Evidence for Follicle-stimulating Hormone Receptor as a Functional Trimer

Background: A carbohydrate of follicle-stimulating hormone (FSH) has been proposed to sterically block other FSH molecules from binding to the putative receptor (FSHR) trimer. Results: FSH increases its receptor binding by 3-fold when the steric hindrance is removed. Conclusion: FSHR forms a functional trimer. Significance: This knowledge may improve designs of therapeutic drugs targeting FSHR.

tion were described previously (11). In brief, the coding sequences of FSH and the full ectodomain of human FSHR ED (Ser 16 -Arg 366 ) were subcloned into pVLAD6. Initial virus stock was produced by co-infecting Sf9 cells with the constructs and baculovirus DNA. The two viruses were further amplified to co-infect GnTI-HEK293 cells. The recombinantly expressed proteins were captured from the conditioned media and purified with affinity and size exclusion columns. The mutant proteins (␣N52D and ␤T60E) were prepared using the same protocols as the fully glycosylated FSH.
X-ray Crystallography-The FSH-FSHR ED complex was concentrated to 10 mg/ml using a disposable ultrafiltration device for crystallization at 20°C. The P3 1 crystals were grown from hanging drops mixed 1:1 with a reservoir solution of 0.1 M imidazole, pH 8.0, and 20% Jeffamine M-600. Crystals were cryo-protected with 15% (v/v) ethylene glycol. Diffraction data were collected at the 21-ID-D beam line of the Advanced Photon Source at wavelength of 0.979 Å and processed using the HKL3000 suite (19). The structure was determined by molecular replacement using the FSH-FSHR ED complex (PDB code 4AY9) as the search model (20). Reiterated cycles of model building and refinement were carried out using REFMAC and BUSTER with TLS parameterization (21,22). The data collection and refinement statistics are shown in Table 1. Structure figures were made using PyMOL.
CHO-hFSHR Membrane Preparation-CHO-DUKX cells expressing the human FSH receptor were disrupted by nitrogen cavitation in a buffer containing 250 mM sucrose, 25 mM Tris, pH 7.4, 10 mM MgCl 2 , 1 mM EDTA, and protease inhibitors (Sigma). The cells were pressurized with 900 p.s.i. of N 2 gas for 20 min, after which the lysate was centrifuged at 1,000 ϫ g for 10 min at 4°C. The supernatant was then collected and centrifuged at 100,000 ϫ g for 1 h at 4°C. The resulting pellet was re-suspended in binding buffer (10 mM Tris, pH 7.4, 5 mM MgCl 2 ) with a Dounce homogenizer. The protein concentra-tion of the samples was determined using the Bio-Rad protein assay reagent.
FSH Binding to CHO-hFSHR Membranes-Radioligand binding assays were performed in 100 l of 10 mM Tris, pH 7.4, 5 mM MgCl 2 , 0.2% BSA (assay buffer) in 96-well plates (Costar 3365). For the experiments shown in Fig. 1, a fixed amount of 5 g of CHO-FSHR membrane was mixed with increasing concentrations of glycosylated 125 I-FSH or 125 I-N52D-FSH (PerkinElmer Life Sciences). For the experiments shown in Fig.  2, Compound 5 was also added to the membrane at the indicated concentrations. Nonspecific binding was determined in the presence of a 500-fold excess of FSH at each concentration of 125 I-FSH. The reactions were incubated for 90 min at 37°C, with shaking, and terminated by filtering through a low protein binding Durapore filter (Millipore Multiscreen), which had been preincubated in assay buffer. The filters were washed 4 times with ice-cold binding buffer (10 mM Tris, pH 7.4, 5 mM MgCl 2 ) and counted on a ␥ counter. Data were analyzed using the GraphPad Prism software.
FSHR ␤-Arrestin Recruitment Assay-Determination of activated FSHR was performed by measuring ␤-arrestin recruitment according to PathHunter FSHR ␤-arrestin assay protocol (DiscoveRx, product code 93-0517C2) as described previously (11). For the assay shown in Fig. 3, cells were incubated in the presence or absence of FSH and Compound 5 at various concentrations mixed with FSH at an EC 100 concentration of 120 pM, or FSH at EC 20 , EC 50 , EC 70 , and EC 100 concentrations mixed with 1 mM Compound 5, respectively. For the assay shown in Fig. 4, cells were incubated with FSH or ␤T60E-mutant at various concentrations.
