To determine the transcriptional targets of FOXL2, we set up a model of murine primary granulosa cells for performing knockdown and transcriptomic analyses. Specifically, we purified preantral-small antral follicles from the ovaries of 8-week-old female Swiss mice. After an overnight culture, the follicles were dispersed using trypsin and cells were grown for a maximum of 8 days. Follicular cells are known to progressively (trans)differentiate in culture, by upregulating markers of Sertoli cells or by luteinizing (Nekola and Nalbandov, 1971). We therefore verified by immunofluorescence that the expression of FOXL2, which is absent in luteinized cells, was maintained throughout culture and that SOX9 was not substantially upregulated. We observed a similar level of FOXL2 expression in follicles cultured for one night as in 9-day cultured cells (Figure 1A). SOX9 upregulation was observed in about 20% of the cells. However, we could check that the known FOXL2-mediated repression of Sox9 was still operant (Figure 1B). Indeed, upregulation of SOX9 could be detected as soon as 24 hr after treating the cells with a commercial siRNA pool targeting FOXL2, and its expression continued to increase 48 hr after treatment. This stresses the importance of limiting the interval between knockdown and RNA analysis to assess the direct and indirect targets of FOXL2, as transcriptional changes can also occur because of SOX9 upregulation. We also verified that we could detect the expression of six NRs of interest (Figure 1—figure supplement 1). We observed that Ar, Esr2 and Nr5a1 were the most expressed NRs, and in particular that Esr2 was at least four times more expressed than Esr1 in these cells, in agreement with previous observations (Lenie and Smitz, 2008). Since progesterone receptor was very poorly expressed, we removed it from the analysis. We focused instead on the targets of the androgen receptor, which is required for terminal folliculogenesis (Walters et al., 2012).

Figure 1 with 2 supplements see all

Download asset Open asset

Figure 1—source data 1

This file contains all Log2 transcriptional effects and transcriptional targets analyzed in Figure 1, which were calculated as indicated in methods.

The different tabs contain respectively (all Log2 transcriptional effects, list of FOXL2 targets, list of FOXL2 ‘in vivo’ targets, list of ESR1 targets, list of ESR2 targets, list of AR targets, list of NR5A1 targets, list of NR2C1 targets, list of AR targets in the absence of FOXL2, and list of NR5A1 targets in the absence of FOXL2). Red indicates activated transcriptional targets (more than twofold). Dark blue indicates repressed transcriptional targets (more than twofold). Orange indicates activated targets (between 1.5-fold and twofold). Light blue indicates repressed transcriptional targets (between 1.5-fold and twofold).

https://doi.org/10.7554/eLife.04207.004

Download elife-04207-fig1-data1-v1.xlsx

Next, we treated the cultured primary follicular cells with commercial siRNA pools targeting Foxl2 and/or Esr1, Esr2, Ar, Nr5a1 or Nr2c1. We verified siRNA efficiency by qPCR (Figure 1—figure supplement 2). We observed a systematic decrease of more than 70% of the targeted mRNAs. We also noticed that Esr2 expression was reduced by about 50% following Foxl2 knockdown, suggesting that FOXL2 activates Esr2 expression. As expected, Sox9 expression was repressed by both FOXL2 and ESR2, as it increased in the absence of either of them, and even more in the absence of both. On the contrary, NR5A1 was required to activate Sox9 expression. These results recapitulate well-known in vivo findings (Kato et al., 2012; Uhlenhaut et al., 2009) and confirm these primary cells are suitable to study the transcriptional regulation by FOXL2 and its interplay with NRs.

