SUMMARY
Reliably producing a competent oocyte entails a deeper comprehension of ovarian follicle maturation, a very complex process that includes meiotic maturation of the female gamete, the oocyte, together with the mitotic divisions of the hormone-producing somatic cells. In this report, we investigate mice ovarian folliculogenesis in vivo using publically available time-series microarrays from primordial to antral stage follicles. Manually curated protein interaction networks were employed to identify autocrine and paracrine signaling between the oocyte and the somatic cells (granulosa and theca cells) and the oocyte and cumulus and mural cells at multiple stages of follicle development. We established protein binding interactions between expressed genes that encoded secreted factors and expressed genes that encoded cellular receptors. Some of computationally identified signaling interactions are well established, such as the paracrine signaling from the oocyte to the somatic cells through the secreted oocyte growth factor Gdf9; while others are novel connections in term of ovarian folliculogenesis, such as the possible paracrine connection from somatic secreted factor Ntn3 to the oocyte receptor Neo1. Additionally, we identify several of the likely transcription factors that might control the dynamic transcriptome during ovarian follicle development, noting that the YAP/TAP signaling is very active in vivo. This novel dynamic model of signaling and regulation can be employed to generate testable hypotheses regarding follicle development, guide the improvement of culture media to enhance in vitro ovarian follicle maturation and possibly as novel therapeutic targets for reproductive diseases.
INTRODUCTION
The production of a competent female germ cell line, oocyte, that can undergo fertilization requires a highly orchestrated paracrine, autocrine, endocrine and juxtracine signaling that has to occur between the oocyte and the supporting somatic cells, granulosa and theca cells. This complex biological structure formed by the oocyte and surrounding somatic cells is called ovarian follicle. During ovarian follicle maturation, a primordial follicle (50 μm diameter in the mouse) that is composed from a handful of cells, has to grow into an antral follicle (500 μm), in order to attain a competent oocyte. Attempts to separate the different cellular follicular components to study ovarian follicle maturation lead to different behavior of the individual cell types. Unfortunately, little can be controlled in vivo to learn the effect of different biological variables (e.g., effects of hormones, extracellular matrix stiffness) on the follicle maturation. Thus, several in vitro systems that mimic in vivo ovarian follicle development (Eppig and OBrien 1996; O’Brien et al. 2003; Xu et al. 2006) has led to some of the most significant advances in reproductive biology (Edson et al. 2009). Up to date, only organ-on-a-chip technology that combines ovarian tissue, fallopian tubes and utero(Xiao et al. 2017)— EVATAR™—has been able to mimic the ovulation period. While with EVATAR™ the effects of hormones or the extracellular matrix stiffness could be systematically study, it suffers from the same limitations as in vivo to understand inter-cellular and intra-cellular communications between the different ovarian follicle cell types. One of the main hindrance is the difficulty to determine how paracrine (e.g., between the oocyte and granulosa cells) and autocrine (e.g., oocyte ligands that affect the oocyte) communication, i.e., inter-cellular communication, between the different follicle cell types occurs. While some of the inter-cellular ligands such as GDF-9 and BMP-15 have been established (Knight and Glister 2006), not all of them are known. Similarly, once that a given ligand binds to its corresponding receptors, a complex signaling takes place through several biochemical mechanisms (e.g., phosphorylation, protein binding, calcium release) that ends in the activation or deactivation of transcriptional programs, i.e., intra-cellular communication. Transcription factors (TFs) are the mediators between the cytoplasmic to the nucleus signaling, by translocating between the two cellular compartments. Once that a TF is in the nucleus, directly or in form of a protein complex, it binds DNA and starts or inhibits transcription. Therefore, TFs are potent regulators of the cellular phenotype. For instance, at the follicle level, FOXL2 is a marker of granulosa cells and essential for proper ovarian follicle development(Uhlenhaut and Treier 2006).
In the recent years, advances in high-throughput techniques have allowed obtaining large amount of information about the ovarian follicle transcriptome (Chronowska 2014; Pan et al. 2005b; Skory et al. 2013; Wigglesworth et al. 2015a; Yoon et al. 2006). Analysis of these large biological datasets requires statistical and computational methods to identify the processes that are associated with the manifest phenotypes. Yet these transcriptional data have not been explored to their maximum potential. Currently, there are methods to computationally identify the more plausible TFs that are regulating a given phenotype (Grant et al. 2011; Zhao and Stormo 2011). Similarly, given a set of genes that are expressed in a given cell, the most likely genes that encode for ligands and receptors that present in a cell type could be identified, using well-curated biological databases (e.g., DIP(Xenarios et al. 2000), MetaCore). Here, we propose a system biology approach to computationally reveal the key intra- and inter-cellular dynamic processes during mice ovarian folliculogenesis in vivo between and among the different follicular cell types (e.g., oocyte, granulosa cells, cumulus cells) involved in each developmental stage (e.g., primordial to primary ovarian follicles).
