A variant of human growth differentiation factor-9 that improves oocyte developmental competence

Growth differentiation factor-9 (GDF9) and bone morphogenetic protein-15 (BMP15) are co-expressed exclusively in oocytes throughout most of folliculogenesis and play central roles in controlling ovarian physiology. Although both growth factors exist as homodimers, recent evidence indicates that GDF9 and BMP15 can also heterodimerize to form the potent growth factor cumulin. Within the cumulin complex, BMP15 “activates” latent GDF9, enabling potent signaling in granulosa cells via type I receptors (i.e. activin receptor-like kinase-4/5 (ALK4/5)) and SMAD2/3 transcription factors. In the cumulin heterodimer, two distinct type I receptor interfaces are formed compared with homodimeric GDF9 and BMP15. Previous studies have highlighted the potential of cumulin to improve treatment of female infertility, but, as a noncovalent heterodimer, cumulin is difficult to produce and purify without contaminating GDF9 and BMP15 homodimers. In this study we addressed this challenge by focusing on the cumulin interface formed by the helix of the GDF9 chain and the fingers of the BMP15 chain. We demonstrate that unique BMP15 finger residues at this site (Arg301, Gly304, His307, and Met369) enable potent activation of the SMAD2/3 pathway. Incorporating these BMP15 residues into latent GDF9 generated a highly potent growth factor, called hereafter Super-GDF9. Super-GDF9 was >1000-fold more potent than WT human GDF9 and 4-fold more potent than cumulin in SMAD2/3-responsive transcriptional assays in granulosa cells. Our demonstration that Super-GDF9 can effectively promote mouse cumulus cell expansion and improve oocyte quality in vitro represents a potential solution to the current challenges of producing and purifying intact cumulin.

During folliculogenesis, GDF9 and BMP15 signaling stimulates granulosa cell (GC) proliferation, whereas preventing follicle-stimulating hormone (FSH)-induced differentiation toward the mural GC phenotype (8,9). In antral follicles, GDF9 and BMP15 direct GCs immediately surrounding the oocyte toward the specialized cumulus cell phenotype, which, in turn, facilitates oocyte maturation and the acquisition of developmental competence (10). In mice, in vitro studies have suggested that the above functions are primarily dependent upon activation of the GDF9-Smad2/3 axis (11,12) and, in support, GDF9 knockout mice are infertile (13), whereas BMP15 knockout mice display only a mild reproductive phenotype (14). In monoovulatory species, however, both the GDF9-Smad2/3 axis and the BMP15-Smad1/5/8 axis appear to play prominent roles in GC function, with sheep homozygous for inactivating mutations in either GDF9 or BMP15 displaying infertility (15). Similarly in women, mutations in GDF9 and BMP15 have been identified in cases of primary ovarian insufficiency, polycystic ovary syndrome, and dizygotic twinning (16).
The species differences in GDF9 and BMP15 identified through genetic studies have been further elaborated by biochemical analysis of these unique TGF-␤ proteins (Table S2). For example, mouse GDF9 is produced in an active form, whereas human GDF9 is latent due to a high affinity interaction with its N-terminal prodomain, which blocks receptor binding (3,17). Furthermore, human BMP15 is produced in an active form, whereas ovine BMP15 is inactive, and mouse BMP15 is misfolded and typically not secreted from cells (18).
The discrepancy between the genetic requirement for both GDF9 and BMP15 in monoovulatory species and the latency of recombinant human and ovine GDF9 (as well as ovine BMP15) was recently resolved by two separate groups through the generation of a highly bioactive human GDF9:BMP15 heterodimer, named cumulin (19,20). Both groups showed that cumulin potently activates Smad2/3 signaling, stimulating cultured mouse GC to proliferate (19) and inducing genes required for cumulus cell expansion (20). Additionally, porcine cumulus-oocyte complexes (COCs) supplemented with recombinant human cumulin during in vitro maturation (IVM) better sup-This article contains supporting information. * For correspondence: Craig A. Harrison: craig.harrison@monash.edu.
cro ARTICLE ported embryo development to at least day 7, whereas human GDF9 and BMP15 homodimers provided no apparent benefit in this model (19). Together, these studies indicate that activation of latent human GDF9 is likely mediated through heterodimerization with BMP15, leading to potent Smad2/3 signaling (19,20).
Based on its in vitro activity, cumulin has great potential as an IVM supplement to improve assisted reproductive technology (ART) (19). However, the structure of cumulin as a noncovalent heterodimer creates challenges during production and purification. Specifically, co-transfecting mammalian cells with both GDF9 and BMP15 cDNA results in production of cumulin, together with GDF9 and BMP15 homodimers. This approach has typically resulted in a BMP15 homodimer production bias, relative to cumulin (19,20). To date, attempts to isolate cumulin free of contaminating homodimers has resulted in loss of activity on cumulus granulosa cells, likely due to prodomain separation during purification (19). To circumvent these issues, we sought to activate homodimeric human GDF9. Specific BMP15 residues predicted to activate GDF9 within the cumulin complex were substituted into the GDF9 cDNA. The resultant Super-GDF9 protein was expressed at high levels and stimulated Smad2/3 signaling with at least 4-fold greater potency than cumulin itself.