We next performed a microarray analysis of the transcriptional changes induced by FOXL2 or NR knockdowns. Specifically, we treated cells on the one hand with control or FOXL2-targeting siRNAs and on the other hand with control siRNAs or siRNAs targeting each of the NRs. This experimental design gave us six different conditions in which we could study the transcriptional targets of FOXL2 independently of the action of the NRs. We thus used a paired Significance Analysis of Microarrays (Tusher et al., 2001). This analysis detected 863 genes whose expression was modified by FOXL2 with a median false discovery rate <0.1%. Among them, we selected the genes undergoing an expression fold-change of at least 1.5. We thus obtained a list of 224 FOXL2-activated targets (i.e., with decreased expression in the absence of FOXL2) and 139 repressed genes (Figure 1C and Figure 1—source data 1). This list contained previously known targets of FOXL2 such as Fst (follistatin); Ptgs2, encoding a cyclo-oxygenase required for folliculogenesis; Fos, coding for a subunit of the AP-1 transcription factor; Cyp19a1, encoding the aromatase enzyme; and Bmpr1b, coding for a BMP receptor. Many other genes known to be critical for folliculogenesis and/or ovulation were also among the FOXL2-activated targets, including Foxo1, Gja1 (a.k.a. connexin 43), Ereg (coding for the EGF-like Epiregulin), Ptger2, or Has2. Some FOXL2-repressed genes are involved in testis development, such as Sox9, Dhh, and Sox8 (Yao et al., 2002; Georg et al., 2012). An analysis of Gene Ontology (GO) term enrichment showed that FOXL2 targets were characterized by keywords such as cell adhesion (p = 3.10⁻⁹), regulation of cell proliferation (p = 7.10⁻⁷), and regulation of hormone levels (p = 1.10⁻⁸) (Table 1). Other enriched terms were reproductive structure development (p = 3.10⁻⁵) and developmental growth (p = 2.10⁻⁵). The Panther Classification system showed that FOXL2 targets were enriched in genes involved in cell communication (p = 6.10⁻⁶) and signal transduction (p = 1.10⁻⁵). In particular, some of the most strongly activated targets are major players of important signal transduction pathways such as Grem1 and Grem2, encoding BMP antagonists; Rgs2 and Rgs4, encoding regulators of G-protein signaling, and the NRs Esr2 and Pparg.

We compared the direct and indirect FOXL2 targets from our study with the genes modified by a twofold factor or more 3 weeks after the induction of Foxl2 deletion in the conditional knockout mouse model of Uhlenhaut et al. (Uhlenhaut et al., 2009). Out of 345 FOXL2 target genes also present in the data set of Uhlenhaut et al., 71 were similarly affected following Foxl2 deletion in the ovary (Figure 1D and Figure 1—source data 1). These notably included the critical factors for folliculogenesis Fst, Cyp19a1, Esr2, Foxo1, Pparg, Gja1, and Bmpr1b, which required FOXL2 for their expression, as well as the testis-determining or testis-specific factors Sox9, Sox8, and Dhh, which were repressed by FOXL2. FOXL2 ablation in granulosa cells results in a loss of their identity and in the appearance of testis-like structures, making difficult to evaluate which genes are first affected by the loss of FOXL2. These common targets may thus represent the early responders to FOXL2 deletion and play a critical role in the maintenance of granulosa cell identity, whereas other differentially expressed genes in the ovaries of mutant mice may be affected following the transdifferentiation of granulosa cells. GO term enrichment of the former genes reveals in particular their role in the regulation of hormone levels (p = 2.10⁻⁵, Table 1). The intersection of both data sets provides strong candidates for in vivo FOXL2 targets in murine ovaries.

We then explored the transcriptional changes owing to the knockdowns of each of the five NRs. To ensure a good reliability of the potential targets obtained, we focused on genes whose expression displayed a twofold change or more (see ‘Materials and methods’ for details). We found that NR5A1, ESR2, and AR were the NRs had the highest numbers of potential direct and indirect targets (according to our inclusion criteria), whereas ESR1 and NR2C1 had very few targets (Figure 1E and Figure 1—source data 1). This suggests that the knockdown of either Esr1 or Nr2c1 may be compensated by their respective paralogs Esr2 and Nr2c2, which are more strongly expressed in follicular cells (Figure 1—figure supplement 1). ESR2 only had activated targets, whereas AR and NR5A1 had as many activated as repressed targets.

We used the same strategy to explore how the absence of FOXL2 modified the expression of the targets of the NRs defined above. Interestingly, ESR2 and AR kept very few targets in the absence of FOXL2, whereas NR5A1 conserved a similar number of targets (Figure 1E). Consistently, the absence of FOXL2 led to halving the average transcriptional effect of ESR2 and AR on their target genes, whereas target modulation by NR5A1 underwent only a slight decrease (Figure 1F). This suggests that FOXL2 is required directly or indirectly for efficient target regulation by ESR2 and AR, but not by NR5A1.

