transcriptome during expansion (Figs 4 and 7). Thus, PGCLCs expanded in culture free from gonadal cues for sex differentiation can be considered a recapitulation of a unique epigenetic “blank slate” with essentially no DNA methylation and with abundant bivalency. In good agreement with this conclusion, we have shown that cultured female PGCLCs, although sometimes losing a single X, appear to re‐activate an Xi that is inactivated upon floating aggregate formation of EpiLCs (Hayashi et al, 2012; Shirane et al, 2016). Thus, female PGCLC induction and expansion culture will serve as a unique system to analyze the mechanism for X re‐activation.

The expansion of PGCs in culture has been difficult and has met with only limited success, which has been a major barrier in promoting PGC biology. The capacity to faithfully expand PGCLCs up to ~50‐fold in culture would be instructive for addressing key questions in PGC biology, including the mechanism of the regulation of genome function during epigenetic reprogramming, the mechanism for male versus female differentiation of germ cells, and accordingly, the mechanism for meiotic entry in female germ cells. Moreover, after further modifications/improvements, our strategy could be applied to PGCs/PGCLCs in other mammalian species, including humans, opening the possibility of expanding and maturing human PGCLCs for their further differentiation in culture.

Materials and Methods

Mice

All the animal experiments were performed under the ethical guidelines of Kyoto University. The BVSC (Acc. No. BV, CDB0460T; SC, CDB0465T: http://www.cdb.riken.jp/arg/TG%20mutant%20mice%20list.html) and Stella‐EGFP transgenic mice were established as reported previously (Payer et al, 2006; Seki et al, 2007; Ohinata et al, 2008) and maintained on a largely C57BL/6 background. The WBB6F1‐W/W ^v, C57BL/6, DBA/2, C3H, BDF1, and ICR mice were purchased from SLC (Shizuoka, Japan). Noon of the day when a copulation plug was identified was designated as embryonic day (E) 0.5. All mice were housed in a specific pathogen‐free animal facility under a 14‐h light/10‐h dark cycle.

Derivation and culture of ESCs

The BVSC R8, H14, and H18 were reported previously (Hayashi et al, 2011, 2012). The female BVSC mice (largely C57BL/6 background) were mated with male DBA/2 or C3H mice to obtain BDF1 or BCF1 embryos. The blastocysts were placed and cultured in a well of a 96‐well plate in N2B27 medium with 2i (PD0325901, 0.4 μM: Stemgent, San Diego, CA; CHIR99021, 3 μM: Stemgent) and LIF (1,000 U/ml; Merck Millipore) on mouse embryonic feeders (MEFs) (Ying et al, 2008; Hayashi et al, 2011). The expanded colonies were passaged by dissociating with TrypLE (Thermo Fisher Scientific). Until passage 2, the ESCs were maintained on MEFs. Thereafter, male ESCs were cultured and maintained feeder‐free on a dish coated with poly‐L‐ornithine (0.01%; Sigma) and laminin (10 ng/ml; BD Biosciences).

Induction of EpiLCs and PGCLCs

Induction of EpiLCs and PGCLCs was performed as reported previously (Hayashi et al, 2011). Briefly, the EpiLCs were induced by plating 1 × 10⁵ ESCs on a well of a 12‐well plate coated with human plasma fibronectin (16.7 mg/ml) in N2B27 medium containing activin A (20 ng/ml), bFGF (12 ng/ml), and KSR (1%). The PGCLCs were induced from d2 EpiLCs under a floating condition in a well of a low‐cell‐binding U‐bottom 96‐well plate (Thermo Scientific) in serum‐free medium [GK15; GMEM (Thermo Fisher Scientific) with 15% KSR, 0.1 mM NEAA, 1 mM sodium pyruvate, 0.1 mM 2‐mercaptoethanol, 100 U/ml penicillin, 0.1 mg/ml streptomycin, and 2 mM l‐glutamine] in the presence of the cytokines BMP4 (500 ng/ml; R&D Systems), LIF (1,000 U/ml; Merck Millipore), SCF (100 ng/ml; R&D Systems), and EGF (50 ng/ml; R&D Systems). To prepare large numbers of PGCLCs for the chemical library screening, PGCLCs were induced in AggreWell400 (STEMCELL Technologies) using the same medium.