Smad signaling by human GDF9, BMP15, and cumulin
Following synthesis and proteolytic cleavage, GDF9, BMP15, and cumulin are secreted from cells noncovalently associated with their prodomains (Fig. 1A) (19). In this study, each of these proteins was produced in HEK-293T cells and Co-IMAC was used to purify the His-tagged pro-mature complexes. To assess activity, COV434 human GCs were transfected with Smad-responsive transcriptional reporters and treated overnight with increasing doses of GDF9, BMP15, or cumulin, all in their proforms. As expected, GDF9 induced no transcriptional activity at any of the doses tested ( Fig. 1, B and C). In contrast, treatment with BMP15 dose-dependently stimulated the Smad1/5/8-responsive transcriptional reporter, leading to high levels of luciferase activity (Fig. 1C). BMP15 also induced a low level of Smad2/3-responsive transcriptional activity (Fig. 1B). Consistent with previous studies, cumulin dose-dependently stimulated the Smad2/3-responsive reporter and the Smad1/5/8-responsive reporter, leading to high levels of luciferase activity in both assays (Fig. 1, B and C). To support these transcriptional assays, we assessed Smad1/5 and Smad2 phosphorylation in COV434 cells stimulated for 1 h with GDF9, BMP15, or cumulin, all in their pro-forms. GDF9 failed to activate either Smad pathway (Fig. 1D). However, both BMP15 and cumulin stimulated phosphorylation of Smad1/5 and Smad2 (albeit to different extents) (Fig. 1D).

How does BMP15 activate latent GDF9 in the cumulin heterodimer?
The luciferase and Smad phosphorylation assays indicated that, within the cumulin structure, BMP15 activates GDF9, as previously hypothesized (19). To provide molecular insights into this activation mechanism, we used homology modeling to obtain a structural model of cumulin. Similar to the models of mature BMP15 and GDF9 (Fig. 2, A and B), cumulin displays the typical butterfly-shaped architecture of TGF-␤ superfamily ligands (Fig. 2C). Based on the crystal structures of TGF-␤ superfamily ligands bound to their type I and/or type II receptors (21)(22)(23)(24), cumulin would interact with its type II receptor (BMPRII) in the "knuckle" region formed by its finger domains (Fig. 2C). Type I receptor (ALK4/5/6) binding, in contrast, would occur in the "wrist" regions of cumulin, which are formed by residues from both the BMP15 and GDF9 chains (Fig. 2C). As a heterodimer, two distinct type I receptor interfaces (wrist regions) are formed in cumulin, relative to homodimeric GDF9 and BMP15. In this study, we focused on the cumulin wrist epitope formed by the helix of the GDF9 chain (Fig. 2C, gold) and the fingers of the BMP15 chain (Fig. 2C, cyan). Unique BMP15 finger residues at this site likely enable high affinity ALK4/5 binding and activation of the Smad2/3 pathway, a process not achievable by the latent GDF9 homodimer. Sequence alignment of the finger domains of BMP15 and GDF9 identified 6 conservative and 10 nonconser- and C, dose-response curves in COV434 human GC transfected with (B) a Smad2/3-responsive luciferase reporter (A3-Luc; ligand concentrations 1.6 to 100 ng/ml), or (C) a Smad1/5/8-responsive luciferase reporter (BRE-Luc; ligand concentrations 0.3 to 20 ng/ml). Luciferase activity is presented as the mean Ϯ S.D. of at least triplicates from representative experiments, relative to an adjusted value of 1.0 for the mean of the control wells. Experiments were repeated Ͼ3 times. D, COV434 cells were treated for 1 h (50 ng/ml). Phosphorylated Smad proteins were detected using anti-phospho-Smad2 or antiphospho-Smad1/5. Anti-GAPDH, anti-Smad2, and anti-Smad5 were used as internal controls.