To better understand how FOXL2 regulates its direct transcriptional targets, we performed a chromatin immunoprecipitation coupled to high-throughput sequencing (anti-FOXL2 ChIP-seq) in primary follicular cells. Using MACS as the peak detection algorithm, we detected about 14,000 peaks suggestive of FOXL2 binding to chromatin (Figure 2—source data 1). The analysis of these peaks revealed a highly enriched sequence corresponding to the characteristic motif of forkhead factors (5′-G/A-T-A-A-A-C/T-A-3′), with a very narrow enrichment window centered at the peak summits (Figure 2A). We analyzed the positions of FOXL2 peaks with respect to the annotated loci on the genome and found that transcription start sites (TSSs) were strongly enriched in FOXL2 peaks (Figure 2—figure supplement 1A). They were especially present within −300 bp and +100 bp around the TSSs (Figure 2B). However, to our surprise, when we correlated the presence of FOXL2 peaks with the transcriptional effects of the Foxl2 knockdown, we observed that the former were not over-represented near the TSSs (or the promoter regions) of its transcriptional targets (Figure 2C). However, we detected a strong enrichment of FOXL2 peaks in the regions containing the transcription units (TUs, extending from the TSS to the annotated polyA signal) of FOXL2-activated targets, but not in the TUs of the repressed ones (Figure 2D). This observation was even more striking considering only the 71 ‘in vivo’ FOXL2 targets (Figure 2—figure supplement 1B). Since the TUs of protein coding genes are largely composed of introns (in terms of sequence length), this suggests that FOXL2 acts mainly through intronic regulatory elements (as previously described for Fst) (Blount et al., 2009). Indeed, a majority of FOXL2-activated targets had FOXL2 peaks in their introns (examples in the Figure 2—figure supplement 2), suggestive of the existence of potential regulatory elements recognized by FOXL2, although some target genes had no peak within their TUs (Figure 2—figure supplement 3). The enrichment of FOXL2 peaks in the TSSs may thus be a consequence of the decondensed chromatin state of these regions (i.e., nonfunctional binding). This observation is worth noting because the presence of ChIP peaks for transcription factors at TSSs is widely used as a criterion suggesting active transcriptional regulation (Ouyang et al., 2009), although as shown here for FOXL2, it may not always be the case. Interestingly, FOXL2-repressed genes were not enriched in FOXL2 peaks, in or at the proximity (50 kb) of their TUs. This notably included the testis determining genes Sox9, Sox8, and Dhh, suggesting that their repression by FOXL2 may be mostly indirect, at least in these cultured cells (Figure 2—figure supplement 3).

Figure 2 with 3 supplements see all

Download asset Open asset

Figure 2—source data 1

MACS results providing the list of FOXL2-binding regions in mm9 canonical female genome coordinates.

https://doi.org/10.7554/eLife.04207.009

Download elife-04207-fig2-data1-v1.xlsx

As previously mentioned, we observed that Esr2 expression dropped following Foxl2 knockdown, showing that the latter is required for the expression of ESR2 in follicular cells. Previous studies have shown that Esr2 expression is dependent on Activin/SMAD3 in granulosa cells (Kipp et al., 2007). Since cooperation between FOXL2 and Activin/SMAD3 has already been identified on some targets, we tested whether FOXL2 was required for Esr2 induction by Activin. For this, we treated primary follicular cells with siRNAs targeting FOXL2 or control siRNAs for 24 hr, and then with recombinant Activin A for 18 hr (Figure 3A). As expected, we observed an increase of Esr2 expression in the presence of Activin A and a decrease following the Foxl2 knockdown. Interestingly, no increase in Esr2 expression could be observed in cells treated with Activin A when Foxl2 was knocked-down. This clearly shows that FOXL2 is required for efficient Activin induction of Esr2 expression.

