Cumulin, an Oocyte-secreted Heterodimer of the Transforming Growth Factor-β Family, Is a Potent Activator of Granulosa Cells and Improves Oocyte Quality*

Background: Cumulin is a newly identified heterodimeric member of the TGF-β family. Results: Mature cumulin potently stimulates granulosa cell signaling and function, whereas pro-cumulin promotes oocyte quality. Conclusion: Formation of cumulin and its potent actions are likely to be central to oocyte paracrine signaling and mammalian fecundity. Significance: The discovery of cumulin provides unique opportunities to improve female fertility in mammals. Growth differentiation factor 9 (GDF9) and bone morphogenetic protein 15 (BMP15) are oocyte-specific growth factors with central roles in mammalian reproduction, regulating species-specific fecundity, ovarian follicular somatic cell differentiation, and oocyte quality. In the human, GDF9 is produced in a latent form, the mechanism of activation being an open question. Here, we produced a range of recombinant GDF9 and BMP15 variants, examined their in silico and physical interactions and their effects on ovarian granulosa cells (GC) and oocytes. We found that the potent synergistic actions of GDF9 and BMP15 on GC can be attributed to the formation of a heterodimer, which we have termed cumulin. Structural modeling of cumulin revealed a dimerization interface identical to homodimeric GDF9 and BMP15, indicating likely formation of a stable complex. This was confirmed by generation of recombinant heterodimeric complexes of pro/mature domains (pro-cumulin) and covalent mature domains (cumulin). Both pro-cumulin and cumulin exhibited highly potent bioactivity on GC, activating both SMAD2/3 and SMAD1/5/8 signaling pathways and promoting proliferation and expression of a set of genes associated with oocyte-regulated GC differentiation. Cumulin was more potent than pro-cumulin, pro-GDF9, pro-BMP15, or the two combined on GC. However, on cumulus-oocyte complexes, pro-cumulin was more effective than all other growth factors at notably improving oocyte quality as assessed by subsequent day 7 embryo development. Our results support a model of activation for human GDF9 dependent on cumulin formation through heterodimerization with BMP15. Oocyte-secreted cumulin is likely to be a central regulator of fertility in mono-ovular mammals.

Growth differentiation factor 9 (GDF9) and bone morphogenetic protein 15 (BMP15) are oocyte-specific growth factors with central roles in mammalian reproduction, regulating species-specific fecundity, ovarian follicular somatic cell differentiation, and oocyte quality. In the human, GDF9 is produced in a latent form, the mechanism of activation being an open question. Here, we produced a range of recombinant GDF9 and BMP15 variants, examined their in silico and physical interactions and their effects on ovarian granulosa cells (GC) and oocytes. We found that the potent synergistic actions of GDF9 and BMP15 on GC can be attributed to the formation of a heterodimer, which we have termed cumulin. Structural modeling of cumulin revealed a dimerization interface identical to homodimeric GDF9 and BMP15, indicating likely formation of a stable complex. This was confirmed by generation of recombinant heterodimeric complexes of pro/mature domains (pro-cumulin) and covalent mature domains (cumulin). Both pro-cumulin and cumulin exhibited highly potent bioactivity on GC, activating both SMAD2/3 and SMAD1/5/8 signaling pathways and promoting proliferation and expression of a set of genes associated with oocyte-regulated GC differentiation. Cumulin was more potent than pro-cumulin, pro-GDF9, pro-BMP15, or the two combined on GC. However, on cumulus-oocyte complexes, pro-cumulin was more effective than all other growth factors at notably improving oocyte quality as assessed by subsequent day 7 embryo development. Our results support a model of activation for human GDF9 dependent on cumulin formation through heterodimerization with BMP15. Oocyte-secreted cumulin is likely to be a central regulator of fertility in mono-ovular mammals.
In this study, we consider the unique interaction of two members of the TGF-␤ family: growth differentiation factor 9 (GDF9) 2 and bone morphogenetic protein 15 (BMP15). These two closely related proteins are oocyte-secreted factors essential for fertility in mammals (1). The primary role of GDF9 and BMP15 is to act on the cells surrounding the oocyte, known as granulosa cells (GC), which in turn supply the oocyte with the support necessary for future healthy embryo/fetal development (2). As with all members of the TGF-␤ family, GDF9 and BMP15 are produced as pro-proteins, consisting of an N-terminal proregion and a receptor binding C-terminal mature region. Activation of these factors requires proteolytic processing by a furin-like protease (3,4). For most TGF-␤ mem-bers, the processed mature domains signal as covalently linked homodimers, although covalent heterodimers are also known (inhibin A and B (5)). GDF9 and BMP15 are unusual within the TGF-␤ family in that they (together with GDF3 and Lefty1/2) lack the cysteine residue involved in forming the mature region intersubunit disulfide bond (6). Despite this, mature GDF9 and BMP15 are thought to maintain the canonical dimer architecture of other TGF-␤ ligands and signal as noncovalent dimers.
As with all members of the TGF-␤ family, GDF9 and BMP15 signal via cell surface receptors, which activate intracellular Sma-and Mad-related (SMAD) proteins by phosphorylation of critical residues, which subsequently regulate the transcription of various genes (1). The receptors are Ser/Thr kinases and are, on the basis of distinct features in the cytoplasmic domain, referred to as either type I (activated by the relevant type II receptor) or type II (constitutively active) receptors. Following prodomain displacement, BMP15 binds to complexes of BMP receptor type II (BMPRII) and activin receptor-like kinase 6 (ALK6), thereby leading to activation of the SMAD1/5/8 intracellular pathway (7,8). Mature GDF9 binds BMPRII and ALK5 and signals through the SMAD2/3 pathway (9 -11).