Primary Granulosa Isolation and Determination of Estradiol Production-Primary granulosa cells from immature rats were used to determine the ability of FSH to induce estradiol secretion, as described previously (11). Briefly, 21-day-old female CD rats were implanted with diethylstilbestrol pellets and euthanized after 72 h for isolation of granulosa cells from the ovaries. The isolated granulosa cells were cultured overnight at 37°C and subsequently treated with serially diluted FSH for 24 h to determine estradiol production.
Molecular Modeling-Homology modeling was performed using the software MOE from the Chemical Computing Group. Structures were analyzed using CCP4 suite software (20), and the conformations were further analyzed against known amino acid conformational tendencies (23). The glycan model in Fig.  1A (left panel) was constructed by directly linking the bi-antennary glycan in human fibrinogen (PDB entry 3GHG) to an FSH Asn␣ 52 residue. The glycan branches were extended to the central cavity of the FSH-FSHR ED complex (PDB entry 4AY9). The other two FSH molecules in the complex were removed due to atomic conflicts. The FSHR 7-TM models were described previously (18). The TSHR trimer model in Fig. 6 was constructed by replacing the FSHR N-terminal residues of PDB entry 4AY9 with the N-terminal structure of TSHR (PDB entry 2XWT) and homology modeling for the TSHR "hinge domain" based on the FSHR crystal structure. The positions of the M22 agonist antibody and K1-70 antagonist antibody were located by superimposing the truncated TSHR in their respective receptor com-

FSH-FSH ED crystal structure
This crystal was grown from a Jeffamine M-600 crystallization solution, different from the PEG4000 buffer used to grow the P1 crystal (11). The Ramachandran allowed region was analyzed by the Molprobilty software (47). Values in parentheses are those for the highest resolution shell.

Data collection
Space

RESULTS AND DISCUSSION
FSHR Binds Three Times the Amount of Asn␣ 52 -Mutant-FSH Than Fully Glycosylated FSH-FSH is a glycosylated heterodimer, with four N-linked glycosylation sites, two at Asn 52 and Asn 78 of the ␣-subunit and the other two at Asn 7 and Asn 24 of the ␤-subunit. In our trimeric activation model, a constitutive FSHR trimer is only able to bind one fully glycosylated FSH molecule, in contrast to three deglycosylated FSH molecules binding to one FSHR trimer (Fig. 1A) (18). This is because the occupied full-length glycan at Asn␣ 52 -FSH at the central cavity of trimeric FSHR sterically prevents the binding of additional fully glycosylated FSH molecules. Accordingly, our model predicts that the mutant FSH, lacking the glycan at Asn␣ 52 , would bind to the cell-surface FSHR by a factor of 3 times that of fully glycosylated FSH (18). To test this hypothesis, we mutated FSH residue Asn␣ 52 to aspartate (N52D) so that the site would no longer be glycosylated. We then tested the receptor binding capacity of the mutant and compared it to that of glycosylated FSH. The left panel of Fig. 1B depicts representative data measuring receptor binding of the mutant-N52D and fully glycosylated FSH. The experiment was repeated three times and the binding ratio of the mutant versus the fully glycosylated FSH is shown in right panel of Fig. 1B for all four experiments. The binding ratio fluctuated around 3:1 across a broad range of FSH concentrations, consistent with the hypothesis.
We have considered the difference of binding affinities as an alternative explanation. The explanation was ruled out for three reasons. First, the crystal structures (11,24) have shown that the glycosylation sites are not in contact with the receptorbinding surface. Second, FSH binds to FSHR at subnanomole affinity. At this high affinity, the ratio of receptor binding between the two forms of FSH would not be higher than 1.5fold within the tested concentration range, as shown in the calculations in Table 2; therefore, a 3-fold FSH binding is not possible without an increase of binding sites. Finally, in the case of a higher affinity for N52D FSH, it would be expected that lower concentrations would be required to reach receptor binding saturation. However, at the respective saturating doses, the total number of bound fully glycosylated and N52D FSH molecules should be the same, which is not the case.