Fluorescence‐activated cell sorting

The sample preparations from cell aggregates were performed as described previously (Hayashi et al, 2011). FACS was performed with a FACSAriaIII (BD) cell sorter. BV and SC fluorescence were detected with the FITC and AmCyan Horizon V500 channel, respectively. Data were analyzed with FACSDiva (BD) software.

Establishment of m220 sublines

The m220 cell line (Majumdar et al, 1994) were cultured in a gelatin‐coated plate in DMEM with 10% FCS. Since m220 cells were highly vulnerable to mitomycin C (MMC) treatment, we established sublines of m220 resistant to MMC. Briefly, single m220 cells were plated on wells of 96‐well plates (6 plates) by FACS. One week after the plating, cell growth was observed in approximately half of the wells. After passaging the cells to one well of each of the two 96‐well plates, one plate was frozen as a replica and the other plate was treated with MMC (4 μg/ml, 2 h). At 10 days after MMC treatment, MMC resistance was evaluated by microscopic observation. In total, 242 m220 sublines were established and seven sublines showed high MMC resistance. The m220‐5 subline was primarily used for experiments.

Detection of BV (+) PGCLCs by a cell analyzer

d4 PGCLCs were plated on m220‐5 feeders in a 96‐well plate by FACS, and BV fluorescence was monitored by a cell analyzer (Cellavista; SynenTec). The fluorescence photos for BV were taken by a Cellavista cell analyzer with the following settings: 10× objectives; exposure time: 140 μsec; gain: 4×; binning: 4 × 4; excitation: 500/24 nm; and emission: 542/27 nm. The BV fluorescence was detected by using the following algorithm/attribute parameters: sensitivity: 10; region merging: 200; min. granule intensity: 50; well edge distance: 200; contrast: 1; size: 3,000; intensity: 255; roughness: 500; granularity: 100; granule intensity: 255; granule count: 10,000; longishness: 100; and compactness: 1. The values of “cell nuclei” were used for the detection of BV fluorescence.

Chemical library screening for PGCLC proliferation

We screened the chemical libraries listed in Fig EV1C at concentrations of 10 and 1 μM. We used 96‐well plates with MMC‐treated m220‐5 cells. In each 96‐well plate, negative (DMSO only) and positive (LIF) controls were allocated to both sides and the compounds were added to 80 wells (Fig EV1B). Two hundred BV (+) d4 PGCLCs induced from BDF1‐2 ESCs were plated in a well of 96‐well plates and BV fluorescence was measured on culture day 1 (c1), c3, c5, and c7 by a Cellavista cell analyzer. The values of “cell nuclei” were used for the detection of BV fluorescence. Because the proliferation rates of d4 PGCLCs differed slightly among the experiments, the values from different experiments were adjusted based on the average values of negative controls obtained from the first experiment. For each compound, the fold difference of the BV fluorescence between c1 and c7 was calculated and the compounds with fold difference values > 3 SDs of the mean values of the negative controls were identified as those enhancing the proliferation of PGCLCs.

Expansion culture of d4 PGCLCs and E9.5 PGCs

m220‐5 cells were cultured for three passages with 10 μM of both Forskolin and Rolipram to permit adaptation to these compounds prior to MMC treatment (1–2 μg/ml, 2 h), since the MMC‐treated m220‐5 feeders were vulnerable to Forskolin and Rolipram. d4 PGCLCs or E9.5 PGCs (BDF1×Stella–EGFP) were sorted by FACS and plated on m220‐5 cells in GMEM containing 10% KSR, 0.1 mM NEAA, 1 mM sodium pyruvate, 0.1 mM 2‐mercaptoethanol, 100 U/ml penicillin, 0.1 mg/ml streptomycin, 2 mM l‐glutamine, 2.5% FCS, 100 ng/ml SCF, 10 μM Forskolin, and 10 μM Rolipram. Half of the culture medium was changed every 2 days.