GDF9 and BMP15 are also functionally unusual within the TGF-␤ family, in that they show remarkable species differences in their activity. Notably, human GDF9 is latent, because it remains associated with its prodomain, whereas mouse GDF9 is particularly potent, because of its prodomain being readily displaced (3,12,13). By contrast, human BMP15 is recombinantly produced in an active form in mammalian cell culture, whereas mouse BMP15 is not expressed under the same conditions (4). The species-specific differences in expression and activity of these oocyte-secreted growth factors is principally due to alterations in a small number of amino acids in the prodomains (BMP15) or mature domains (GDF9), which affect complex formation or stability, respectively (3,4,13). Hence, the relative levels of oocyte expression of BMP15 and GDF9 vary markedly across species, whereby the mono-ovular phenotype (i.e. low fecundity) is characterized by balanced BMP15:GDF9 ratios, whereas polyovular species (high fecundity) express predominately GDF9 (14). As such, GDF9, but not BMP15, is essential for fertility in the mouse (15,16), whereas both are required in mono-ovular species, such as sheep (17,18).
Interactions at the genetic level between these two proteins have been observed in both mice (16) and sheep (1,18). Evidence for GDF9/BMP15 physical interactions have also been provided by co-immunoprecipitation results (19,20), and a number of functional studies have demonstrated marked synergistic interactions between these two proteins (20 -23). Recently, Peng et al. (24) have provided evidence that heterodimerization of GDF9 and BMP15 gives rise to a more potent ligand that signals solely via the SMAD2/3 pathway.
In the current study, we hypothesized that human GDF9 and BMP15 form a stable heterodimer that has potent actions on mural GC, cumulus GC, and oocyte development. We provide the first evidence of a purified BMP15-GDF9 heterodimer. We find that both the promature form and a covalent mature form of the heterodimer are particularly potent and are dual pathway activators, signaling via the SMAD2/3 and SMAD1/5/8 path-ways. Interestingly, only the promature heterodimer is capable of improving oocyte quality, indicative of the crucial function that the prodomain plays in the functioning of these proteins. In deference to its potent activation of cumulus cells and to align with the existing nomenclature for the other major reproductive TGF-␤ superfamily heterodimer, inhibin, we have called this new heterodimer cumulin.

Experimental Procedures
Expression Plasmids, Stable Cell Lines, and Protein Production-Expression cassettes encoding wild type human GDF9 or BMP15 or mouse GDF9 (mGDF9) were synthesized (Genscript USA, Inc., Piscataway, NJ) incorporating the rat serum albumin signal sequence at the 5Ј end followed by sequences encoding for His 8 and StrepII affinity tags at the N terminus of the respective proregion. The various human GDF9 and BMP15 Ser-Cys mutants (Ser 418 in GDF9 and Ser 356 in BMP15) used in this study were derived from expression cassettes (Genscript USA, Inc., Piscataway, NJ) incorporating the glutamate receptor GluR-D subunit signal sequence (MRIICRQIVLLFSGFWGLAMG) at the 5Ј end followed by sequences encoding for the FLAG tag (a design based on Kuusinen et al. (25)) and His 6 affinity tag at the N terminus of the respective proregion (as used previously (8)). All the cDNA fragments were cloned into the pEF-IRES expression vector (26).
HEK293T cells were obtained from the American Type Culture Collection and cultured in DMEM (Invitrogen) containing 10% FCS, 2 mM L-glutamine, 100 IU/ml penicillin (Sigma), and 100 g/ml streptomycin. The different expression plasmids were transfected into HEK293T cells using Lipofectamine 2000 (Invitrogen) following the manufacturer's instructions. Stable cell lines were established via puromycin selection. For production of cells producing pro-cumulin, HEK293T cells were cotransfected with both an expression plasmid for human GDF9 of wild type sequence, incorporating no epitope tags, along with the expression plasmid for wild type human BMP15 (His 8 /Stre-pII-tagged), as described above. For production of cells producing cumulin (as a covalently stabilized form), HEK293T cells were co-transfected with expression plasmids encoding human GDF9 and BMP15 as the Ser-Cys mutant forms, as described above.
Chromatography-Production medium containing pro-GDF9, pro-BMP15, or pro-cumulin was modified to include 10 mM imidazole, 500 l of protease inhibitor mixture (Thermo Scientific), 500 l of phosphatase inhibitor mixture (Calbiochem, Alexandria, Australia) and subsequently adjusted to pH 7.4. Invitrogen nickel-nitrilotriacetic acid-agarose resin was incubated with conditioned medium on a rotary agitator at 4°C for 2 h and centrifuged at 900 rpm for 5 min, and the supernatant was removed. The His-tagged target protein was eluted from the nickel-nitrilotriacetic acid-agarose using elution buffer (PBS/500 mM imidazole) and stored in LoBind tubes (Eppendorf, Hamburg, Germany) at Ϫ80°C. Cumulin (the covalently stabilized form) was produced and purified as above, with the exception that the mature region was eluted from the IMAC resin via a 7 M urea wash. This material was then loaded onto a reverse phase HPLC column equilibrated with 0.1% TFA, and the different mature region forms (i.e. GDF9, cumulin and finally BMP15) were eluted via a linear gradient of acetonitrile. The fractions were stored in LoBind tubes at Ϫ80°C.
Silver Staining, Western Blotting, and Protein Quantification-Fractions from the cumulin reverse phase HPLC were prepared for electrophoresis by lyophilizing 5 l and resuspending in SDS-PAGE loading buffer Ϯ DTT. All samples were heated at 95°C for 5 min and then loaded onto a Mini PROTEAN TGX 4 -15% gel (Bio-Rad). After 40 min at 150 V, the gels were used for silver staining or Western blotting. Silver staining was carried out following the manufacturer's instructions using a silver quest staining kit (Invitrogen). For Western blotting, proteins were transferred to a Hybond-ECL membrane (GE Healthcare) for 75 min at 100 V. The membranes were blocked with 2% blocking reagent (supplied in an ECL Advance kit; GE Healthcare) diluted in TBS containing 0.1% (v/v) Tween 20 for 1 h at room temperature. Subsequently the membranes were incubated with anti-GDF9 antibody (mAb53 (28)) or anti-BMP15 antibody (mAb28 (27)) diluted 1:5000 at 4°C overnight, followed by incubation with goat anti-mouse IRDye TM 800 CW antibody diluted 1:10,000, for 1 h in the dark. Fluorescence intensity was determined using an Odyssey infrared imaging system (Licor Bioscience, Lincoln, NE). The various purified forms of GDF9 or BMP15 were quantified via Western analysis relative to recombinant mGDF9 or hBMP15 standards, respectively (R&D Systems).