We noticed that the ratio of mutant FSH to fully glycosylated FSH appears to drop slightly as the FSH concentration increases. This drop approached but did not reach statistical significance when the data were tested for a statistical significance of a non-zero slope straight line (slope: Ϫ0.29; intercept: 2.9; p value: 0.06). However, such a drop could be consistent with the fact that fully glycosylated FSHs are heterogeneous regarding the lengths and conformations of carbohydrates, such that at higher FSH concentrations, two smaller glycans with suboptimal conformations may fit into the central cavity of the FSHR trimer. Such a mechanism could explain the negative cooperativity observed for glycoprotein hormones binding to their receptors (25).
An Allosteric Modulator Increases FSH Binding 3-Fold-LMW modulators have been observed to increase FSH binding to cell-surface receptors from their normal level by ϳ3-fold (15)(16)(17)(18). These observations led us to propose that a LMW modulator binds to the FSHR 7-TM domain and induces a conformational change of the receptor. A dramatic conformational change, such as the 14-Å dislocation for the helix TM6 in ␤2-adrenergic receptor (␤2AR) (4), may disrupt the trimeric configuration, resulting in each of the dissociated monomers to bind one FSH molecule. To test this hypothesis, we utilized the LMW FSHR modulator, designated Compound 5 (3-((2S,5R)-5-(2-((3-ethoxy-4-methoxyphenethyl)amino)-2-oxoethyl)-4 -oxo-2-(4-(phenylethynyl)phenyl)thiazolidin-3-yl)benzamide), which has been demonstrated to bind to an allosteric site in the FSHR transmembrane domain (26). We performed 125 I-FSH binding assays in the presence and absence of Compound 5, by measuring the specific binding of 125 I-labeled human FSH to the human FSHR. Fig. 2A shows the binding data in both the presence and absence of the LMW compound. The maximal binding (B max ) reached 20,860 disintegrations per minute (dpm) in the presence of Compound 5 (at 10 M), as compared with 7,723 dpm in the absence of Compound 5 ( Fig. 2A, right  panel). To reach the ideal state of total separation of trimer, the ratio has to be extrapolated to a maximum concentration of the LMW modulator. The saturated ratio of 2.8 is consistent with the theoretical limit of 3 when every FSHR trimer is fully separated into three FSHR monomers (Fig. 2B). Again, we considered the alternative explanation. The 3-fold increase of FSH binding is not due to an increase of ligand affinity. As the calculations in Table 3 show, FSH binding in the presence of the LMW modulator would not be higher than that in the absence of the modulator without an increase of binding sites. Moreover, the number of FSH molecules bound at approaching saturation concentrations in the absence of the LMW modulator is less than in its presence, consistent with a difference in binding site number rather than in affinity. Furthermore, our model predicts that the LMW modulator should have little effect on the deglycosylated FSH binding to    (18), because all of the binding sites would already be fully occupied. Fig. 2C shows the experimental results that confirm the prediction for Compound 5.

An Allosteric Modulator Increases FSHR Intracellular Signaling Levels by 3-Fold-
To determine whether the increased FSH binding caused by the modulator can increase the level of the associated activation proteins to the receptor, we assessed the level of intracellular signaling immediately following receptor activation. Although several assays are available to measure GCPR activation, G-protein-mediated assays, including ones for cAMP production and [ 35 S]GTP␥S binding, can lead to an overestimation of the potency and efficacy of compounds in recombinant, overexpressing systems, where different LMW modulators may produce the same maximal response (27). In contrast, the ␤-arrestin assay can measure GPCR activity with a linear relationship to ␤-arrestin occupancy (27). In the present study, a ␤-arrestin recruitment assay was used to assess FSHR activation following stimulation with FSH alone or in combination with Compound 5.