Generation and characterization of Super-GDF9
Using site-directed mutagenesis, we incorporated the identified BMP15 residues, in various combinations, into latent human GDF9. As it has previously been shown that substituting Gly 391 in the pre-helix loop of human GDF9 to arginine (present in both mouse GDF9 and human BMP15) reduces latency (17,25), we also included this change in our mutant panel ( Table 1). The BMP15 residues were incorporated into both chains of the GDF9 homodimer. HEK-293T cells were transiently transfected with expression vectors for the GDF9 variants and the conditioned medium was subjected to Western blotting. Using a GDF9-targeted antibody (mAb-53/1), the monomeric mature (20 kDa) and precursor (60 kDa) forms of GDF9 were detected for each mutant (Fig. 3A), confirming they were secreted normally. Therefore, we carried out large-scale production of each variant and used Co-IMAC to purify the N terminally tagged pro-GDF9 complexes.
To assess activity, COV434 granulosa cells, transfected with Smad-responsive luciferase reporters, were treated with the purified pro-GDF9 variants. The base mutation (G391R) marginally activated human GDF9 in the Smad2/3-responsive assay (Fig. 3, B and C). Incorporating various combinations of BMP15 residues from finger 1 into human GDF9 led to a progressive increase in reporter activity. Specifically, mutant 2 (S363R, N369H,G391R), mutant 1 (K366G,N369H,G391R), and mutant 3 (S363R,K366G,G391R) were 3-, 11-, and 14-fold more potent, respectively, than the G391R variant alone (Fig. 3B). The high activity of mutants 1 and 3, relative to mutant 2, identified the K366G mutation as a key determinant for enhanced type I receptor binding. Surprisingly, incorporating all four of these BMP15 residues into human GDF9 (mutant 4: S363R,K366G,N369H,G391R) had a synergistic effect, dramatically increasing both potency (50-fold) and the maximal showing the canonical dimer architecture. The two monomer subunits of the homodimer are colored in gold and orange. The putative binding epitopes for the type I receptors (wrist and fingers) are indicated. C, ribbon plot of human mature cumulin showing the canonical butterfly-shaped dimer architecture, and then rotated around the x axis by 90°. The monomer subunits of the heterodimer are colored in cyan (BMP15) and gold (GDF9). The putative binding epitopes for the type I (wrist) and type II (knuckle) receptors are indicated. Four BMP15 finger residues (Arg 301 , Gly 304 , His 307 , and Met 369 ), which likely enable high affinity ALK4/5 binding are shown. D, sequence alignment of the mature domains of human GDF9 and BMP15 (excluding the disordered N terminus). The residues are numbered according to the first residue of the signal peptide. Residues forming the fingers are indicated. Residues modified in this study are highlighted in cyan (fingers) and yellow (wrist). Identical (*) and conserved (dot and colon) amino acid residues between GDF9 and BMP15 are indicated.

Activation of human GDF9
response (Fig. 3C, note the difference in y axes scale between B and C). Finally, our modeling also identified a BMP15 residue from finger 2 (Met 369 ) that could potentially contribute to type I receptor binding. Incorporating this residue into mutant 4 generated a particularly potent GDF9 variant (Super-GDF9: S363R,K366G,N369H,G391R,T431M; Fig. 3D), which was Ͼ100-fold more potent than the G391R variant alone and induced a remarkable 40-fold increase in Smad2/3-responsive luciferase activity (Fig. 3C). Incorporating five BMP15 residues activated latent human GDF9, enabling potent stimulation of the Smad2/3 pathway.
Because BMP15 typically activates the alternative Smad1/5 pathway, we assessed whether the GDF9 variants might also activate these transcription factors. COV434 granulosa cells were transfected with a Smad1/5/8-responsive reporter and then treated with increasing doses of BMP15, GDF9 (mutant 4), or Super-GDF9. Only BMP15 induced a luciferase response in this Smad1/5/8 assay (Fig. 3E). Consistent with this, no Smad1/5 phosphorylation was detectable by Western blotting in cells treated with any of the GDF9 variants (Fig. 3F). Thus, Super-GDF9 and the other GDF9 variants only activated the Smad2/3 pathway. Finally, we examined how Super-GDF9 was A, BMP15 residues were substituted into GDF9 using in vitro site-directed mutagenesis. Conditioned media from HEK-293T cells transfected with WT or mutant constructs were analyzed by Western blotting using mAb-53/1, targeted to the GDF9 mature domain. B, C, and E, dose-response curves in COV434 human GC transfected with (B and C) a Smad2/3-responsive luciferase reporter (A3-Luc; ligand concentrations 1.6 to 100 ng/ml). Note the difference in y axes scale between B and C, or E, a Smad1/5/8-responsive luciferase reporter (BRE-Luc; ligand concentrations 0.3 to 20 ng/ml). Luciferase activity is presented as the mean Ϯ S.D. of at least triplicates from representative experiments, relative to an adjusted value of 1.0 for the mean of the control wells. Experiments were repeated Ͼ3 times. D, ribbon plot of mature Super-GDF9 showing the canonical butterfly-shaped dimer architecture, and then rotated around the x axis by 90°. The two monomer subunits of the homodimer are colored in gold and orange. The putative binding epitopes for the type I receptors (wrist) are indicated. Residues that were substituted to the corresponding residues in BMP15 are shown in blue (S363R,K366G,N369H,G391R,T431M). F, COV434 cells were treated for 1 h (50 ng/ml). Phosphorylated Smad proteins were detected using anti-phospho-Smad2 or anti-phospho-Smad1/5. Anti-GAPDH, anti-Smad2, and anti-Smad5 were used as internal controls.