Granulosa Cell Bioassays-Bioactivities of pro-forms and/or mature forms of BMP15, GDF9, and cumulin were compared using primary mouse mural GC and a human granulosa tumor cell line (COV434). The mouse granulosa cell [ 3 H]thymidine incorporation bioassay was performed using standard procedures as previously described (28,29). In brief, mural GC were collected from 129/Sv mice 44 -46 h after gonadotropin priming. Cells were cultured at 37°C in 5% CO 2 in protein-free medium at 2 ϫ 10 5 cells/ml with treatments for 18 h followed by a further 6 h with 15.4 kBq [ 3 H]thymidine (PerkinElmer Life Sciences). Granulosa cell [ 3 H]thymidine incorporation was assessed using a Wallac Microbeta plate counter as an indicator of cell DNA synthesis. Each treatment was performed in duplicate, and the experiments were replicated three to six times.
RT-PCR-Mural GC were collected from 21-26-day-old C57BL/6 ϫ DBA F1 hybrid female mice 46 h after intraperitoneal administration of gonadotropin priming. Mice were maintained in accordance with the Australian Code for the Care and Use of Animals for Scientific Purposes and with the approval of the University of New South Wales Animal Ethics Committee. Mural GC (1 ϫ 10 5 live cells per 24-plate well) were cultured for 18 h in bicarbonate-buffered ␣MEM (Gibco Life Technologies) supplemented with 3 mg/ml BSA (Sigma-Aldrich). The cells were then treated for 6 h with fresh medium containing recombinant human FSH (50 mIU/ml; Puregon; Organon, Oss, The Netherlands) plus the following treatments, as indicated: 100 ng/ml cumulin, 100 ng/ml pro-cumulin, 100 ng/ml pro-mature GDF9, 100 ng/ml pro-mature BMP15, or 50 ng/ml each of promature GDF9 ϩ pro-mature BMP15 when added together. RNA extraction, reverse transcription, and quantitative PCR were performed as previously described (31) using their reported primers; however, PCRs were performed on a Roche LightCycler 480, and results are expressed as raw gene expression levels relative to the geometric means of two housekeeping genes (Mrpl19 and Ppia) using the 2 Ϫ(⌬⌬CT) method. The data are means Ϯ S.E. of four replicate experiments.
Oocyte Quality Assay-The effects of treatments on oocyte quality (oocyte developmental competence) were assessed by treating cumulus-oocyte complexes with the different growth factors during the oocyte in vitro maturation (IVM) phase, followed by in vitro fertilization and embryo culture and assessment of subsequent blastocyst yield and quality. We chose to use a well established porcine model of low oocyte developmental competence, because this experimental model better approximates the complexities of human oocyte developmental competence than that of murine oocytes. This model uses oocytes from small antral follicles collected from abattoir-derived ovaries from gilts (peripubertal female pig) (13,(32)(33)(34). The methods used were as per recently described in detail (13,34). In brief, cumulus-oocyte complexes were aspirated from small antral follicles (2-4 mm) and matured in protein-free porcine oocyte maturation basic IVM medium (35), supplemented for the first 22 h of IVM with 1 mM dibutyryl-cAMP and 100 ng/ml amphiregulin (R&D Systems) and treated with either vehicle (control), 20 or 100 ng/ml pro-cumulin, 20 or 100 ng/ml mature cumulin, 100 ng/ml pro-GDF9, 100 ng/ml pro-BMP15, or 100 ng/ml of each of pro-GDF9 ϩ pro-BMP15. Cumulus expansion was assessed at 22 h of IVM and scored as per standard criteria (36,37). At 22 h, cumulus-oocyte complexes were washed and matured for a further 22 h in porcine oocyte maturation without supplements or treatments (13,33,34). Thereafter, in vitro fertilization, in vitro embryo production, and assessment of blastocysts was performed as previously described (13). Five replicate experiments were performed with ϳ180 oocytes/replicate. Molecular Modeling-To obtain three-dimensional structure models for mature human GDF9, mature human BMP15, and cumulin, homology modeling was performed using either the software SWISS-Modeler or a manual approach. To identify the most suitable structure template, the sequences of GDF9 and BMP15 were blasted against the protein sequences present in the Protein Data Bank, revealing that both oocyte factors exhibit the highest sequence homology (32-33% identity) to either BMP2 or GDF5. For manual modeling the sequences of the mature region of GDF9 and BMP15 were then aligned to the amino acid sequence of the identified templates, i.e. BMP2 (Protein Data Bank entry 3BMP and 1REW (chains A and B) and 3BK3 (chains A and B)) and GDF5 (Protein Data Bank entry 1WAQ) using the software CLUSTALW. A model of human GDF9 comprising residues Asn 327 -Arg 430 was then built manually by replacing amino acid residues in the BMP2 template (chains A and B of Protein Data Bank entry 3BK3) employing the tool ProteinDesign of software package Quanta2008 (MIS Accelrys, San Diego, CA). A single amino acid insertion at the end of helix ␣1 found in GDF9 (when compared with the template BMP2) was rebuilt manually. For modeling of human BMP15 (comprising residues Asn 289 -Arg 392 ), the same approach was applied but using the structure of human GDF5 (Protein Data Bank entry 1WAQ) as BMP15, similar to GDF5, binds specifically the BMP receptor BMPR1b. Both initial models were subsequently refined first by performing rotamer searches using the tool X-BUILT of the software Quanta2008 to identify side chain conformations not leading to van der Waals clashes. Then energy minimization runs were performed employing only geometrical but no electrostatic energy terms using Quanta2008 and the forcefield CHARMm27. First, all backbone atoms were kept fixed, and only side atoms were allowed to move but restrained by harmonic positional restraints (5 kcal mol Ϫ1 Å Ϫ2 ) to minimize atom movements. After energy minimization of side chain atoms converged, these restraints were removed, and the positions of the backbone atoms were refined by applying a strong harmonic potential (25 kcal mol Ϫ1 Å Ϫ2 ). The final models of GDF9 and BMP15 exhibit good backbone and side chain geometry. The structure models for both homodimeric factors are highly similar, yielding a root mean square deviation of only 1.1Å for 164 C␣ carbons. Only 40 residues, which are located either in the fingertips or the loop N-terminal of helix ␣1, exhibit differing conformations, which may be due to selecting different template structures for the models and does not reflect inherent structural differences between GDF9 and BMP15.