To facilitate the data interpretation, we assessed the theoretical ratio of the FSHR trimer in complex with ␤-arrestin. Although the ␤2AR-G s complex structure is known (4), no crystal structure of any GPCR in complex with ␤-arrestin is available. Fortunately, the crystal structures of active and inactive ␤-arrestin are available and the major interaction site of ␤-arrestin with GPCR 7-TM domains is known (28 -30). The ratio of the long dimension of a rectangular prism-like shaped ␤-arrestin to diameter of a 7-TM is ϳ2:1, and the 7-TM binds approximately to the center of ␤-arrestin along the long dimension. This mode of interaction would prevent ␤-arrestin from binding more than one molecule to the 7-TM trimer, assuming the FSHR trimer possesses a 3-fold or pseudo-3-fold symmetry (Fig. 3A). Therefore, we hypothesized that one intact FSHR trimer can only accommodate one ␤-arrestin. Once the receptor trimer is dissociated, each activated receptor would then be able to bind one ␤-arrestin molecule. This prediction was tested using Compound 5 to modulate the receptor. As shown in Fig.  3B, addition of Compound 5 to a maximally stimulating concentration of FSH (normalized to 100%) resulted in further activation, approaching a plateau of ϳ280% of FSH alone.
Compound 5 alone can activate FSHR and recruit ␤-arrestin to a greater extent than FSH alone (Fig. 3B, left panel). Addition of glycosylated FSH to a high concentration of Compound 5  resulted in recruitment of ␤-arrestin at levels approximately to the same 280% of FSH alone (Fig. 3C, right panel). Crystallographic and Mutagenesis Studies Are Consistent with the Trimer Model-The trimer structure was crystallized from a PEG solution and determined in the P1 space group (11). To test whether trimer configuration depended on the particular crystallization condition, we crystallized and determined the complex structure in the P3 1 space group from a Jeffamine M-600 crystallization solution. As shown in Fig. 4A, the trimeric arrangements are almost identical in both structures, supporting the proposed FSHR constitutive trimer.
Mutagenesis studies also support the FSHR trimer model. As shown in the inset of Fig. 4B, the FSH residue Thr␤ 60 , at the potential exosite of FSH, does not interact with its primary binding monomer but potentially interact with the neighboring monomer. Thr␤ 60 makes hydrophobic contacts with Val 85 and Tyr 110 of the neighboring FSHR. A ␤T60E mutation would then disrupt the hydrophobic interface and create charge-charge repulsion against the neighboring FSHR residue Glu 87 (Fig. 4B,  inset). The additional disruption of the trimeric interface in the extracellular domains might create enough room for a second FSH to bind, resulting in an enhancement of FSHR signaling. To test this hypothesis, we made the ␤T60E mutant. The FSH ␤T60E mutant indeed enhanced signaling in both the ␤-arres-tin and estradiol production assays, as measured by the maximum percent of receptor response (Fig. 4C).
Proposed Activation Mechanism of the FSHR Trimer-Based on these data, we now further extend our previously proposed activation model for the FSHR trimer (18). For the trimeric receptor to be activated from its extracellular domains, these domains must undergo rearrangement and at least one of the "hinge" hairpin loops has to be shifted. As shown in Fig. 5 (middle), FSH normally activates FSHR asymmetrically with the whole trimer acting as a single monomer. Addition of LMW modulators results in the separation of the trimer into monomers. Each separated monomer is then fully functional. Alternatively, FSHR may take a different pathway by activating the 7-TM domains directly with the binding of LMW modulators alone (Fig. 5, left route). Depending on the strength of the LMW compound to dissociate the trimer, the number of ␤-arrestins recruited to one trimeric receptor would vary. As part of an internal screening program targeting FSHR, we have subjected hundreds of LMW hits in ␤-arrestin recruitment assays, but they alone did not achieve greater than 2-fold of ␤-arrestin recruitment over the FSH control. This is consistent with the model that insertion of the full-length Asn␣ 52 glycan into the central cavity is required for the complete separation of the trimer. Additionally, three deglycosylated FSH molecules can bind to the trimeric receptor (Fig. 5, right).