Figure 3

Download asset Open asset

The anti-FOXL2 ChIP-seq showed the presence of three main peaks in Esr2 introns, which we named peaks 1 to 3 based on their distance from the TSS (Figure 3B). A sequence analysis of these elements showed that peak 3 contained a 20-bp DNA sequence composed of contiguous FOXL2, GATA, and SMAD elements (Figure 3C) and that peak 2 contained several potential FOXL2 elements. Conversely, peak 1 contained no obvious sequences corresponding to FOXL2 or SMAD3 motifs. To test whether these peaks could be regulatory elements, we cloned each of them upstream of a minimal SV40 promoter driving the expression of the luciferase gene. Next, we co-transfected these constructs with plasmids driving the expression of FOXL2 and SMAD3 in COV434 cells, human granulosa-derived cells, which are advantageous in this case as they do not express FOXL2 and very poorly express SMAD3. These cells were then treated with Activin A or a control solution (Figure 3D). We found that neither FOXL2 nor SMAD3 could activate the expression of luciferase through the peak 1 sequence, whereas only FOXL2 could activate luciferase expression through the peak 2 sequence. Interestingly, FOXL2 and SMAD3 synergistically activated luciferase expression through the peak 3 sequence (located in the 8^th intron of Esr2), and this effect was enhanced when the cells were treated with Activin A. This observation was confirmed when we expressed a constitutively active Activin receptor (TGFBR1-T204D), suggesting that the observed effect is mediated by the canonical TGF-β pathway (Figure 3E). This is coherent with the presence of a FOXL2/SMAD3 composite element in the peak 3 sequence. We therefore created constructs where we mutated either the FOXL2 or the SMAD3 site in the composite element, and a construct in which the whole element was removed. As expected, we observed a strong reduction of luciferase expression when the SMAD3 or FOXL2 elements were mutated or when the whole composite element was removed (Figure 3E, ‘full peak 3’ series). However, a weaker synergistic activation of luciferase expression by FOXL2 and SMAD3 remained even in the absence of FOXL2 and SMAD3 elements, suggesting that other non-canonical DNA elements or interactions may account for part of FOXL2/SMAD3 synergy.

We observed that a canonical estrogen receptor-binding site was present nearby FOXL2 and SMAD3 sites in the peak 3 sequence. We therefore tested whether ESR2 could activate its own expression, alone or in combination with FOXL2 and TGFβ/SMAD3 (Figure 3F). Interestingly, we observed that the overexpression of ESR2 alone or in combination with SMAD3 and TGFBR1 (T204D) had little effect on luciferase activity. However, ESR2 was able to slightly induce luciferase in the presence of FOXL2 and more strongly in the presence of FOXL2, SMAD3, and TGFBR1 (T204D). Therefore, this experiment suggests that the induction of Esr2 expression due to FOXL2 and TGFβ signals may be amplified by ESR2 action on its own enhancer, although ESR2 itself may not be able to stabilize alone its own expression. This further supports the idea that FOXL2 is required to establish a robust ESR2 expression in the ovary. Several observations corroborate this finding. Indeed, the transcriptional effect of Foxl2 knockdown is well correlated with the one of ESR2 on its target genes (Figure 4—figure supplement 1). Thirteen out of the 20 ESR2 target genes detected here are also (direct and indirect) targets of FOXL2. These shared targets represent slightly more than 10% of FOXL2 targets, confirming that FOXL2 acts upstream of ESR2. In line with this, a hierarchical clustering of FOXL2-activated targets yields two main clusters (Figure 4A). The first cluster is mainly characterized by a strong effect of Esr2 knockdown and includes ESR2 transcriptional targets, whereas no particular signature was observed for the second cluster (Figure 4B and Figure 4—figure supplement 2). These results show that gene regulation via, or along with, ESR2 may represent an important part of the transcriptional effects of FOXL2. Moreover, a third of the ESR2 targets in cultured cells, and Esr2 itself, had a decreased expression in the Foxl2 conditional-knockout ovaries, suggesting that FOXL2 regulates ESR2 (and its targets) in vivo (Figure 1—source data 1). Finally, it is interesting to note that although FOXL2 has a strong positive transcriptional effect on ESR2 targets in control conditions, it has in average no effect on the same genes when Esr2 is knocked down (Figure 4C). FOXL2 still activates a few genes, such as Fam84a, although to lower extents (Figure 4—source data 1). This suggests that the overall positive effect of FOXL2 on ESR2 targets is indeed due to the activation by FOXL2 of ESR2 expression, although FOXL2 may also have some positive or negative effects on individual ESR2 targets through independent pathways.

Figure 4 with 5 supplements see all

Download asset Open asset

Figure 4—source data 1

This file contains all Log2 transcriptional effects and transcriptional targets analyzed in Figure 4, which were calculated as indicated in methods.