RESULTS
Identification of the ligands and receptors that lead ovarian follicle development inter-cellular signaling
Understanding inter-cellular communications during ovarian follicle maturation in vivo entails the identification of the secreted proteins (i.e., ligands) and available receptors in oocyte and in the somatic cells (e.g., granulosa and theca cells) that support the oocyte growth and maturation. We established the most likely ligands and receptors during ovarian follicle development by characterizing the set of statistically significant genes that encode for ligands and receptors during ovarian follicle maturation. We mined several publically available time-series transcriptomics—i.e., oocytes (Pan et al. 2005a), somatic cells—e.g. granulosa and theca cells(Peñalver Bernabé et al.)—cumulus and mural cells collected during antrum formation (Wigglesworth et al. 2015b) and cumulus cells during oocyte competence acquisition (Charlier et al. 2012). Combination of all these data sets led to a list of the significant transcribed ligands and receptors in each individual cell type (e.g., oocyte, somatic, cumulus granulosa cells).
We identified 100 genes that encode for ligands and 95 genes that encode for receptors that could potentially regulate the inter-cellular communication during ovarian follicle development (File S1). Some of the genes that encode for ligands and receptors were active in multiple cell lines, e.g., Dnc in the oocyte and somatic cells, Efna2 in mural and cumulus granulosa cells; yet others were very specific (e.g., Wnt10a in cumulus cells). More than half of the genes that encode for ligands, a total of 59, were cell-specific: 12 to the oocyte, 19 in somatic cells, 8 in cumulus granulosa cells, 10 mural granulosa cells, and 10 in cumulus granulosa cells during the oocyte transition from a chromatin non-surrounded nucleolus (NSN) to a surrounded chromatic nucleolus (SN). More than 60% of the receptors were specific (i.e., 12, 24, 3, 12, and 7 in the oocyte and somatic, cumulus and mural granulosa cells and in cumulus cells during the oocyte transition from NSN to SN, respectively). In terms of the number of stages that a given gene that encode for ligand or a receptor was active, some intercellular signaling proteins were more ubiquitous as they were active at multiple ovarian follicle stages (e.g. Apoa4 from secondary to large antral follicles, Igf1 from primary to large antral follicles), while others were very specific, such as Bmp15 and Wnt6, which were only active during the primordial to primary transition or in the primary to secondary transition, respectively. Finally, we identified several genes that encode for ligands that have been reported to bind to multiple genes that encode for receptors, e.g., Thbs1 or Vegfa (55 and 37 connections during the small to large antral transition, respectively). Yet some were very specific as only bind 1 or 2 receptors, e.g., Shh or Rspo2 (File S1).
Constructions of inter-cellular signaling networks during ovarian follicle maturation
Combination of multiple datasets and manually curated databases—i.e., Metacore and DIP(Xenarios et al. 2000)—led to the identification of a total of 1,663 connections between the 100 ligands and 95 receptors (File S1). Out of those interactions, 46% of the connections were autocrine signaling, mostly between somatic cells (45%)—note that these somatic autocrine connections could be within the same somatic cell or between two different somatic cells within the same follicle or even between two different follicles. In terms of the possible paracrine signaling, more than 22% were initiated from a ligand produced by the oocyte. Interestingly, 290 of the 1,663 connections that we identified were between specific genes that encode for ligands and receptor (i.e., only significant in one cell type) and 40% of them were autocrine signaling, mostly within somatic cells (65 somatic ligands to somatic receptors).
Inter-cellular signaling networks in vivo from primordial to primary
The primordial to primary transition was the most complex of all the stages during ovarian follicle development. The majority of inter-cellular communications occurred during the transition from primordial to primary ovarian follicles (22% of the connections). Multiple hallmarks during this transition were present in the transcriptomics data (e.g., zona pellucida formation, gap connections) and multiple genes known to be involved in the development of primary ovarian follicles from primordial ovarian follicles were identified in the transcriptional data (e.g., Zp1, Gja4, Amhr, Supplemental Note 1).