Immunofluorescence

The following antibodies were used at the indicated dilutions: rabbit anti‐MVH (1/250; Abcam ab13840); rabbit anti‐DAZL (1/250; Abcam ab34139); mouse anti‐OCT4 (1/250; BD 611203); mouse anti‐5mC (1/500; Abcam ab10805); rabbit anti‐H3K27me3 (1/500; Millipore 07‐449); rabbit anti‐H3K9me2 (1/500; Millipore 07‐441); rabbit anti‐DNMT1 (1/100; Santa Cruz Biotechnology sc‐20701); mouse anti‐DNMT3A (1/200; Abcam ab13888); mouse anti‐DNMT3B (1/200; Novus Biologicals NB100‐56514); rabbit anti‐UHRF1 (1/100; Santa Cruz Biotechnology sc‐98817); and chicken anti‐GFP (1/500; Abcam ab13970). The following secondary antibodies from Thermo Fisher Scientific were used at a 1/500 dilution: Alexa Fluor 568 goat anti‐rabbit IgG; Alexa Fluor 568 goat anti‐mouse IgG; and Alexa Fluor 488 goat anti‐chicken IgG. For staining F‐actin, Alexa Fluor 568‐conjugated phalloidin (1/40, Thermo Fisher Scientific A12380) was used.

The protocol for immunofluorescence staining was described previously (Hayashi et al, 2011; Nakaki et al, 2013). For the MVH, DAZL, and OCT4 staining, d4c7 PGCLCs (BDF1‐2) were sorted by FACS, mixed with E13.5 male PGCs at a ratio of 1:1 and spread onto MAS‐coated glass slides with Cyto Spin 4 (Thermo Fisher Scientific). The E13.5 male germ cells (ICR) were sorted using SSEA1 antibody conjugated with Alexa Fluor 647 by FACS. For the 5mC, H3K27me3 and H3K9me2 staining, d4c7 PGCLCs (BDF1‐2) were sorted by FACS, mixed with d2 EpiLCs at a ratio of 1:1 and spread onto MAS‐coated glass slides with Cyto Spin 4 (Thermo Fisher Scientific). Images were captured with a confocal microscope (Zeiss, LSM780) and the signal intensities were analyzed by ImageJ (NIH).

Measurement of cAMP concentration

The intracellular cAMP concentration was measured using a cAMP‐Glo Max assay kit (Promega) according to the manufacturer's instructions. The standard curve using purified cAMP was generated by calculating the Δ relative light units (RLU [0 nM] – RLU [X nM]). For each sample, 1 × 10⁴ d4 PGCLCs were pretreated with Forskolin and/or Rolipram for 30 min at RT, and the Δ RLU (RLU [untreated sample] – RLU [treated sample]) was calculated. The increase in intracellular cAMP levels by the chemical treatment was inferred from the cAMP standard curve. Three biological replicates were analyzed for each sample.

Cell‐cycle analysis

The cell‐cycle statuses of ESCs, EpiLCs, d4, d4c3, d4c5, and d4c7 PGCLCs (BDF1‐2) and male germ cells at E13.5, E14.5, and E15.5 were examined as reported previously (Kagiwada et al, 2013). To label cultured cells, cells were incubated with BrdU (10 μM) for 30 min. To label germ cells, female mice (ICR) were mated with Stella–EGFP males, pregnant females were intraperitoneally injected with 1 mg of BrdU, and embryos were collected after 30 min. The cultured cells or male gonads were dispersed into single cells by TrypLE treatment. For the detection of BrdU incorporation, we used an APC‐BrdU Flow Kit (BD Biosciences) according to the manufacturer's instructions. The stained samples were analyzed using BD FACSAriaIII (BD) with FACSDiva (BD) software with PGCLCs and male germ cells identified by BV or Stella‐EGFP fluorescence, respectively. Three biological replicates were analyzed for each sample.

Transplantation of PGCLCs into the testis of W/W ^v mice

After PGCLCs were purified by FACS, 1 × 10⁴ to 1 × 10⁵ cells per testis were injected into the testes of randomly selected neonate (7 days post‐partum) or adult W/W ^v mice as previously described (Chuma et al, 2005). Anti‐mouse CD4 antibody (50 mg per dose, clone GK1.5; Biolegend) was injected intraperitoneally on day 0, 2, or 4 for immunosuppression when necessary (Kanatsu‐Shinohara et al, 2003). For the assessment of fertility, the recipients at 10 weeks after transplantation were mated with BDF1 females. The genotyping of offspring for BVSC transgenes was performed as reported previously (Ohinata et al, 2008). For HE staining, the testes or epididymis was fixed with Bouin's solution, embedded in paraffin, and sectioned.