The model of cumulin was then built by superimposing the three-dimensional models of homodimeric GDF9 and BMP15 and selecting one chain of each structure to form a heterodimeric assembly. Because of the almost identical amino acid composition in the dimer interface, the model of cumulin required very minor minimization, which was performed by rotamer searches for side chains with close contacts. The heterodimeric noncovalent form of cumulin was then converted into the covalent form by exchanging the serine residues Ser 356 in BMP15 and Ser 394 in GDF9 against cysteines and connecting both sulfur atoms to form the thioether bond. The disulfide bond was subsequently refined by a short energy minimization run to obtain a thioether bond of correct length.
Statistical Analysis-For mRNA expression data, differences between means were assessed by one-way analysis of variance followed by Tukey's multiple comparison test. Similarly, effects of treatments on embryo yield were assessed by one-way analysis of variance on arc-sine transformed data followed by a least significant difference post hoc test. Cumulus expansion data were assessed by a Kruskal-Wallis one-way analysis of variance on ranks followed by a Dunnett's post hoc test. p Ͻ 0.05 was considered statistically significant.

Results
The various forms of GDF9, BMP15, and cumulin analyzed in this study were as follows: 1) pro-forms: noncovalent homoand heterodimers of the promature regions of BMP15 and GDF9 (the form in which these proteins are secreted from the cell); 2) mature forms: noncovalent homo-and heterodimers of the mature regions of BMP15 and GDF9 (the form in which these proteins bind to their signaling receptors); and 3) covalent forms: covalent homo-and heterodimers of the mature regions of BMP15 and GDF9 (stabilized versions of the mature growth factors).
GDF9/BMP15 Synergism Is Mediated by Heterodimer (i.e. Cumulin) Formation-We have consistently observed GDF9 and BMP15 interacting in a synergistic manner, whether in their pro-forms (i.e. as isolated pro/mature complexes; Fig. 1) or as purified mature proteins (23,38). The extracellular formation of a BMP15/GDF9 heterodimer, which we have termed cumulin, could underlie this potent synergism. To directly test this possibility, we generated a covalent BMP15 homodimer (by replacing Ser 356 with the corresponding cysteine found in other TGF-␤ proteins), which would be incapable of forming cumulin in the presence of wild type GDF9. Covalent human BMP15 mature region homodimer (BMP15 S:C ) was purified and tested for its ability to synergize with human GDF9 (mature region homodimer) in a mural granulosa cell proliferation assay. At the low doses tested, GDF9, BMP15, and BMP15 S:C did not promote proliferation; however, the GDF9ϩBMP15 combination showed potent synergism ( Fig. 2A). In contrast, this synergism was not observed with the GDF9ϩBMP15 S:C combination. Because we have previously shown that GDF9/BMP15 synergism is dependent upon Smad2/3 activation (23, 38), we examined the ability of our BMP15 variants to affect this response in COV434 cells. Although wild type BMP15 promoted GDF9-dependent Smad2/3 transcriptional activity in a dose-dependent manner, the BMP15 S:C variant could not potentiate a GDF9 effect (Fig.  2B). To ensure the BMP15 S:C variant had bioactivity, we tested its ability to stimulate the SMAD1/5/8 pathway in COV434 cells transfected with a BMP-responsive luciferase reporter. In this system, the BMP15 S:C protein was equally potent as wild type BMP15 (Fig. 2C). As expected, GDF9 did not activate the SMAD1/5/8 pathway. Together, these results indicate that extracellular heterodimerization of mature GDF9 and BMP15 is possible, because simple co-addition of GDF9ϩBMP15, but not of GDF9ϩBMP15 S:C , resulted in potent synergism.
To further explore extracellular associations between GDF9 and BMP15, we mixed conditioned media from cells expressing untagged pro-GDF9 with conditioned media from cells expressing N-terminally poly-His-tagged pro-BMP15 and carried out IMAC purification of the pro-BMP15. Under the mild conditions used, the untagged GDF9 mature region was capable of associating and co-purifying with pro-BMP15, as evidenced by GDF9 immunoreactivity co-eluting off the IMAC resin (Fig. 3A). No GDF9 immunoreactivity was detected when BMP15 was purified alone (data not shown), and the untagged GDF9 did not bind to the IMAC resin in the absence of the poly-His-tagged BMP15 (Fig. 3B). It is clear from these results that, like the mature proteins, the pro forms of GDF9 and BMP15 are capable of associating extracellularly to form pro-cumulin.