The transcriptional activity and the number of paracrine and autocrine communications of the somatic cells surpassed those of the oocyte (Table S1). Out of all the autocrine and paracrine communications, only a few of them were cell specific, i.e., between ligands and receptors predicted as uniquely present in one cell type (File S1). Precisely, only 11 oocyte autocrine, 15 somatic autocrine and 12 oocyte-somatic and 28 somatic-oocyte paracrine communications were specific. Our results recovered well-known ligands in this stage, such as Gdf9, Bmp15 or Amh as well as ligands that have not been previously reported in the literature of their presence and role during the primordial to primary ovarian follicle transition, such as Adam2 and Ntn3 (Fig. 2)
The intricacy and complexity of the primordial to primary transition is clearly depicted in the inter-cellular networks, which were dived into several subnetworks. The largest subnetwork (shown in green in Fig. 1) included well-known ligands and receptors from the Tgf family (e.g., Gdf9, Bmp4, Inha) and from the Bmp family and also contained a substantial core of diverse extracellular binding protein families, such as integrins (e.g., Itga6, Itgb1), laminin (e.g., Lama1, Lamab1) and collagen (e.g., Col18a1). Only the integrins and laminins transcribed in the oocyte (i.e., Itga5, Itgb1, Lamb1) significantly changed in this transition (p-value<0.01, File S1). Some ligands, such as Dcn—proteoglycan identified in the later stages of follicle development (Adam et al. 2012)—and Nrp1 were produced by both the oocyte and the somatic cells. Additionally, this green subnetwork contained connections not previously study in relationship to ovarian follicle development, such as Ntn3 and their corresponding receptors, e.g., Unc5b, Unc5c and Neo1. According to our computational model, Ntn3 was a somatically-produced ligand that interacted with receptors in oocyte and the somatic cells. Ntn3 functions have been described in other developmental processes, such as axonal growth (Kang et al. 2004; Wang et al. 1999). Similarly, its receptors Unc5b and Unc5 are known to participate in angiogenesis(Larrivee et al. 2007; Lu et al. 2004) and are anti-apoptotic (Ozmadenci et al. 2015) and Neo1 is related with cellular growth(Wilson and Key 2007).
Other medium size networks (e.g., purple) encoded the Ephrin and Wnt families. Oocyte secreted Efna5 ligand interacted with Epha1 and Gji somatic receptors, which were among the top genes whose transcriptional abundance changed the most (File S1). Mice who lack Efna5 ligand are subfertile (Buensuceso et al. 2016). Also, this purple subnetwork encompassed paracrine communication from somatic ligands Wnt4 and Wnt5 to oocyte specific receptors Lrp6 and Ryk. Wnt4 signaling regulate the expression of Amh and mice that lack Wnt4 suffer from premature ovarian failure(Prunskaite-Hyyrylainen et al. 2014). Rspo2 somatic specific ligand also bind to the oocyte specific receptor Lrp6. The red highlighted subnetwork contained a somatic autocrine communication between B2m and Fcgrt and the later was also involved in a specific oocyte-somatic communication with the oocyte ligand Ighg1. B2m and Igh1 have been previously studied in relation to ovarian cancer biology(Qian et al. 2018; Yang et al. 2009), but this research has not been extended to ovarian follicle development.
Finally, two other smaller and disconnected networks were related with the Ramp and the Robo families of receptors. Ramp2 was identified as a specific somatic receptor that could bind to the oocyte specific ligand Adm2—pink subnetwork. Adm2 prevents oocyte atresia by regulating cell-cell interactions in cumulus-oocyte complexes (Chang et al. 2011). The specific somatic receptor Silt2 could bind to the non-specific somatic ligand Robo4—yellow subnetwork. This interaction between Silt2 and Robo is known to occur at the time of the formations of the ovarian follicles and diminished the rate of oocyte proliferation(Dickinson et al. 2010). Finally, neurotrophins soluble growth factors were also identified as important during the primordial to primary transition, corroborating previous studies (Dissen et al. 1995) and their involvement on the formation of squamous somatic cells (Dissen et al. 2001) through the oocyte-somatic specific paracrine communication between Ntf5 and Ngfr.
Inter-cellular signaling networks in vivo from primary to secondary
During the transition between primary to secondary follicles, the follicle start acquiring up to 10 layers of granulosa cells (Pedersen and Peters 1968), the formation of the theca layer commences and the follicles are capable to produce estrogenic hormones. Several of the transcripts involved in those developmental pathways could be recapitulated from the recompiled ensemble transcriptional data, such as Cyp17a1 (Supplemental Note 2). Numerous transcripts were changing during the primary to secondary transition, most likely due to the addition of the theca cells, yet the complexity of the inter-cellular signaling network was reduced compared with the primordial to primary transition (Table S1, Fig. S1). The majority of the intercellular communications during the primary to secondary transition were autocrine communications between somatic cells, followed by paracrine signaling between ligands secreted by the somatic cells to receptors in the oocyte. Only 21 of the inter-cellular communications were specific, i.e., there were between ligands or receptors only expressed in the oocyte or somatic cells. For instance, Angpt2 was produced by the somatic cells and interacted with integrins Itga5 and Itgb5 that were present only in somatic cells and with an oocyte specific receptor, Tek. Angpt2, Igf2, Pros1, and Thbs2 were the only cell-specific genes that encode for ligands and H2-D1, H2-L, Tgfbr2 and Sdc3 the only transcripts encoding for cell-specific receptors that were significantly altered during the primary to secondary transition.