The different tabs contain respectively (all Log2 transcriptional effects from both arrays, a list of E2 targets and a list of E2 targets in the absence of FOXL2). Red indicates activated transcriptional targets (more than twofold). Dark blue indicates repressed transcriptional targets (more than twofold). Orange indicates activated targets (between 1.5-fold and twofold). Light blue indicates repressed transcriptional targets (between 1.5-fold and twofold).

https://doi.org/10.7554/eLife.04207.015

Download elife-04207-fig4-data1-v1.xlsx

The above-mentioned findings support our hypothesis that FOXL2 plays an important role in estrogen signaling. We therefore analyzed the transcriptomic changes following a 10 hr 17β-estradiol (E2) treatment of primary granulosa cells pre-treated for 24 hr with either anti-FOXL2 or control siRNAs. We considered as potential targets those genes in the microarrays whose expression displayed a fold-change >2 in the presence of E2. We found 41 genes responding to estradiol according to our inclusion threshold, 39 of which were activated and only 2 were repressed. This is in agreement with previous reports suggesting that ERs are mainly transcriptional activators (Kininis et al., 2007) (Figure 4D and Figure 4—source data 1). The E2-activated targets comprised granulosa markers such as Hsd11b2 (Xu et al., 2011), but also some genes involved in cancer such as Tacc1, Angpt2, or Cbs (Ghayad et al., 2009, 1; Avraham et al., 2014; Weiner et al., 2012). In agreement with previous works, E2 also repressed, at a lower level (<2-fold), the expression of Sox9 and Sox8, involved in testis development.

We next compared the effect of E2 with the effect of Esr1 and Esr2 knockdowns on the same genes, as determined in our previous experiments. We found that the response to E2 was strongly correlated with the transcriptomic effect of Esr2 knockdown, but not to that of Esr1 (Figure 4E). This result suggests that ESR2, which is more expressed than ESR1 in these cells, is the main vector of estrogen signaling.

We then determined the number of genes affected by E2 treatment in the absence of FOXL2 and we observed a strong reduction of E2 effect in this case. Indeed, only 27 genes were E2 targets in the absence of FOXL2 (Figure 4D). In average, E2-dependent gene activation (in terms of mRNA levels) was reduced by about 35% in the absence of FOXL2 (Figure 4F). This supports the idea that FOXL2 is required for normal E2 signaling in follicular cells.

To have a broader view of the effect of FOXL2 on E2-induced transcriptional changes, we conducted an unsupervised hierarchical clustering of genes modified by >1.5-fold in the presence of E2 (Figure 4—figure supplement 3A). This analysis yielded three main groups of E2-activated genes and two groups of E2-repressed genes. These clusters were mainly characterized by the effect of the Foxl2 knockdown (Figure 4—figure supplement 3B). However, the average effect of E2 in the presence or absence of FOXL2 was remarkably similar between the groups. Indeed, E2 effect was reduced as much (≈40%) for FOXL2-repressed genes than for FOXL2-activated ones. This suggests an indirect effect of FOXL2 on E2-dependent transcriptional changes, that is, that FOXL2 is required for the expression of a mediator of E2-dependent transcriptional regulation. However, a direct cooperation or antagonism between FOXL2 and ERs may occur on some genes.

Next, we selected 27 E2-activated genes among those having relatively high basal expression levels and tested by qPCR whether E2 could elicit their activation in the absence of ESR1, ESR2, or FOXL2 (Figure 4F and Figures 4—figure supplement 4 and 5). Not surprisingly, E2 activation was mostly maintained in the absence of ESR1, but it was strongly decreased upon Esr2 knockdown. This is coherent with the correlation of the effect of E2 with only the effect of ESR2 and confirms that ESR2 is the main effector of estradiol signaling in these primary cells. The same experiment in the presence of anti-FOXL2 siRNAs showed a 40% reduction of E2-dependent gene activation, similar to what was observed on microarrays (Figure 4F). Altogether, these results show that FOXL2, mainly through its regulation of Esr2 expression, is required for the establishment of normal estradiol signaling in follicular cells.

To better understand the combined effect of FOXL2 and ESR2 on granulosa cell maintenance, we analyzed SOX9 expression at the protein level 48 hr after Foxl2 and/or Esr2 knockdown. As expected considering our transcriptomic analyses, we observed that SOX9 expression was upregulated in the absence of either FOXL2 or ESR2 (Figure 5A). SOX9 expression was much higher in the absence of both ESR2 and FOXL2. This confirms that estrogen receptors and FOXL2 co-regulate SOX9 expression (Uhlenhaut et al., 2009). Interestingly, our transcriptomic analyses indicate that FOXL2 is required for the expression of Cyp19a1 (encoding the rate-limiting enzyme in estradiol production) both in the presence (Figure 5B) and in the absence of hormones (Figure 5C). Thus, FOXL2 may allow both estradiol production and enable estrogen signaling in the same cells. These data suggest the existence of a coherent feed-forward loop in which FOXL2 stimulates both estradiol production and receptivity (i.e., ESR2 expression) that might be responsible at least in part for the maintenance of granulosa cell identity by repressing the testis determinant Sox9.