In vitro fertilization

Sperm were retrieved from the cauda epididymis and pre‐incubated in HTF medium (Kyudo Co., Ltd.) at 37°C for 1 hr. Oocytes were recovered from super‐ovulated BDF1 females by injecting PMSG and hCG, and fertilized with the sperm in HTF medium. The resultant two‐cell embryos were transferred into the oviducts of pseudopregnant ICR females at 0.5 days post‐coitum (dpc). Pups were delivered by cesarean section at 18.5 dpc.

LacZ staining

The seminiferous tubules were fixed with 2% paraformaldehyde and 0.2% glutaraldehyde in PBS at 4°C for 1 h. After washing three times with PBS, the seminiferous tubules were incubated in X‐gal solution (0.1% X‐gal, 0.1% Triton X‐100, 1 mM MgCl₂, 3 mM K₄[Fe(CN)₆], and 3 mM K₃[Fe(CN)₆] in PBS) at 37°C for 2–3 h.

DNA FISH and immunofluorescence–DNA FISH on PGCLCs

The ESCs, EpiLCs, and female MEFs were dissociated with TrypLE, and d4, d4c3, and d4c7 PGCLCs were purified with FACS. The cell samples were transferred onto a poly‐L‐lysine (Sigma)‐coated glass coverslip in a drop of PBS and allowed to adhere to the coverslip by aspirating the excess medium prior to fixation. For DNA FISH, cell samples on coverslips were fixed for 10 min in 3% paraformaldehyde (PFA) (pH 7.4), permeabilized on ice for 3 min in 0.5% Triton X‐100/PBS, and stored in 70% ethanol at −20°C. After dehydrating through an ethanol series, they were denatured in 70% formamide/2× SSC (pH 7.4) for 30 min at 80°C and dehydrated again through an ice‐cold ethanol series. They were then hybridized with fluorescent BAC probes (RP24‐157H12) for Huwe1 at 37°C overnight. Coverslips were counterstained with DAPI (1 μg/ml) and mounted in Vectashield (Vector Laboratories).

Immunofluorescence followed by DNA FISH was carried out as described previously (Chaumeil et al, 2004, 2008). The cell samples on coverslips were fixed for 10 min in 3% PFA (pH 7.4) at room temperature. Permeabilization of cells was performed on ice in 0.5% Triton X‐100/PBS for 3 min. After rinsing in PBS, the preparation was blocked in 1% BSA (Sigma)/PBS for 30 min, incubated with anti‐H3K27me3 (1/200; Millipore) overnight at 4°C and then washed in PBS three times and incubated with an Alexa Fluor 488 anti‐rabbit secondary antibody (1/500; Thermo Fisher Scientific) for 30 min at room temperature. After washing in PBS, preparations were post‐fixed in 4% PFA for 10 min at room temperature and rinsed in PBS. Preparations were incubated in 0.7% Triton X‐100, 0.1 M HCl for 10 min on ice. They were then washed twice in 2× SSC for 10 min at RT. Finally, the preparations were denatured in 70% formamide/2× SSC (pH 7.4) for 30 min at 80°C, dipped in ice‐cold 2× SSC, and hybridized with fluorescent BAC probes for Huwe1 as described above.

RNA sequencing (RNA‐seq)

Total RNAs were purified from ESCs, EpiLCs, and BVSC (+) d4, d6, d4c3, d4c5, and d4c7 PGCLCs (two biological replicates for each) using an RNAeasy Micro Kit (Qiagen). 10 ng RNAs (equivalent to 1,000 cells) were subjected to the previously described cDNA amplification method (Kurimoto et al, 2006), and their 3′ termini were then subjected to deep sequencing on a SOLiD5500xl system as described previously (Nakamura et al, 2015). 1 ng RNAs from E10.5 PGCs and E12.5 and E13.5 male/female germ cells, and amplified cDNA from E9.5 PGCs (two biological replicates for each) prepared in the previous study (Kagiwada et al, 2013) were also subjected to the RNA‐seq.