In terms of sub-networks, communities highlighted in green, pink, and blue during the primordial to primary transition (Fig. 1) were diminished in terms of the number of connections between the transcripts and the yellow and purple communities were not present at all (Fig. S1). Similarly, the somatic-oocyte paracrine and somatic autocrine communications of Amh were not identified either. Interestingly, a somatic-specific subnetwork of the Edn family appeared during the primary to the secondary transition, in agreement with prior studies of the role of endothelin in ovarian follicle development(Bridges et al. 2011)
Inter-cellular signaling networks in vivo from secondary to small antral transition
Only a few secondary follicles sensitive to endocrinal hormones FSH and LH will transition into small antral follicles, avoiding atresia, the default pathway (McGee and Hsueh 2000). Follicles start producing androgens in the theca cells and estrogens in the granulosa cells (Wood and Strauss 2002), the antrum cavity emerges, filled with hyaluronic acid and proteoglycans(Gebauer et al. 1978; Jensen and Zachariae 1958), such as versican and perlican (Eriksen et al. 1999), and theca cells become vascularized(Young and McNeilly 2010). Multiple of the genes that are known to play a role during this transition were also significantly changing, e.g., Fshr, Vcan (File S2, Supplementary Note 3), although the number of downregulated genes exceeded the number of upregulated transcripts (Table S1). The complexity of the inter-cellular signaling network during the secondary to small antral transition was similar to the primary to secondary transitions, and thus with less inter-cellular connections than the primordial to primary transition (Fig. S2, Table S1, File S1). The majority of inter-cellular communications were somatic autocrine interactions and only 12 of them were specific. Out of the active ligands, Inha was the only one whose transcriptional abundance increased in this stage; Rspo2 and Wnt9 were significantly downregulated (File S2).
There were distinct changes during the secondary to small antral transition compared with the two other prior transitions. For instance, the major blue subnetwork during the primordial to primary transition (Fig. 1) was divided into two smaller subnetworks, one of them highly enriched in members of the Tgf family (Fig. S2). Somatic cells started been sensitive to INS2 and FSH and Wnt signaling intercellular communications were more prevalent that in the primary to secondary transition. While the Ephrin family networks appeared again at this stage, the Edn subnetwork, important in the primary to secondary transition disappeared.
Inter-cellular signaling networks in vivo from small to large antral transition
At the end of this transition, the oocyte is competent to resume meiosis (Mehlmann 2005) and the large antral cavity that allow enough oxygen supply to the oocyte is fully formed. Transcripts from genes that are known to participate during the antrum formation were present in the transcriptomic data that we collected (e.g., Hspg2, Star, Hsd3b1, Supplemental Note 4, File S2). The oocyte was mostly transcriptionally silent, yet the somatic cells were very active—even more so than in any other prior stage during ovarian follicle development, which the number of downregulated genes excessed the number of upregulated transcripts (Table S1). Opposite to the two prior transitions—from primary to small antral follicles—the complexity of the signaling network highly raised and communications were led by somatic cells (Fig. 2). Indeed, all the autocrine communications emerged between somatic cells and the majority of the paracrine signaling was through somatically-produced ligands. Only 5 inter-cellular communications were between actively changed and cell specific transcripts (i.e., oocyte or somatic cells) and all of them were somatic autocrine connections (e.g., Col18a1 and Gpc1).
The large and complex green subnetwork that involved members of the Tgf family, integrins and vascular signals during the primordial to primary transition (Fig. 1) appeared again during the formation of the antral cavity (Fig.2). The Eph family subnetwork (highlighted in purple in Fig.1) contained more nodes and more connections at this stage compared to the primordial to the primary transition. On the other hand, the blue subnetwork—mostly enriched in Wnt genes that encode for ligands—and the red subnetwork—associated with Ramp genes that encode for receptors—decreased their importance during the antral cavity formation and the subnetwork associated with the Robo family (yellow subnetwork) had completely disappeared. The connections pertaining to the Edn families were significant again at this stage, as they were during the primary to secondary transition (Fig. S1), and the transcriptional levels of the gene that encode for the Lhcgr receptor were significantly increasing for the first time during ovarian follicle development. Also, other endocrinal led communications, such as from Fsh to Fshr or from Ins2 to its somatic receptors, were present at this stage as well.