The trimer model may explain the delayed and lower receptor binding of fully glycosylated FSH than the hypoglycosylated FSH, as demonstrated recently (31). In comparison to the fully glycosylated FSH, the hypoglycosylated FSH had a 2-fold increase of receptor binding. The two forms also differ in their receptor binding behavior. The fully glycosylated FSH displayed a sigmoid binding curve, with a slow start in the first 50 min, followed by a rapid period of 2 h before it reached the saturation level. In contrast, hypoglycosylated FSH showed a hyperbolic curve, with almost no signs of delay. Because deglycosylation does not significantly change the amino acid structure of glycoprotein hormones (18,(32)(33)(34)(35), the glycans must be attributed to the hampering kinetic effect of receptor binding. It is unclear, however, which of the four glycans plays the most important role, due to the fact that both ␣ and ␤ chains of their hypoglycosylated FSH heterodimer are less glycosylated. The Asn␣ 52 -glycan of their hypoglycosylated FSH adopts a compact helical shape, due to its higher mannose content than that in the fully glycosylated FSH (31). The smaller and compact glycan at Asn␣ 52 in the hypoglycosylated FSH would readily fit into the central cavity of the FSHR trimer, rendering a smooth, hyperbolic binding curve. In contrast, the more bulky and extended glycan of the fully glycosylated FSH would require more time to fit into the central cavity, resulting in a delayed, sigmoid binding curve. Finally, the central cavity can also accommodate more of the compact glycan, allowing more hypoglycosylated FSH binding to the receptor trimer.
The trimer model is also consistent with the observation that ligand binding is increased for TSHR or FSHR with the constitutively active mutation D6.30G (i.e. D619G and D567G, respectively). Several constitutively active mutations in the TSHR 7-TM domain cause increases in TSH binding (36). Among these mutants, the D6.30G mutation is most interesting. This negatively charged residue is well conserved in GPCR family members. The equivalent residue (E6.30, i.e. Glu 268 ) in ␤2AR plays a central role in receptor activation (37), and moves dramatically during the activation (14 Å outward from the inactive state) without causing significant conformational changes for the residues in the top half (i.e. toward the extracellular side) of the 7-TM domain (4,38). When normalized for receptor number expressed on live cells, a D6.30G mutation of TSHR and FSHR resulted in a 3-fold increase in TSH and FSH binding, respectively (36). Although without further investigation a change in ligand affinity caused by this mutation cannot be formally ruled out, the low likelihood of this mutated residue causing a change in the binding affinity of the anti-TSHR or FSHR antibody used to normalize receptor number, together with the fact that the mutation occurs in a site with the potential to destabilize receptor oligomers, are consistent with our FSH: FSHR binding model.
Remaining Open Questions-The proposed mechanism postulates the existence of FSHR as a functional trimer in the native state, which has not yet been demonstrated by direct evidence. Direct evidence might come from the crystal structure of a fulllength FSHR in the ground state, an electron or atomic force microscope image of FSHR on a membrane surface, or a superresolution single-molecule optical image on a live cell. All of these approaches would require specialized capabilities. Nevertheless, the observed electrophoresis band of molecular mass ϳ240 kDa in harsh SDS-containing solutions (12) does support the existence of strong FSHR trimers in the ground state. It is not unprecedented for a membrane protein to exist exclusively as a non-covalently linked oligomer in both the native functional form as well as in the presence of SDS, as shown for SKC1 (39,40).
The proposed model also does not address the mechanism of how the binding of LMW modulators causes the conformational change of the 7-TM domains that leads to subsequent separation to monomers. Although receptor activation is known to change the 7-TM conformation dramatically (4), it is unclear how an FSHR antagonist (ADX68692) also increased FSH binding by 3-fold (41), whereas a partial agonist (Org 42599) was ineffective in the binding increase (17,42). As LMW modulators can bias FSHR activation (43), the details of the conformational changes upon bindings of the LMW modulators await the crystal structures of such modulators bound to the 7-TM domains.
Earlier reports did not explicitly note a 3-fold increase of FSH binding to FSHR by allosteric modulators (15)(16)(17), nor was the mechanism of action consistent with the model proposed herein. Rather, in these studies, the increased binding was attributed to tighter receptor affinity or to enhanced receptor expression. The results in the current report demonstrating a 3-fold increase in binding of FSH to FSHR by Compound 5 cannot be explained by either increased intrinsic affinity or enhanced expression, as (i) no increase in ligand binding affinity was observed, and (ii) current studies were performed on receptor-expressing membranes rather than on viable cells. In the absence of a direct comparison between Compound 5 and other modulators in our model system, it remains speculative as to whether the earlier tested modulators mediate different mechanisms of action. However, it should be noted that the presence of Org 214444 resulted in a 2-fold increase in B max at 1 g but no results were reported at higher concentrations (17).
More studies on LMW modulators will be needed to understand the details of the receptor activation mechanism, as called for recently (41).