Intracellular Formation of Cumulin-GDF9 and BMP15 are co-expressed in the oocyte throughout folliculogenesis and so do not naturally exist in isolation (reviewed in Ref. 39). Thus, heterodimerization of these two TGF-␤ family members to form cumulin is likely to occur during synthesis within the oocyte, as well as after secretion into the extracellular space. In the in vivo context, cumulin will be in a form where the mature regions are noncovalently associated and will most likely act as a pro/mature complex. To produce such a form (i.e. pro-cumulin), we co-expressed N-terminally poly-His-tagged pro-BMP15 with untagged pro-GDF9 and purified the noncovalent pro-cumulin complex by IMAC. Theoretically, this approach could lead to the purification of a mixture of pro-BMP15 homodimers and pro-cumulin. Quantification of the levels of both the processed GDF9 and BMP15 mature regions following IMAC purification indicated a 3:1 ratio of pro-BMP15 homodimer:pro-cumulin heterodimer (data not shown).
Pro-cumulin Is a Potent Activator of Granulosa Cells via Both SMAD Pathways-We next examined the potency of pro-cumulin in ovarian GC bioassays. We compared pro-cumulin to pro-GDF9, pro-BMP15, and pro-GDF9 ϩ pro-BMP15, after quantitating each protein form relative to the level of the GDF9 mature region (data not shown). Pro-GDF9 alone was inactive in promoting GC DNA synthesis, as shown previously (3), as was pro-BMP15 alone (Fig. 4A). In contrast, pro-cumulin was exceptionally potent on primary mouse mural GC, stimulating a 3.8-fold increase in [ 3 H]thymidine incorporation at the lowest dose tested (1 ng/ml; Fig. 4A). Pro-cumulin was notably more potent (6 -12-fold) than pro-GDF9ϩpro-BMP15 at all doses examined.
Engineering Covalent Mature Cumulin-The pro-cumulin preparation described above and that previously produced by Peng et al. (24,40) contain significant amounts of contaminating pro-BMP15 homodimer. This was due to the almost equal propensity of monomeric BMP15 to homodimerize or heterodimerize (with GDF9) and the fact that the purification strategies were directed against tagged BMP15 prodomains. To overcome these limitations and to more conclusively examine the biological function of cumulin, we undertook to engineer a covalent form of cumulin, which could be fully purified from contaminating BMP15 and GDF9. To produce such a protein, we replaced Ser 418 in GDF9 and Ser 356 in BMP15 with the corresponding cysteine found in other TGF-␤ proteins and co-ex-pressed the resultant cDNAs in HEK293T cells. We have used this approach before to produce human covalent mature BMP15 (8). Both the pro-BMP15 S:C and pro-GDF9 S:C mutants were produced with N-terminal poly-His tags; hence IMAC FIGURE 3. GDF9 associates and co-purifies with BMP15 to form procumulin. Conditioned media containing untagged pro-GDF9 and N-terminally His 8 -tagged pro-BMP15 were combined and subjected to IMAC chromatography. A, Western blot of resin load probed with Mab28 (specific for human BMP15), or the resin load, unbounded material, wash, and eluted peak fraction, probed with Mab53 (specific for GDF9). B, Western blot for IMAC chromatography of conditioned media containing only untagged pro-GDF9. Shown are the resin load, unbound material, wash, and eluate, probed with Mab53. purification of conditioned media from cells co-expressing these proteins theoretically could result in the co-purification of pro-GDF9 or pro-BMP15 homodimers, as well as the sought after pro-cumulin. We decided to target the GDF9 and BMP15 mature regions for purification to avoid any potentially misfolded disulfide linked pro/mature complexes. This was accomplished by a two-step procedure of IMAC chromatography followed by reverse phase HPLC. This approach succeeded in resolving three distinct covalent species: mature GDF9, mature cumulin, and mature BMP15 (Fig. 5A). This can be concluded because Western analysis of the fractions from the reverse phase HPLC purification indicated that fractions 31-32 contained GDF9, but no BMP15; fractions 33-34 contained both GDF9 and BMP15 (i.e. cumulin), whereas fractions 36 -39 contained only BMP15 (Fig. 5B). The silver-stained gels confirmed the presence of disulfidelinked dimers, as the reduced bands at 15-18 kDa (Fig. 5C, fractions 32-34 and 37) that migrated at 25-35 kDa without reduction (Fig. 5D, fractions 32-34 and 37).
Covalent Mature Cumulin Is Potently Bioactive-Because purified mature cumulin has not previously been produced, it was important to determine its bioactivity, especially given its form as a stabilized disulfide linked covalent mature region heterodimer. We initially tested the fractions across the reverse phase HPLC profile for bioactivity in the mural GC assay. The peak of bioactivity (Fig. 5E, fraction 34) coincided with the fraction containing the peak of covalent cumulin protein (Fig. 5,  B-D, fraction 34).
We next directly compared the bioactivity of covalent mature cumulin to the previously characterized noncovalent pro-cumulin on GC. Mature cumulin was highly potent in the range of different granulosa cell assays performed; in most cases, it was more potent than pro-cumulin (Fig. 6). In the primary mouse mural granulosa cell assay of [ 3 H]thymidine incorporation, covalent mature cumulin had an ED 50 of 0.6 ng/ml, whereas pro-cumulin had an ED 50 of 4 ng/ml (Fig. 6A). Mature covalent cumulin was also more effective than pro-cumulin at promoting SMAD2 signaling in human COV434 cells, as assessed by A3-Lux activity (Fig. 6B), whereas the two cumulin forms were equipotent at promoting SMAD1/5/8 signaling (Fig. 6C). Together, our analysis of noncovalent pro-cumulin and covalent mature cumulin allows us to propose a model whereby cumulin activates both SMAD pathways but SMAD2/3 in particular (Fig. 7).
One of the physiological roles of oocyte-secreted GDF9 and BMP15 is to direct granulosa cell differentiation toward the cumulus cell phenotype (41), which can be characterized by expression of genes required for cumulus cell mucification (Ptx3, Has2, Tnfaip6, and Ptgs2) (42). Pro-cumulin and mature covalent cumulin were notably more effective at promoting mRNA expression of these genes in primary mouse mural GC than pro-GDF9, pro-BMP15, or both combined (Fig. 6D). Mature cumulin was more effective than pro-cumulin, particularly at promoting Ptx3 mRNA expression (p Ͻ 0.05).