Inter-cellular signaling networks in vivo from small antral to large antral between oocyte and mural and cumulus granulosa cells
At the phenotypical level, one of the most important biological processes during antral formation is the differentiation of the granulosa cells into mural and cumulus granulosa cells (Mehlmann 2005). Several genes involve in this stage were presented in the publically available transcriptomic data of mural and cumulus cells(Wigglesworth et al. 2015b), such as Cd34 and Has2 (Supplemental Note 5, File S2). The number of downregulated and upregulated genes was very comparable in cumulus granulosa cells and in mural granulosa cells (Table S1). Interestingly, the number of total transcripts that were significantly changed in the cumulus cells far exceeded those of the oocyte, somatic cells or mural granulosa cells (Table S1). More than a third of the significantly altered genes in the cumulus cell transcriptomic data were specifically produced bynonly cumulus granulosa cells. The transcripts from mural granulosa cells exhibited a similar ratio of specificity.
The number of paracrine and autocrine signals was substantial, with almost all the autocrine signaling equally divided between mural or cumulus granulosa cells (Table S1, Fig. 3). Several paracrine communications were initiated by non-significantly changing oocyte ligands to receptors in both mural and cumulus cells and the order of magnitude of paracrine communications for cumulus and mural granulosa cells were comparable, with a limited number towards non-significantly changing receptors in the oocyte.
The network inter-cellular signaling pathways specificity seemed to agree with the distinct and specific functionality that mural and cumulus granulosa cells play during ovarian follicle development. For instance, while Gdf9 transcript was detected in the cumulus, mural and oocyte, it was only actively changing in cumulus cells. Several ligands (e.g., Wnt11, Tgfa, Inhba) and receptors (e.g., Pdgfrb, Acvr1c and Acvr2a) were specifically produced by the cumulus cells, while ligands such Epha5, Garba1, Sort1 and Pfgfd and receptors such as Fzd4, Il10rb, Cd44 or Gabrb2 were specific to mural cells. The autocrine signaling between the ligand Pdgf and the receptor Pdgfrb only occurred in cumulus cells and the paracrine communication between cumulus ligand Wnt11 to mural receptor Fzd4 was specific as well. At least at the transcriptional level, the mural autocrine communication between the ligand Lamc1 and the receptor Cd44 was important, with abrupt transcriptional increases for both genes (FC=265, p-value (fdr)=5.48*10-17 and FC=135, p-value (fdr)=8.68*10-18, respectively).
At the subnetwork level, the green subnetwork during the primordial to primary transition (Fig. 1) gained new members of the Tgf, Pdgf and integrin families. More members of the Apoa family were detected, although fragmented from the large green subnetwork (Fig. 3). Interestingly, more members of the Fgf family emerged as well as multiple small subnetworks related with Kitl, Il10, Il13ra1 and Gdnf. The Wnt family become mostly focused in Fdz receptors and also gained more connections. On the contrary, Amh signaling completely vanished at this mural-cumulus-oocyte network stage (Fig. 3).
Inter-cellular signaling networks in vivo during the NSN to SN transition in the oocyte
Oocyte maturation is required for adequate egg fertilization. One of the hallmarks required for achieving oocyte maturation is chromatin condensation (Albertini et al. 2003), i.e., chromatin condensation in the oocyte nucleolus, from a non-surrounded nucleolus (NSN) to a surrounded nucleolus (SN) stage (Zuccotti et al. 1995). This transition cannot be achieved by nude oocytes; they required the present of cumulus granulosa cells (De la Fuente and Eppig 2001). Analysis of the available transcriptomics data during the NSN to SN transition agreed with the current understating of the changes associated with oocyte maturation (Supplemental Note 6).
Cumulus cells were orchestrating the NSN to SN transition with a total of 237 inter-cellular signaling communications, out of which 162 were autocrine communications between cumulus cells, 55 were paracrine communications led by cumulus cell ligands. There was no autocrine communication between the oocyte and no paracrine communication led by ligands secreted by the oocyte. While multiple genes that encode for cumulus ligands increased their transcription rates during this stage (e.g., Wnt10a, Ltb, Il6, Ereg, Camp and Apln), their associated receptors did not significantly change their transcription levels (i.e., Aplnr, Cd14, Erbb2, Erbb4, Il6ra, Il6st, Ltbr, Robo1, Robo2, Fzd1 and Lrp5). Additionally, Gpr182 and Epha8, cumulus cell specific genes according to our models, were significantly upregulated in cumulus cells. As expected, the oocyte was completely silent at the transcriptional level (File S2).