Until the crystal structures of the full-length FSHR in free form and in complex with LMW modulators are available, it is an open question whether the mechanism of action is truly caused by conformational changes in the 7-TM domains; thus, the proposed model remains a work in progress. Despite these uncertainties, this trimer model may help stimulate new ideas and motivate new research in this field.
Implication for the Mechanism of TSHR Autoantibody Agonist and Antagonist Activities-The TSHR is a major autoantigen in autoimmune thyroid disease. Two types of TSHR autoantibodies have been discovered. Antibodies with thyroid- . TSHR (2) is shown in a magenta surface. C, same representation as in B except the autoantibody is K1-70. Note that there is no clash between the autoantibody and its neighboring TSHR. stimulating (agonist) activity are responsible for the hyperthyroidism of Graves disease, whereas antagonist antibodies cause hypothyroidism by preventing the binding of TSH to TSHR. Unlike TSH, which requires sulfation of tyrosine 385 on TSHR for receptor activation, the tyrosine is not required for stimulating autoantibodies to activate the TSHR (44). It was expected that the binding of the stimulating autoantibodies would cause a conformational change in the TSHR. The crystal structures of the hormone-binding portion of TSHR (amino acids 22-260) (TSHR260) in complex with a Fab fragment of thyroid-stimulating autoantibody (M22) (PDB entry 3G04) and a blocking type TSHR antoantibody (K1-70) (PDB entry 2XWT) were determined (45,46). These crystal structures showed no conformational difference for that portion of TSHR. Thus, how the binding of M22 causes receptor activation has remained poorly understood (46).
Our model may explain the activities of these two different classes of autoantibodies. As both the stimulating antibody M22 and the blocking antibody K1-70 bind to nearly identical epitopes on the concave surface of TSHR, the resulting opposite bioactivities have been difficult to explain. However, as the bound antibodies have differing orientations, corresponding to a rotation of ϳ155 o along their respective longitudinal axes (46), the M22 (green colored), but not K1-70 (red colored) clashes with the hinge hairpin loop in the current model (Fig.  6A). Therefore, M22 would have shifted the hinge hairpin loop on the 7-TM domain. Consistent with our model, such a shift, caused by an antibody or a ligand, would be critical in the activation of the GPCR. As our trimer model suggests, and the deglycosylated hormones demonstrate, the ligand-hairpin loop interaction constitutes one of the two requirements in receptor activation via the extracellular domain. The other requirement is the disruption or perturbation of trimeric configuration of the extracellular domains. According to our model, the agonist M22 clashes with its neighboring receptor TSHR (2) , whereas the antagonist K1-70 does not (Fig. 6, B and C). The clashing area of M22 on TSHR (2) is 300 Å 2 (Fig. 6B, right panel). If we assume TSHRs adopt the same trimeric configuration as that of the FSH-FSHR ED complex, M22 binding would encounter steric hindrance, pushing the neighboring receptor aside. Indeed, steric hindrance to thyroid-stimulating antibody binding to the TSHR on the cell surface was observed (49). Essentially, the mechanism by which M22 activates the TSHR mimics that of FSH to FSHR, dislocating the hairpin loop and disturbing the trimeric configuration. In contrast, K1-70 does neither of these two actions.
Closing Remarks-The central piece of the proposed hypothesis is the existence of FSHR as a constitutive trimer, which is normally capable of binding a single fully glycosylated FSH, leading to the activation of a single G protein and binding of ␤-arrestin. The results from our designed experiments confirm the predicted 3:1 stoichiometric ratio based on the receptor binding of Asn␣ 52 -deglycosylated FSH versus the fully glycosylated FSH, the binding of FSH and subsequent ␤-arrestin recruitment following stimulation in the presence and absence of a LMW modulator, and by mutagenesis studies demonstrating that disruption of the hydrophobic interaction at the FSH exosite enhances receptor stimulation efficacy. The model is further supported by our crystallographic studies that reveal that the FSHR trimeric structural configuration is not dependent on the crystallization conditions and space groups (as in the cases of P1 and P3 1 ). As GPCR oligomerization may be a general phenomenon, conclusions from our studies may shed light on the activation mechanism of other oligomeric GPCRs. The knowledge of the FSHR activation mechanism may be used in improving therapeutic drugs targeting FSHR and the related receptors.