Activation of human GDF9
able to induce such a potent transcriptional response, relative to GDF9 G391R (Fig. 3C), despite both proteins promoting Smad2 phosphorylation following 1 h of treatment (Fig. 3F). Time course analysis demonstrated the capacity of Super-GDF9 to sustain a high level of Smad2 phosphorylation for an extended period of time, relative to GDF9 G391R (Fig. S1).
Next, we assessed the capacity of exogenous type II receptor extracellular domains (ECD) to antagonize the activity of Super-GDF9 on COV434 granulosa cells transfected with a Smad2/3-responsive luciferase reporter (Fig. 4C). When used alone, Super-GDF9 stimulated high levels of luciferase activity. Co-treatment with BMPRII-ECD, fused to an IgG Fc domain for stability, antagonized Super-GDF9 activity in a dose-dependent manner. In contrast, addition of activin receptor type IIB (ActRIIB) ECD-Fc had no effect on Super-GDF9 activity (Fig.  4C). Thus, BMPRII is the type II receptor for Super-GDF9.

Super-GDF9 stimulates granulosa cell functions with greater potency than cumulin
To assess whether Super-GDF9 works as effectively as cumulin, we initially compared the purified proteins in the COV434 Smad2/3-responsive luciferase assay (Fig. 5A). In this assay, Super-GDF9 was ϳ4-fold more potent than cumulin and stimulated a higher maximum response. One of the physiological functions of oocyte-secreted factors (OSFs) on GCs is to drive development of the cumulus cell phenotype (33,34). During the peri-ovulatory period, cumulus cells undergo a highly coordinated process, known as cumulus expansion, where they produce a complex extracellular matrix (ECM) critical to successful ovulation and fertilization (35). Critical genes involved with ECM production and stability during cumulus expansion (Ptx3, Has2, and Ptgs2) are downstream targets of mouse GDF9, human BMP15, and cumulin (19,20). Therefore, we assessed whether Super-GDF9 could stimulate expression of Ptx3, Has2, and Ptgs2 in cultured mouse GC (Fig. 5, B-D). In agreement with the response in the luciferase assay (Fig. 5A), Super-GDF9 was ϳ5-fold more potent than cumulin in terms of inducing Ptx3 and Has2 expression (Fig. 5, B and C), whereas Ptgs2 expression was similarly regulated by both growth factors (Fig. 5D).

Super-GDF9 improves oocyte developmental competence
To characterize Super-GDF9 and cumulin in a model closer to the physiological role of OSFs, we assessed whether treatment of isolated mouse cumulus-oocyte complexes during IVM could enhance cumulus expansion. At 50 ng/ml, both Super-GDF9 and cumulin significantly improved cumulus expansion after 17 h of IVM (Fig. 5E). To investigate whether these effects on cumulus cells also translated to an improvement in oocyte developmental competence, the oocytes were subsequently fertilized and their capacity to support embryo development to the blastocyst stage was assessed (Fig. 5F). Relative to untreated controls, oocytes matured in the presence of 50 ng/ml of Super-GDF9 or cumulin showed significantly improved blastocyst development rates at 5 days post-fertilization.