At a more granular level, the large subnetwork during the primordial to primary transition (marked in blue in Fig. 1) contained less nodes and less interactions during the NSN to SN transition (Fig. S3). For instance, the cumulus specific ligand Fd6 and its associated connections were non-present. Similarly, Serpine1, Ins2, Erg, and Tgf families were separated from the majority of the components of the core blue subnetwork, which still contained a large number of inter-cellular communications through integrins (e.g. Itga7). Other parts of the primordial-to-primary blue subnetwork completely vanishes, such as Gdf9, Bmp and Inhibin families. Additionally, multiple small subnetworks only appeared during this stage (i.e., Ifnr1, Cd47 and Gpr182; Il6; Camp); while others disappeared (e.g., Ramp1, Akrc receptor and Fgf ligand families). Interesting, Robo, which was only present during primordial to primary transition (Fig. 1), became significant again during NSN to SN transition.
Identification of the most likely TFs that control the significant genes in each follicular cell type during in vivo follicle maturation
Finally, we identified the most likely TFs that could regulate the significant genes during ovarian follicle development in vivo for each cell type and each follicular stage (Fig. 4, File S3) by determining the targets for a given TFs from experimental, manually curated databases, e.g., Metacore, Ovarian kaleidoscope (Leo et al. 2000), or computationally, e.g., FIMO (Grant et al. 2011) and BEEML (Zhao and Stormo 2011).
Based on our results, TF ALX3 regulated oocyte transcriptional program from primordial to antral follicles, with the exception of secondary to small antral transition—most likely due to the lack of power from the close similarity between secondary follicles and small antral follicles. Other TFs that regulated oocyte development included SPZ1, from primordial to the secondary follicles and HSF4 from primary to antral follicles. Some TFs were specific to ovarian follicle transitions. For instance, TF MHGA2 was significant during the primordial to primary transition, while the TF EBOX was identified as a very likely regulator during the last ovarian follicle stage, from small antral to large antral transition.
For somatic cells, several zinc finger TFs (e.g., ZFP281 and ZFP740) regulated the initial activation of primordial follicles, including YAP signaling—also identified as a possible activated TF during the small antral to large antral transition. NR2F2 and RORA—RORA is a theca TF(Young and McNeilly 2010)—were among the most significant TFs from the primordial to the small antral transition and NR2F2 regulated the transcription of a large group of genes during antral formation. While some TFs were common in the mural and cumulus cells (e.g., SNA and MEF2), others were uniquely active either in cumulus cells (e.g., CTNNB1 and ZFP128) or in mural cells (e.g., THRB and ZBTB18), as expected due to the different functions that these two cells types have during ovarian follicle development.
DISCUSSION
Signals that control ovarian follicle development and enable the formation of a competent oocyte are not fully understood yet due to the difficulty of disentangled the intra-cellular and inter-cellular communications among and between the oocyte and its surrounding somatic cells. In this report, we have presented the most plausible inter-cellular communication networks during in vivo follicle maturation, as well as the most likely TFs that controlled and regulated ovarian follicle development at each follicular state and in each individual follicle cell using available transcriptomic data. These inter-cellular networks and intra-cellular regulation can help to generate testable hypothesis in the laboratory that can enable to better understand this complex system.
Not only our computational approach has been able to recapitulate the presence of well-known ligands and receptors including the cell type that produce them (e.g., oocyte, mural cells) and the exact ovarian follicle developmental stage (e.g., from primordial to primary stage), but we also identified novel ligands and receptors that were non-known to play a role during ovarian follicle development. For instance, several families of ligands and receptors that are well-known to intervene during follicle development such as the Bmp, Inh, and Tgfb families and their corresponding receptors, the Bmpr, Acvr and Tgfbr, as well as mechano-transduction receptors such as integrins were included in our inter-cellular signaling networks. Yet, our transition specific networks also portrait other families that have been little explored during ovarian follicle development. For instance, the role of the Efn family and its receptors Eph (Buensuceso and Deroo 2013), or the functions associated with the binding of the Thbs family (Hatzirodos et al. 2014) to integrins present in the oocyte plasma membrane, especially during the primordial to the primary transition. Additionally, there are several unique genes such as Gnr, Ighg1, Ndp, Ntn3, Pibf1, Pros1, Sct, App, Neo1, Tyro3, Ptprz1 or Phtr1, just to name a few, which have been barely examined in their connection to ovarian follicle development. Moreover, the combinatorial possibilities are enormous: there are ligands capable to bind to the same receptor in the oocyte and somatic cells such as Amh, while others bind to multiple receptors such as Vegfa; there are ligands whose genes are expressed in both cell types and have receptors in both cell types as well, such as Dnc. All this complex inter-cellular communication supports previous observations related with the difficulty to grow primordial follicles in 3D alginate gels, while it is possible to growth them in ovarian tissue(Eppig and OBrien 1996; O’Brien et al. 2003) or groups of primordial follicles(Hornick et al. 2013), as the somatic and ovarian cells proportionate all those ligands.