Chromatin immunoprecipitation sequencing (ChIP‐seq)

ChIP‐seq was performed as described previously (Kurimoto et al, 2015). Briefly, 1 × 10⁵~1 × 10⁶ BV (+) d4c7 PGCLCs were purified with FACS, and fixed with 1% formalin (Sigma) for 10 min at room temperature followed by quenching with 150 mM glycine. The fixed cells were washed with PBS, dissolved in lysis buffer containing 1% SDS, and sonicated using a Bioruptor UCD250 for 10 cycles of 30 s at high power. Solubilized chromatin fraction was incubated with mouse monoclonal antibodies for histone H3K4me3, H3K27ac, or H3K27me3 (Hayashi‐Takanaka et al, 2011) in complex with M280 Dynabeads Sheep anti‐mouse IgG (Life Technologies) at 4°C overnight with rotation (two biological replicates for each). Chromatins were eluted in a buffer containing 1% SDS and 10 mM DTT after washing. The eluents were reverse‐crosslinked at 65°C overnight, treated with 4 μg proteinase K at 45°C for 1 h, and purified with a Qiaquick PCR purification column (Qiagen). The ChIP‐ed and input DNAs were then sheared to an average size of about 150 bp by ultrasonication (Covaris, Woburn, MA), and subjected to library preparation methods for deep sequencing on a SOLiD5500xl system as previously described (Kurimoto et al, 2015).

Whole‐genome bisulfite sequencing (WGBS)

WGBS was performed as described previously (Shirane et al, 2016). Briefly, purified BV‐positive d4c3 and d4c7 PGCLCs (two biological replicates for each) were lysed with 10 mM Tris–Cl (pH 8.0) containing 150 mM NaCl, 10 mM EDTA, 0.5% SDS, and 1 mg/mL proteinase K at 55°C overnight with shaking. The lysate was incubated with 1.32 μg/ml RNase A at 37°C for 1 h, and extracted once with TE‐saturated phenol, twice with phenol–chloroform, and once with chloroform. Genomic DNA was precipitated with an equal volume of isopropanol, washed twice with 70% ethanol, air‐dried, and then dissolved in 10 mM Tris–Cl (pH 8.0). The purified genomic DNA (50 ng) was spiked with 0.5 ng unmethylated lambda phage DNA (Promega) and subjected to bisulfite conversion and library construction using the post‐bisulfite adaptor tagging (PBAT) method (Miura et al, 2012) for deep sequencing on an Illumina HiSeq 1500/2500 system as described previously (Shirane et al, 2016).

Data analysis for RNA‐seq

The RNA‐seq read data were mapped on the mouse mm10 genome and annotated to reference genes with extended transcription terminal sites as described previously (Nakamura et al, 2015), using cutadapt v1.3 (Martin, 2011), tophat v1.4.1/bowtie v1.0.1 (Kim et al, 2013), and cufflinks v2.2.0 (Trapnell et al, 2012). Expression levels were normalized to reads per million‐mapped reads (RPM). Significant expression levels were defined as log₂ (RPM+1) > 3. Genes were considered as differentially expressed if the fold changes of expression levels were > 2 [i.e. if the difference of log₂ (RPM+1) was > 1]. Genes significantly expressed in at least one sample and differentially expressed in at least one pairwise comparison (10,437 genes) were used in principal component analysis (PCA) and unsupervised hierarchical clustering (UHC). Gene ontology (GO) (Ashburner et al, 2000) of differentially expressed genes was analyzed using the DAVID program (Huang da et al, 2009).

Data analysis for ChIP‐seq

Read data of ESCs, EpiLCs, d6 PGCLCs (Kurimoto et al, 2015), and d4c7 PGCLCs were mapped on the mouse mm10 genome and analyzed as described previously (Kurimoto et al, 2015) using bowtie v1.1.2 (Langmead et al, 2009), picard‐tools v2.1.0 (http://broadinstitute.github.io/picard/), IGVtools v2.3.52 (Robinson et al, 2011), samtools v1.3 (Li et al, 2009), and MACS v2.1.0 (Zhang et al, 2008). Read patterns were visualized by IGV (Robinson et al, 2011).

H3K4me3 peaks with P‐values < 10⁻⁵ detected in proximity (within 1 kb) were combined as a single peak, and the read densities of the peaks within 500 bp from the center were normalized by those of Input (the larger of those within 500 bp and within 5 kb) (IP/Input levels). H3K4me3 peaks with the greatest IP/Input levels among those located within 2 kb from TSSs were considered TSS‐associated peaks. The IP/Input levels of TSS‐associated H3K4me3 peaks were further normalized to that associated with a gene with the 95^th percentile of the significant expression levels, and defined as the H3K4me3 levels.