Discussion
The major role of the oocyte-specific growth factors GDF9 and BMP15 is to regulate the growth and differentiation of GCs, which in turn supply the oocyte with the support necessary for future healthy embryo/fetal development (10, [36][37][38][39]. Human BMP15 binds to complexes of type I (ALK6) and type II (BMPRII) receptors on the surface of GCs (Fig.  6A) and leads to the activation of Smad1/5 transcription factors (7). In contrast, human GDF9 remains associated with

Activation of human GDF9
its prodomain in a latent complex (Fig. 6A) (3). Even following prodomain removal, mature hGDF9 has very low signaling capacity via Smad2/3 transcription factors (17). For many years the latency of human GDF9 was a major conundrum in the field, as many of the important actions of these growth factors on GCs are dependent upon Smad2/3 activation (11,12,20,40). Then, two groups discovered that GDF9 and BMP15 readily form a heterodimer, termed cumulin (19,20), which binds two distinct type I receptors (ALK4/5 and ALK6) and potently activates both Smad2/3 and Smad1/5 transcription pathways (Fig. 6A).
In this study, we sought to understand the molecular mechanisms underlying the potent Smad2/3-activating capacity of cumulin, relative to latent human GDF9. One possible mechanism could be that the conformation of cumulin differs from latent GDF9 homodimers in a way that allows the ␣-helix region of the GDF9 chain to bind ALK4/5 with much higher affinity. A second possible mechanism is that cumulin may be in a conformation that limits the inhibitory effect of the GDF9 prodomain, as has been discussed previously by Peng and colleagues (25). A third possible mechanism is that BMP15 finger residues directly bind ALK4/5 with high affinity. Although we cannot rule out a reduction in the inhibitory effect of the GDF9 prodomain as being a contributing factor toward the enhanced activity of cumulin, our development of Super-GDF9 suggests direct binding by BMP15 finger residues is likely to be the most crucial factor. As described previously (19), type I receptorbinding epitopes are distinct in cumulin, relative to GDF9 and BMP15 (type II receptor-binding sites, in contrast, are maintained across the three proteins). One of these unique epitopes, formed by the ␣-helix of the GDF9 chain and residues on the convex surface of the fingers of the BMP15 chain, was identified as the most likely ALK4/5 receptor-binding site. As the ␣-helix at this site is common to both cumulin and latent GDF9, we reasoned that enhanced signaling capacity via Smad2/3 must be conveyed by BMP15 finger residues. In particular, modeling identified three finger 1 residues (Arg 301 , Gly 304 , and His 307 ) and one finger 2 residue (Met 369 ) within BMP15 distinct from the corresponding residues in GDF9. Arg 301 , Gly 304 , and His 307 form a contiguous interface on the fingertip of the BMP15 chain, and the corresponding residues on BMP2 (22, 41) and GDF5 (21) have been implicated in type I receptor binding. In addition, residues in BMP2, GDF11, and GDF5, analogous to Met 369 in BMP15, interact with their respective type I receptors (21,22,42). Introducing these four BMP15 residues, together with Arg 329 (18), into latent GDF9 generated a novel, highly potent growth factor, which we termed Super-GDF9. Super-GDF9 was Ͼ1000-fold more potent than WT human GDF9 and Ͼ100-fold more active than GDF9 G391R , which was the most active human GDF9 homodimer previously described (17,25). Moreover, Super-GDF9 even displayed a greater maximal Smad2/3 response than cumulin, likely because the BMP15 residues were incorporated into both chains of the GDF9 homodimer, forming two high affinity ALK5-binding sites. Our results support the idea that in vivo activation of latent hGDF9 occurs via heterodimerization with BMP15 (19), but also indicate that hGDF9 can be activated in vitro by the introduction of specific BMP15 residues.
In response to the luteinizing hormone surge, cumulus GCs immediately surrounding the oocyte expand and produce a thick extracellular matrix, which is critical for ovulation, fertilization, and ensuing embryo development (10, 35). Peng and colleagues (20) showed that cumulin, via Smad2/3, potently induced the expression of "expansion" genes (Ptx3, Has2, and Ptgs2) in GCs, as well as the full process of expansion in mouse oocytectomized cumulus cell complexes. Subsequently, we demonstrated that porcine cumulus-oocyte complexes supple-