Pro-cumulin, but Not Mature Covalent Cumulin, Is a Potent Stimulator of Oocyte Developmental Competence-To test our human proteins in a model closer to the in vivo situation, we isolated porcine cumulus-oocyte complexes from small unstimulated ovarian follicles, treated them for 24 h with our different proteins, and examined cumulus cell expansion and oocyte developmental competence (capacity of the oocyte to support embryo development to day 7). With best practice IVM and embryo production, this model of low oocyte developmental competence typically yields 10 -20% blastocysts (13,32). Interestingly, when porcine cumulus-oocyte complexes are used as the target, rather than cells in a monolayer (either mural granulosa or COV434 cells), the most bioactive protein was noncovalent pro-cumulin and not covalent mature cumulin (Fig. 8). For cumulus expansion, the only protein that caused a significant increase was pro-cumulin (Fig. 8A). Exposure of oocytes to 20 ng/ml pro-cumulin for 24 h improved oocyte quality as assessed by a substantial improvement in capacity to support embryo development to the blastocyst (2.3-fold; 28-63%; p Ͻ 0.05; Fig. 8B) and hatched blastocyst stages (3.8-fold; 5-19%; p Ͻ 0.05; Fig. 8C). There was no significant difference in blastocyst rates when oocytes were treated with 20 or 100 ng/ml pro-cumulin (63 and 55%, respectively) or mature cumulin (32 and 39%, respectively). No other treatment improved oocyte developmental competence, although pro-GDF9 ϩ pro-BMP15 tended to increase blastocyst and hatched blastocyst rates (p Ͼ 0.05; Fig. 8C). There was no effect of any treatments FIGURE 6. The mature domain of covalent cumulin is more potent than pro-cumulin on GC. The pro-cumulin preparation was produced in HEK-293T cells with a N-terminal His tag in the prodomain allowing purification by IMAC. To generate purified mature cumulin, the IMAC product was subsequently subjected to reverse phase HPLC, generating a highly pure mature domain protein with fixed heterodimer architecture (mature covalent cumulin). Bioactivity was assessed by [ 3 H]thymidine incorporation in primary mouse mural GC (A), Smad2/3-responsive luciferase reporter activity (B), and Smad1/5/8 luciferase reporter activity in COV434 human GC (C), and mouse primary mural GC mRNA expression of genes associated with cumulus cell differentiation (D), Ptx3, Tnfaip6, Has2, and Ptgs2 (all proteins at 100 ng/ml, except pro-GDF9 ϩ pro-BMP15, which were added at 50 ng/ml each). The bars with different letters indicate significant differences at p Ͻ 0.05 within the same graph.
on subsequent blastocyst quality as assessed by blastocyst cell numbers (Fig. 8D).
Molecular Modeling and Analysis of Cumulin Assembly-To provide molecular insights into cumulin assembly, stability, and potential receptor-binding properties, we performed homology modeling to obtain structural models of homo-and heterodimeric forms of GDF9 and BMP15. Structure templates were selected on the basis of sequence homology to BMPs and GDFs with known structure in the Protein Data Bank. The models of mature GDF9 (Fig. 9, A and B) and BMP15 (Fig. 10, A  and B) exhibit the canonical butterfly-shaped architecture of TGF-␤ ligands. The structure of the monomeric subunits is FIGURE 7. Hypothesized model of human cumulin formation and signaling. BMP15 and GDF9 are co-expressed in the oocyte throughout most of oogenesis. a, during synthesis, the prodomains of BMP15 and GDF9 direct folding and dimerization of their respective mature domains. b, dimeric precursors are cleaved by furin-like proteases, and then BMP15 and GDF9 are secreted from the oocyte noncovalently associated with their prodomains, forming their respective pro-forms. The pro-cumulin heterodimer is likely to assemble within the oocyte but can also form in the extracellular space (ECM). In isolation, human pro-GDF9 is latent, and activation is dependent upon heterodimerization with BMP15 to form cumulin. Following prodomain displacement (c), cumulin activates both SMAD2/3 and SMAD1/5 transcription factors (e), presumably through a receptor complex involving two BMPRII receptors, an ALK6 receptor and an ALK5/4 receptor, on cumulus GC and mural GC (d). Pro-cumulin and mature cumulin exhibit differing potencies on different granulosa cell lineages. defined by the cysteine knot, which results in three loops emanating from this central motif. The two loops running in parallel form ␤-sheets, whereas the third loop, which emanates from the cysteine core in the opposite direction, folds into an ␣-helix. The monomer architecture is also often compared with an open hand, with the ␤-sheets forming fingers 1 and 2 and the ␣-helix forming the wrist (Figs. 9, A and B, and 10, A and B). The dimer assembly can then be envisioned as the inner sides of the two hands coming together at palm and wrist.
Unlike most TGF-␤ ligands, GDF9 and BMP15 lack the cysteine residue that forms an intermolecular disulfide bond, locking the dimer assembly (this cysteine is replaced by an isosteric serine in both GDF9 and BMP15). This unique feature, together with the very high sequence conservation across the GDF9 and BMP15 dimer interfaces (Figs. 9C and 10C), enables these growth factors to heterodimerize to form cumulin (Fig. 11, A and B). Importantly, the cumulin model shows a dimerization interface (Fig. 11C) identical to homodimeric GDF9 and BMP15 (Figs. 9C and 10C, respectively), indicating that formation of the heterodimer is possible and that the stability of such a complex is likely indistinguishable from the homodimeric forms. Production of covalent cumulin also appears likely based upon our modeling, because in silico exchange of the two serine residues (Ser 418 in GDF9 and Ser 356 in BMP15; Fig. 11B) locates both sulfur atoms sufficiently close to form a thioether bond without requiring conformational or large structural rearrangements of the dimer architecture (Fig. 11D).