Figure 5

Download asset Open asset

As previous work showed that FOXL2 and ESR1 were both required for SOX9 repression in the ovary, we tested whether E2 and ESR1 could have an influence on SOX9 expression in our system. We therefore deprived follicular cells of hormones for 24 hr before treating them with siRNA pools for 24 hr, and finally with E2 for 10 hr before lysis. We observed that E2 treatment induced a repression of SOX9 expression, in coherence with our transcriptomic analysis (Figure 5D). Esr1 knockdown, on the other hand, had no clear effect on SOX9 expression or on the efficiency of SOX9 repression following E2 treatment. However, we observed that ESR2 knockdown resulted in an increase of SOX9 expression even in the virtual absence of hormones and that E2-induced repression was reduced in the absence of ESR2. Finally, SOX9 expression was much increased in the absence of FOXL2, but E2 was still efficient in repressing SOX9 in this case. Altogether, our data show that FOXL2 and ESR2/E2 repress somewhat independently SOX9 expression in follicular cells. It has previously been suggested that FOXL2 and ESR1 bind together to a Testis-specific enhancer of Sox9 to repress its expression. Here, we have detected no binding of FOXL2 to this sequence or to any other close sequence in our ChIP-Seq analysis. It is possible that the binding was lost during cell culture or that it occurs at a different stage of follicular development. However, the fact that FOXL2/E2 repression of SOX9 is still operant in our case suggests that several pathways orchestrated by FOXL2 cooperate to repress SOX9 expression in these cells (Figure 5E).

The rabbit anti-FOXL2 antibody has previously been described. The rabbit anti-SOX9 antibody was a kind gift of Dr Brigitte Boizet. The mouse anti-GAPDH antibody was purchased from ABM (ref G041) and the anti ER-a (HC-20) antibody was purchased from Santa Cruz Biotechnology (SCBT, Dallas, Texas) at a concentration of 2 µg/µl (ref SC-543X).

The full-length coding sequence of Esr2 was amplified from primary follicular cells cDNA and cloned in frame with a V5 tag using the pcDNA3.1 TOPO-TA/V5-His cloning kit (Life Technologies, Carlsbad, California) according to manufacturer's instruction. FOXL2 and ESR1 potential regulatory elements identified from ChIP-Seq were amplified from primary follicle cell genomic DNA and inserted between the KpnI and NheI sites of the pGL3-promoter vector (Promega, Madison, Wisconsin). Vectors driving the expression of FOXL2, SMAD3 have been described previously (Anttonen et al., 2014), and the vector expressing constitutively active TGFBR1 (with T204D mutation) was a kind gift of Dr Peter Scheiffele (Gore et al., 2005).

The ovaries of 8-week-old female Swiss mice (pools of 5 mice per replicate for transcriptome analysis, 20 mice per condition for ChIP experiments) were dissected and digested in DMEM/F-12 medium (Life Technologies) containing 1 mg/ml collagenase (Sigma–Aldrich, Saint-Louis, Missouri), 0.5 mg/ml BSA (Sidma–Aldrich), and 2000 U/ml DNase I (Thermo-Fisher Scientific), in a ratio of 500 µl per ovary, at 37°C for 30 min. Ovaries were progressively dispersed during digestion using pipette tips of decreasing size to flush the organ debris. The mix was then diluted in 10 vol of DMEM/F-12 medium, filtered using a 500-µm membrane, spun at 300×g. The pellet was rinsed in cold PBS and spun again, and the pelleted debris was resuspended in a PBS/Ficoll mix of density 1.10. A PBS/Ficoll gradient was made by depositing PBS/Ficoll layers of density 1.08, 1.07, 1.06, 1.05, 1.03, and 1. The gradient was spun at 1500×g for 20 min, and 1 ml fractions were collected and diluted in 5-ml culture medium (i.e., DMEM/F-12 supplemented with 4% FBS, 1% Insulin-Transferrin-Selenium-X supplement, 1% Penicillin-Streptomycin [Life Technologies], and 10⁻⁷M androstenedione (Sigma–Aldrich). We then manually collected intact preantral follicles, characterized by the presence of several layers of granulosa cells around the oocyte and the absence of antrum. They mainly migrated to the 1.07/1.06 and 1.06/1.05 interfaces. Follicles were then rinsed, transferred to culture plates, and grown overnight. On day 1, they were dispersed using trypsin and then cultured in monolayers. Culture medium was changed on day 3, and cells were passed on days 5 and 7. For estradiol induction experiments, we used Phenol Red-free DMEM/F-12 and Optimem, charcoal-treated FBS (Life Technologies), and no androstenedione was added to culture medium.