H3K27ac peaks with P‐values < 10⁻²⁰ detected in proximity (within 1 kb) were combined as a single peak. The read densities of the peaks within 500 bp from the center were normalized to the average of the log₂ IP/Input levels, and defined as the H3K27ac levels. The H3K27ac peaks were considered biased to d6‐ or d4c7, if the fold changes of the H3K27ac levels were > 2.

The read densities of H3K27me3 for the regions around TSSs (within 1 kb) and for the TSS‐associated H3K4me3 peaks were normalized by Input, and were further normalized to the average IP/Input levels of H3K27me3 around the TSSs of the genes with expression levels of log₂ (RPM+1) > 2.5 and smaller than 3.5, to define H3K27me3 levels.

Data analysis for WGBS

Adaptor trimming, mapping on the mouse mm10 genome, and analyses of WGBS data were performed as described previously (Shirane et al, 2016) using the programs Trim Galore! v0.4.1 (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), cutadapt v1.9.1, and Bismark v0.15.0 (Krueger & Andrews, 2011), bowtie v1.1.2 and R. The bisulfite conversion rate was estimated as > 99.5% using the lambda phage genome as a positive control.

Promoters were defined as regions 0.9 kb upstream and 0.4 kb downstream from the transcription start sites and classified into three categories depending on their GC content and CpG density as described previously (HCPs, ICPs, and LCPs) (Borgel et al, 2010). Promoters with at least 5 CpGs were used for methylation analysis. Coordinates of the ICRs that were defined in E12.5 embryos were obtained from a previous publication (Tomizawa et al, 2011). For the methylation analysis of repetitive elements, the processed reads were mapped to the repetitive consensus sequences (Shirane et al, 2016) and CpG sites that covered at least four reads were used. The uniquely mapped regions overlapped with repeats were defined using RepeatMasker (Smit, AFA, Hubley, R & Green, P. RepeatMasker Open‐4.0. 2013‐2015 http://www.repeatmasker.org). For analysis of the uniquely mapped whole‐genomic regions, the 5mC levels in 2‐kb sliding windows with 1‐kb overlap were calculated.

For the methylation analysis of uniquely mapped regions, CpG sites that covered less than four reads and more than 200 reads were excluded; thus, the minimum sequence depth to call the methylated/unmethylated cytosine was 4. For the methylation analysis of repetitive elements, the processed reads were mapped to the repetitive consensus sequences (RepBase19.0.4) and CpG sites that covered at least five reads were used.

The sequencing and mapping statistics for RNA‐seq, ChIP‐seq, and WGBS are shown in Dataset EV1, Tables EV3 and EV4, respectively, and summarized in Dataset EV7.

Accession numbers

The accession number for the RNA‐seq data of d4c3/d4c5/d4c7 PGCLCs and E9.5/E12.5 germ cells is GSE87644 (the GEO database). The RNA‐seq data of ESCs/EpiLCs/d4/d6 PGCLCs [BVSC (+)] (GSE67259) and of E10.5/E11.5/E13.5 germ cells (GSE74094) were downloaded from the GEO database.

The accession number for the ChIP‐seq data of H3K4me3, H3K27ac, and H3K27me3 for d4c7 PGCLCs is GSE87645 (the GEO database). The ChIP‐seq data of H3K4me3, H3K27ac, and H3K27me3 for ESCs/EpiLCs/d2/d4/d6 PGCLCs (GSE60204) were downloaded from the GEO database.

The accession number for the WGBS‐seq data of d4c3/d4c7 PGCLCs is DRA005166 (the DDBJ database). The WGBS‐seq data of ESCs/EpiLCs/d2/d4/d6 PGCLCs (DRA003471) and E10.5/E13.5 PGCs (DRA000607) were downloaded from the DDBJ database.

Author contributions

HO conducted all the experiments regarding PGCLC expansion and transplantation. KK conducted WGBS and ChIP‐seq experiments and analyzed the data. IO analyzed the X chromosome state in female PGCLCs. TN, HM, and TY contributed to the RNA‐seq, YY contributed to the analyses of RNA‐seq/WGBS data, and KS and HS contributed to the WGBS. YO and MH contributed to the chemical screening. MS conceived the project, and HO, KK, and MS designed the experiments and wrote the manuscript.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Table EV1

Table EV2

Table EV3

Table EV4

Dataset EV1

Dataset EV2

Dataset EV3

Dataset EV4

Dataset EV5

Dataset EV6

Dataset EV7

Review Process File

The EMBO Journal (2017) 36: 1888–1907