In line with the experimental observations of the difficulties of growing primordial follicles in vitro by themselves, the most entangled communication between the different cell types occurs between primordial to primary states through ligands that were mostly secreted from somatic cells. While the oocyte autocrine and oocyte-somatic paracrine communication proportions decreased during follicle maturation, from 64 to 0; somatic autocrine and somatic-oocyte paracrine communications, were maintained or slightly decreased (Table S1, Fig. S4). These results highlight the growing importance of somatic cells in controlling the intercellular communications between the oocyte and the somatic cells as the follicle matures from the primordial to the antral stage. Additionally, our inter-cellular networks between the primordial to primary transition indicated that this early stage may entitle several very convoluted communications between the oocyte and the surrounding somatic cells or very likely the stromal cells in the ovary as primordial follicles can grow in ex-vivo ovaries. Importantly, from the inter-cellular networks, there are several possible candidates that maybe further explored experimentally by adding them to the follicle culture media to grow in vitro primordial follicles, such as Ntn3 or Omd.
Finally, using available experimental data and computational methods, we were able to also recovered TFs that have been studied before during ovarian follicle maturation as well as others not that well understood, such as zinc finger proteins. Our results indicate that the YAP/TAP pathway is indeed active in somatic cells (i.e., granulosa and theca cells) in the primordial to primary transition and during the granulosa cell expansion (Fig. 4). YAP/TAP signature is an indication of cell proliferation and growth (Lei et al. 2008), which correlates with the great cellular expansion that the granulosa cells undergo during ovarian follicle development. Yet, primordial follicles are under arrested growth due to the activation of the Hippo pathway, which in turn represses YAP/TAP activation(Kawamura et al. 2013b). In fact, the YAP/TAP pathway is activated spontaneously when arrested primordial follicles are removed from their ovaries (Kawamura et al. 2013a). While it is not clear how YAP/TAP regulation could overcome Hippo repression, one plausible mechanism is through Akt dephosphorization of YAP in activation in primordial follicles (Li et al. 2010) and through a FSH-mediated PKA activation in secondary follicles (Yu et al. 2013). The final size of antral follicle might be regulated by the activation of the Hippo pathway. Expansion of the granulosa cells increases the number of granulosa cell-cell interactions (Aplin et al. 1999) and mechanical stress is capable to activate the Hippo pathway (Gerard and Goldbeter 2014), thus repressing the YAP/TAP activation and subsequently avoiding the continuation of the granulosa cell proliferation.
While our computational approach is very powerful to disentangle the complex inter-cellular communications during ovarian follicle maturation, several pitfalls should be noted. For instance, though the inter-cellular connections apparently only affect a given follicle, secreted molecules, either proteins or metabolites by the oocyte and somatic cells are capable to alter the processes in other surrounding follicles as well, directly (i.e., by binding in the receptors active in other somatic cells) or indirectly (i.e., by altering the endocrine system). These differences cannot be unraveled with the current available experimental data. Moreover, as the transcriptomics data belong to surviving follicles, all the competition effects that lead some secondary follicles undergo atresia and not entering the recruiting pool and subsequently are not represented in the inter-cellular networks depicted in this article. Finally, the current networks are based on transcriptional abundance—we have employed RNA levels as a proxy of protein activation—and they can be highly improved by obtaining proteomic measurements at each follicular transition for each individual cell type (e.g. oocyte, granulosa, theca, mural and cumulus cells).
In summary, systems reproductive biology approaches not only allow to reveal the key ligands and receptors associated with each cell type at each transition during ovarian follicle development, but also understand the complex autocrine and paracrine communications between the oocyte and surrounding supporting cells that allow the production of a competent oocyte. We expect that our computational predictions allow to generate novel data based hypothesis that could be experimentally validated to increase our comprehension of ovarian follicle development and thus the exploration of novel treatments for fertility disorders, such as polycystic ovarian syndrome or fertility preservation.