After 7 days of culture, cells were seeded for transfection in 12-well plates. After an 18-hr incubation, cell medium was replaced with a premade Mix containing 500 µl Optimem (Life Technologies), a total of 12 pmol of siRNAs (2 × 6 pmol in the case of simultaneous knockdown), 200 ng pDsRed plasmid (Clontech Laboratories, Mountain View, California), and 2 µl Lipofectamine 2000 (Life Technologies). 6 hr after transfection, 1.5 ml of culture medium was added to each well, and cells were then grown for 24 hr before lysis in 500 µl Trizol reagent (Life Technologies), for a total culture time of 9 days.

The ON-TARGETPlus Smartpool siRNAs used here were purchased from Thermo Fisher Scientific (now distributed under the Dharmacon brand by GE Healthcare, Little Chalfont, Buckinghamshire, United Kingdom): non targeting pool (ref D-001810-10-05), Mouse anti-FOXL2 siRNAs (L-043309-01-0005), mouse ESR1 (L-058688-01-0005), mouse ESR2 (L-065564-01-0005), mouse AR (L-050296-00-0005), mouse NR5A1 (L-051262-01-0005), mouse NR2C1 (L-061593-01-0005). RNA/protein preparations were performed according to manufacturer's instruction.

For the first experiment, 100 ng total RNA was reverse-transcribed and linearly amplified using the Whole Transcriptome Amplification Transplex kit (Sigma–Aldrich), according to the manufacturer's instructions. The amplified cDNAs from the 12 samples were sent to the Nimblegen expression platform (Roche NimbleGen, Reykjavík, Iceland) in biological duplicates (independent mice pools, primary cultures, amplifications, and hybridizations per RNA sample). Samples were hybridized on two high-density Mus musculus MM9 Gene Expression 12 × 135K array set, each one allowing the study of 12 samples in parallel and interrogating over 45K annotated transcripts of the Mm9 genome assembly. Nimblegen performed sample labeling, hybridization, and fluorescence acquisition and provided us with normalized data files. Data were deposited in Gene Expression Omnibus (Series GSE60764).

For the second experiment, 50 ng total RNA was amplified and labeled with Cy3 using Low input Quick Amp labeling Kit (Agilent technologies, Santa Clara, California) according to manufacturer's instruction. Labeled cRNA was then fragmented and hybridized over 1 Sureprint G3 Mm9 8 × 60K array (Agilent technologies), allowing the study of 8 samples in parallel and interrogating over 40K annotated transcripts of Mm9 assembly. Hybridization was performed according to manufacturer's instructions. Raw fluorescence data were extracted using Genespring software v12 (Agilent technologies) and normalized using the 75th centile method in each array. To allow for an easy comparison between the two experiments, we then took for each gene the average expression value of all transcripts. We removed from further analysis the genes for which median fluorescence was under 100 (NimbleGen data) in the first experiment, under 10 in the second experiment (Agilent data). Values were then log2 normalized and subtracted of the median value for each replicate. Data were deposited in Gene Expression Omnibus (Series GSE60770).

To uncover FOXL2 targets in the first experiment, we conducted a paired Significance Analysis for Microarrays using the Multiexperiment Viewer program v4.9 (http://www.tm4.org/mev.html). Each array treated with a control siRNA pool was paired with the corresponding array treated with a FOXL2 targeting siRNA pool, giving 12 independent pairs of arrays. Delta value was adjusted to the minimum value allowing a median FDR = 0. The differentially expressed genes were then sorted according to the average absolute difference between the control siRNA/control siRNA and control siRNA/FOXL2 siRNA conditions, and all genes above a threshold of 0.584 (i.e., log2(1.5)) were selected as FOXL2 potential targets.