Activation of human GDF9
mented with recombinant human cumulin during in vitro maturation are better able to support embryo development to at least day 7 (19). These studies highlighted the significant potential of cumulin to promote oocyte developmental competence in vitro and, thereby, impact human clinical IVM and advanced breeding in domestic animals. However, as a noncovalent heterodimer, cumulin is difficult to produce and purify free of contaminating GDF9 and BMP15 homodimers. Our development of Super-GDF9, a homodimer with potent Smad2/3-activating capacity, represents a potential solution to the problems surrounding cumulin production and purification.
In support, we showed that Super-GDF9 was 4-to 5-fold more potent than cumulin in activating the Smad2/3 pathway and stimulating the expression of the cumulus expansion genes Ptx3 and Has2 (Ptgs2 expression was similarly regulated by both growth factors). Of importance for human ART, cumulus cells from oocytes that develop into better quality embryos display stronger expression of HAS2 and PTGS2 (43)(44)(45). Importantly, Super-GDF9 was as effective as cumulin at promoting cumulus cell expansion and improving mouse oocyte develop-mental competence. These results provide support for the concept that Smad2/3 activation, without Smad1/5 activation, is sufficient to drive the specialized cumulus cell functions that facilitate mouse oocyte maturation and the acquisition of developmental competence (20). Thus, Super-GDF9 may be an effective substitute for cumulin to improve the efficiency of IVM and make this procedure more clinically viable.
Although the in vitro results for Super-GDF9 are highly encouraging, this protein differs from cumulin in one key aspect: it cannot activate the Smad1/5 pathway. This distinction may be important, because Smad1/5 activation during bovine IVM may contribute to oocyte maturation and subsequent embryo development (37,38). Critically, with the generation of Super-GDF9, we now have the tools to address this question. Indeed, future experiments will compare the effects on oocyte maturation and embryo development of Super-GDF9 alone, or in combination with BMP15.
In summary, the engineering of Super-GDF9 has uncovered the underlying biochemical basis for cumulin's substantial bioactivity. To date, Super-GDF9 has been shown to replicate the Figure 6. Model of human BMP15, GDF9, cumulin, and Super-GDF9 formation and signaling. A, BMP15 and GDF9 are co-expressed in mammalian oocytes throughout most of folliculogenesis. During synthesis, their prodomains direct the folding and dimerization of their mature domains. Dimeric precursors are cleaved by furin-like proteases, and then BMP15, GDF9, and potentially cumulin are secreted from the oocyte noncovalently associated with their prodomains. Following prodomain displacement in the extracellular matrix of nearby granulosa cells, BMP15 binds two BMPRII receptors and two ALK6 receptors to signal predominantly via pSmad1/5, but also weakly via pSmad2/3. Cumulin forms a receptor complex consisting of two BMPRII receptors, one ALK6 receptor, and one ALK4/5 receptor to signal via both Smad pathways, with pSmad2/3 signaling essential for many of its principal functions. Human GDF9 is naturally latent due to a high affinity interaction with its prodomain that prevents receptor binding. B, Super-GDF9 can be expressed in transfected HEK293 cells. During synthesis, its prodomain directs the folding and dimerization of the mature domains. Dimeric precursors are likely cleaved by furin-like proteases and it is then secreted noncovalently associated with its prodomain. Following purification, when pro-Super-GDF9 is added to the media of cultured granulosa cells the prodomain is hypothetically displaced in the extracellular matrix. Super-GDF9 then binds two BMPRII receptors and two ALK5 receptors to potently signal via pSmad2/3.

Activation of human GDF9
biological functions of cumulin and, thus, this protein may provide an expeditious approach toward generating a potent recombinant OSF of clinical-grade purity for use as an IVM supplement.

Production of GDF9/BMP15 variants
Construction of the human (h) GDF9 and BMP15 expression vectors has been described previously (19). The hGDF9 cDNA was amplified by PCR and subcloned into the mammalian expression vector pCDNA3.1(ϩ) (Thermo Fisher Scientific, Waltham, MA, United States) between the restriction sites NheI and EcoRI. For activation of hGDF9, multiple point mutations were incorporated into the cDNA using mutagenic primers (Table S1) with the QuikChange Lightning Multi Site-directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA, United States).
For production of recombinant human GDF9, BMP15, and cumulin, HEK-293T cells were plated at 8 ϫ 10 5 cells/well in 6-well-plates in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal calf serum (FCS) and incubated at 37°C in 5% CO 2 . After overnight incubation, plasmid DNA (2.5 g/construct) was combined with polyethylenimine-MAX (Polysciences, Warrington, PA, United States) and after 10 min DNA-polyethylenimine complexes were added directly to cells and incubated in Opti-MEM (Life Technologies, United States) medium for 4 h before replacing with fresh Opti-MEM for 16 h. The medium was then replaced with production media (DMEM:F-12 medium containing GlutaMAX TM (Life Technologies, United States), 0.02% bovine serum albumin (BSA), 0.005% heparin (Sigma, United States)) and incubated a further 72 h before collection.