Author contributions
BPB designed the computational approached and performed all the computational study. BPB, TKW, LJB and LDS interpreted the results and wrote the manuscript.
METHODS
Inter-cellular networks
Murine genes that encode for secreted proteins and receptors were identified using the GeneGO database (Advance Search 2.0). Protein-protein interactions between secreted proteins and receptors were obtained from manually curated databases, GENEGO and DIP(Xenarios et al. 2000). Autocrine and paracrine connections were deemed possible if either one of the member of the interaction, either the ligand or the receptor, has a statistically significant change for the corresponding transition under consideration—see Table S3 from Peñalver Bernabé and colleagues for more details—and that the corresponding receptor or ligand was at least present in the microarray for the very same transition. The specificity of each genes that encode for a ligand or a receptor was previously identified—i.e., a gene is only transcribed by the oocyte, by the mural cells(Peñalver Bernabé et al.). All the inter-cellular graphs were plotted with Cytoscape (Shannon et al. 2003).
Most likely transcription factors
Computationally predicted targets of TFs were detected by exploring whether the TF position weighted matrices (PWMs) could bind to the consensus mammalian promoter regions of a given gene (Xie et al. 2005) between −2000 to 2000 base pairs with respect to the transcription starting site (TSS) of the given gene. We used to different TF binding site search programs, FIMO (Grant et al. 2011) and BEEML (Zhao and Stormo 2011), to establish the targets of a TF. Agreement between the results of FIMO and BEEML, using cutoff of p-value≤10-4 and E-score≥0.3, respectively, was deemed as an indication that a given TF could bind to the promoter region of a gene. The list of explored TFs using FIMO and BEEML was obtained from a combination of different sources: i) TFs that have experimentally identified position weighted matrices (PWMs) in vertebrates that were reported in TRANSFAC(Matys et al. 2003; Wingender et al. 1996); ii) the non-overlapping PWMs that were in the CIS-BP database (Weirauch et al. 2014) but not in the TRANSFAC, including the non-inferred PWMs in the CIS-BP dataset –a total 3,216 PWMs from 1,164 different TFs. We additionally added connections between TFs and target genes that were in the Ovarian kaleidoscope database (Leo et al. 2000), and from conserved motifs in mammals (Schulz et al. 2012). We established whether a TF was active at a given stage by determining the significance of the ratio between genes that were significantly changing for a given cell type and stage compared with non-significant genes using a hypergeometric distribution (Sui et al. 2005). A total of 500 bootstrapping samples with the same number of the non-significant genes for a given cell type and stage than the number of significant genes were selected. Medians for p-values were reported after been corrected for multiple comparisons using false discovery rate method (Benjamini and Hochberg 1995).
SUPPLEMENTARY FILES
File S1. Inter-cellular connections in vivo during ovarian murine follicle development
File S2. Transcriptional abundance of selected genes in the oocyte, somatic, cumulus and mural cells during ovarian follicle development. This file has been adapted from File S8 from Peñalver Bernabé and colleagues (17)
File S3. The most likely involved TFs during ovarian follicle development.
ACKNOWLEDGMENT
We thank Dr. Lei Lei, Sarah Kiesewetter for their invaluable technical support and Dr. Nadereh Jafari, Director of the Genomics Core Facility (Center for Genetic Medicine, Northwestern University) and Dr. Simon Lin and Dr. Gang Feng.
Footnotes
Beatriz Peñalver Bernabé, Ph.D., Arnold O. Beckman Postdoctoral Fellow Department of Pediatrics, University of California San Diego School of Medicine, 9500 Gilman Drive, La Jolla, CA 92093, Email: bpenalverbernabe{at}ucsd.edu, Phone: 812-760-2280
Teresa K. Woodruff, Ph.D., Thomas J. Watkins Professor Department of Obstetrics and Gynecology Northwestern University, Robert H Lurie Medical Research Center Room 10-119, 303 E Superior, Chicago IL 60611, Email: tkw{at}northwestern.edu, Phone: 312-503-2504
Linda J. Broadbelt, Ph.D., Sarah Rebecca Roland Professor, Department of Chemical and Biological Engineering, Northwestern University, 2145 Sheridan Road Tech L253, Evanston, IL 60208-3109, Email: broadbelt{at}northwestern.edu, Phone:847-467-1751
Lonnie Shea, Ph.D. Chair, William and Valerie Hall Professor Department of Biomedical Engineering University of Michigan 1119 Carl A. Gerstacker Building 2200 Bonisteel Boulevard Ann Arbor, MI 48109-2099 Email: ldshea{at}umich.edu Phone: (734) 764-7149