Finally, analysis of our homology models indicates that the binding sites for type I receptors within the "wrist" regions are distinct in cumulin, relative to its homodimeric ancestors (Figs.  11B, 9B, and 10B). This is due to the fact that, in contrast to the type II receptor epitopes, the wrist or type I receptor-binding FIGURE 9. Homology model of mature GDF9. A, ribbon plot of human mature GDF9 highlighting the canonical dimer architecture. The two monomer subunits of the homodimer are colored in red and orange. The potential binding epitopes for the type I (wrist) and type II (knuckle) receptors are marked. The six cysteine residues forming the characteristic cysteine knot motif, as well as the serine residues replacing the cysteine residues, which are involved in the intermolecular disulfide bond found in other TGF-␤ members, are shown as sticks. B, as in A but rotated around the x axis by 90°. C, detail of the GDF9 dimer interface showing that interfacial amino acid residues are highly conserved between GDF9 and BMP15 (see also Fig. 10), suggesting that heterodimer formation (see Fig. 11) is very likely not impaired by steric or chemical restraints resulting from differences in the dimer interface. The potential binding epitopes for the type I (wrist) and type II (knuckle) receptors are indicated. The six cysteine residues (colored in yellow) forming the characteristic cysteine knot motif, as well as the serine residues replacing the cysteine residues, which are involved in the intermolecular disulfide bond found in other TGF-␤ members, are shown as sticks. B, as in A but rotated around the x axis by 90°. C, detailed view of the BMP15 dimer interface showing that interfacial amino acid residues are highly conserved with those found in GDF9 (see also Fig. 9), suggesting that cumulin formation (see Fig. 11) is very likely not impaired by steric or chemical restraints by differences in the dimer interface.
site is built from both monomer subunits. In cumulin, this epitope therefore presents a chimera with unique amino acid composition not found in either homodimeric growth factor ancestor. These differences may alter type I receptor affinity and/or specificity for cumulin and contribute to its enhanced signaling capacity, particularly in comparison to latent human GDF9.

Discussion
In females, GDF9 and BMP15 are essentially only expressed in oocytes where they are co-expressed throughout most of oogenesis and so do not naturally exist in isolation. Moreover, given their high sequence homology, these key growth factors can be expected to interact with each other and hence should generally be considered as acting together. There is ample evidence for interactions between GDF9 and BMP15 at genetic (1,16,18), biochemical (19,20), and functional levels (20 -23). A putative BMP15-GDF9 heterodimer was first modeled by McNatty et al. (1) in 2004. Most recently, Peng et al. (24) produced GDF9-BMP15 heterodimers with potent bioactivity. We have pointed out a number of potential methodological and interpretation deficiencies in that study (43), including that the preparation contained other protein forms, and the only evidence presented for formation of a heterodimer was co-immunoprecipitation of the two proteins. In the current study, we have conducted detailed molecular modeling of heterodimer assembly and succeeded in producing two heterodimeric forms: pro-cumulin and mature covalent cumulin. Both forms are particularly potent on GCs acting via both the SMAD2/3 and SMAD1/5/8 pathways and show differential actions on mural GC versus cumulus GC.
The co-addition of separately expressed GDF9 and BMP15 homodimers elicits a potent synergistic response on GC (20 -23). The results of the current study provide strong evidence that the basis for such synergism is formation of the cumulin heterodimer. Such GDF9/BMP15 synergism has been described by others as the action of two homodimers (24). Our results do not support this contention. We deduce this from the observation that the BMP15 S:C covalent homodimer, which cannot heterodimerize with GDF9, does not synergize with wild type GDF9, in contrast to wild type BMP15. The BMP15 S:C mutant in all other aspects resembles wild type BMP15. Furthermore, both pro-cumulin and mature covalent cumulin elicited a more potent response than the synergistic effects of co-addition of wild type GDF9 and BMP15 homodimers. Using the novel covalent cumulin alleviates any concerns arising from studying solely a noncovalent heterodimer, which may exist in an equilibrium between homodimers and heterodimers, as we previously discussed (43). This leads us to hypothesize that the process of heterodimerization is the mechanism of activation of the otherwise latent GDF9 in the human.
Although inhibin A and B are the other major naturally occurring TGF-␤ heterodimers (5), it is possible to co-express other TGF-␤ ligands and force their dimerization and in so doing generate more biologically potent proteins. Within the BMP subgroup, BMP2/7 (44,45), BMP2/6 (46), and BMP7/ GDF7 (47) heterodimers are all reported as more potent or as exhibiting altered bioactivity, in comparison with a combination of homodimers. There are a number of possible mechanisms to account for greater potency of heterodimers. First, a heterodimer may have a decreased binding affinity for extracellular BMP binding proteins, which commonly act as antagonists. Many of these antagonists act themselves as obligate or nonobligate dimers and thus require a symmetrical homodimer for maximal binding affinity. Another explanation is that TGF-␤ ligands tend to have a high affinity for either a type I or for a type II receptor, often leading to a strict sequential binding mechanism. A heterodimer, however, could combine a high affinity binding site for a type I and a type II receptor within the one dimeric ligand, thereby greatly enhancing the overall affin- FIGURE 11. Molecular modeling of mature cumulin. A, ribbon plot of the homology model of human mature cumulin showing the butterfly-shaped dimer architecture and with the monomer subunits colored in red and green. Structural features and the potential binding epitopes for the type I (wrist) and type II (knuckle) receptors are indicated. The six cysteine residues forming the characteristic cysteine knot motif are shown as sticks. B, as in A but rotated around the x axis by 90°. C, detail of the cumulin dimer interface showing that, despite lacking the intermolecular disulfide bond present in most other TGF-␤ ligands, similar polar and hydrophobic interactions stabilize dimer assembly. D, production of covalent cumulin appears possible, because in silico exchange of the two serine residues (Ser 418 in GDF9 and Ser 356 in BMP15) with cysteines locates both sulfur atoms sufficiently close to form a thioether bond without requiring conformational or large structural rearrangements of the dimer architecture.
ity to its cell surface receptors. Finally, because of its inherent asymmetry, a heterodimeric ligand is more prone to recruit two different type I receptors into the signaling complex. For example, the cumulin receptor complex (Fig. 7) is predicted to comprise two BMPRII receptors: one ALK4/5/7 receptor, which is necessary for the activation of the SMAD2/3 pathway, and one ALK6 receptor, which is needed for SMAD1/5/8 signaling, as originally hypothesized by McNatty et al. (1).