Purification of GDF9/BMP15 variants
Conditioned media from HEK-293T cells transfected with cDNAs for GDF9 variants (ϳ200 ml) was concentrated (Centricon Plus-70, 5 kDa molecular mass cut-off; Millipore, Billerica, MA, United States) and resuspended in binding buffer (50 mM phosphate buffer, 300 mM NaCl, pH 7.4) to a final volume of 5 ml. The concentrated media was subjected to purification by Co-IMAC as previously described (47). Bound proteins were eluted from the resin with elution buffer (50 mM phosphate buffer, 300 mM NaCl, 500 mM imidazole) and imidazole was removed from the preparation by dialysis against binding buffer using a Slide-A-Lyzer MINI Dialysis Device (2 ml 3.5 K molecular mass cut-off; Thermo Fisher Scientific, United States). Mass estimates for GDF9, BMP15, and cumulin were determined by Western blotting using recombinant hGDF9 or hBMP15 mature proteins (R&D Systems, Minneapolis, MN, United States) as standards. Densitometry was assessed with Image Lab TM software (Bio-Rad).

In vitro COV434 granulosa cell bioassays
COV434 human granulosa tumor cells (7.5 ϫ 10 4 cells/well) were plated in DMEM, 10% FCS into 48-well-plates at 37°C in 5% CO 2 . The following day, cells were transfected using Lipofectamine 3000 (Life Technologies) with 250 ng/well of plasmid DNA. For the Smad2/3-responsive A3-luciferase assay, this consisted of 50 ng/well of pA3-Lux, 100 ng/well of pFAST2, and 100 ng/well of pCDNA3.1. For the Smad1/5/8-responsive BREluciferase assay, 250 ng/well of pBRE-Luc was transfected. The following day, cells were treated with increasing doses of GDF9/ BMP15 variants diluted in low-serum media (DMEM with 50 mM HEPES and 0.2% FCS) and incubated overnight at 37°C in 5% CO 2 . The media was then removed and cells were lysed in solubilization buffer (26 mM glycylglycine (pH 7.8), 16 mM MgSO 4 , 4 mM EGTA, 900 M DTT, 1% Triton X-100). The lysate was transferred to a white 96-well-plate and luminescence was measured immediately after the addition of the substrate D-luciferin (Life Technologies) (48). To determine which type II receptor Super-GDF9 utilized, Fc-fused extracellular domains of BMPRII and ActRIIB (R&D Systems) were used to antagonize Super-GDF9 activity.

Molecular modeling
The GDF9, BMP15, and cumulin structural models were created using the Swiss model comparative protein modeling server (49,50) and the crystal structure of GDF5 (Protein Data Bank entry 1WAQ) as a template (51). Subsequently the UCSF Chimera package from the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (52), was used to overlay the GDF9, BMP15, and cumulin dimers onto the GDF5 template and to produce the molecular graphics images.

Activation of human GDF9
Type I receptor utilization assays R1B L17 mink lung epithelial cells (6 ϫ 10 4 cells/well) were plated in DMEM, 10% FCS into 48-well-plates at 37°C in 5% CO 2 . The following day, cells were transfected using Lipofectamine 3000 with 250 ng/well of plasmid DNA, consisting of: 50 ng/well of pA3-Lux; 100 ng/well of pFAST2; 0 to 10 ng/well of ALK4 expression vector, or 0 to 10 ng/well of ALK5 expression vector; and 90 to 100 ng/well of empty pCDNA3.1 vector to normalize the total quantity of DNA transfected. After 24 h, transfected cells were treated with Super-GDF9, activin A, or myostatin (both from R&D Systems) diluted in low-serum media, and incubated overnight at 37°C in 5% CO 2 . The media was then removed and cells were lysed in solubilization buffer. The lysate was transferred to a white 96-well-plate and luminescence was measured immediately after the addition of the substrate D-luciferin.