To infer NR targets in the first experiment (NimbleGen data) and E2 targets in the second experiment (Agilent data), we sorted genes according to the average absolute difference between the control siRNA/control siRNA and control siRNA/NR siRNA conditions (or control siRNA/E2 treatment condition). All genes above a threshold of 1 (i.e., log2(2)) were retained. We then verified for each gene that this difference was superior to twice the sum of the standard deviations for each condition. The retained genes were considered as NRs or E2 potential targets.

After 9 days of culture, cells grown to confluence in three 150-mm plates were fixed in situ with 1% formaldehyde for 10 min at room temperature, fixation was then blocked with 125 mM glycine, cells were rinsed and scrapped in PBS. Cells were first lysed in Lysis buffer A (50 mM Hepes pH 7.5, NaCl 140 mM, EDTA 1 mM, glycerol 10% NP-40 0.5%, and Triton X-100 0.25%), then spun and rinsed in lysis buffer B (Tris 10 mM pH 7.5, NaCl 200 mM, EDTA 1 mM, EGTA 0.5 mM), centrifugated again, and finally resuspended in lysis buffer C (Tris 10 mM pH 7.5, NaCl 100 mM, EDTA 1 mM, EGTA 0.5 mM, Na–deoxycholate 0.5%, N-lauryl sarcosine 0.5%). Chromatin was sheared in a Bioruptor sonicator, used at high power with 30/30 s cycles for 20 min, then 1% Triton X-100 was added to lysates, and cell debris were centrifugated. In parallel, 30 μg anti-FOXL2 or anti-ESR1 was incubated with 200 μl ProteinG Dynabeads for 4 hr in PBS containing 0.1% BSA. Lysates were incubated with the beads overnight, beads were then washed five times in wash buffer (50 mM Hepes pH 7.6, 500 mM LiCl, 1 mM EDTA, 0.7% Na–deoxycholate, 1% NP-40) and once in TBS. DNA–protein complexes were then eluted in 10 mM Tris pH 8, 1 mM EDTA, 1% SDS at 65°C for 15 min. Cross links were reversed at 65°C overnight, then recovered DNA was treated with RNaseA, proteinase K, and purified using phenol–chloroform. Input samples underwent to the same procedure.

ChIP-Seq was performed by the Platform IMAGIF from the french Centre National de la Recherche Scientifique (www.imagif.cnrs.fr) on an Illumina HiSeq1000 instrument using standard protocols (single reads). The reads in fastq files were imported on Galaxy server (https://usegalaxy.org/). Reads were mapped on mm9 Canonical Female genome using Map Bowtie for Illumina tool with commonly used settings. Unmapped reads were then removed using the Filter SAM tool. Peak calling was performed using MACS tool using the following settings: Tag size: 50; Band width: 160 (first experiment) or 174 (second experiment); Mfold: 20; new peak detection method: yes; and other settings unchanged. Input DNA analyzed in parallel was used as a control for peak detection. Peaks with an FDR <0.05 were kept for further analysis. De novo motif discovery was conducted using the MEME-chip pipeline on MEME server (http://meme.nbcr.net/meme/). Data were deposited in Gene Expression Omnibus (Series GSE60858).

COV434 cells were seeded 16 hr prior to transfection at a density of 4.10⁴ cells/cm² and transfected using the calcium–phosphate method. The biological activity of luciferase reporter constructs (see mains text) was assessed by the Dual-Luciferase Reporter Assay System (Promega). Relative luciferase units represent the ratio of firefly luciferase activity over Renilla luciferase activity in the samples. Each value is the mean of six independent experiments, and standard error bar represent the standard error of the mean (SEM).

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

This work was supported by Centre National de la Recherche Scientifique (CNRS), La Ligue Nationale contre le Cancer and l’Université Paris Diderot-Paris7. We thank Claude Thermes, Delphine Naquin, Erwin van Dijck, Maud Silvain, and Yan Jaszczyszyn from the CNRS' IMAGIF platform for their most kindful help. Animal studies were conducted according to the guidelines of the Animal Experimentation Ethical Committee Buffon (CEEA-40), project CEB-25-2012. All high-throughput data are available at the Gene Expression Omnibus database and can be accessed through the SuperSeries GSE60859.

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Centre National pour la Recherche Scientifique and the Université Paris Diderot/Paris VII. Animal studies were conducted along the guidelines of the Animal Experimentation Ethical Committee Buffon (CEEA-40), project CEB-25-2012.

© 2014, Georges et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.