An open question prior to this study was whether cumulin could only form inside the oocyte or whether it might also spontaneously assemble from homodimeric GDF9 and BMP15 present in the extracellular space. The latter seemed feasible because GDF9 and BMP15 naturally form transient dimers, whose dimer architecture is not fixed by an intermolecular disulfide bond. Indeed, we observed that the two wild type homodimeric proteins, separately produced and purified, synergize via a process of heterodimer formation. High resolution structure studies of different BMPs provide a possible explanation for GDF9 and BMP15 disassembly and subsequent cumulin formation. Specifically, the structures show that the interface between the monomeric halves of TGF-␤ proteins is considerably hydrated, which is unusual compared with the water-free hydrophobic core of soluble proteins (48 -51). The localization of these conserved water molecules in the dimerization interface around the central cysteine knot likely limits the stability of the dimer; however, in the case of covalent BMPs, a full disassembly of the dimer is prevented because of the covalent linkage via the intermolecular disulfide. However, because GDF9 and BMP15 lack this bond, both factors potentially undergo full disassembly and reassembly, which, when both factors are simultaneously present (usual situation in most mammals), would yield an equilibrium-based mixture of homo-and heterodimers unless steric restraints specifically interfere with heterodimer formation. The latter seems highly unlikely, however, given the fact that the interface residues are highly conserved between GDF9 and BMP15. Together, this evidence suggests that the formation of cumulin in vivo might, at least in part, occur postsecretion from the oocyte in the extracellular matrix of the granulosa and cumulus cells. If so, it seems plausible that the nature and composition of the extracellular matrix, which in turn is affected by in vivo and in vitro context, is likely to affect cumulin assembly and bioactivity.
Two remarkable features of cumulin bioactivity is its capacity to activate both the SMAD2/3 and SMAD1/5/8 signaling pathways, as well as its enhanced signaling potency, in particular when compared with the GDF9 and BMP15 homodimers. On isolated mural GCs, mature covalent cumulin was even more potent than pro-cumulin, particularly in terms of GC proliferation, which is consistent with its enhanced SMAD2/3 signaling (29). Dual SMAD pathway activation by cumulin is in disagreement with Peng et al. (24), who claimed that the GDF9-BMP15 heterodimer binds to, but does not activate, the ALK6 receptor. The form of heterodimer that Peng et al. (24) produced has an epitope tag (either FLAG or myc) incorporated into the N terminus of the mature region. Although this appears to not impede BMP15 homodimer bioactivity (52), the effects of the GDF9 (myc) and BMP15 (FLAG) tags on het-erodimer activity are unknown. In contrast, covalent mature cumulin does not contain any tags, and pro-cumulin incorporates a tag only at the N terminus of the BMP15 prodomain. Hence, in all forms of GDF9, BMP15, or cumulin used in the current study, the amino acid sequence of the mature region is identical to the wild type sequence (apart from the Ser-Cys mutation to form covalent dimers). The lack of any modifications to our GDF9 and BMP15 mature regions may also explain why we are able to detect synergism between separately expressed and purified pro-GDF9 and pro-BMP15 (current study), whereas others do not (24,53).
An intriguing and important observation from the current study is that when cumulus-oocyte complexes were treated, only the pro-cumulin form improved oocyte quality, as assessed by subsequent embryo yield, whereas mature covalent cumulin did not, despite being exceptionally potent on mural GC. Clearly, pro-cumulin contains the prodomains of GDF9 and BMP15 as part of a complex, whereas mature cumulin is an isolated mature region heterodimer (Fig. 3). These results are entirely consistent with our recent observation that pro-BMP15, but not mature domain BMP15 nor mature domain mouse GDF9, enhances oocyte developmental competence (54,55). It is well established that prodomains of TGF-␤ superfamily growth factors have important roles in protein folding and regulation of bioactivity (56,57). In particular, interactions of the prodomains of cumulin with the specialized extracellular matrix of cumulus GC may facilitate presentation of the heterodimeric mature domain to its receptors. In support of these concepts, we have shown that pro-GDF9 binds strongly to heparin, suggesting that certain heparan sulfate proteoglycans may act as co-receptors for pro-GDF9, pro-BMP15, and pro-cumulin (58). Hence, the differential activities of pro-cumulin and mature cumulin on mural GC versus cumulus GC may be attributed to differential expression of heparan sulfate proteoglycans and/or type I or II receptors in these GC.
Finally, the current study also has significant practical implications for reproductive medicine and advanced breeding in domestic animals. The development of purified and highly potent pro-cumulin is likely to prove an important additive for oocyte IVM (59). IVM is an assisted reproductive technology that generates viable embryos using reduced or no exogenous ovarian hormone stimulation of the patient. IVM is thus an important adjunct technology to in vitro fertilization. To make IVM more clinically viable, including in fertility preservation in cancer survivors, an additive such as pro-cumulin is needed to improve the efficiency of IVM (60). We and others have shown that addition of "native" oocyte-secreted factors (e.g. pro-BMP15 or pro-GDF9) to IVM enhances oocyte quality, having profound effects on embryo development and fetal survival (13,54,55,(61)(62)(63)(64). This study is the first report on the effects of cumulin on oocyte quality. We carefully chose to use an established porcine model of low oocyte developmental competence using oocytes from growing antral follicles to better recapitulate the human scenario. The addition of pro-cumulin led to a major improvement in oocyte quality, more than doubling embryo yield, which, if translatable to human oocytes, would have a large impact on clinical IVM and oncofertility.