Isolation and culture of primary mouse GC
Young female mice (129/T2/Sv, 21-28 days old) were primed with 5 IU pregnant mare's serum gonadotropin (PMSG) by intraperitoneal injection and ovaries were collected 46 h later. 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 Monash University Animal Ethics Committee. Antral follicles were punctured with a 27-gauge needle releasing GC into collection media (Medium-199 containing GlutaMAX TM (Life Technologies) ϩ 3 mg/ml of BSA (Sigma-Aldrich) ϩ 1% antibiotic-antimycotic (Life Technologies) ϩ 25 mM HEPES). The GC pool was subsequently filtered using a 40-m nylon cell strainer (Falcon, Corning Inc., Corning, NY, United States) to remove oocytes. GC were plated at 10 6 cells/well in 24-well plates in media supplemented with 10% FCS. After overnight culture at 37°C with 5% CO 2 , media was removed and GC was treated with Super-GDF9 or cumulin (diluted in fresh serum-free media) for 6 h at 37°C in 5% CO 2 . Media was then removed and the plate was snap frozen and stored at Ϫ80°C prior to RNA extraction.

Reverse-transcription PCR (RT-PCR)
RNA was extracted from GC using a GenElute TM Total RNA Purification Kit (Sigma-Aldrich) and subsequently subjected to DNase treatment with a TURBO DNA-free TM Kit (Thermo Fisher Scientific). Total RNA (500 ng) was then reverse transcribed to cDNA in a 20-l reaction using random hexamers with the SuperScript III First-Strand Synthesis System for RT-PCR (Life Technologies) on a T100 TM Thermal Cycler (Bio-Rad). In 10 l of quantitative PCR (qPCR) reactions, 1 l of cDNA was used as template, carried out on a Mastercycler epgradient realplex 4 S (Eppendorf, Hamburg, Germany) using TaqMan TM Fast Universal PCR Master Mix (2 times), no AmpErase TM UNG (Thermo Fisher Scientific) and TaqMan TM Gene Expression assays (Ptx3 Mm00477267_g1; Has2 Mm00515089_m1; Ptgs2 Mm00478374_m1). Each qPCR was carried out in duplicate and normalized to the housekeeping gene Gapdh (Mm99999915_g1). Relative mRNA levels were determined using the 2 Ϫ⌬⌬CT method (53, 54).

Oocyte quality assay
IVM and embryo cultures were performed as previously described (55) with minor modifications. Briefly, 28-day-old C57/Bl6 female mice were given 5 IU PMSG intraperitoneal injection. Mice were killed by cervical dislocation 46 h post-PMSG and ovaries were harvested. 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. Large antral follicles were punctured in HEPES-buffered ␣MEM medium containing 3 mg/ml of BSA (CellMaxx, MP Biomedicals, New Zealand) and 100 M 3-isobutyl-1-methylxanthine to liberate COCs. COCs were washed with culture medium (bicarbonate-buffered ␣MEM containing 3 mg/ml of BSA, 1 mg/ml of fetuin (Sigma-Aldrich), and 50 ng/ml each of recombinant mouse amphiregulin and epiregulin (R&D Systems)) and matured in this culture medium Ϯ 50 ng/ml Super-GDF9 or cumulin for 17 h at 37°C with 5% CO 2 in air. Following 17 h of IVM, cumulus expansion was scored blinded using criteria described by Vanderhyden and colleagues (56).
For embryo production, all media were purchased from IVF Vet Solutions (Adelaide, Australia). Following IVM as described above, COCs were washed (Research Wash, supplemented with 4 mg/ml of BSA) and co-incubated with capacitated sperm for 3.5 h at 37°C with 5% O 2 , 6% CO 2 , and 89% N 2 in fertilization medium (Research Fert) supplemented with 4 mg/ml of BSA. Sperm were recovered from hybrid CBB6F1 male mice Ͼ12 weeks of age. Presumptive zygotes were washed three times in wash medium and then once in cleavage medium (Research Cleave) supplemented with 4 mg/ml of BSA and cultured in cleavage medium drops overlayered with mineral oil at a density of one embryo per 2 l of medium. Embryo development was assessed every 24 h over 6 days.

Statistical analysis
For gene expression assays, data were assessed with ordinary one-way analysis of variance then Tukey's multiple comparisons test. Cumulus expansion scores were assessed with a Kruskal-Wallis test followed by Dunn's multiple comparisons test. Blastocyst development data were arcsine transformed and then assessed using one-way analysis of variance followed by Tukey's multiple comparison's test. p Ͻ 0.05 was considered statistically significant.